CN111008335A

CN111008335A - Information processing method, device, equipment and storage medium

Info

Publication number: CN111008335A
Application number: CN201911333011.0A
Authority: CN
Inventors: 胡乐
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2019-12-20
Filing date: 2019-12-20
Publication date: 2020-04-14
Anticipated expiration: 2039-12-20
Also published as: CN111008335B

Abstract

The embodiment of the application discloses an information processing method, an information processing device, information processing equipment and a storage medium, wherein the method comprises the following steps: acquiring feature information of a target user, wherein the feature information comprises feature information of at least one first resource associated with the target user; respectively determining the weight of the characteristic information of the at least one first resource under a plurality of resource categories; weighting the weight corresponding to the at least one first resource to obtain the weighted weight corresponding to the at least one first resource; and determining the push information of the target user according to the weighting weight corresponding to the at least one first resource. By the aid of the method and the device, reliability of the determined user push information is improved.

Description

Information processing method, device, equipment and storage medium

Technical Field

The present application relates to the field of computers, and in particular, to an information processing method, apparatus, device, and storage medium.

Background

With the continuous development of internet technology, more and more network users are provided, and there are often situations that resources need to be pushed to users. The inventor finds that the situation that the pushed resources are unreliable exists when the resources are pushed at present, and particularly when the behavior characteristics of the user are not obviously biased, the reliability of the determined pushed resources is poor.

Disclosure of Invention

The embodiment of the application provides an information processing method, device, equipment and storage medium, which are beneficial to improving the reliability of the determined user push information.

In one aspect, an embodiment of the present application provides an information processing method, including:

acquiring feature information of a target user, wherein the feature information comprises feature information of at least one first resource associated with the target user;

respectively determining the weight of the characteristic information of the at least one first resource under a plurality of resource categories;

weighting the weight corresponding to the at least one first resource to obtain the weighted weight corresponding to the at least one first resource;

and determining the push information of the target user according to the weighting weight corresponding to the at least one first resource.

In another aspect, an embodiment of the present application provides an information processing apparatus, including:

an obtaining unit, configured to obtain feature information of a target user, where the feature information includes feature information of at least one first resource associated with the target user;

the processing unit is used for respectively determining the weight of the characteristic information of the at least one first resource under a plurality of resource categories;

the processing unit is further configured to perform weighting processing on the weight corresponding to the at least one first resource to obtain a weighted weight corresponding to the at least one first resource;

the processing unit is further configured to determine, according to the weighting corresponding to the at least one first resource, push information for the target user.

a processor adapted to implement one or more instructions; and the number of the first and second groups,

a computer (readable) storage medium having stored thereon one or more instructions adapted to be loaded by the processor and to carry out the method described above.

In a possible design, the information processing device may further include an input device and an output device.

In yet another aspect, an embodiment of the present application provides a computer-readable storage medium, which stores a computer program, where the computer program includes program instructions, and the program instructions, when executed by a processor, cause the processor to execute the above-mentioned method.

According to the embodiment of the application, the weight of the feature information of the associated resource under a plurality of resource categories can be determined by acquiring the feature information of the resource associated with the target user, the weight corresponding to the associated resource is weighted, and then the pushing information of the target user is determined according to the weighted weight corresponding to the associated resource after the weighting processing, so that the accuracy of the determined pushing information of the user is improved, and the reliability is high.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings in the description of the embodiments will be briefly introduced below.

Fig. 1 is a schematic diagram of an information acquisition method provided in an embodiment of the present application;

FIG. 2 is a schematic structural diagram of a double tower model provided in an embodiment of the present application;

fig. 3 is a schematic view of a block chain network according to an embodiment of the present disclosure;

fig. 4 is a schematic flowchart of an information processing method provided in an embodiment of the present application;

FIG. 5 is a schematic flow chart diagram of another information processing method provided in the embodiments of the present application;

FIG. 6a is a schematic diagram of a manner of obtaining weights of features provided by an embodiment of the present application;

FIG. 6b is a schematic diagram illustrating a manner of obtaining weighting weights according to an embodiment of the present disclosure;

fig. 6c is a schematic diagram of a weight splicing manner provided in the embodiment of the present application;

fig. 7 is a schematic structural diagram of an information processing apparatus according to an embodiment of the present application;

fig. 8 is a schematic structural diagram of an information processing apparatus according to an embodiment of the present application.

