CN107993085A

CN107993085A - Model training method, the user's behavior prediction method and device based on model

Info

Publication number: CN107993085A
Application number: CN201710980102.8A
Authority: CN
Inventors: 马书超
Original assignee: Alibaba Group Holding Ltd
Current assignee: Advanced Nova Technology Singapore Holdings Ltd
Priority date: 2017-10-19
Filing date: 2017-10-19
Publication date: 2018-05-04
Anticipated expiration: 2037-10-19
Also published as: CN107993085B; CN113409084A

Abstract

This specification embodiment provides a kind of model training method, the user's behavior prediction method and device based on model, this specification embodiment can be used for the prediction model for predicting the specific behavior of user according to training sample and special algorithm training, when needing to predict the specific behavior of user, obtain the user with the relevant data of specific behavior, from this and the relevant extracting data of specific behavior and the relevant feature of specific behavior, and the prediction model will be inputted with the relevant feature of specific behavior and obtain output valve, according to the output valve, the specific behavior of the user is predicted.Due to user and the relevant data of specific behavior, it can largely reflect that the behavior of user is intended to, therefore this specification embodiment can relatively accurately predict the specific behavior of user, afterwards according to the specific behavior push relevant information predicted, so as to improve the precision of pushed information.

Description

Model training method, the user's behavior prediction method and device based on model

Technical field

This application involves field of computer technology, more particularly to a kind of model training method, the user behavior based on model Forecasting Methodology and device.

Background technology

With the fast development of Internet technology, the service platform based on Internet technology is more next for bringing for people More facility, for example, class application of going on a tour can push some information of going on a tour to people.If the specific of user can be predicted Behavior, relevant information is pushed according to the specific behavior, can undoubtedly make it that the information of push is more accurate.

The content of the invention

The purpose of this specification embodiment is to provide a kind of model training method, the user's behavior prediction method based on model And device, relatively accurately to predict the specific behavior of user, afterwards according to the related letter of specific behavior push predicted Breath, so as to improve the precision of pushed information.

In order to solve the above technical problems, what this specification embodiment was realized in：

First aspect, there is provided a kind of model training method, the described method includes：

Training sample set is obtained, the training sample concentrates the training sample included for training pattern, the trained sample This include user with the relevant data of specific behavior；

Extract that the training sample concentrates training sample with the relevant feature of specific behavior；

It is trained according to special algorithm to described with the relevant feature of specific behavior, obtains prediction model, the prediction Model is used to establish and the mapping relations between the relevant feature of specific behavior and the specific behavior information of forecasting of user.

Second aspect, there is provided a kind of user's behavior prediction method based on model, the described method includes：

Obtain targeted customer with the relevant data of specific behavior；

From the described and relevant extracting data of specific behavior and the relevant feature of specific behavior；

By the described and relevant feature input prediction model of specific behavior, output valve is obtained, the prediction model is use Special algorithm is trained obtained model to training sample and the relevant feature of specific behavior, and the prediction model is used for Foundation and the mapping relations between the relevant feature of specific behavior and the specific behavior information of forecasting of user；

According to the output valve, the specific behavior of the targeted customer is predicted.

The third aspect, there is provided a kind of model training apparatus, described device include：

First acquisition unit, for obtaining training sample set, the training sample concentrates the instruction included for training pattern Practice sample, the training sample include user with the relevant data of specific behavior；

First extraction unit, for extract the training sample that the first acquisition unit gets concentrate training sample with The relevant feature of specific behavior；

Training unit, for being extracted according to special algorithm to first extraction unit and the relevant spy of specific behavior Sign is trained, and obtains prediction model, and the prediction model is used to establishing the specific of feature relevant with specific behavior and user Mapping relations between behavior prediction information.

Fourth aspect, there is provided a kind of user's behavior prediction device based on model, described device include：

Second acquisition unit, for obtaining targeted customer and the relevant data of specific behavior；

Second extraction unit, for being got from the second acquisition unit and the relevant extracting data of specific behavior With the relevant feature of specific behavior；

Processing unit, for extracting second extraction unit and the relevant feature input prediction mould of specific behavior Type, obtains output valve, and the prediction model is that training sample feature relevant with specific behavior is carried out using special algorithm The obtained model of training, the prediction model are used to establishing feature relevant with specific behavior and pre- to the specific behavior of user Mapping relations between measurement information；

Predicting unit, for the output valve obtained according to the processing unit processes, predicts that the targeted customer's is specific Behavior.

5th aspect, there is provided a kind of electronic equipment, including：

Processor；And

The memory of storage computer executable instructions is arranged to, the executable instruction makes the place when executed Manage device and perform following operation：

6th aspect, there is provided a kind of electronic equipment, including：

Processor；And

Obtain targeted customer with the relevant data of specific behavior；

7th aspect, there is provided a kind of computer-readable storage medium, the computer-readable recording medium storage one or more A program, one or more of programs by the electronic equipment including multiple application programs when being performed so that the electronics is set It is standby to perform following operation：

Eighth aspect, there is provided a kind of computer-readable storage medium, the computer-readable recording medium storage one or more A program, one or more of programs by the electronic equipment including multiple application programs when being performed so that the electronics is set It is standby to perform following operation：

Obtain targeted customer with the relevant data of specific behavior；

The technical solution provided by above this specification embodiment is as it can be seen that this specification embodiment can be according to training sample And the prediction model of specific behavior of the special algorithm training for predicting user, when needing to predict the specific behavior of user, obtain Take the user with the relevant data of specific behavior, it is relevant with the relevant extracting data of specific behavior and specific behavior from this Feature, and the prediction model will be inputted with the relevant feature of specific behavior and obtain output valve, according to the output valve, predict the user Specific behavior.Due to user and the relevant data of specific behavior, it can largely reflect that the behavior of user is anticipated Figure, therefore this specification embodiment can relatively accurately predict the specific behavior of user, it is specific according to predicting afterwards Behavior pushes relevant information, so as to improve the precision of pushed information.

Brief description of the drawings

In order to illustrate more clearly of this specification embodiment or technical solution of the prior art, below will to embodiment or Attached drawing needed to be used in the description of the prior art is briefly described, it should be apparent that, drawings in the following description are only Some embodiments described in this specification, for those of ordinary skill in the art, in not making the creative labor property Under the premise of, other attached drawings can also be obtained according to these attached drawings.

Fig. 1 is the flow chart of one embodiment model training method of this specification；

Fig. 2 is the flow chart of user's behavior prediction method of the one embodiment of this specification based on model；

Fig. 3 is the instance graph of user's behavior prediction method of the one embodiment of this specification based on model；

Fig. 4 is the instance graph of user's behavior prediction method of another embodiment of this specification based on model；

Fig. 5 is the structure diagram of one embodiment electronic equipment of this specification；

Fig. 6 is the structure diagram of another embodiment electronic equipment of this specification；

Fig. 7 is the structure diagram of one embodiment model training apparatus of this specification；

Fig. 8 is the structure diagram of user's behavior prediction device of the one embodiment of this specification based on model.

Embodiment

In order to make those skilled in the art more fully understand the technical solution in this specification, below in conjunction with this explanation Attached drawing in book embodiment, is clearly and completely described the technical solution in this specification embodiment, it is clear that described Embodiment be only this specification part of the embodiment, instead of all the embodiments.Based on the embodiment in this specification, sheet Field those of ordinary skill all other embodiments obtained without making creative work, should all belong to this The scope of specification protection.

