CN107423442B - Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment - Google Patents
Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment Download PDFInfo
- Publication number
- CN107423442B CN107423442B CN201710666989.3A CN201710666989A CN107423442B CN 107423442 B CN107423442 B CN 107423442B CN 201710666989 A CN201710666989 A CN 201710666989A CN 107423442 B CN107423442 B CN 107423442B
- Authority
- CN
- China
- Prior art keywords
- data
- recommendation
- user
- application
- basic
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 84
- 238000004458 analytical method Methods 0.000 title claims abstract description 38
- 230000006399 behavior Effects 0.000 claims abstract description 101
- 239000013598 vector Substances 0.000 claims abstract description 80
- 230000004927 fusion Effects 0.000 claims abstract description 58
- 238000012549 training Methods 0.000 claims abstract description 44
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 43
- 238000012545 processing Methods 0.000 claims abstract description 28
- 238000000605 extraction Methods 0.000 claims abstract description 25
- 230000008569 process Effects 0.000 claims description 19
- 238000005070 sampling Methods 0.000 claims description 16
- 230000015654 memory Effects 0.000 claims description 14
- 238000004140 cleaning Methods 0.000 claims description 11
- 238000004590 computer program Methods 0.000 claims description 11
- 238000007477 logistic regression Methods 0.000 claims description 10
- 210000002569 neuron Anatomy 0.000 claims description 8
- 238000007637 random forest analysis Methods 0.000 claims description 7
- 238000012706 support-vector machine Methods 0.000 claims description 7
- 230000003993 interaction Effects 0.000 claims description 6
- 238000012360 testing method Methods 0.000 claims description 6
- 230000003044 adaptive effect Effects 0.000 claims description 5
- 238000009434 installation Methods 0.000 claims description 5
- 238000002790 cross-validation Methods 0.000 claims description 4
- 238000010801 machine learning Methods 0.000 claims description 4
- 230000009467 reduction Effects 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 238000013480 data collection Methods 0.000 claims description 3
- 238000010187 selection method Methods 0.000 claims description 2
- 238000004141 dimensional analysis Methods 0.000 abstract description 6
- 230000006403 short-term memory Effects 0.000 abstract description 4
- 230000007787 long-term memory Effects 0.000 abstract description 3
- 230000008901 benefit Effects 0.000 description 6
- 238000010586 diagram Methods 0.000 description 4
- 230000006870 function Effects 0.000 description 4
- 230000008859 change Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 239000000284 extract Substances 0.000 description 3
- 238000003672 processing method Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 238000010276 construction Methods 0.000 description 2
- 230000000694 effects Effects 0.000 description 2
- 238000001914 filtration Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 230000002452 interceptive effect Effects 0.000 description 2
- 238000013178 mathematical model Methods 0.000 description 2
- 238000007781 pre-processing Methods 0.000 description 2
- 230000000306 recurrent effect Effects 0.000 description 2
- 230000003542 behavioural effect Effects 0.000 description 1
- 238000004364 calculation method Methods 0.000 description 1
- 239000000969 carrier Substances 0.000 description 1
- 230000007812 deficiency Effects 0.000 description 1
- 238000011161 development Methods 0.000 description 1
- 230000008451 emotion Effects 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 230000014509 gene expression Effects 0.000 description 1
- 230000007774 longterm Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 239000011159 matrix material Substances 0.000 description 1
- 230000007246 mechanism Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000001105 regulatory effect Effects 0.000 description 1
- 238000013519 translation Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/90—Details of database functions independent of the retrieved data types
- G06F16/95—Retrieval from the web
- G06F16/953—Querying, e.g. by the use of web search engines
- G06F16/9535—Search customisation based on user profiles and personalisation
Landscapes
- Engineering & Computer Science (AREA)
- Databases & Information Systems (AREA)
- Theoretical Computer Science (AREA)
- Data Mining & Analysis (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
The invention provides a game recommendation method and system based on user portrait behavior analysis. And processing the user portrait data, the application list data and the data reported by the client by constructing a feature collector to obtain normalized feature vectors meeting the mathematical modeling requirements. Predicting by utilizing a plurality of basic recommendation models to generate a preliminary user application recommendation list and corresponding download probability; and training a fusion model by combining the downloading probability and the actual label to generate a final application recommendation list. And performing feature extraction to construct a user portrait data warehouse through multi-dimensional analysis on the user historical behavior log. For the basic recommendation model, the time sequence relation of the user behaviors is innovatively introduced into the long-term and short-term memory network learning, the preference degree of the user to the articles is better described, and the recommended game application is high in matching degree with the requirements of the user. And model fusion is carried out by adding ensemble learning, the learning results of all models are integrated, and the stability and generalization capability of the recommendation algorithm are improved.
Description
Technical Field
The invention belongs to the technical field of network information, and particularly relates to an application recommendation method based on user portrait behavior analysis, an application recommendation system based on user portrait behavior analysis, a computer-readable storage medium and computer equipment.
Background
In recent years, with the rapid development of the mobile internet industry, the amount of information carried by the internet has also been increasing dramatically. Various mobile internet information carriers provide diversified information content obtaining modes for users, but the information overload trouble is brought, users change from actively searching internet content to passively receiving a large amount of information such as internet subscription and pushing, and meanwhile, the cost for obtaining information by the users is correspondingly increased. Under a specific service scene, the behavior habits of the user are combined for analysis, the content and service meeting the user preference are provided as core requirements, and the recommendation method and the recommendation system in the internet field are produced for solving the user requirements.
Currently, the number of applications of iOS and Android two-large mobile system platforms has exceeded the scale of millions. The enormous application size has brought economic prosperity to the application market, where game application revenue occupies a large portion of the application market. The application market provides users with rich application choices, and simultaneously brings trouble to application choices. Therefore, the application recommendation service meeting the personalized requirements of the user is provided for the user, the user experience can be improved, and the platform benefit can be increased.
At present, widely applied application recommendation methods mainly include collaborative filtering recommendation, content-based recommendation, implicit model recommendation and the like. The collaborative filtering recommendation method is mainly applied to the e-commerce industry, generally adopts a nearest neighbor technology, calculates the distance between users by using historical preference information of the users, then predicts the preference degree of a target user for a specific article by using a weighted evaluation value of the nearest neighbor user of the target user for article evaluation, and recommends the target user according to the preference degree. However, this recommendation method is directed to a user having history data. In particular, when the user is a new user, there is a certain difficulty in recommending products for the user at this time because there is no historical operation data. The recommendation method based on the content characterizes the user and the article through related characteristic attributes, learns the user interests based on the characteristics of the user and the article, and accordingly recommends according to the interest matching degree of the user and the article. However, the content-based recommendation method needs to manually extract meaningful features, and is required to describe the preference degree of the user for the articles as much as possible. Implicit semantic model recommendation, the core idea is to link user interests and items through implicit features. The implementation process of the model generally comprises three parts, namely firstly mapping the articles to the implicit classification, secondly determining the interest of the user in the implicit classification, and finally selecting the articles in the classification which the user is interested in and recommending the articles to the user. The method is based on statistics of user behavior data, and then automatic clustering is carried out to find out potential subjects or classifications.
However, the data sources of the above models are relatively single, and in different context environments, the data sources have the disadvantages of poor stability and poor generalization capability, so that the accuracy of the application recommendation result is reduced.
Disclosure of Invention
Based on this, the present invention provides an application recommendation method based on user portrait behavior analysis, an application recommendation system based on user portrait behavior analysis, a computer-readable storage medium, and a computer device, which can improve stability and generalization capability of application recommendation and provide personalized application recommendation for users.
The invention is realized by the following scheme:
an application recommendation method based on user portrait behavior analysis comprises the following steps: acquiring a user behavior log reported by a client, and storing the user behavior log in a basic database of a server; by constructing a feature collector, carrying out data collection, cleaning, standardization processing and feature combination and extraction on user portrait data, original application list data and user behavior log data to obtain feature vectors which are unified and standard and meet the requirement of mathematical modeling; calling a plurality of preset basic recommendation models to respectively calculate the characteristic vectors to obtain a preliminary application recommendation list of corresponding users under each basic recommendation model and download probabilities of the corresponding users to various applications in the preliminary application recommendation list; inputting the download probability obtained by each basic recommendation model as a new feature vector, and training a preset fusion recommendation model by taking whether the application is actually downloaded or not as a label of the fusion recommendation model; calling the fusion recommendation model to process the newly added feature vectors of the users to obtain a final application recommendation list of the corresponding users;
sampling the user portrait data, the original application list data and the user behavior log data by adopting a sliding window method to obtain sampling data; in the step of calling a plurality of preset basic recommendation models to respectively operate the feature vectors, a machine learning method adopted by the basic recommendation models comprises the following steps: logistic regression, adaptive enhancement, support vector machine and random forest, and also includes learning methods of long-short term memory networks; and when a long-short term memory network is adopted for training, on the basis of a data set of a sampled sliding window sliding once, the data set is used as the input of a corresponding basic recommendation model, and the basic recommendation model is trained by adopting a training algorithm of a back propagation algorithm.