Detailed Description

With the continuous development of internet technology, there are many scenarios in which resources need to be pushed to users to realize reliable pushing of resources. At present, the method for pushing resources is determined to have poor pushing effect, targeted pushing cannot be performed on personalized users, and the pushing reliability is low. The method and the device can use the attention mechanism for pushing the resources, for example, by obtaining the characteristic information of the resources associated with the user, for example, when a user request is received, obtaining the characteristic information corresponding to the user, further determining the weights of the characteristic information of the associated resources under a plurality of resource categories, performing weighting processing, and then determining the pushing information of the user according to the weighted weights after the weighting processing, so that different weights can be given to the resources of different categories (classification, category) according to the characteristics of the user, thereby enabling the pushing information of the user to be more personalized and emphasized, and being helpful for improving the reliability of the determined pushing information of the user. Optionally, the resource may be a commodity, an advertisement, other information, and the like, and the application is not limited.

Among them, the attention mechanism is also called attention mechanism. When performing the feature processing, for example, in the processing process of the multi-valued features such as the features of the resources viewed (browsed or clicked) by the user, the push result is lost by directly adding or averaging. Therefore, the method and the device can combine with an attention mechanism and adopt a weighting method to perform feature processing so as to realize reliable pushing of resources.

For example, taking resources as goods, from the perspective of the final result of recommendation, when the behavior characteristics of the user have no obvious bias (such as browsing a mobile phone and beer), the final recommended goods may not be related to both the mobile phone and the beer without attention, and the reliability is poor; and by adopting the attention method, the recommended result can be concentrated on two blocks of mobile phones and wines, and the reliability is higher.

It is to be understood that the technical solution of the present application is applicable to an information processing apparatus or an information processing device, and the information processing device may be disposed in the information processing apparatus. Optionally, the information processing apparatus may be a server, a terminal, or other apparatuses, and the information processing apparatus may be a third party platform such as an advertisement platform, an advertisement system, and the like, which is not limited in this application. The following description will be given taking an information processing apparatus as an example.

For example, please refer to fig. 1, which is a schematic diagram of an information obtaining method provided in an embodiment of the present application. As shown in fig. 1, for example, when an advertiser puts a relarketingdpa advertisement, the advertiser may send back user behavior data of an advertiser site (e.g., a shopping website, a platform, etc.) through feedback, and after obtaining the user behavior data, the advertisement system may recommend a product to the user according to the user behavior data, i.e., historical behaviors, in combination with an attention mechanism solution provided by the product. Optionally, when the user selects a "similar recommendation" mode (i.e., reproducing), the technical scheme of the present application may be adopted to obtain the recommendation result, or the original recommendation result may be optimized to obtain the recommendation result. In some embodiments, the advertiser may send the user behavior Data of the advertiser's own website to the advertising system side through an interface of a Data Management Platform (DMP), and the advertising system receives the user behavior Data. Alternatively, the advertisement system may maintain a list of behavior records for each user, as shown in FIG. 1. For example, user behavior data may be stored to a key-value system. The remarkering refers to re-marketing recommendation, such as a method of recommending goods based on user behavior data of an advertiser. The DPA (Dynamic Product Ads) means that the advertisement can be selected by an algorithm according to the preference and characteristics of the user, and the most suitable Product is assembled with an advertisement template to form an advertisement, so that the effect of thousands of people is achieved.

For example, when an online request occurs, such as a request by user 3, the history list of user 3 may be queried. Based on the history list of the user 3The system recommends a list of recommended goods for the user 3. Because the system uses the attention mechanism, each commodity in the recommendation result has a bias towards a part of commodities in the original behavior of the user. As an example, for example, the article a in the figure may be associated with an article₃₁Similar, belonging to 1 category; the article b may be related to the article₃₂More closely, belong to another category.

In some embodiments, the present application may determine user push information by combining a double-tower model, that is, an attention mechanism may be applied to the double-tower model to implement resource push. The double-tower model can change a fully-connected network into two independent fully-connected networks by adapting the deep neural network to large-scale data, and connects the two networks at the uppermost layer, thereby realizing the requirement of large-scale data processing.