This specification embodiment provides a kind of model training method, the user's behavior prediction method based on model and dress Put.

A kind of model training method provided first below this specification embodiment is introduced.

Fig. 1 is the flow chart of one embodiment model training method of this specification, as shown in Figure 1, this method can wrap Include following steps：

In S102, training sample set is obtained, which concentrates the training sample included for training pattern, the instruction Practice sample include user with the relevant data of specific behavior.

In this specification embodiment, specific behavior can include：Tourist behavior, buying behavior or body-building behavior etc..For Readily appreciate, below mainly this specification embodiment will be described by taking tourist behavior as an example, other behaviors are with going out parade To be similar.

In this specification embodiment, training sample, which is concentrated, includes multiple training samples, and the plurality of training sample includes：It is multiple Positive sample and multiple negative samples, it is generally the case that positive sample includes：Had that the user's of specific behavior is relevant with specific behavior Data, negative sample include：Always there is no a user of specific behavior with the relevant data of specific behavior.When specific behavior is Parade for when, positive sample includes：There is including with relevant data of going on a tour, negative sample for the user of tourist behavior：Always do not have Cross the user of tourist behavior with relevant data of going on a tour.

In this specification embodiment, any group of one or more of can be included with the relevant data of specific behavior Close：Search record, purchaser record, collection record, browse record and application program number of clicks, in addition, in any of the above-described combination On the basis of, it can also include with the relevant data of specific behavior：User basic information and user tag.Wherein, application program All application programs installed in the terminal device that can be user, such as installed in Android operation system mobile phone all Application program, or all application programs in all application programs in an application, such as " Alipay application ", are used Family label be for characterize user whether be specific behavior fan information.

When specific behavior is tourist behavior, in one example, the positioning for the user for having tourist behavior can be believed It is determined as positive sample with relevant data of going on a tour in 30 days before breath (positional information on ground of going on a tour), go on a tour time T and T time.

In S104, extraction training sample concentrate training sample with the relevant feature of specific behavior.

Alternatively, in one embodiment, training sample includes with the relevant data of specific behavior in S102：Search Record, at this time, above-mentioned S104 can include：

Extracted and the relevant keyword of specific behavior in being recorded from search；

One of using the keyword as feature relevant with specific behavior or with the relevant feature of specific behavior.

When specific behavior is tourist behavior, in one example, record has search time and search sequence in search record Row, the search sequence in being recorded to search carry out cutting word, and relevant keyword of excavating and go on a tour, keyword is made in itself For with relevant feature of going on a tour.Alternatively, can also handle some special words, assemblage characteristic is formed, assemblage characteristic is made For with relevant feature of going on a tour, wherein, assemblage characteristic includes two classes：

The first kind, carries out specially treated to the national vocabulary in search sequence, for this kind of word, manually establishes vocabulary；Example Such as, if user search for " Hong Kong micro-wave oven ", since Hong Kong is in the vocabulary of specially treated, followed by be category word, institute Assemblage characteristic is used as using Hong Kong+category.

Second class, carries out specially treated to words such as the travelling in search sequence, wifi, for this kind of word, manually establishes word Table, concrete processing procedure are as follows：Judge whether user searched for national relevant word the proximal segment time, if any, these words Feature is used as with country together combination；For example, a user has searched for " travelling ", if this user searches the proximal segment time again " Thailand ", Thailand+travelling is as assemblage characteristic, if only having searched for travelling, this word is directly filtered out not as feature.

Alternatively, in another embodiment, training sample includes with the relevant data of specific behavior in S102：Purchase Record is bought, at this time, above-mentioned S104 can include：

Extraction and the relevant Item Number of specific behavior and goods categories from purchaser record；

Using Item Number and goods categories as feature relevant with specific behavior or with the relevant feature of specific behavior it One.

When specific behavior is tourist behavior, in one example, purchaser record includes time buying and purchase article Relevant information, can from purchase article relevant information in extract Item Number id and goods categories category, wherein, should Article can include：Actual object and virtual objects, wherein, virtual objects can be order etc. of going on a tour.

Alternatively, in another embodiment, training sample includes with the relevant data of specific behavior in S102：Receive Record is hidden, at this time, above-mentioned S104 can include：

Extraction and the relevant Item Number of specific behavior and goods categories from collection record；

Alternatively, in another embodiment, training sample includes with the relevant data of specific behavior in S102：It is clear Look at record, at this time, above-mentioned S104 can include：

Extracted and the relevant Item Number of specific behavior and goods categories from browsing in record；

It should be noted that due to collection record and the characteristic extraction procedure for browsing record, carried with the feature of purchaser record Take process similar, therefore details are not described herein.

Alternatively, in another embodiment, training sample includes with the relevant data of specific behavior in S102：Should With program number of clicks, at this time, above-mentioned S104 can include：

According to application program number of clicks, the accounting value with the relevant application program number of clicks of specific behavior is calculated；

One of using accounting value as feature relevant with specific behavior or with the relevant feature of specific behavior.

When specific behavior tourist behavior, in one example, above-mentioned S104 can include：Clicked on according to application program secondary Number, the accounting value for the relevant application program number of clicks that calculates and go on a tour；Using accounting value as with go on a tour relevant feature or with Go on a tour one of relevant feature.

For example, filtering out in advance in " Alipay application " with the relevant application program app that goes on a tour, " paid according to user To the click logs of app in Baoying County's use ", user is obtained to the number of clicks of above-mentioned app and to all app in " Alipay application " Total number of clicks, obtain the accounting value between one [0-1].

Alternatively, in another embodiment, in S102 training sample with the relevant data of specific behavior above-mentioned It can also include on the basis of any embodiment：User basic information and user tag, the user's label are used to characterize user Whether it is specific behavior fan；At this time, above-mentioned S104 can include：

User basic information feature is extracted from user basic information, the user's essential information feature and user tag are made For one of with the relevant feature of specific behavior；Wherein, the user's essential information feature can include：Age and occupation etc..

When specific behavior is tourist behavior, in one example, user tag is used to characterize whether user is love of going on a tour Good person, at this time, using the user's essential information feature and user tag as with one of relevant feature of going on a tour.

Preferably as an example, training sample includes with the relevant data of specific behavior in S102：Search note Record, purchaser record, collection record, browse record and application program number of clicks, at this time, is carried from the search record of training sample Take keyword relevant with specific behavior, from the purchaser record of training sample extraction with the relevant Item Number of specific behavior and Goods categories, from the collection record of training sample extraction with the relevant Item Number of specific behavior and goods categories, from training Sample browses extraction and the relevant Item Number of specific behavior and goods categories and the application according to training sample in record Program number of clicks calculates the accounting value with the relevant application program number of clicks of specific behavior, by the keyword extracted, thing Product numbering and goods categories and accounting value are as training sample and the relevant feature of specific behavior.

Preferably, as an example, training sample includes with the relevant data of specific behavior in S102：Search note Record, purchaser record, collection record, browse record, application program number of clicks, user basic information and user tag, at this time, from Extraction is extracted and spy with the relevant keyword of specific behavior, from the purchaser record of training sample during the search of training sample records Determine the relevant Item Number of behavior and goods categories, extraction and the relevant article of specific behavior from the collection record of training sample Numbering and goods categories, from training sample browse extraction and the relevant Item Number of specific behavior and goods categories in record, Application program number of clicks according to training sample calculates with the accounting value of the relevant application program number of clicks of specific behavior, with And the user basic information feature such as age and occupation, the key that will be extracted are extracted from the user basic information of training sample Word, Item Number and goods categories, accounting value, user basic information feature and user tag are as specific with training sample The relevant feature of behavior.