The application recommendation method based on user portrait behavior analysis is characterized in that a user portrait data warehouse is constructed by performing feature extraction on a log of user historical behavior through multi-dimensional analysis on the log. The method has the advantages that a fusion recommendation model is added, learning results of all basic recommendation models are integrated, stability and generalization capability of a recommendation algorithm are improved, matching degree of recommended applications and requirements of users is high, in the process of generating a primary application recommendation list by the basic recommendation models, time sequence factors are considered, and dependency on time sequence can be learned due to the fact that sample characteristics acquired by a sliding window method span a certain time period.
In one embodiment of the invention, the user behavior log data includes user application installation list, device information, game login time, game recharge and consumption information.
By comprehensively collecting user behavior log data, the interest of the user can be more accurately grasped, and the accuracy of recommending the application to the user by the user is improved.
In an embodiment of the present invention, the step of performing data acquisition, cleaning, standardization processing, and feature combination and extraction on the user portrait data, the original application list data, and the user behavior log data to obtain a feature vector that is unified and meets a mathematical modeling requirement includes:
sampling the user portrait data, the original application list data and the user behavior log data to obtain various sampling data, wherein the sampling data comprises: numerical data, text data, time sequence data and enumeration data;
separating time sequence data into dimensions required by each basic operation model according to different time units; carrying out Z-score standardization processing on the numerical data; performing semantic analysis on the text data; processing the enumerated classified data by adopting one-hot coding;
performing feature extraction on each processed item of data, and performing dimension reduction on the extracted feature vector;
generating a feature vector having a uniform specification and meeting mathematical modeling requirements, comprising: user feature vectors, application feature vectors, behavior feature vectors, and interaction feature vectors.
By constructing a feature collector to sample the user portrait data, the original application list data and the user behavior log data, numerical data, text data, time sequence data and enumeration data are obtained, and each data is expanded from different dimensions, the stability and generalization capability of a recommendation algorithm can be improved.
In an embodiment of the present invention, the step of calling a plurality of preset basic recommendation models to respectively perform operations on the feature vectors to obtain a preliminary application recommendation list of a corresponding user under each basic recommendation model, and the step of obtaining the download probability of each application in the preliminary application recommendation list by the corresponding user includes:
and training each basic recommendation model by adopting a k-fold cross verification method, in a training stage, performing parameter tuning on each basic recommendation model by adopting a grid search method to obtain optimal parameters, and generating a preliminary application recommendation list of each user under each basic recommendation model and the download probability of each application in the preliminary application recommendation list by the user.
The feature vectors of the users are respectively calculated through the basic recommendation model to obtain a preliminary application recommendation list and download probabilities of various applications, and the fusion recommendation model of the second layer can be effectively trained. And by combining the results of all the basic recommendation models, the recommendation results are more accurate, which is equivalent to the fact that a plurality of basic recommendation models participate in recommendation.
In one embodiment of the invention, training each basic recommendation model by adopting a k-fold cross validation method comprises the following steps:
dividing a training sample set of each basic recommendation model into k subsets which are identical in size and mutually exclusive in content;
and performing k iterations, wherein each iteration adopts a union of k-1 subsets as a training set, the rest subsets are used as a test set, and the training of the basic recommendation model is performed on the k groups of training sets and the test set.
The generalization capability of a single basic recommendation model is improved by training each basic recommendation model by adopting a k-fold cross validation method.
In one embodiment of the invention, the back propagation algorithm comprises:
calculating the output value of each neuron in a forward direction;
calculating the value of an error term of each neuron in a backward direction, wherein the error term comprises two directions, one direction is the backward propagation of the error term along the time, and the other direction is the upward propagation of the error term;
the gradient of each weight is calculated according to the corresponding error term.
The established long-term and short-term memory network can be effectively trained through a back propagation algorithm, and accurate model data can be obtained.
In an embodiment of the present invention, there is also provided a system for application recommendation, including:
the behavior data acquisition module is used for acquiring a user behavior log reported by a client and storing the user behavior log in a basic database of the server;
the characteristic extraction module is used for carrying out data acquisition, cleaning, standardization processing and characteristic combination and extraction on user portrait data, original application list data and the user behavior log data by constructing a characteristic collector to obtain a characteristic vector which is unified and standard and meets the requirement of mathematical modeling;
the basic model operation module is used for calling a plurality of preset basic recommendation models to respectively operate the feature vectors to obtain a preliminary application recommendation list of corresponding users under each basic recommendation model and download probabilities of the corresponding users to various applications in the preliminary application recommendation list;
the fusion recommendation model module is used for inputting the download probability obtained by each basic recommendation model as a new feature vector, and training a preset fusion recommendation model by taking whether the application is actually downloaded or not as a label of the fusion recommendation model;
and the fusion recommendation model operation module is used for calling the fusion recommendation model to process the newly added feature vectors of the users to obtain a final application recommendation list of the corresponding users.
According to the application recommendation system based on user portrait behavior analysis, a behavior data acquisition module acquires user behavior log data, and a feature extraction module acquires, cleans, standardizes and combines and extracts data of user portrait data, original application list data and the user behavior log data by constructing a feature collector to obtain uniform and standard feature vectors; the basic model operation module adopts a plurality of basic recommendation models to process the feature vector to obtain a preliminary application recommendation list and the download probability of the user to various applications in the preliminary application recommendation list, the fusion recommendation model module trains to obtain a fusion recommendation model, and the fusion recommendation model operation module calls the fusion recommendation model to process to obtain the application recommendation list of the user. And performing multi-dimensional analysis on the user historical behavior log, and performing feature extraction on the log to construct a user portrait data warehouse. And a fusion recommendation model is added, the learning results of all basic recommendation models are integrated, the stability and the generalization capability of a recommendation algorithm are improved, and the recommended application has high matching degree with the requirements of users.
In an embodiment of the present invention, there is further provided a computer-readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of the application recommendation method based on user portrait behavior analysis as described in any one of the above.
In an embodiment of the present invention, there is further provided a computer device, including a storage, a processor, and a computer program stored in the storage and executable by the processor, wherein the processor, when executing the computer program, implements the steps of the application recommendation method based on user portrait behavior analysis according to any one of the above.
The invention provides an application recommendation method based on user portrait behavior analysis based on the traditional application recommendation method based on user portrait behavior analysis and by combining the characteristics of a game platform application environment.
The method comprises the steps of processing an original characteristic data modeling framework, namely constructing a characteristic collector, so that the characteristic collector has the function of processing different types of original data characteristics into normalized characteristic vectors which can be used for mathematical model training, wherein the original data can be numerical data, text data, time sequence data, enumeration data and the like; and secondly, a recommendation model is fused, namely a plurality of single classifiers are fused, so that the model obtains more excellent generalization performance than the single classifiers, and thus, the recommendation is generated. Due to the diversity of application scenarios faced by the recommendation system, under different scenarios, the prediction obtained by a single recommendation algorithm may have poor generalization capability, so the combination strategy of multiple recommendation algorithms provided by the invention is particularly important. The fusion recommendation model is constructed and combined with the advantages of a plurality of recommendation algorithms, and the advantages of the recommendation algorithms are made up for the deficiencies of the recommendation algorithms and combined to form a strong recommendation frame. In addition, the behavior habits of the users may be changed due to the influence of time series factors, so on the aspect of time series behavior characteristics, the invention adds a long-short term memory network (LSTM) to process time series data in a basic prediction algorithm of a first layer of a fusion recommendation model, takes the prediction results of a plurality of recommendation algorithms as new characteristics, trains the prediction algorithm of a second layer and obtains a final game application list.
Therefore, the time sequence behavior of the user is effectively processed. The user time sequence behavior, namely the time sequence information of the product consumed by the user hides the rules of data change, and the association between the user and the product can be mined by utilizing the rules. The time sequence behavior plays an important role in predicting whether the user clicks to download the corresponding application. After the user clicks to view an application of a certain category, it is likely that the user will continue to click to view applications of the same category. In recent years, a Recurrent Neural Networks (RNN) has a sequence modeling capability, and thus is rapidly applied in a large amount in the fields of natural language processing, image recognition, voice recognition, and the like. For example, google has achieved a dramatic improvement in machine translation quality using RNNs, which have gained increased attention, and which have also begun to attempt to be used in the recommendation field.