For example, please refer to fig. 2, which is a schematic structural diagram of a double tower model provided in the embodiment of the present application. As shown in fig. 2, the entire model is divided into two sides (one for each tower). The left tower represents a model of a user, which is called a user tower, the lowest layer of the user tower is user features (such as browsed commodities, sex, age, city, purchased commodities and the like), each feature becomes a vector after passing through an embedding layer (vector layer), all feature vectors can be spliced together, and then all information of the user can be converted into a vector through a full-connection network with 3 layers (other structures can also be used as well), such as a 32-dimensional vector. The right tower represents information of resources such as commodities, and can be called a target tower, the lowest layer is commodity characteristics (such as the merchants, categories, names, brands and the like of the commodities), and similarly, the commodities are converted into a vector through a series of conversion, such as a 32-dimensional vector. And then calculating an inner product through the vector of the user and the vector of the commodity, wherein the obtained value is the score of the commodity for the user, the larger the score is, the higher the recommendation degree of the commodity is, and the pushing information for the user is determined based on the score.

The current attribute technology is mainly applied to the Natural Language Processing (NLP) field, and is more applied to the ranking stage in the recommendation field. In the sequencing stage, the candidate range is small, only dozens of resources such as commodities are usually available, and the calculation time is not sensitive, so that the full-connection network containing the attention structure can be estimated for each resource and user, and the obtained results are sequenced. However, in the recommendation field, in the resource triggering stage, especially when the resource candidate range is large, tens of millions of resources exist, if the resources are processed according to the same full-connection network method, calculation of a full-connection network needs to be performed on each resource, each resource needs up to millions of times of arithmetic operation, and time consumption is large. In order to further reduce time consumption and operation complexity, an attention mechanism and a double-tower model can be combined, and a resource vector can be calculated for resources of ten million levels by using the model in advance. In the actual resource recommendation process, the pushed information of the user can be determined according to the user vector and the resource vector by calculating the user vector once, for example, the pushed resource can be determined by searching in the resource vector based on the user vector. In this way, the average calculation amount on each resource is less than 1 arithmetic operation, so that the calculation amount is greatly reduced, the calculation complexity is reduced, the retrieval speed is increased, and the time consumption is reduced.

Optionally, when the goods need to be triggered, for example, the goods are triggered according to the received request, a retrieval algorithm, such as a neighbor retrieval algorithm or other retrieval algorithms, may be used to determine the pushed goods. Wherein, the neighbor retrieval may refer to that given a vector, in the vector set of the large data set, k vectors with the closest cosine distance to the vector are searched. It can be appreciated that the method of traversing the cosine distances of each vector of the set may be time consuming, and therefore, other search algorithms may also be employed to return more reliable search results in a shorter time, which is not limited in this application.

By combining the attention mechanism and the double-tower model, the model can automatically learn a plurality of classifications, different weights are given to different types of resources according to user characteristics, and the attention mechanism can be used in tens of millions of resource retrieval, so that the reliability of resource pushing is improved, the operation complexity is reduced, and the time consumption is reduced.

It should be understood that the structure of the double tower model shown in fig. 2 is only an example, and is not intended to limit the structure of the double tower model, and the double tower model related to the present application may also be other structures, and the present application is not limited thereto.

Optionally, the information processing apparatus may be a blockchain node in a blockchain network, or may be an apparatus independent of all blockchain nodes in the blockchain network, which is not limited in this application. It is understood that in other embodiments, the block link points may be referred to by other names, such as block link point devices, nodes, and the like, but are not limited thereto.

In some embodiments, after obtaining user information (e.g., feature information, a weight corresponding to an associated resource, a weighted weight, etc.) of a user, such as a target user, the information processing apparatus may further uplink the user information, so as to avoid the user information from being tampered by an illegal user, ensure the authenticity and reliability of the user information, and facilitate a blockchain node in a blockchain network to process the user based on the user information, such as determining a user tag, quickly determining push information, and the like. For example, please refer to fig. 3, which is an architecture diagram of a blockchain network according to an embodiment of the present application. One or more block chain nodes may be included in the block chain network, which shows 3 block link points including the information processing apparatus, block link point 1, and block link point 2, as shown in fig. 3, as an example. It is to be understood that, in other alternative embodiments, the information processing apparatus may be an apparatus other than a blockchain network, so that the information processing apparatus may perform uplink processing on the user information and the like of the target user by uploading the user information and the like of the target user to a blockchain node in the blockchain network, where the uplink processing includes generating a block corresponding to the user information and distributing the block to the blockchain network and the like, and details of the processing are not described herein.