It should be noted that since the gap of positive negative sample is larger, usually require to sample negative sample, specifically The ratio of sampling, can be adjusted according to actual needs, and this specification embodiment is not construed as limiting this.

In S106, training sample and the relevant feature of specific behavior are trained according to special algorithm, obtained pre- Model is surveyed, which is used to establish reflecting between the relevant feature of specific behavior and the specific behavior information of forecasting of user Penetrate relation.

In this specification embodiment, special algorithm can include:Logistic regression algorithm or neural network algorithm.

Preferably, in this specification embodiment, special algorithm is logistic regression algorithm, correspondingly, prediction model is LR moulds Type.Logistic regression is a kind of sorting technique, and being mainly used for two classification problems of solution, (i.e. only two kinds of output, represents two respectively Classification), logistic regression algorithm utilizes Logistic functions (or being Sigmoid functions), and the curve form of the function is bent for S types Line, functional form are：

In the case of linear barrier, boundary regime is as follows：

Utilize formula (1) and formula (2) structure forecast function：

Wherein, θ_iFor weighted value, x_iIt is characterized the corresponding characteristic values of i, θ^T=[θ₁,θ₂,...,θ_n], x=[x₁,x₂,..., x_n], under normal conditions for a training sample, if the feature i in the training sample meets certain condition, x_iTake It is worth for 1, otherwise value is 0；In addition, characteristic value can also be other natural numbers, this specification embodiment is not construed as limiting this.

It is numerical value by the characteristic quantification of the training sample for each training sample in this specification embodiment, that is, trains One feature of sample corresponds to a numerical value (namely characteristic value), and the characteristic value of each training sample is carried out vectorization table Show, finally by vectorization expression as a result, bringing into formula (3).A large amount of training samples are processed as above, can be obtained Big flow function, is iterated solution to foregoing big flow function afterwards, θ is calculated^T=[θ₁,θ₂,...,θ_n], so as to obtain LR Model, i.e., one probability function being made of the corresponding characteristic value of multiple features and the corresponding weighted value of each characteristic value：

For the process by the characteristic quantification of training sample for numerical value, by taking accounting value as an example, accounting value is before [0,1] One value, is multiplied by accounting value 5, carries out rounding, obtains the numeral between [0,5], therefrom chooses a numeral as characteristic value.

It should be noted that by the characteristic quantification (conversion) of training sample into the process of numerical value, can according to actual needs, It is suitable numerical value by each Feature Conversion of training sample using rational rule.In addition, when using LR models, will treat Prediction user's is input to LR models with the relevant feature of specific behavior, and the output of LR models is a probable value, the probable value Value range be 0~1.

In this specification embodiment, the different features of training sample can be used, train the LR models of different purposes, Specifically, can train for predicting whether user there are the LR models for the purpose for making specific behavior, use can also be trained In the LR models of the purpose specific behavior of prediction user.

When specific behavior is tourist behavior, can train for predicting whether user there are the LR models for the purpose of going on a tour, The LR models of the destination of going on a tour for predicting user can also be trained.

Alternatively, in one embodiment, when needs training is used to predict whether user has the meaning for making specific behavior To LR models when, can use LR algorithm and training sample the relevant feature of following and specific behavior at least two： Keyword, Item Number and goods categories, accounting value, are trained to obtain LR models.

Preferably as an example, using LR algorithm to keyword, Item Number and goods categories, accounting value, user Essential information feature and user tag are trained, and are obtained for predicting whether user has the LR for the purpose for making specific behavior Model.

Alternatively, in another embodiment, training is being needed to be used for the LR moulds for predicting the purpose specific behavior of user During type, mainly using search characteristics.

When specific behavior is tourist behavior, since country is relatively more, user multiple destinations that are possible to go on a journey (can be Country), a LR model is established for each destination of going on a tour, when estimating destination, mainly using search characteristics.With training One is used for exemplified by predicting the LR models for destination of going on a tour of user, and uses the following related to going on a tour of LR algorithm and training sample Feature：Keyword, is trained to obtain LR models.The training process such as above process of the corresponding LR models in multiple destinations, Details are not described herein.

So far training has obtained two class LR models, when needing to be predicted the tourist behavior of user to be predicted, if Need to predict whether the user goes on a tour purpose, then using for predicting whether user there are the LR models for the purpose of going on a tour, from the use Family with relevant data of going on a tour, extraction meet the LR models input requirements feature, the feature extracted is input to In the LR models, output valve is obtained, wherein, output valve illustrates that the user may more go on a tour purpose, output valve is got over closer to 1 Close to 0, illustrate the more unlikely purpose of going on a tour of the user.

If necessary to predict which place the user goes go on a tour, then using multiple destinations of going on a tour for being used to predict user LR models, each LR models correspond to a destination of going on a tour, from the user with relevant data of going on a tour, extraction meets this kind of The feature of the input requirements of the LR models, the feature extracted is input in the plurality of this kind of LR models, obtains multiple outputs Value, wherein, output valve illustrates, the user is more possible to the corresponding destination of the output valve, and output valve is closer closer to 1 0, then explanation the user is more unlikely goes to the corresponding destination of the output valve.

As seen from the above-described embodiment, which can be used to predict user's according to training sample and special algorithm training The prediction model of specific behavior, when needing to predict the specific behavior of user, obtain the user with the relevant number of specific behavior According to, from this and the relevant extracting data of specific behavior and the relevant feature of specific behavior, and will be with the relevant spy of specific behavior Sign inputs the prediction model and obtains output valve, according to the output valve, predicts the specific behavior of the user.Due to user with it is specific The relevant data of behavior, can largely reflect that the behavior of user is intended to, therefore this specification embodiment can compare The specific behavior of user is accurately predicted, afterwards according to the specific behavior push relevant information predicted, is pushed away so as to improve Deliver letters the precision of breath.

In addition, with international development, overseas trip (specific behavior of going abroad) user is selected to increase, according to user's Behavioral data, the departure of intelligent acquisition user are intended to, it will the significantly more efficient dispensing for carrying out marketing activity, lifts user's Physical examination, can reduce the artificial making time of operation, improve the effect of operation activity, and then increase the viscosity of product, promote across Swim the development of business in border.Based on this situation, in this specification embodiment, specific behavior can be specific behavior of going abroad, at this time, with The relevant data of specific behavior be with relevant data of going abroad, with the relevant relevant feature that is characterized as and goes abroad of specific behavior. Train to obtain LR models using LR algorithm and with relevant feature of going abroad, can be used for predicting whether user goes abroad purpose, And which country prediction user will go, due to the training of LR models in the training process and embodiment illustrated in fig. 1 of its LR model Process is similar, therefore details are not described herein.

The establishment process of the prediction model of the specific behavior for predicting user is described above, below to how to use institute The specific behavior of the prediction model prediction user of establishment is introduced.

Fig. 2 is the flow chart of user's behavior prediction method of the one embodiment of this specification based on model, such as Fig. 2 institutes Show, this method may comprise steps of：

In S202, obtain targeted customer with the relevant data of specific behavior.

In this specification embodiment, targeted customer is the user of specific behavior to be predicted, targeted customer's and specific behavior Relevant data are targeted customer and the relevant historical data of specific behavior.