The invention provides an automatic feature acquisition device, namely a device which can clean numerical data, text data, time sequence data, enumeration data and the like and process the numerical data, the text data, the time sequence data, the enumeration data and the like into a standardized format so as to extract feature vectors.
The invention provides a fusion recommendation model, and provides a long-short term memory network (LSTM) to mine time sequence behavior characteristics in an application layer of an application recommendation system, so that the potential relation between a user and a product is analyzed through time sequence information.
Drawings
FIG. 1 is a flowchart of an application recommendation method based on user portrait behavior analysis in an embodiment of the present invention;
FIG. 2 is a schematic diagram illustrating a principle of data acquisition by a sliding window method according to an embodiment of the present invention;
FIG. 3 is a schematic diagram of a feature collector constructed in an embodiment of the present invention;
FIG. 4 is a schematic diagram of a fusion recommendation model constructed in an embodiment of the present invention;
FIG. 5 is a schematic structural diagram of an application recommendation system based on user portrait behavior analysis according to an embodiment of the present invention.
Detailed Description
Please refer to fig. 1, which is a flowchart illustrating an application recommendation method based on user portrait behavior analysis according to an embodiment of the present invention, wherein the application recommendation method based on user portrait behavior analysis includes the following steps:
s101, acquiring a user behavior log reported by a client, and storing the user behavior log in a basic database of a server;
s102, performing data acquisition, cleaning, standardization processing, feature combination and extraction on user portrait data, original application list data and user behavior log data by constructing a feature collector to obtain feature vectors which are unified and standard and meet the requirement of mathematical modeling;
s103, calling a plurality of preset basic recommendation models to respectively calculate the feature vectors, and obtaining a preliminary application recommendation list of corresponding users under each basic recommendation model and download probabilities of the corresponding users to various applications in the preliminary application recommendation list;
s104, inputting the downloading probability obtained by each basic recommendation model as a new feature vector, and training a preset fusion recommendation model by taking whether the application is actually downloaded or not as a label of the fusion recommendation model;
and S105, calling the fusion recommendation model to process the newly added feature vectors of the users to obtain a final application recommendation list of the corresponding users.
The application recommendation method based on user portrait behavior analysis comprises the steps of carrying out data acquisition, cleaning, standardization processing, feature combination and extraction on user portrait data, original application list data and user behavior log data by constructing a feature collector to obtain unified and standard feature vectors, processing the feature vectors by adopting a plurality of basic recommendation models to obtain a primary application recommendation list, and obtaining download probabilities of various applications in the primary application recommendation list by a user, so as to train and obtain a fused recommendation model, and calling the fused recommendation model to process and obtain an application recommendation list of the user. And performing multi-dimensional analysis on the user historical behavior log, and performing feature extraction on the log to construct a user portrait data warehouse. And a fusion recommendation model is added, the learning results of all basic recommendation models are integrated, the stability and the generalization capability of a recommendation algorithm are improved, and the recommended application has high matching degree with the requirements of users.
In one embodiment of the present invention, in step S101, the user behavior log data includes a user application installation list, device information, game login time, game recharge and consumption information.
The client can obtain and report user behavior logs, the reported logs comprise information such as user application installation lists, equipment information, game login time, game recharging and consumption, and the reported logs are stored in the basic database of the server.
Through multi-dimensional analysis of the user historical behavior log, a user portrait data warehouse is constructed, and generalization capability of a recommendation algorithm can be improved.
In an embodiment of the present invention, in step S102, the step of performing data acquisition, cleaning, standardization processing, and feature combination and extraction on the user portrait data, the original application list data, and the user behavior log data to obtain a feature vector that is unified and meets a mathematical modeling requirement includes:
sampling the user portrait data, the original application list data and the user behavior log data to obtain various sampling data, wherein the sampling data comprises: numerical data, text data, time sequence data and enumeration data;
separating time sequence data into dimensions required by each basic operation model according to different time units; carrying out Z-score standardization processing on the numerical data; performing semantic analysis on the text data; processing the enumerated classified data by adopting one-hot coding;
performing feature extraction on each processed item of data, and performing dimension reduction on the extracted feature vector;
generating a feature vector having a uniform specification and meeting mathematical modeling requirements, comprising: user feature vectors, application feature vectors, behavior feature vectors, and interaction feature vectors.
By constructing a feature collector to sample the user portrait data, the original application list data and the user behavior log data, numerical data, text data, time sequence data and enumeration data are obtained, and each data is expanded from different dimensions, the stability and generalization capability of a recommendation algorithm can be improved.
According to the reported data obtained in step S101, in step S102, a feature collector may be constructed. A feature collector is a device which mainly cleans original user image data, original application list data and data reported by a client and processes the data into a standardized format so as to extract a data feature vector. The feature collector constructed in the invention can be customized and generalized to various data application scenarios, and converts a data source into a feature vector capable of performing mathematical model training based on an input data source, wherein the data source can be numerical data, text data, time sequence data, enumeration data and the like.
The steps of constructing the feature collector are as follows:
(1) writing an interface of an input data source, and using the interface to transmit basic data;
(2) processing numerical data, text data, time sequence data, enumeration data and the like into normalized feature vectors;
(3) an interface for outputting the feature vector is written, and the normalized feature vector can be obtained by using the interface.
The data processing method comprises data preprocessing, dimensioning, feature binarization, missing value replacement and the like.
The invention establishes a universal characteristic extraction method in the original time sequence log data recorded in the data reporting layer.
In the present invention, user characteristics (F) are defined and useduser) Application characteristic (F)app) Behavior characteristics (F)act) And interactive features (F)inter) Wherein:
user characteristics (F)user) Including user registration duration, registration time, age, gender, device system, liveness, and consumption ability index, etc.;
application characteristic (F)app) The method comprises the steps of game application category, shelf time, shelf duration, starting times, user viscosity, recharging amount of orders and the like;
behavioral characteristics (F)act) The method comprises the steps of clicking a game application detail page, clicking time, subscribing or not subscribing the game application, downloading the game application, adding a game group or not, evaluating the game and the like;
interaction feature (F)inter) I.e., the manner in which features are combined, two-by-two, or more, generates interactive features, including user and game applications, user and game categories, game applications and game categories, and so on.
Preferably, in the feature processing, different dimensions such as time, count and ratio are added to the four types of sampling data to expand, and further derived into a near n-day order number, a near n-day active user, a near n-day newly added user, a near n-day average consumption amount, and the like (n is 1, 2, 3, 5, 7, 15, and the like), and the data processing method includes data preprocessing, dimensioning, feature binarization, and missing value replacement.
Specifically, in the feature construction process of step S102, for the time stamp class such as registration time, time on shelf, and other features, the time is separated into dimensions required by a plurality of models such as month, week, day, hour, and the like; carrying out Z-score standardization processing on numerical value types such as characteristics of age, liveness, starting times and the like in order to reduce interference of numerical values on an algorithm; for text type characteristics, such as game evaluation, emotion analysis is mainly adopted to evaluate the popularity of the game application and serve as the characteristics of the popularity; for the data characteristics of the classification type, such as game application category, user gender and the like, a One-Hot Encoding (One-Hot Encoding) processing mode is adopted.
Based on the feature construction and feature processing, in the feature extraction and dimension reduction processing, a feature selection method of a penalty term is adopted, a Logistic Regression (LR) model is selected to calculate the weight coefficient of each feature, and feature terms lower than a threshold coefficient are filtered.
In a preferred embodiment, a sliding window method is adopted for data sampling, sample data in a time interval is selected as a characteristic, sample data in the next time period is used as a label, and the time length of the label slides to the next time period for next sampling. As shown in fig. 2.
In the invention, after the data source is processed by the data feature collector, four types of feature vectors, namely user features, application features, behavior features and interaction features, can be generated. The specific framework flowchart is shown in fig. 3.
In a preferred embodiment, the invention stores the processed feature vectors in a feature database that employs a combination of relational and non-relational databases.
After the data feature collector is constructed, the processed feature vectors need to be stored in a database. Specifically, the invention uses relational databases and non-relational databases, including MongoDB, MySQL and Redis databases. The MongoDB and the MySQL are used for storing daily offline data characteristics, and the fusion recommendation model is updated and maintained according to daily offline characteristic vectors; redis is used for storing newly added feature vectors on the current date in real time, and a recommendation system can perform real-time personalized recommendation on users.
In one embodiment of the present invention, step S103 includes:
and training each basic recommendation model by adopting a k-fold cross verification method, in a training stage, performing parameter tuning on each basic recommendation model by adopting a grid search method to obtain optimal parameters, and generating a preliminary application recommendation list of each user under each basic recommendation model and the download probability of each application in the preliminary application recommendation list by the user.