In some embodiments, the feature information of the user may be information stored in the blockchain network, that is, the feature information of a plurality of users may be stored in the blockchain network. The information processing device can further acquire the feature information of the target user from the blockchain network to determine the push information of the target user based on the feature information, so that the reliability of the acquired feature information can be improved, and the reliability of the determined push information can be further improved.

In some embodiments, the information processing apparatus may further determine a user tag corresponding to the user according to the user information, and may further perform uplink processing on the user tag. So that the block chain node in the block chain network pushes information or performs other processing on the user based on the user label.

For example, taking the user information of the user is uplink, and the information processing apparatus is an apparatus independent of all the blockchain nodes in the blockchain network as an example, the process of the information processing apparatus uplink the user information may be: the information processing device sends the user information to a target block chain node in the block chain network, the target block chain node generates a block according to the user information, the block comprises the user information, and the target block chain node can issue the block to the block chain network. The target blk may be any blk node in the blk network, or may be a designated node in the blk network, such as a blk associated with the user, and so on, which is not limited in this application.

The blockchain is a novel application mode of computer technologies such as distributed data storage, point-to-point transmission, a consensus mechanism and an encryption algorithm. A Block chain (Block chain), which is essentially a decentralized database, is a series of data blocks associated by using a cryptographic method, and each data Block contains information of a batch of network transactions, so as to verify the validity (anti-counterfeiting) of the information and generate a next Block. The blockchain may include a blockchain underlying platform, a platform product services layer, and an application services layer.

The block chain underlying platform can comprise processing modules such as user management, basic service, intelligent contract and operation monitoring. The user management module is responsible for identity information management of all blockchain participants, and comprises public and private key generation maintenance (account management), key management, user real identity and blockchain address corresponding relation maintenance (authority management) and the like, and under the authorization condition, the user management module supervises and audits the transaction condition of certain real identities and provides rule configuration (wind control audit) of risk control; the basic service module is deployed on all block link point devices and used for verifying the validity of the service request, recording the valid request after consensus is completed on storage, for a new service request, the basic service firstly performs interface adaptation analysis and authentication processing (interface adaptation), then encrypts service information (consensus management) through a consensus algorithm, transmits the encrypted service information to a shared account (network communication) completely and consistently, and performs recording and storage; the intelligent contract module is responsible for registering and issuing contracts, triggering the contracts and executing the contracts, developers can define contract logics through a certain programming language, issue the contract logics to a block chain (contract registration), call keys or other event triggering and executing according to the logics of contract clauses, complete the contract logics and simultaneously provide the function of upgrading and canceling the contracts; the operation monitoring module is mainly responsible for deployment, configuration modification, contract setting, cloud adaptation in the product release process and visual output of real-time states in product operation, such as: alarms, monitoring network conditions, monitoring node device (e.g., blockchain node) health status, and the like.

The platform product service layer provides basic capability and an implementation framework of typical application, and developers can complete block chain implementation of business logic based on the basic capability and the characteristics of the superposed business. The application service layer provides the application service based on the block chain scheme for the business participants to use.

The embodiment of the application discloses an information processing method, an information processing device, information processing equipment and a storage medium, which are beneficial to improving the reliability of determined user push information. The details are described below.

Referring to fig. 4, fig. 4 is a schematic flowchart of an information processing method according to an embodiment of the present application. As shown in fig. 4, the information processing method of the present embodiment includes the steps of:

401. feature information of a target user is obtained, wherein the feature information comprises feature information of at least one first resource associated with the target user.

In some embodiments, the target user may be a user that triggers resource pushing, or may be a user that initiates a resource request, or may be any user, and the like, which is not limited in this application. Wherein triggering may refer to retrieving resources, such as advertisements or goods, etc., that satisfy a condition (e.g., a targeting condition, etc.) on a per request basis (e.g., as determined by a user, an ad slot, etc.).