In this specification embodiment, targeted customer's can include following at least one with the relevant data of specific behavior： Search record, purchaser record, collection record, browse record and application program number of clicks.

When specific behavior is tourist behavior, targeted customer with go on a tour relevant data for targeted customer and phase of going on a tour The historical data of pass.Targeted customer's can include following at least one with relevant data of going on a tour：Search record, purchase note Record, collection record, browse record and application program number of clicks.

Alternatively, in one embodiment, above-mentioned S202 can include：

Obtain before current time that targeted customer with the relevant data of specific behavior, M is natural number in M days.When specific When behavior is tourist behavior, obtain current time before in M days targeted customer with relevant data of going on a tour.

In present embodiment, M can be a natural number for being empirically worth setting, such as M values are 8；M can also be A numerical value is gone out according to neural metwork training.

In S204, from targeted customer and the relevant extracting data of specific behavior and the relevant feature of specific behavior.

In S206, by targeted customer and the relevant feature input prediction model of specific behavior, output valve is obtained, this is pre- It is to be trained obtained model to training sample and the relevant feature of specific behavior using special algorithm to survey model, this is pre- Survey model and be used for foundation and the mapping relations between the relevant feature of specific behavior and the specific behavior information of forecasting of user.

Preferably, in this specification embodiment, special algorithm is LR algorithm, and prediction model is LR models.Using LR moulds During type, it is necessary to from targeted customer's feature corresponding with the input requirements of LR models with the relevant extracting data of specific behavior, And by extract targeted customer, it is meeting input requirements, with the relevant feature of specific behavior input the LR models, exported Value.The output of LR models is a probable value, and the value range of the probable value is 0~1.

In S208, according to output valve, the specific behavior of targeted customer is predicted.

As seen from the above-described embodiment, when needing to predict the specific behavior of user, which can obtain the user's With the relevant data of specific behavior, from this and the relevant extracting data of specific behavior and the relevant feature of specific behavior, and incite somebody to action Output valve is obtained with the relevant feature input prediction model of specific behavior, according to the output valve, predicts the specific behavior of the user. Due to user and the relevant data of specific behavior, it can largely reflect that the behavior of user is intended to, therefore this theory Bright book embodiment can relatively accurately predict the specific behavior of user, related according to the specific behavior push predicted afterwards Information, so as to improve the precision of pushed information.

This specification provide another embodiment in, can according to targeted customer and the relevant data of specific behavior, Predict whether the targeted customer has the purpose for making specific behavior, in order to achieve the above object, used prediction model is root According to the search record of training sample, purchaser record, collection record, browse in record and application program number of clicks at least two Constructed model, the prediction model are used to predict whether user has the purpose for making specific behavior；

At this time, the S202 in above-mentioned embodiment illustrated in fig. 2 can specifically include：

Targeted customer, and specific behavior relevant data corresponding with used prediction model are obtained, wherein, mesh The type with the relevant data of specific behavior of user is marked, it is related with specific behavior to the training sample for training LR models Data type it is identical.

At this time, the S204 in above-mentioned embodiment illustrated in fig. 2 can specifically include：

Perform at least two in following characteristics extraction operation：Extraction and specific behavior from the search record of targeted customer Relevant keyword, extraction and the relevant Item Number of specific behavior and goods categories from the purchaser record of targeted customer, from Extraction and the relevant Item Number of specific behavior and goods categories, note is browsed from targeted customer in the collection record of targeted customer Extraction and the relevant Item Number of specific behavior and goods categories in record, and the application program number of clicks according to targeted customer Calculate the accounting value with the relevant application program number of clicks of specific behavior；

Using the content that feature extraction operation is extracted as targeted customer and the relevant feature of specific behavior.

At this time, the S206 in above-mentioned embodiment illustrated in fig. 2 can specifically include：

It is that features described above extraction operation is extracted, meet prediction model input requirements, with the relevant spy of specific behavior Sign inputs the prediction model, obtains output valve.

At this time, the S208 in above-mentioned embodiment illustrated in fig. 2 can specifically include：

If output valve reaches preset first threshold value, prediction targeted customer has the purpose for making specific behavior；

If the output valve is not up to preset first threshold value, prediction targeted customer does not make the purpose of specific behavior.

When specific behavior is tourist behavior, with the relevant data of specific behavior be with relevant data of going on a tour, it is and specific The relevant relevant feature that is characterized as and goes on a tour of behavior；When special algorithm is LR algorithm, prediction model is LR models；At this time, In one example, as shown in figure 3, being according to search record, purchase for predicting whether user has the LR models for the purpose of going on a tour Record, collection record, browse record and the model constructed by application program number of clicks, then targeted customer's is related to going on a tour Data should include：The search record of targeted customer, purchaser record, collection record, browse record and application program is clicked on time Number, is extracted with relevant keyword of going on a tour, purchaser record, collection note from targeted customer from the search record of targeted customer afterwards Extract and go on a tour in record relevant Item Number and goods categories are recorded and browse, the application program according to targeted customer is clicked on secondary Number calculate with the accounting values of relevant application program numbers of clicks of going on a tour, and by the keyword extracted, Item Number and article Classification, accounting value are input to the LR models, obtain output valve, wherein, for output valve closer to 1, illustrating that the targeted customer more has can Can be gone on a tour purpose, and output valve illustrates the more unlikely purpose of going on a tour of the targeted customer closer to 0.Its prediction result is specific For：If output valve reaches preset first threshold value (such as 0.7), prediction targeted customer goes on a tour purpose；If the output valve does not reach To preset first threshold value, then predict that targeted customer does not go on a tour purpose.

This specification provide another embodiment in, can according to targeted customer and the relevant data of specific behavior, Predict the purpose specific behavior of the targeted customer, in order to achieve the above object, used prediction model includes：According to training sample This search record, purchaser record, collection record, browse in record and application program number of clicks at least two constructed by First kind prediction model, and the second class prediction model of multiple search record structures according to training sample, first kind prediction mould Type is used to predict whether user has the purpose for making specific behavior, and the second class prediction model is used for the purpose particular row for predicting user For a second class prediction model corresponds to a purpose specific behavior；

Targeted customer, and specific behavior relevant data corresponding with used prediction model are obtained, wherein, mesh The type with the relevant data of specific behavior of user is marked, with the training sample for training prediction model and specific behavior phase The type of the data of pass is identical.

From targeted customer search record in extract targeted customer with the relevant keyword of specific behavior；

If the search record of training sample involved in the building process of first kind prediction model, performs following characteristics extraction At least one of operation：Extraction and the relevant Item Number of specific behavior and article class from the purchaser record of targeted customer Not, extraction and the relevant Item Number of specific behavior and goods categories from the collection record of targeted customer, from targeted customer's Browse extraction and the relevant Item Number of specific behavior and goods categories in record, and the application program point according to targeted customer Hit number calculating and the accounting value of the relevant application program number of clicks of specific behavior；

If being not directed to the search record of training sample in the building process of first kind prediction model, perform following characteristics and carry At least two in extract operation：Extraction and the relevant Item Number of specific behavior and article class from the purchaser record of targeted customer Not, extraction and the relevant Item Number of specific behavior and goods categories from the collection record of targeted customer, from targeted customer's Browse extraction and the relevant Item Number of specific behavior and goods categories in record, and the application program point according to targeted customer Hit number calculating and the accounting value of the relevant application program number of clicks of specific behavior；The keyword and feature extraction operation are carried The content got is as targeted customer and the relevant feature of specific behavior.