And generating a game application list for each sample data by using a recommendation algorithm module from the basic characteristics, the historical behaviors, the application characteristics and the like of the user. Specifically, given a user, feature vectors such as an application list and user historical behaviors are obtained from a database and are spliced, namely [ Fuser,Fapp,Fact,Finter]Judging whether the user downloads the game, converting the game into a classification problem, and calculating the probability p (y is 1| F) that the user downloads the gameuser,Fapp,Fact,Finter) I.e., the amount of probability that the user is likely to download the gaming application.
In a preferred embodiment, the basic recommendation model mainly adopts a machine learning method, which includes a combination of two or more of the following: combinational Logistic Regression (LR), Adaptive Boosting (AdaBoost), Support Vector Machine (SVM), and Random Forest (RF) and long-short term memory network (LSTM). The flow of the model of the method provided by the invention is that basic data is processed by a feature collector to obtain an output feature vector, and the feature vector is input to a classifier for training, so that game content and service to be recommended are obtained.
The feature vectors of the users are respectively calculated through the basic recommendation model to obtain a preliminary application recommendation list and download probabilities of various applications, and the fusion recommendation model of the second layer can be effectively trained. And by combining the results of all the basic recommendation models, the recommendation results are more accurate, which is equivalent to the fact that a plurality of basic recommendation models participate in recommendation.
The application scenes that a recommendation system needs to face are often greatly different, under different scenes, the prediction obtained by a single recommendation algorithm may have the condition of poor generalization performance, and meanwhile, recommendation problems under various scenes cannot be well processed, and the fusion algorithm which reasonably utilizes the model can have obvious effect improvement compared with the single model algorithm, so that the combination strategy which fuses a plurality of recommendation algorithms is particularly important.
In an embodiment of the present invention, the step of training each basic recommendation model by using k-fold cross-validation comprises:
dividing a training sample set of each basic recommendation model into k subsets which are identical in size and mutually exclusive in content;
and performing k iterations, wherein each iteration adopts a union of k-1 subsets as a training set, the rest subsets are used as a test set, and the training of the basic recommendation model is performed on the k groups of training sets and the test set.
By adopting a k-fold cross verification method, data characteristic training can be better utilized to adjust parameters, and the generalization performance of a single basic recommendation model classifier is improved.
In the present invention, the base learning algorithm (base learning algorithm) mainly includes Logistic Regression (LR), adaptive boosting (AdaBoost), Support Vector Machine (SVM), Random Forest (RF) and long short term memory network (LSTM) classifiers. In the LR model, the probability model of the user downloading the game application is:
where y is {0,1} classification data, p is the probability corresponding to that classification in y, w is the weight matrix, x is the feature vector, and b is the bias term.
In the training stage, the classifier method of each basic recommendation model adopts a grid search method to conduct parameter tuning, optimal parameters are obtained, and a game application list to be recommended and corresponding downloading probability are generated for each user.
In an embodiment of the invention, a sliding window method is adopted to sample the user portrait data, the original application list data and the user behavior log data, so as to obtain sampled data; in the step of calling a plurality of preset basic recommendation models to respectively operate the feature vectors, a machine learning method adopted by the basic recommendation models comprises the following steps: logistic regression, adaptive enhancement, support vector machine and random forest, and also includes learning methods of long-short term memory networks; and when a long-short term memory network is adopted for training, on the basis of a data set of a sampled sliding window sliding once, the data set is used as the input of a corresponding basic recommendation model, and the basic recommendation model is trained by adopting a training algorithm of a back propagation algorithm.
The algorithm of the recommendation system meets the problem of temporary sequence feature processing, for example, long-term and short-term interests of users can change.
In the process of generating the preliminary application recommendation list by the basic recommendation model, time sequence factors are considered, sample characteristics acquired by a sliding window method span a certain time period, dependency on time sequence can be learned, a long-short memory network (LSTM) is introduced, the time sequence relation of user behaviors is learned, and the personalized recommendation model is obtained.
In the generation of the recommendation list to be primarily applied by the basic recommendation model, the time sequence factor is considered, as the sample characteristics acquired by the sliding window method span a certain time period, in order to learn the dependency relationship on the time sequence, a long-short term memory network (LSTM) is added to the data set sampled by the sliding window in the algorithm strategy layer for training, namely, the data set is used as the input of the model on the basis of the data set which slides once, and the training algorithm of the model is mainly a back propagation algorithm.
In one embodiment of the invention, the back propagation algorithm comprises:
calculating the output value of each neuron in a forward direction;
calculating the value of an error term of each neuron in a backward direction, wherein the error term comprises two directions, one direction is the backward propagation of the error term along the time, and the other direction is the upward propagation of the error term;
the gradient of each weight is calculated according to the corresponding error term.
The established long-term and short-term memory network can be effectively trained through a back propagation algorithm, and accurate model data can be obtained.
The back propagation algorithm mainly comprises the following calculation steps:
(1) calculating the output value of each neuron in the forward direction, i.e. ft，it，ct，ot，htValues of five vectors, ftIndicating the state output of the ForgetGate at time t, itIndicating the state output of the Input Gate at time t, ctIndicating the state value at time t, otIndicating the state Output of the Output Gate at time t, htRepresenting the output of the long-short term memory network (LSTM) model at time t;
(2) calculating the value of an error term of each neuron in a backward direction, wherein one direction is backward propagation along time, namely from the current t moment, calculating the error term of each moment, and the other direction is propagation of the error term to the upper layer;
(3) the gradient of each weight is calculated according to the corresponding error term.
The model may be defined as:
ft＝σ(Wf*[ht-1,xt]+bf)
it＝σ(Wi*[ht-1,xt]+bi)
ot＝σ(Wo*[ht-1,xt]+bo)
ht＝ot*tanh(Ct)
wherein, the σ function represents a Sigmoid layer, the tanh function is a tanh layer, and the expressions are respectively:
in long-short term memory networks (LSTM), the advantage of the long-short term memory network over the Recurrent Neural Network (RNN) is that information can be persisted, and the model operates through Gate mechanisms, i.e., Input Gate, Forget Gate, and output Gate, where the Forget Gate represents C at the previous timet-1How much to keep current time CtInputGate represents the input x at the current timetHow much to save to CtIn, Output Gate is used to control CtHow much output value h is output to the current momenttThe vector values are compressed into 0 or 1 through a Sigmoid layer, wherein 0 represents eliminating input values, and 1 represents allowing the input values to pass through and participate in subsequent operations, and the method comprises the following specific steps:
(1) forget Gate's Sigmoid layer decides which information to get from CtMiddle rejection, Forget Gate will be according to [ h ]t-1,xt]I.e. output h at the previous momentt-1A vector (comprising user characteristics, application characteristics, behavior characteristics and interaction characteristics) formed by connecting the sample characteristics sliding to the current time t is used as characteristic input of the time t;
(2) the Sigmoid layer and the tanh layer of the Input Gate decide which information to update. Specifically, the Sigmoid layer of the Input Gate determines which information is to be updated, and the tanh layer calculates a new one
Will be provided with
And Ct-1Are combined to obtain updated t's Ct；
(3) Output Gate of Sigmoid layer decides h to be OutputtI.e. CtOutput o of Sigmoid layer with Output Gate after processing through tanh layer (making Output value between-1 and 1)tAnd multiplying to obtain a final result.
For step S104, inputting a preliminary application recommendation list obtained by each basic recommendation model and the download probability of the corresponding user to each application in the preliminary application recommendation list as new feature vectors, and training a preset fusion recommendation model by taking whether the application is actually downloaded or not as a label of the fusion recommendation model;
and in the fusion stage, the preliminary application recommendation list generated in the step S103 is subjected to second fusion recommendation to generate a final application recommendation list. The fusion recommendation model can select and fuse the optimal models aiming at various service scenes, and fully utilizes the advantages that different models learn different characteristics. The fusion recommendation model processing method comprises the following specific steps:
(1) predicting each basic model, and generating recommendation for each user, wherein the recommendation result comprises an application list to be recommended of the user and a corresponding download probability;
(2) taking the download probability output by the basic model in the last step as characteristic input, taking whether a game is actually downloaded as a label of the fusion recommendation model, training the fusion recommendation model, and evaluating the prediction effect of the fusion recommendation model;
(3) and generating a fusion recommendation model according to the two steps, recommending each user, and generating a final application recommendation list.
In the fusion phase, in order to avoid the risk of overfitting, the sample data not used in step S103 is used to generate training samples in step S104, and the principle framework of the ensemble-learned fusion recommendation model is shown in fig. 4. Meanwhile, in the current service scene, based on the product and the operation rule, some rejection rules can be set in the finally generated application recommendation list, for example, applications to be off-shelf due to time attenuation factors are filtered, and recommendation results of the part which do not meet the conditions are filtered; rules for regulating control may also be added, such as setting the application that downloads the most amount of data within a certain time, or game applications within the operating rules.