The characteristic information of the at least one (one or more) first resources associated with the target user may refer to the characteristic information of the resources viewed by the target user, that is, the first resources. Optionally, the feature information may include behavior data of the target user, or may include a feature vector corresponding to the behavior data of the target user, or may also be other forms of information for indicating a feature on the user side. Further optionally, the first resource may be a resource within a time range, and/or may be a resource of an ordering website, a platform, and/or may be a resource of a designated price interval, and/or may be a resource of a designated region, and the like, which is not limited in this application.

In other optional embodiments, the characteristic information of the at least one first resource associated with the target user may also be characteristic information of a resource purchased by the target user, that is, the first resource.

402. The weight of the characteristic information of the at least one first resource under a plurality of resource categories is respectively determined.

In some embodiments, the characteristic information of the target user may include behavior data of the target user. Furthermore, when determining the weights of the feature information of the at least one first resource under the multiple resource categories, the information processing apparatus may obtain, after embedding (vectorizing) the feature vector corresponding to the first resource (or referred to as the feature vector corresponding to the target user), a plurality of feature vectors for the target user behavior data, such as the feature of the historically viewed product of the user, and may obtain the feature vectors corresponding to the multiple resource categories. Further, the weights of the feature information of each first resource under the plurality of resource categories, or the weights of each first resource under the plurality of resource categories, may be determined based on the feature vector corresponding to the first resource and the feature vectors corresponding to the plurality of resource categories, so as to obtain the plurality of weights. Alternatively, the plurality of weights may be stored in a matrix form.

In some embodiments, the feature information of the at least one first resource is a feature vector, that is, the feature information of the target user may include a feature vector corresponding to target user behavior data (referred to as a feature vector corresponding to the target user), for example, the feature vector corresponding to the target user may be obtained from stored user information, as if the feature vector is calculated in real time through the above-mentioned manner. The information processing apparatus may further determine, by obtaining the feature vectors corresponding to the plurality of resource categories, a weight of the feature information of the at least one first resource under the plurality of resource categories according to the feature vector of the at least one first resource and the feature vectors corresponding to the plurality of resource categories.

403. And performing weighting processing on the weight corresponding to the at least one first resource to obtain the weighting weight corresponding to the at least one first resource.

After determining the weight of the characteristic information of the at least one first resource under the plurality of resource categories, the weight can be weighted. For example, weighting each weight according to the feature vector corresponding to the target user to obtain a weighted weight; for another example, the weights may be weighted according to the feature vector corresponding to the target user and the viewing duration of each first resource, so as to obtain weighted weights, and the like. Alternatively, the plurality of weighting weights may be stored in a matrix form, such as a weighting vector matrix.

404. And determining the push information of the target user according to the weighting weight corresponding to the at least one first resource.

In some embodiments, when determining the push information for the target user according to the weighting weight, the weighting weight may be input into a pre-trained recommendation model to determine the push information for the target user. Optionally, the recommendation model may be obtained by training in advance according to weighting weights corresponding to a plurality of users and push information thereof.

In some embodiments, after the weighting weight is obtained, the weighting weight of the feature information of each first resource under the multiple resource categories may be further spliced to obtain a weight vector corresponding to each first resource, and then the push information for the target user may be determined according to the weight vector corresponding to the at least one first resource. For example, the weight vector corresponding to the first resource is input into a pre-trained recommendation model to determine push information for the target user, and the recommendation model may be obtained by pre-training according to the weight vectors corresponding to a plurality of users and the push information thereof; in another example, push information for the target user may be determined based on the weight vector in combination with a two-tower model.

Optionally, the feature information further includes viewing information of the at least one first resource, and the viewing information includes viewing order and/or viewing time information of the at least one first resource. The viewing time information may include a viewing time point, a viewing time period, and/or a viewing duration, among others. When the information processing device performs the splicing processing, the information processing device may further perform the splicing processing on the weighted weights of the feature information of each first resource under the multiple resource categories according to the viewing order and/or the viewing time information of the at least one first resource, so as to obtain a weight vector corresponding to each first resource. Therefore, the reliability of the determined push information is further improved, and the recommendation effect is improved.

In some embodiments, after determining push information, such as a push resource, for the target user, an interface including the push resource may also be generated and presented to the target user.