If the building process of first kind prediction model is related to the search record of training sample, by from the search of targeted customer The content input first kind prediction model that the keyword and features described above extraction operation extracted in record extracts, is exported Value；

If the building process of first kind prediction model is not directed to the search record of training sample, features described above is extracted and is grasped Make the content input first kind prediction model extracted, obtain output valve；

If the output valve of first kind prediction model is more than default second threshold, will be carried from the search of targeted customer record The keyword got inputs multiple second class prediction models respectively, obtains multiple output valves.

From multiple output valves corresponding with the second class prediction model, the output valve of value ranking top N is determined；

By specific behavior corresponding to the output valve of value ranking top N, it is determined as the purpose specific behavior of targeted customer, N For natural number.In practical applications, the value of N can be 5.

When specific behavior is tourist behavior, with the relevant data of specific behavior be with relevant data of going on a tour, it is and specific The relevant relevant feature that is characterized as and goes on a tour of behavior；When special algorithm is LR algorithm, prediction model is LR models；At this time, In one example, as shown in figure 4, targeted customer and relevant feature input of going on a tour are used to predict whether user goes on a tour The process of the LR models of purpose, it is similar with upper one embodiment, this is repeated no more.4 the second class LR moulds are shown in Fig. 4 Type, is respectively：LR models A for the destination of going on a tour of predicting user, the LR models of the destination of going on a tour for predicting user B, will be from mesh for the LR MODEL Cs for destination of going on a tour and the LR model D of the destination of going on a tour for predicting user of predicting user The keyword extracted in the search record of mark user is inputted in above-mentioned 4 the second class LR models, 4 output valves is obtained, for example, 4 A output valve is respectively：0.9th, 0.95,0.6 and 0.5, by the corresponding destination of the output valve of value front two, i.e., by output valve 0.9th, 0.95 corresponding destination, is predicted as the destination of going on a tour of targeted customer.

With international development, overseas trip is selected (to go abroad and go on a tour) the increasing of user, according to the behavioral data of user, The departure of the acquisition user of intelligence is intended to, it will the significantly more efficient dispensing for carrying out marketing activity, lifts the physical examination of user, can be with The artificial making time of operation is reduced, improves the effect of operation activity, and then increases the viscosity of product, promotes cross-border trip business Development.

Based on this situation, in this specification embodiment, it can be to go abroad to go on a tour to go on a tour, and at this time, this specification provides another In one embodiment, used LR models are for the LR for the tourist behavior of going abroad for predicting targeted customer, are carrying out model instruction When practicing, training sample includes with relevant data of going on a tour：The training sample with relevant data of going abroad, training sample with Relevant feature of going on a tour includes：The training sample with relevant feature of going abroad, using LR algorithm and training sample with going out The relevant feature of state trains to obtain LR models, can be used for predicting whether user goes abroad purpose, and which prediction user will go A country, since the training process of this LR model is similar with the training process of LR models in embodiment illustrated in fig. 1, herein not Repeat again.

Using this LR model prediction targeted customer go abroad tourist behavior when, it is necessary to obtain targeted customer with going abroad out Swim relevant data, extract targeted customer with relevant feature of going on a tour of going abroad, input this LR model, obtain output valve, foundation The output valve, predicts the tourist behavior of going abroad of targeted customer, since it predicts the prediction process class of process and embodiment illustrated in fig. 2 Seemingly, therefore details are not described herein.

In this specification embodiment, prediction result can be used for the intelligent marketing scheme before cross-border go on a tour, to ensure to excavate Outbid to be worth potential departure crowd, further, different countries can be directed to, different operation activities is configured, effectively carry The efficiency and effect of high operation activity.

It can be seen from the above that this specification embodiment can carry out the prediction of potential departure crowd based on LR models, so as to reach Very high precision, and LR algorithm is simple, and computation complexity is small, can tackle ultra-large user behavior data.In addition, in base After potential departure crowd result, with further predicting its trip purpose, the intention of the crowd further refined, more added with Improve to effect the precision and effect of operation activity.

Fig. 5 is the structure diagram of one embodiment electronic equipment of this specification.Fig. 5 is refer to, should in hardware view Electronic equipment includes processor, alternatively further includes internal bus, network interface, memory.Wherein, memory may include interior Deposit, such as high-speed random access memory (Random-Access Memory, RAM), it is also possible to further include non-volatile memories Device (non-volatile memory), for example, at least 1 magnetic disk storage etc..Certainly, which is also possible that other The required hardware of business.

Processor, network interface and memory can be connected with each other by internal bus, which can be ISA (Industry Standard Architecture, industry standard architecture) bus, PCI (Peripheral Component Interconnect, Peripheral Component Interconnect standard) bus or EISA (Extended Industry Standard Architecture, expanding the industrial standard structure) bus etc..The bus can be divided into address bus, data/address bus, control always Line etc..For ease of representing, only represented in Fig. 5 with a four-headed arrow, it is not intended that an only bus or a type of Bus.

Memory, for storing program.Specifically, program can include program code, and said program code includes calculating Machine operational order.Memory can include memory and nonvolatile memory, and provide instruction and data to processor.

Processor reads corresponding computer program into memory and then runs from nonvolatile memory, in logical layer Model training apparatus is formed on face.Processor, performs the program that memory is stored, and specifically for performing following operation：

The method that model training apparatus disclosed in the above-mentioned embodiment illustrated in fig. 5 such as this specification performs can be applied to handle In device, or realized by processor.Processor is probably a kind of IC chip, has the disposal ability of signal.Realizing During, each step of the above method can pass through the integrated logic circuit of the hardware in processor or the instruction of software form Complete.Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit, CPU), network processing unit (Network Processor, NP) etc.；It can also be digital signal processor (Digital Signal Processor, DSP), it is application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), existing Field programmable gate array (Field-Programmable Gate Array, FPGA) or other programmable logic device, divide Vertical door or transistor logic, discrete hardware components.It can realize or perform and is in this specification embodiment disclosed Each method, step and logic diagram.General processor can be microprocessor or the processor can also be any conventional Processor etc..The step of method with reference to disclosed in this specification embodiment, can be embodied directly in hardware decoding processor execution Complete, or completion is performed with the hardware in decoding processor and software module combination.Software module can be located at random storage Device, flash memory, read-only storage, this area such as programmable read only memory or electrically erasable programmable memory, register into In ripe storage medium.The storage medium is located at memory, and processor reads the information in memory, is completed with reference to its hardware The step of stating method.

The method that the electronic equipment can also carry out Fig. 1, and implementation model training device is in the function of embodiment illustrated in fig. 1, Details are not described herein for this specification embodiment.

Fig. 6 is the structure diagram of one embodiment electronic equipment of this specification.Fig. 6 is refer to, should in hardware view Electronic equipment includes processor, alternatively further includes internal bus, network interface, memory.Wherein, memory may include interior Deposit, such as high-speed random access memory (Random-Access Memory, RAM), it is also possible to further include non-volatile memories Device (non-volatile memory), for example, at least 1 magnetic disk storage etc..Certainly, which is also possible that other The required hardware of business.