In an embodiment of the present invention, there is also provided an application recommendation system based on user portrait behavior analysis, as shown in fig. 5, the application recommendation system based on user portrait behavior analysis includes:
a behavior data obtaining module 10, configured to obtain a user behavior log reported by a client, and store the user behavior log in a server basic database;
the feature extraction module 20 is used for acquiring, cleaning, standardizing and combining and extracting features of the user portrait data, the original application list data and the user behavior log data by constructing a feature collector to obtain feature vectors which are unified and standard and meet the mathematical modeling requirement;
the basic model operation module 30 is configured to invoke a plurality of preset basic recommendation models to respectively operate the feature vectors, so as to obtain a preliminary application recommendation list of a corresponding user under each basic recommendation model and download probabilities of the corresponding user for various applications in the preliminary application recommendation list;
the fusion recommendation model module 40 is configured to input the download probability obtained by each basic recommendation model as a new feature vector, and train a preset fusion recommendation model by using whether the application is actually downloaded or not as a label of the fusion recommendation model;
and the fusion recommendation model operation module 50 calls the fusion recommendation model to process the newly added feature vectors of the users to obtain an application recommendation list of the corresponding users.
According to the application recommendation system based on user portrait behavior analysis, a behavior data acquisition module acquires user behavior log data, and a feature extraction module acquires, cleans, standardizes and combines and extracts data of user portrait data, original application list data and the user behavior log data by constructing a feature collector to obtain uniform and standard feature vectors; the basic model operation module adopts a plurality of basic recommendation models to process the feature vector to obtain a preliminary application recommendation list and the download probability of the user to various applications in the preliminary application recommendation list, the fusion recommendation model module trains to obtain a fusion recommendation model, and the fusion recommendation model operation module calls the fusion recommendation model to process to obtain the application recommendation list of the user. And performing multi-dimensional analysis on the user historical behavior log, and performing feature extraction on the log to construct a user portrait data warehouse. And a fusion recommendation model is added, the learning results of all basic recommendation models are integrated, the stability and the generalization capability of a recommendation algorithm are improved, and the recommended application has high matching degree with the requirements of users.
The application recommendation method and system based on user portrait behavior analysis provided by the invention can be divided into the following 5 modules: the system comprises a client, a server, a database, a feature collector and a recommendation algorithm module. The client can obtain and report a user behavior log, the reported log content comprises information such as a user application installation list, equipment information, game login time, game recharging and consumption and the like, and the reported log content is stored in a server basic database; the feature collector performs the functions of data cleaning, standardization, feature combination, extraction and the like, processes the reported data and the sample data of the original user picture database into a normalized feature vector and stores the normalized feature vector in the feature database; the recommendation algorithm module calls the features in the feature database for modeling and provides an interface for the server to call. When a user uses a client to make a request to a server, the server calls a recommendation algorithm module to return the content and service of the game application to the client.
In an embodiment of the present invention, there is further provided a computer-readable storage medium, on which a computer program is stored, wherein the computer program, when executed by a processor, implements the steps of any one of the above-mentioned application recommendation methods based on user portrait behavior analysis.
In an embodiment of the present invention, there is further provided a computer device, including a storage, a processor, and a computer program stored in the storage and executable by the processor, where the processor implements the steps of any one of the above application recommendation methods based on user portrait behavior analysis when executing the computer program.
The above-mentioned embodiments only express several embodiments of the present invention, and the description thereof is more specific and detailed, but not construed as limiting the scope of the invention. It should be noted that, for a person skilled in the art, several variations and modifications can be made without departing from the inventive concept, which falls within the scope of the present invention.
Claims (9)
1. An application recommendation method based on user portrait behavior analysis is characterized in that: the method comprises the following steps:
acquiring a user behavior log reported by a client, and storing the user behavior log in a basic database of a server;
by constructing a feature collector, carrying out data collection, cleaning, standardization processing and feature combination and extraction on user portrait data, original application list data and user behavior log data to obtain feature vectors which are unified and standard and meet the requirement of mathematical modeling;
calling a plurality of preset basic recommendation models to respectively calculate the characteristic vectors to obtain a preliminary application recommendation list of corresponding users under each basic recommendation model and download probabilities of the corresponding users to various applications in the preliminary application recommendation list;
inputting the download probability obtained by each basic recommendation model as a new feature vector, and training a preset fusion recommendation model by taking whether the application is actually downloaded or not as a label of the fusion recommendation model;
calling the fusion recommendation model to process the newly added feature vectors of the users to obtain a final application recommendation list of the users;
sampling the user portrait data, the original application list data and the user behavior log data by adopting a sliding window method to obtain sampling data;
in the step of calling a plurality of preset basic recommendation models to respectively operate the feature vectors, a machine learning method adopted by the basic recommendation models comprises the following steps: logistic regression, adaptive enhancement, support vector machine and random forest, and also includes learning methods of long-short term memory networks;
and when a long-short term memory network is adopted for training, on the basis of a data set of a sampled sliding window sliding once, the data set is used as the input of a corresponding basic recommendation model, and the basic recommendation model is trained by adopting a training algorithm of a back propagation algorithm.
2. The application recommendation method based on user portrait behavior analysis of claim 1, wherein:
the user behavior log data comprises a user application installation list, equipment information, game login time, game recharge and consumption information.
3. The user representation behavior analysis-based application recommendation method of claim 1, wherein the step of performing data collection, cleaning, standardization processing, and feature combination and extraction on the user representation data, the raw application list data, and the user behavior log data to obtain feature vectors that are unified and meet mathematical modeling requirements comprises:
sampling the user portrait data, the original application list data and the user behavior log data to obtain various sampling data, wherein the sampling data comprises: numerical data, text data, time sequence data and enumeration data;
separating time sequence data into dimensions required by each basic operation model according to different time units; carrying out Z-score standardization processing on the numerical data; performing semantic analysis on the text data; processing the enumerated classified data by adopting one-hot coding;
carrying out feature extraction on each processed data, adopting a feature selection method of a punishment item to carry out dimension reduction processing, and generating a feature vector which has unified specification and accords with the requirement of mathematical modeling, wherein the method comprises the following steps: user feature vectors, application feature vectors, behavior feature vectors, and interaction feature vectors.
4. The application recommendation method based on user portrait behavior analysis of claim 3, wherein:
calling a plurality of preset basic recommendation models to respectively calculate the feature vectors to obtain a preliminary application recommendation list of corresponding users under each basic recommendation model, wherein the step of obtaining the download probability of each application in the preliminary application recommendation list by the corresponding users comprises the following steps:
and training each basic recommendation model by adopting a k-fold cross verification method, in a training stage, performing parameter tuning on each basic recommendation model by adopting a grid search method to obtain optimal parameters, and generating a preliminary application recommendation list of each user under each basic recommendation model and the download probability of each application in the preliminary application recommendation list by the user.
5. The application recommendation method based on user portrait behavior analysis of claim 4, wherein: the method for training each basic recommendation model by adopting a k-fold cross validation method comprises the following steps:
dividing a training sample set of each basic recommendation model into k subsets which are identical in size and mutually exclusive in content;
and performing k iterations, wherein each iteration adopts a union of k-1 subsets as a training set, the rest subsets are used as a test set, and the training of the basic recommendation model is performed on the k groups of training sets and the test set.
6. The application recommendation method based on user portrait behavior analysis of claim 1, wherein the back propagation algorithm comprises:
calculating the output value of each neuron in a forward direction;
calculating the value of an error term of each neuron in a backward direction, wherein the error term comprises two directions, one direction is the backward propagation of the error term along the time, and the other direction is the upward propagation of the error term;
the gradient of each weight is calculated according to the corresponding error term.
7. A system for application recommendation, comprising:
the behavior data acquisition module is used for acquiring a user behavior log reported by a client and storing the user behavior log in a basic database of the server;
the characteristic extraction module is used for carrying out data acquisition, cleaning, standardization processing and characteristic combination and extraction on user portrait data, original application list data and the user behavior log data by constructing a characteristic collector to obtain a characteristic vector which is unified and standard and meets the requirement of mathematical modeling;
the basic model operation module is used for calling a plurality of preset basic recommendation models to respectively operate the feature vectors to obtain a preliminary application recommendation list of corresponding users under each basic recommendation model and download probabilities of the corresponding users to various applications in the preliminary application recommendation list;
the fusion recommendation model module is used for inputting the download probability obtained by each basic recommendation model as a new feature vector, and training a preset fusion recommendation model by taking whether the application is actually downloaded or not as a label of the fusion recommendation model;
and the fusion recommendation model operation module is used for calling the fusion recommendation model to process the newly added feature vectors of the users to obtain a final application recommendation list of the corresponding users.
8. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the steps of the method for application recommendation based on user representation behavior analysis according to any one of claims 1 to 6.
9. A computer device comprising a storage, a processor and a computer program stored in the storage and executable by the processor, the processor implementing the steps of the application recommendation method based on user representation behavior analysis of any of claims 1 to 6 when executing the computer program.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710666989.3A CN107423442B (en) | 2017-08-07 | 2017-08-07 | Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN201710666989.3A CN107423442B (en) | 2017-08-07 | 2017-08-07 | Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN107423442A CN107423442A (en) | 2017-12-01 |
| CN107423442B true CN107423442B (en) | 2020-09-25 |
Family
ID=60436664
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN201710666989.3A Active CN107423442B (en) | 2017-08-07 | 2017-08-07 | Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN107423442B (en) |
Families Citing this family (129)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108256548A (en) * | 2017-12-04 | 2018-07-06 | 北京大学 | A kind of user's portrait depicting method and system based on Emoji service conditions |
| CN110019163A (en) * | 2017-12-05 | 2019-07-16 | 北京京东尚科信息技术有限公司 | Method, system, equipment and the storage medium of prediction, the recommendation of characteristics of objects |
| CN107977864A (en) * | 2017-12-07 | 2018-05-01 | 北京贝塔智投科技有限公司 | A kind of customer insight method and system suitable for financial scenario |
| CN108062375B (en) * | 2017-12-12 | 2021-12-10 | 百度在线网络技术（北京）有限公司 | User portrait processing method and device, terminal and storage medium |
| CN108053120B (en) * | 2017-12-15 | 2020-09-01 | 阿里巴巴集团控股有限公司 | Model integration method and device |
| CN109934369A (en) * | 2017-12-15 | 2019-06-25 | 北京京东尚科信息技术有限公司 | Method and device for information push |
| CN108133013B (en) * | 2017-12-22 | 2021-02-09 | 平安养老保险股份有限公司 | Information processing method, information processing device, computer equipment and storage medium |
| CN108038237A (en) * | 2017-12-27 | 2018-05-15 | 广州市云润大数据服务有限公司 | A kind of information recommendation method and system |
| CN107944986B (en) * | 2017-12-28 | 2022-02-15 | 广东工业大学 | Method, system and equipment for recommending O2O commodities |
| CN108021715B (en) * | 2017-12-29 | 2020-07-28 | 西安交通大学 | Heterogeneous label fusion system based on semantic structure feature analysis |
| CN108154401B (en) * | 2018-01-15 | 2022-03-29 | 阿里巴巴（中国）有限公司 | User portrait depicting method, device, medium and computing equipment |
| CN108470023A (en) * | 2018-01-18 | 2018-08-31 | 阿里巴巴集团控股有限公司 | The recommendation method and device of business function |
| CN108446297B (en) * | 2018-01-24 | 2021-03-26 | 北京三快在线科技有限公司 | Recommendation method and device and electronic equipment |
| CN108460489A (en) * | 2018-03-15 | 2018-08-28 | 重庆邮电大学 | A kind of user behavior analysis based on big data technology and service recommendation frame |
| CN110309414B (en) * | 2018-03-16 | 2021-08-13 | 阿里巴巴（中国）有限公司 | Content recommendation method, content recommendation device and electronic equipment |
| CN108491720B (en) * | 2018-03-20 | 2023-07-14 | 腾讯科技（深圳）有限公司 | Application identification method, system and related equipment |
| CN108805383A (en) * | 2018-03-20 | 2018-11-13 | 东华大学 | A kind of user's portrait platform and application for washing shield big data based on clothes |
| CN108510320A (en) * | 2018-03-21 | 2018-09-07 | 四川斐讯信息技术有限公司 | A kind of method and device that business is launched |
| CN110334720A (en) * | 2018-03-30 | 2019-10-15 | 百度在线网络技术（北京）有限公司 | Feature extracting method, device, server and the storage medium of business datum |
| CN108520058B (en) * | 2018-03-30 | 2021-09-24 | 维沃移动通信有限公司 | Merchant information recommendation method and mobile terminal |
| CN108744524B (en) * | 2018-04-10 | 2021-10-15 | 浙江工业大学 | Recommendation method suitable for cognitive training platform |
| CN108648049B (en) * | 2018-05-03 | 2022-03-01 | 中国科学技术大学 | Sequence recommendation method based on user behavior difference modeling |
| CN108763502B (en) * | 2018-05-30 | 2022-03-25 | 腾讯科技（深圳）有限公司 | Information recommendation method and system |
| US11853306B2 (en) * | 2018-06-03 | 2023-12-26 | Apple Inc. | Techniques for personalizing app store recommendations |
| CN108897790B (en) * | 2018-06-11 | 2022-09-20 | 南京邮电大学 | Recommendation method based on robust collaborative filtering algorithm |
| CN108920213B (en) * | 2018-06-29 | 2021-09-17 | 超级魔方(北京)科技有限公司 | Dynamic configuration method and device of game |
| US10885905B2 (en) * | 2018-07-13 | 2021-01-05 | Samsung Electronics Co., Ltd. | Predicting user actions on ubiquitous devices |
| CN109033309B (en) * | 2018-07-17 | 2023-04-07 | 腾讯科技（深圳）有限公司 | Data processing method, device, equipment and storage medium |
| CN109144721B (en) * | 2018-07-18 | 2022-08-16 | Oppo广东移动通信有限公司 | Resource sorting method, resource display method, related device and storage medium |
| CN109241408A (en) * | 2018-08-14 | 2019-01-18 | 深圳开黑科技有限公司 | User's portrait processing method, equipment and the system for playing platform are accompanied based on game |
| CN109063188A (en) * | 2018-08-28 | 2018-12-21 | 国信优易数据有限公司 | A kind of entity recommended method and device |
| JP6573205B2 (en) * | 2018-09-10 | 2019-09-11 | 澪標アナリティクス株式会社 | PROCESSING DEVICE, INFORMATION PROCESSING METHOD, PROGRAM, AND MARKETING INFORMATION PROCESSING DEVICE FOR Calculating Predictive Data Regarding Use Of Application Program By One User |
| CN109460440B (en) * | 2018-09-18 | 2023-10-27 | 平安科技（深圳）有限公司 | Image processing method, device and equipment based on weight value |
| CN109214585B (en) * | 2018-09-25 | 2023-04-18 | 平安科技（深圳）有限公司 | User consumption prediction method and device, electronic equipment and storage medium |
| CN109272164B (en) * | 2018-09-29 | 2021-09-28 | 清华大学深圳研究生院 | Learning behavior dynamic prediction method, device, equipment and storage medium |
| CN109358877B (en) * | 2018-09-30 | 2023-08-18 | 上海碳蓝网络科技有限公司 | Method and equipment for upgrading application in user equipment |
| CN109359137B (en) * | 2018-10-16 | 2021-03-26 | 大连理工大学 | User growth portrait construction method based on feature screening and semi-supervised learning |
| CN109635079A (en) * | 2018-10-25 | 2019-04-16 | 北京中关村科金技术有限公司 | A kind of determination method, apparatus, computer equipment and storage medium that user is intended to |
| CN109525665A (en) * | 2018-11-16 | 2019-03-26 | 济南浪潮高新科技投资发展有限公司 | A kind of wound visitor's cloud center configuration recommended method based on crowdsourcing |
| CN109902852A (en) * | 2018-11-28 | 2019-06-18 | 北京三快在线科技有限公司 | Grouping of commodities method, apparatus, electronic equipment and readable storage medium storing program for executing |
| CN111291253B (en) * | 2018-12-06 | 2023-09-19 | 北京嘀嘀无限科技发展有限公司 | Model training method, consultation recommendation method and device and electronic equipment |
| CN109583659A (en) * | 2018-12-07 | 2019-04-05 | 南京富士通南大软件技术有限公司 | User's operation behavior prediction method and system based on deep learning |
| CN109325640B (en) * | 2018-12-07 | 2022-04-26 | 中山大学 | User value prediction method, device, storage medium and equipment |
| CN109767264A (en) * | 2018-12-20 | 2019-05-17 | 深圳壹账通智能科技有限公司 | Product data method for pushing, device, computer equipment and storage medium |
| CN109886719B (en) * | 2018-12-20 | 2023-06-20 | 平安科技（深圳）有限公司 | Data mining processing method and device based on grid and computer equipment |
| CN109710890B (en) * | 2018-12-20 | 2023-06-09 | 四川新网银行股份有限公司 | Method and system for identifying false material in real time based on constructed behavior portrait model |
| CN109635204A (en) * | 2018-12-21 | 2019-04-16 | 上海交通大学 | Online recommender system based on collaborative filtering and length memory network |