Processor, network interface and memory can be connected with each other by internal bus, which can be ISA (Industry Standard Architecture, industry standard architecture) bus, PCI (Peripheral Component Interconnect, Peripheral Component Interconnect standard) bus or EISA (Extended Industry Standard Architecture, expanding the industrial standard structure) bus etc..The bus can be divided into address bus, data/address bus, control always Line etc..For ease of representing, only represented in Fig. 6 with a four-headed arrow, it is not intended that an only bus or a type of Bus.

Processor reads corresponding computer program into memory and then runs from nonvolatile memory, in logical layer The user's behavior prediction device based on model is formed on face.Processor, performs the program that memory is stored, and specifically for holding The following operation of row：

Obtain targeted customer with the relevant data of specific behavior；

The method that the user's behavior prediction device based on model disclosed in the above-mentioned embodiment illustrated in fig. 6 such as this specification performs It can be applied in processor, or realized by processor.Processor is probably a kind of IC chip, has the place of signal Reason ability.During realization, each step of the above method can by the integrated logic circuit of the hardware in processor or The instruction of software form is completed.Above-mentioned processor can be general processor, including central processing unit (Central Processing Unit, CPU), network processing unit (Network Processor, NP) etc.；It can also be Digital Signal Processing Device (Digital Signal Processor, DSP), application-specific integrated circuit (Application Specific Integrated Circuit, ASIC), field programmable gate array (Field-Programmable Gate Array, FPGA) or other can Programmed logic device, discrete gate or transistor logic, discrete hardware components.It can realize or perform this specification reality Apply disclosed each method, step and the logic diagram in example.General processor can be that microprocessor or the processor also may be used To be any conventional processor etc..The step of method with reference to disclosed in this specification embodiment, can be embodied directly in hardware Decoding processor performs completion, or performs completion with the hardware in decoding processor and software module combination.Software module can With positioned at random access memory, flash memory, read-only storage, programmable read only memory or electrically erasable programmable memory, post In the storage medium of this areas such as storage maturation.The storage medium is located at memory, and processor reads the information in memory, knot Close the step of its hardware completes the above method.

The method that the electronic equipment can also carry out Fig. 2, and realize the user's behavior prediction device based on model shown in Fig. 2 The function of embodiment, details are not described herein for this specification embodiment.

Certainly, in addition to software realization mode, the electronic equipment of this specification is not precluded from other implementations, such as Mode of logical device or software and hardware combining etc., that is to say, that the executive agent of following process flow is not limited to each Logic unit or hardware or logical device.

This specification embodiment additionally provides a kind of computer-readable recording medium, the computer-readable recording medium storage One or more programs, the one or more program include instruction, and the instruction is when the portable electric for being included multiple application programs When sub- equipment performs, the method for portable electric appts execution embodiment illustrated in fig. 1 can be made, and it is following specifically for performing Method：

This specification embodiment additionally provides a kind of computer-readable recording medium, the computer-readable recording medium storage One or more programs, the one or more program include instruction, and the instruction is when the portable electric for being included multiple application programs When sub- equipment performs, the method for portable electric appts execution embodiment illustrated in fig. 2 can be made, and it is following specifically for performing Method：

Obtain targeted customer with the relevant data of specific behavior；

Fig. 7 is the structure diagram of one embodiment model training apparatus of this specification, refer to Fig. 7, a kind of soft In part embodiment, model training apparatus 700 may include：

First acquisition unit 701, for obtaining training sample set, the training sample, which is concentrated, to be included for training pattern Training sample, the training sample include user with the relevant data of specific behavior；

First extraction unit 702, training sample is concentrated for extracting the training sample that the first acquisition unit 701 is got This with the relevant feature of specific behavior；

Training unit 703, for being extracted according to special algorithm to first extraction unit 702 and specific behavior phase The feature of pass is trained, and obtains prediction model, and the prediction model is used to establish and the relevant feature of specific behavior and user Specific behavior information of forecasting between mapping relations.

Alternatively, it is described to include following at least one with the relevant data of specific behavior as one embodiment：Search note Record, purchaser record, collection record, browse record and application program number of clicks.

Alternatively, it is described to include with the relevant data of specific behavior as one embodiment：Search record；

First extraction unit 702, including：

First extraction subelement, for extraction in being recorded from described search and the relevant keyword of specific behavior；

Fisrt feature determination subelement, for using the keyword as feature relevant with specific behavior or and particular row For one of relevant feature.

Alternatively, it is described to include with the relevant data of specific behavior as one embodiment：Purchaser record；

First extraction unit 702, including：

Second extraction subelement, for the extraction from the purchaser record and the relevant Item Number of specific behavior and article Classification；

Second feature determination subelement, for using the Item Number and goods categories as with the relevant spy of specific behavior Sign or one of with the relevant feature of specific behavior.

Alternatively, it is described to include with the relevant data of specific behavior as one embodiment：Collection record；

First extraction unit 702, including：

3rd extraction subelement, for the extraction from the collection record and the relevant Item Number of specific behavior and article Classification；

Third feature determination subelement, for using the Item Number and goods categories as with the relevant spy of specific behavior Sign or one of with the relevant feature of specific behavior.

Alternatively, it is described to include with the relevant data of specific behavior as one embodiment：Browse record；

First extraction unit 702, including：

4th extraction subelement, for browsing extraction and the relevant Item Number of specific behavior and article in record from described Classification；

Fourth feature determination subelement, for using the Item Number and goods categories as with the relevant spy of specific behavior Sign or one of with the relevant feature of specific behavior.

Alternatively, it is described to include with the relevant data of specific behavior as one embodiment：Application program number of clicks；

First extraction unit 702, including：

5th extraction subelement, for according to the application program number of clicks, calculating and the relevant application of specific behavior The accounting value of program number of clicks；

Fifth feature determination subelement, for using the accounting value as feature relevant with specific behavior or and particular row For one of relevant feature.

Alternatively, it is described to be further included with the relevant data of specific behavior as one embodiment：User basic information and use Family label, the user tag are used to characterize whether the user is specific behavior fan；

First extraction unit 702, further includes：

6th extraction subelement, for extracting user basic information feature from the user basic information；

Sixth feature determination subelement, for using the user basic information feature and user tag as with specific behavior One of relevant feature.

Alternatively, as one embodiment, the special algorithm is logistic regression LR algorithm, and the prediction model is LR moulds Type.

Alternatively, as one embodiment, the specific behavior is tourist behavior, the described and relevant data of specific behavior For with relevant data of going on a tour, it is described with the relevant relevant feature that is characterized as and goes on a tour of specific behavior.

The method that model training apparatus 700 can also carry out embodiment illustrated in fig. 1, and implementation model training device is in Fig. 7 institutes Show the function of embodiment, details are not described herein for this specification embodiment.

Fig. 8 is the structure diagram of user's behavior prediction device of the one embodiment of this specification based on model, please be joined Fig. 8 is examined, in a kind of Software Implementation, the user's behavior prediction based on model is practiced device 800 and be may include：

Second acquisition unit 801, for obtaining targeted customer and the relevant data of specific behavior；

Second extraction unit 802, for being got from the second acquisition unit 801 and the relevant data of specific behavior Middle extraction and the relevant feature of specific behavior；

Processing unit 803, for feature relevant with the specific behavior input for extracting second extraction unit 802 Prediction model, obtains output valve, and the prediction model is to training sample and the relevant spy of specific behavior using special algorithm Sign is trained obtained model, and the prediction model is used to establish feature relevant with specific behavior and to the specific of user Mapping relations between behavior prediction information；

Predicting unit 804, for handling obtained output valve according to the processing unit 803, predicts the targeted customer Specific behavior.

Alternatively, as one embodiment, the prediction model is according to search record, purchaser record, collection record, clear Look in record and application program number of clicks at least two constructed by model, the prediction model is used for whether predicting user There is the purpose for making specific behavior；

Second extraction unit 802, including：

7th extraction subelement, for performing at least two in following characteristics extraction operation：From described search record Extraction and the relevant keyword of specific behavior, extraction and the relevant Item Number of specific behavior and article from the purchaser record Classification, extraction and the relevant Item Number of specific behavior and goods categories, browse in record from described from the collection record Extraction with the relevant Item Number of specific behavior and goods categories, and according to the application program number of clicks calculate with it is specific The accounting value of the relevant application program number of clicks of behavior；

Seventh feature determination subelement, for using the content that the feature extraction operation is extracted as with specific behavior phase The feature of pass.

Alternatively, as one embodiment, the predicting unit 804, including：

First prediction is alone, in the case of reaching preset first threshold value in the output valve, predicts the target User has the purpose for making specific behavior；In the case where the output valve is not up to the preset first threshold value, described in prediction Targeted customer does not make the purpose of specific behavior.

Alternatively, include as one embodiment, the prediction model：According to search record, purchaser record, collection note Record, browse in record and application program number of clicks at least two constructed by first kind prediction model, and multiple bases search Second class prediction model of Suo Jilu structures, the first kind prediction model make specific behavior for predicting whether user has Purpose, the second class prediction model are used for the purpose specific behavior for predicting user, and a second class prediction model corresponds to one Purpose specific behavior；

Second extraction unit 802, including：

8th extraction subelement, for extraction in being recorded from described search and the relevant keyword of specific behavior；And

If searching for record involved in the building process of the first kind prediction model, perform in following characteristics extraction operation At least one：Extraction and the relevant Item Number of specific behavior and goods categories from the purchaser record, from the collection Extraction and the relevant Item Number of specific behavior and goods categories in record, extraction and specific behavior phase in record are browsed from described The Item Number and goods categories of pass, and calculated and specific behavior is relevant applies journey according to the application program number of clicks The accounting value of sequence number of clicks；

If being not directed to search record in the building process of the first kind prediction model, following characteristics extraction operation is performed In at least two：Extraction and the relevant Item Number of specific behavior and goods categories from the purchaser record, from the receipts Extraction and the relevant Item Number of specific behavior and goods categories in record are hidden, extraction and specific behavior in record are browsed from described Relevant Item Number and goods categories, and calculated and the relevant application of specific behavior according to the application program number of clicks The accounting value of program number of clicks；

Eighth feature determination subelement, for using the content that the keyword and the feature extraction operation are extracted as With the relevant feature of specific behavior；

The processing unit 803, including：

First processing subelement, for being related to the situation of search record in the building process of the first kind prediction model Under, the content that the keyword and the feature extraction operation are extracted inputs the first kind prediction model, is exported Value；And

In the case where the building process of the first kind prediction model is not directed to search record, the feature extraction is grasped The content for making to extract inputs the first kind prediction model, obtains output valve；

Second processing subelement, in the case of being more than default second threshold in the output valve, by the keyword The multiple second class prediction model is inputted respectively, obtains multiple output valves.

Alternatively, as one embodiment, the predicting unit 804, including：

Output valve screening is alone, for from multiple output valves corresponding with the second class prediction model, determining to take It is worth the output valve of ranking top N；

Second prediction subelement, for by specific behavior corresponding to the output valve of the value ranking top N, being determined as institute The purpose specific behavior of targeted customer is stated, N is natural number.

Alternatively, as one embodiment, the second acquisition unit 801, including：

Obtain before current time that targeted customer with the relevant data of specific behavior, the M is natural number in M days.

Alternatively, as one embodiment, the special algorithm is LR algorithm, and the prediction model is LR models.

The method that user's behavior prediction device 800 based on model can also carry out embodiment illustrated in fig. 2, and realize and be based on mould The user's behavior prediction device of type is in the function of embodiment illustrated in fig. 8, and details are not described herein for this specification embodiment.

In short, the foregoing is merely the preferred embodiment of this specification, the protection of this specification is not intended to limit Scope.For all spirit in this specification with principle, any modification, equivalent replacement, improvement and so on, should be included in this Within the protection domain of specification.

System, device, module or the unit that above-described embodiment illustrates, can specifically be realized by computer chip or entity, Or realized by having the function of certain product.One kind typically realizes that equipment is computer.Specifically, computer for example may be used Think personal computer, laptop computer, cell phone, camera phone, smart phone, personal digital assistant, media play It is any in device, navigation equipment, electronic mail equipment, game console, tablet PC, wearable device or these equipment The combination of equipment.

Computer-readable medium includes permanent and non-permanent, removable and non-removable media can be by any method Or technology come realize information store.Information can be computer-readable instruction, data structure, the module of program or other data. The example of the storage medium of computer includes, but are not limited to phase transition internal memory (PRAM), static RAM (SRAM), moves State random access memory (DRAM), other kinds of random access memory (RAM), read-only storage (ROM), electric erasable Programmable read only memory (EEPROM), fast flash memory bank or other memory techniques, read-only optical disc read-only storage (CD-ROM), Digital versatile disc (DVD) or other optical storages, magnetic cassette tape, the storage of tape magnetic rigid disk or other magnetic storage apparatus Or any other non-transmission medium, the information that can be accessed by a computing device available for storage.Define, calculate according to herein Machine computer-readable recording medium does not include temporary computer readable media (transitory media), such as data-signal and carrier wave of modulation.

It should also be noted that, term " comprising ", "comprising" or its any other variant are intended to nonexcludability Comprising so that process, method, commodity or equipment including a series of elements not only include those key elements, but also wrapping Include other elements that are not explicitly listed, or further include for this process, method, commodity or equipment it is intrinsic will Element.In the absence of more restrictions, the key element limited by sentence "including a ...", it is not excluded that wanted including described Also there are other identical element in the process of element, method, commodity or equipment.

Each embodiment in this specification is described by the way of progressive, identical similar portion between each embodiment Divide mutually referring to what each embodiment stressed is the difference with other embodiment.It is real especially for system For applying example, since it is substantially similar to embodiment of the method, so description is fairly simple, related part is referring to embodiment of the method Part explanation.

Claims

1. a kind of model training method, the described method includes：

Training sample set is obtained, the training sample concentrates the training sample included for training pattern, the training sample bag Include user with the relevant data of specific behavior；

It is trained according to special algorithm to described with the relevant feature of specific behavior, obtains prediction model, the prediction model For the mapping relations established and between the relevant feature of specific behavior and the specific behavior information of forecasting of user.

2. the method as described in claim 1, described to include following at least one with the relevant data of specific behavior：Search note Record, purchaser record, collection record, browse record and application program number of clicks.

3. method as claimed in claim 2, described to include with the relevant data of specific behavior：Search record；

The extraction training sample concentrate training sample with the relevant feature of specific behavior, including：

Extraction and the relevant keyword of specific behavior from described search record；

4. method as claimed in claim 2, described to include with the relevant data of specific behavior：Purchaser record；

Extraction and the relevant Item Number of specific behavior and goods categories from the purchaser record；

Using the Item Number and goods categories as feature relevant with specific behavior or with the relevant feature of specific behavior it One.