| CN109800790B (en) * | 2018-12-24 | 2022-08-19 | 厦门大学 | Feature selection method for high-dimensional data |
| CN111381053B (en) * | 2018-12-28 | 2024-09-27 | 深圳迈瑞生物医疗电子股份有限公司 | Instrument recommendation method, in-vitro diagnosis analyzer and computer storage medium |
| CN111597330A (en) * | 2019-02-21 | 2020-08-28 | 中国科学院信息工程研究所 | Intelligent expert recommendation-oriented user image drawing method based on support vector machine |
| CN110427548B (en) * | 2019-03-05 | 2021-06-29 | 北京京东尚科信息技术有限公司 | Information pushing method, information pushing device and computer readable storage medium |
| CN110109902A (en) * | 2019-03-18 | 2019-08-09 | 广东工业大学 | A kind of electric business platform recommender system based on integrated learning approach |
| GB2598052B (en) * | 2019-03-22 | 2023-03-15 | Ibm | Unification of models having respective target classes with distillation |
| CN110490685A (en) * | 2019-03-27 | 2019-11-22 | 南京国科双创信息技术研究院有限公司 | A kind of Products Show method based on big data analysis |
| CN109960761B (en) * | 2019-03-28 | 2023-03-31 | 深圳市雅阅科技有限公司 | Information recommendation method, device, equipment and computer readable storage medium |
| CN111753182B (en) * | 2019-03-28 | 2024-05-17 | 北京达佳互联信息技术有限公司 | Multimedia information recommendation method and device, electronic equipment and readable storage medium |
| CN110008408B (en) * | 2019-04-12 | 2021-04-06 | 山东大学 | Session recommendation method, system, device and medium |
| CN110162698B (en) * | 2019-04-18 | 2023-10-20 | 腾讯科技（深圳）有限公司 | User portrait data processing method, device and storage medium |
| CN111832595B (en) * | 2019-04-23 | 2022-05-06 | 北京新唐思创教育科技有限公司 | Teacher style determination method and computer storage medium |
| CN111832787B (en) * | 2019-04-23 | 2022-12-09 | 北京新唐思创教育科技有限公司 | Teacher style prediction model training method and computer storage medium |
| CN110163381A (en) * | 2019-04-26 | 2019-08-23 | 美林数据技术股份有限公司 | Intelligence learning method and device |
| CN111767455A (en) * | 2019-05-09 | 2020-10-13 | 北京京东尚科信息技术有限公司 | Information pushing method and device |
| CN110175171B (en) * | 2019-05-16 | 2023-07-07 | 贵州电网有限责任公司 | System for IT equipment intelligent recommendation of on-shelf position |
| CN110237536A (en) * | 2019-06-03 | 2019-09-17 | 北京金山安全软件有限公司 | Personalized game service providing method and device, electronic equipment and storage medium |
| CN110391010B (en) * | 2019-06-11 | 2022-05-13 | 山东大学 | Food recommendation method and system based on personal health perception |
| CN110347917A (en) * | 2019-06-14 | 2019-10-18 | 北京纵横无双科技有限公司 | A kind of medical information method for pushing and device |
| CN110489453B (en) * | 2019-07-02 | 2023-04-14 | 广东工业大学 | User game real-time recommendation method and system based on big data log analysis |
| CN110489639B (en) * | 2019-07-15 | 2022-10-25 | 北京奇艺世纪科技有限公司 | Content recommendation method and device |
| US11250322B2 (en) * | 2019-07-15 | 2022-02-15 | Sony Interactive Entertainment LLC | Self-healing machine learning system for transformed data |
| CN110503459B (en) * | 2019-07-19 | 2023-09-15 | 平安科技（深圳）有限公司 | User credibility assessment method and device based on big data and storage medium |
| CN110457543B (en) * | 2019-07-25 | 2021-10-22 | 中国科学院软件研究所 | Entity resolution method and system based on end-to-end multi-view matching |
| CN110503468A (en) * | 2019-08-19 | 2019-11-26 | 珠海天燕科技有限公司 | A kind of resource recommendation method and device based on user behavior |
| CN110689040B (en) * | 2019-08-19 | 2022-10-18 | 广州荔支网络技术有限公司 | Sound classification method based on anchor portrait |
| CN110533515A (en) * | 2019-09-04 | 2019-12-03 | 深圳创新奇智科技有限公司 | A kind of the electric business personalized recommendation method and device of high low latency of handling up |
| CN110674100B (en) * | 2019-09-28 | 2023-02-10 | 上海浦东发展银行股份有限公司信用卡中心 | User demand prediction method and framework based on full-channel operation data |
| CN110765352B (en) * | 2019-10-11 | 2022-11-11 | 上海上湖信息技术有限公司 | User interest identification method and device |
| CN110795582A (en) * | 2019-10-31 | 2020-02-14 | 京东方科技集团股份有限公司 | Image recommendation method, system, terminal device and server |
| CN111080338B (en) * | 2019-11-11 | 2024-05-24 | 建信金融科技有限责任公司 | User data processing method and device, electronic equipment and storage medium |
| CN112948704B (en) * | 2019-11-26 | 2024-06-04 | 北京达佳互联信息技术有限公司 | Model training method and device for information recommendation, electronic equipment and medium |
| CN111192602A (en) * | 2019-12-03 | 2020-05-22 | 广州荔支网络技术有限公司 | White noise audio content value evaluation method based on audio content portrait system |
| CN110929209B (en) * | 2019-12-06 | 2023-06-20 | 北京百度网讯科技有限公司 | Method and device for transmitting information |
| CN111091351A (en) * | 2019-12-16 | 2020-05-01 | 北京政信1890智能科技有限公司 | User portrait construction method and device, electronic equipment and readable storage medium |
| CN111210335B (en) * | 2019-12-16 | 2023-11-14 | 北京淇瑀信息科技有限公司 | User risk identification method and device and electronic equipment |
| CN111210336A (en) * | 2019-12-16 | 2020-05-29 | 北京淇瑀信息科技有限公司 | User risk model generation method and device and electronic equipment |
| CN111191122A (en) * | 2019-12-20 | 2020-05-22 | 重庆邮电大学 | Learning resource recommendation system based on user portrait |
| CN111143684B (en) * | 2019-12-30 | 2023-03-21 | 腾讯科技（深圳）有限公司 | Artificial intelligence-based generalized model training method and device |
| CN111191092B (en) * | 2019-12-31 | 2023-07-14 | 腾讯科技（深圳）有限公司 | Label determining method and label determining model training method |
| CN113127727A (en) * | 2019-12-31 | 2021-07-16 | 国信优易数据股份有限公司 | Information recommendation model determination method, information recommendation method and device |
| CN111143689A (en) * | 2019-12-31 | 2020-05-12 | 青梧桐有限责任公司 | Method for constructing recommendation engine according to user requirements and user portrait |
| CN111539782B (en) * | 2020-01-21 | 2024-04-30 | 中国银联股份有限公司 | Deep learning-based merchant information data processing method and system |
| CN111275350B (en) * | 2020-02-08 | 2021-06-04 | 支付宝(杭州)信息技术有限公司 | Method and device for updating event evaluation model |
| CN111260448B (en) * | 2020-02-13 | 2023-10-20 | 平安科技（深圳）有限公司 | Medicine recommendation method based on artificial intelligence and related equipment |
| CN113377018A (en) * | 2020-03-09 | 2021-09-10 | 青岛海尔空调器有限总公司 | Intelligent control method and intelligent control equipment of air conditioner |
| CN113375300A (en) * | 2020-03-09 | 2021-09-10 | 青岛海尔空调器有限总公司 | Intelligent control method and intelligent control equipment of air conditioner |
| CN112287228A (en) * | 2020-03-20 | 2021-01-29 | 张明 | Online learning recommendation method, online learning system and server |
| CN111461841B (en) * | 2020-04-07 | 2023-04-07 | 腾讯云计算（北京）有限责任公司 | Article recommendation method, device, server and storage medium |
| CN113515689A (en) * | 2020-04-10 | 2021-10-19 | 阿里巴巴集团控股有限公司 | Recommendation method and device |
| CN111652735A (en) * | 2020-04-17 | 2020-09-11 | 世纪保众(北京)网络科技有限公司 | Insurance product recommendation method based on user behavior label characteristics and commodity characteristics |
| CN111723287B (en) * | 2020-06-03 | 2021-03-05 | 开普云信息科技股份有限公司 | Content and service recommendation method and system based on large-scale machine learning |
| CN111695035B (en) * | 2020-06-10 | 2023-05-05 | 广东数果科技有限公司 | Recommendation system and multi-algorithm fusion recommendation processing flow |
| CN111737575B (en) * | 2020-06-19 | 2023-11-14 | 北京字节跳动网络技术有限公司 | Content distribution method, content distribution device, readable medium and electronic equipment |
| CN111767461B (en) * | 2020-06-24 | 2024-02-06 | 北京奇艺世纪科技有限公司 | Data processing method and device |
| CN111816300A (en) * | 2020-06-30 | 2020-10-23 | 平安科技（深圳）有限公司 | Data classification method, device, equipment and medium based on deep reinforcement learning |