5. method as claimed in claim 2, described to include with the relevant data of specific behavior：Collection record；

Extraction and the relevant Item Number of specific behavior and goods categories from the collection record；

6. method as claimed in claim 2, described to include with the relevant data of specific behavior：Browse record；

Extraction and the relevant Item Number of specific behavior and goods categories in record are browsed from described；

7. method as claimed in claim 2, described to include with the relevant data of specific behavior：Application program number of clicks；

According to the application program number of clicks, the accounting value with the relevant application program number of clicks of specific behavior is calculated；

One of using the accounting value as feature relevant with specific behavior or with the relevant feature of specific behavior.

It is 8. described to be further included with the relevant data of specific behavior such as claim 2-7 any one of them methods：User believes substantially Breath and user tag, the user tag are used to characterize whether the user is specific behavior fan；

The extraction training sample concentrate training sample with the relevant feature of specific behavior, further include：

User basic information feature is extracted from the user basic information, by the user basic information feature and user tag As one of with the relevant feature of specific behavior.

9. the method as described in claim 1, the special algorithm is logistic regression LR algorithm, and the prediction model is LR moulds Type.

10. the method as described in claim 1, the specific behavior is tourist behavior, the described and relevant data of specific behavior For with relevant data of going on a tour, it is described with the relevant relevant feature that is characterized as and goes on a tour of specific behavior.

11. a kind of user's behavior prediction method based on model, the described method includes：

Obtain targeted customer with the relevant data of specific behavior；

By the described and relevant feature input prediction model of specific behavior, output valve is obtained, the prediction model is using specific Algorithm is trained obtained model to training sample and the relevant feature of specific behavior, and the prediction model is used to establish Mapping relations between feature relevant with specific behavior and the specific behavior information of forecasting of user；

12. method as claimed in claim 11, described to include following at least one with the relevant data of specific behavior：Search note Record, purchaser record, collection record, browse record and application program number of clicks.

13. method as claimed in claim 12, the prediction model is according to search record, purchaser record, collection record, clear Look in record and application program number of clicks at least two constructed by model, the prediction model is used for whether predicting user There is the purpose for making specific behavior；

It is described from it is described with the relevant extracting data of specific behavior with the relevant feature of specific behavior, including：

Perform at least two in following characteristics extraction operation：Extraction and the relevant key of specific behavior from described search record Word, extraction and the relevant Item Number of specific behavior and goods categories, carry from the collection record from the purchaser record Relevant with specific behavior Item Number and goods categories are taken, extraction and the relevant article of specific behavior in record are browsed from described Numbering and goods categories, and calculate application program click time relevant with specific behavior according to the application program number of clicks Several accounting values；

Using the content that the feature extraction operation is extracted as with the relevant feature of specific behavior.

14. method as claimed in claim 13, described according to the output valve, the specific behavior of the targeted customer is predicted, Including：

If the output valve reaches preset first threshold value, predict that the targeted customer has the purpose for making specific behavior；

If the output valve is not up to the preset first threshold value, predict that the targeted customer does not make the meaning of specific behavior To.

15. method as claimed in claim 12, the prediction model includes：According to search record, purchaser record, collection note Record, browse in record and application program number of clicks at least two constructed by first kind prediction model, and multiple bases search Second class prediction model of Suo Jilu structures, the first kind prediction model make specific behavior for predicting whether user has Purpose, the second class prediction model are used for the purpose specific behavior for predicting user, and a second class prediction model corresponds to one Purpose specific behavior；

If searching for record involved in the building process of the first kind prediction model, perform in following characteristics extraction operation extremely Few one kind：Extraction and the relevant Item Number of specific behavior and goods categories from the purchaser record, from the collection record It is middle extraction with the relevant Item Number of specific behavior and goods categories, from it is described browse record in extraction it is relevant with specific behavior Item Number and goods categories, and calculated and the relevant application program point of specific behavior according to the application program number of clicks Hit the accounting value of number；

If being not directed to search record in the building process of the first kind prediction model, perform in following characteristics extraction operation At least two：Extraction and the relevant Item Number of specific behavior and goods categories from the purchaser record, from the collection note In record extraction with the relevant Item Number of specific behavior and goods categories, from it is described browse record in extraction it is related to specific behavior Item Number and goods categories, and according to the application program number of clicks calculate with the relevant application program of specific behavior The accounting value of number of clicks；

Using the content that the keyword and the feature extraction operation are extracted as with the relevant feature of specific behavior；

It is described to incite somebody to action the described and relevant feature input prediction model of specific behavior, output valve is obtained, including：

If the building process of the first kind prediction model is related to search record, the keyword and the feature extraction are grasped The content for making to extract inputs the first kind prediction model, obtains output valve；

If the building process of the first kind prediction model is not directed to search record, the feature extraction operation is extracted Content inputs the first kind prediction model, obtains output valve；

If the output valve is more than default second threshold, the keyword is inputted into the multiple second class respectively and predicts mould Type, obtains multiple output valves.

It is described according to the output valve 16. method as claimed in claim 15, predict the specific behavior of the targeted customer, Including：

By specific behavior corresponding to the output valve of the value ranking top N, it is determined as the purpose particular row of the targeted customer For N is natural number.

It is described obtaining targeted customer with the relevant data of specific behavior 17. method as claimed in claim 11, including：

18. method as claimed in claim 11, the special algorithm is LR algorithm, and the prediction model is LR models.

19. method as claimed in claim 11, the specific behavior is tourist behavior, the described and relevant data of specific behavior For with relevant data of going on a tour, it is described with the relevant relevant feature that is characterized as and goes on a tour of specific behavior.

20. a kind of model training apparatus, described device include：

First acquisition unit, for obtaining training sample set, the training sample concentrates the training sample included for training pattern This, the training sample include user with the relevant data of specific behavior；

First extraction unit, for extract that the training sample that the first acquisition unit gets concentrates training sample with it is specific The relevant feature of behavior；

Training unit, for first extraction unit is extracted according to special algorithm with the relevant feature of specific behavior into Row training, obtains prediction model, and the prediction model is used to establish the specific behavior with the relevant feature of specific behavior and user Mapping relations between information of forecasting.

21. a kind of user's behavior prediction device based on model, described device include：

Second extraction unit, for being got from the second acquisition unit and the relevant extracting data of specific behavior and spy Determine the relevant feature of behavior；

Processing unit, for second extraction unit is extracted with the relevant feature input prediction model of specific behavior, Output valve is obtained, the prediction model is that training sample and the relevant feature of specific behavior are trained using special algorithm Obtained model, the prediction model are used to establish with the relevant feature of specific behavior and predict letter to the specific behavior of user Mapping relations between breath；

Predicting unit, for the output valve obtained according to the processing unit processes, predicts the specific behavior of the targeted customer.

22. a kind of electronic equipment, including：

Processor；And

The memory of storage computer executable instructions is arranged to, the executable instruction makes the processor when executed Perform following operation：

23. a kind of electronic equipment, including：

Processor；And

Obtain targeted customer with the relevant data of specific behavior；

24. a kind of computer-readable storage medium, the computer-readable recording medium storage one or more program, it is one or Multiple programs by the electronic equipment including multiple application programs when being performed so that the electronic equipment performs following operation：

25. a kind of computer-readable storage medium, the computer-readable recording medium storage one or more program, it is one or Multiple programs by the electronic equipment including multiple application programs when being performed so that the electronic equipment performs following operation：

Obtain targeted customer with the relevant data of specific behavior；