| CN111797322A (en) * | 2020-07-10 | 2020-10-20 | 杭州慕锐科技有限公司 | Intelligent recommendation method for layered pattern design chart of textiles with affiliated style |
| CN112052386B (en) * | 2020-08-10 | 2023-09-29 | 腾讯科技（深圳）有限公司 | Information recommendation method, device and storage medium |
| CN111898031B (en) * | 2020-08-14 | 2024-04-05 | 腾讯科技（深圳）有限公司 | Method and device for obtaining user portrait |
| CN111949884B (en) * | 2020-08-26 | 2022-06-21 | 桂林电子科技大学 | Multi-mode feature interaction-based depth fusion recommendation method |
| CN111767474A (en) * | 2020-09-01 | 2020-10-13 | 上海蜜度信息技术有限公司 | Method and equipment for constructing user portrait based on user operation behaviors |
| CN112148983B (en) * | 2020-09-30 | 2023-02-03 | 神思电子技术股份有限公司 | Content updating and recommending method for tax industry |
| CN112235615B (en) * | 2020-10-14 | 2023-03-21 | 广州欢网科技有限责任公司 | Method and system for communicating and using portrait data of television terminal user |
| CN112307332B (en) * | 2020-10-16 | 2021-08-24 | 四川大学 | Collaborative filtering recommendation method and system based on user portrait clustering and storage medium |
| CN113763093B (en) * | 2020-11-12 | 2024-10-18 | 北京沃东天骏信息技术有限公司 | Article recommending method and device based on user portrait |
| CN112801744B (en) * | 2020-12-01 | 2022-04-29 | 珠海必要工业科技股份有限公司 | Activity recommendation method and device, electronic equipment and storage medium |
| CN112561598B (en) * | 2020-12-23 | 2024-08-02 | 中国农业银行股份有限公司重庆市分行 | Customer loss prediction and retrieval method and system based on customer portrayal |
| CN112685641B9 (en) * | 2020-12-31 | 2024-08-16 | 五八有限公司 | Information processing method and device |
| CN115081501A (en) * | 2021-03-15 | 2022-09-20 | 中国电信股份有限公司 | User classification method and device, cascaded user classification model and equipment |
| CN113204577A (en) * | 2021-04-15 | 2021-08-03 | 北京沃东天骏信息技术有限公司 | Information pushing method and device, electronic equipment and computer readable medium |
| CN113313545B (en) * | 2021-04-19 | 2024-04-23 | 深圳市竹芒信息技术有限公司 | Information recommendation method, device, computer equipment and storage medium |
| CN113076488A (en) * | 2021-06-04 | 2021-07-06 | 明品云(北京)数据科技有限公司 | Method and system for recommending information based on user data |
| CN113449188A (en) * | 2021-06-30 | 2021-09-28 | 东莞市小精灵教育软件有限公司 | Application recommendation method and device, electronic equipment and readable storage medium |
| CN113612639B (en) * | 2021-07-30 | 2022-11-11 | 江苏易安联网络技术有限公司 | Method and device for analyzing and predicting file downloading behavior based on website access record |
| CN113779384B (en) * | 2021-08-23 | 2022-07-05 | 广州百奕信息科技有限公司 | Flight recommendation method based on passenger portrait |
| CN113569557B (en) * | 2021-09-26 | 2022-01-04 | 腾讯科技（深圳）有限公司 | Information quality identification method, device, equipment, storage medium and program product |
| CN113987353B (en) * | 2021-10-29 | 2022-08-09 | 掌阅科技股份有限公司 | Book recommendation method, computing device and storage medium |
| CN114595943B (en) * | 2022-02-14 | 2024-11-08 | 烟台杰瑞石油服务集团股份有限公司 | Image generation method for mechanical equipment |
| CN114549078B (en) * | 2022-02-22 | 2024-05-14 | 中国平安人寿保险股份有限公司 | Client behavior processing method and device based on time sequence and related equipment |
| CN114647721B (en) * | 2022-05-23 | 2022-09-16 | 风林科技(深圳)有限公司 | Educational intelligent robot control method, device and medium |
| CN115531886A (en) * | 2022-10-08 | 2022-12-30 | 广州易幻网络科技有限公司 | User and equipment data management method, system and storage medium |
| CN115830405B (en) * | 2023-02-13 | 2023-09-22 | 国信蓝桥教育科技股份有限公司 | Method and system for analyzing tagged user capability portrayal |
Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101755283A (en) * | 2007-07-24 | 2010-06-23 | 三星电子株式会社 | Method and apparatus for recommending information using hybrid algorithm |
| CN105893609A (en) * | 2016-04-26 | 2016-08-24 | 南通大学 | Mobile APP recommendation method based on weighted mixing |
| CN106056427A (en) * | 2016-05-25 | 2016-10-26 | 中南大学 | Spark-based big data hybrid model mobile recommending method |
Family Cites Families (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US9818409B2 (en) * | 2015-06-19 | 2017-11-14 | Google Inc. | Context-dependent modeling of phonemes |
-
2017
- 2017-08-07 CN CN201710666989.3A patent/CN107423442B/en active Active
Patent Citations (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101755283A (en) * | 2007-07-24 | 2010-06-23 | 三星电子株式会社 | Method and apparatus for recommending information using hybrid algorithm |
| CN105893609A (en) * | 2016-04-26 | 2016-08-24 | 南通大学 | Mobile APP recommendation method based on weighted mixing |
| CN106056427A (en) * | 2016-05-25 | 2016-10-26 | 中南大学 | Spark-based big data hybrid model mobile recommending method |
Non-Patent Citations (1)
| Title |
|---|
| 基于级联过滤和增强模型集成的推荐方法硏究;黄斌;《中国优秀硕士学位论文全文数据库信息科技辑》;第I138-567页;20161015;第24-46页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN107423442A (en) | 2017-12-01 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN107423442B (en) | Application recommendation method and system based on user portrait behavior analysis, storage medium and computer equipment | |
| CN111444428B (en) | Information recommendation method and device based on artificial intelligence, electronic equipment and storage medium | |
| CN110162703B (en) | Content recommendation method, training device, content recommendation equipment and storage medium | |
| CN111177575B (en) | Content recommendation method and device, electronic equipment and storage medium | |
| CN111339433B (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
| CN110503531B (en) | Dynamic social scene recommendation method based on time sequence perception | |
| CN110717098B (en) | Meta-path-based context-aware user modeling method and sequence recommendation method | |
| CN111797321B (en) | Personalized knowledge recommendation method and system for different scenes | |
| CN111611488B (en) | Information recommendation method and device based on artificial intelligence and electronic equipment | |
| CN111966914B (en) | Content recommendation method and device based on artificial intelligence and computer equipment | |
| US12020267B2 (en) | Method, apparatus, storage medium, and device for generating user profile | |
| CN105893609A (en) | Mobile APP recommendation method based on weighted mixing | |
| CN114202061A (en) | Article recommendation method, electronic device and medium based on generation of confrontation network model and deep reinforcement learning | |
| CN111680147A (en) | Data processing method, device, equipment and readable storage medium | |
| CN113051468B (en) | Movie recommendation method and system based on knowledge graph and reinforcement learning | |
| CN113032676B (en) | Recommendation method and system based on micro-feedback | |
| CN115618101A (en) | Streaming media content recommendation method and device based on negative feedback and electronic equipment | |
| CN117217284A (en) | Data processing method and device | |
| CN112800109A (en) | Information mining method and system | |
| CN111209469A (en) | Personalized recommendation method and device, computer equipment and storage medium | |
| CN110083766B (en) | Query recommendation method and device based on meta-path guiding embedding | |
| CN117235264A (en) | Text processing method, apparatus, device and computer readable storage medium | |
| CN116910357A (en) | Data processing method and related device | |
| CN114328992A (en) | Multimedia information recommendation method, device, program product, equipment and medium | |
| CN114547472A (en) | Information recommendation method and device |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| GR01 | Patent grant | ||
| GR01 | Patent grant |