In this embodiment, the information processing apparatus can determine, by acquiring the feature information of the resource associated with the target user, the weight of the feature information of the associated resource in the plurality of resource categories, perform weighting processing on the weight corresponding to the associated resource, and then determine the push information for the target user according to the weighted weight after the weighting processing corresponding to the associated resource, which is beneficial to improving the accuracy of the determined user push information and has high reliability.

Please refer to fig. 5, which is a flowchart illustrating another information processing method according to an embodiment of the present application, wherein the method according to the present application can be applied to a double-tower model, in which an attention mechanism is added. In this embodiment, the method may be divided into several parts, i.e., writing off-line user behavior data, preparing training data, training a double-tower model, and performing on-line retrieval. As shown in fig. 5, the information processing method of the present embodiment may include the steps of:

501. feature information of a plurality of users is obtained, and the feature information of each user comprises resource features of at least one third resource associated with the user.

Wherein the resource characteristics of the at least one (one or more) third resources associated with the user may refer to user behavior data such as characteristics of resources that have been viewed (browsed), i.e. historical browsing records. In some embodiments, the characteristic information of the user may also be characteristic information of at least one third resource associated with the user. Optionally, the user behavior data may be written into the advertisement system through the advertiser calling interface, so that the subsequent service may query the user behavior through the user id and the like, which may specifically refer to the above description, and is not described herein again.

Thereby enabling user behavior data writing.

502. And taking the resource characteristics of a fourth resource in the plurality of pushed resources as a positive sample, taking the resource characteristics of the pushed resources except the fourth resource in the plurality of pushed resources as a negative sample, taking the resource characteristics of the at least one third resource as the characteristics of the user tower, taking the resource characteristics of the plurality of pushed resources as the characteristics of the target tower, and training the double-tower model.

Wherein the plurality of pushed resources may be resources, such as exposed advertising items, pushed to the user. The fourth resource may refer to a pushed resource clicked by the user among the plurality of pushed resources.

For example, an advertising system may have hundreds of millions of exposures per day. The exposure data can be used as training data recommended by the commodity, and modeling is carried out by taking user clicks as targets. Specifically, the record of resource generation such as advertisement click can be used as a training positive sample, the record of exposure without click can be used as a negative sample, the exposed commodity (i.e. the multiple pushed resources) can be used as the commodity feature on the right side of the double-tower model, i.e. the feature of the target tower, and the history browsing record of the user can be used as a feature of the user, i.e. the feature of the user tower. A large amount of training data is thus available.

After the training data is obtained, the two-tower model may be trained based on the obtained training data.

503. Feature information of a target user is obtained, wherein the feature information comprises feature information of at least one first resource associated with the target user.

The feature information of the first resource may include a feature vector after vectorization of the feature of the first resource by a user tower of the double-tower model, such as the feature vector after embedding.

504. The weight of the characteristic information of the at least one first resource under a plurality of resource categories is respectively determined.

The resource categories may be obtained by random initialization, or may be obtained by clustering a plurality of resources, that is, by clustering a large number of existing resources.

For example, assuming that a user has 3 historical browsing behaviors, browses product 1, product 2, and product 3, respectively, and virtualizes 10 resource categories, feature vectors of 10 resource categories may be constructed first and randomly initialized. As shown in fig. 6a, on the user side, after being subjected to vectorization, the historical viewed product features of the user are changed into a plurality of feature vectors (i.e., viewed product 1, viewed product 2, and viewed product 3 in the figure), and then each feature vector can be subjected to inner product calculation with feature vectors corresponding to 10 resource categories, and each feature obtains 10 inner product results, the value of each inner product result is an attribute weight, that is, the weight matrix in fig. 6a can represent the weight (value) of each feature (product) on each virtual category.

505. And performing weighting processing on the weight corresponding to the at least one first resource to obtain the weighting weight corresponding to the at least one first resource.

Further, after obtaining the weight matrix, the weight matrix may be multiplied by each commodity (each weight coefficient is multiplied by the eigenvector, non-matrix multiplication), that is, the weight matrix is weighted according to the eigenvector corresponding to the commodity, so as to obtain the weight vector matrix, as shown in fig. 6 b. Through the weighting process, 3 vectors originally representing 3 commodities become 3 × 10 vectors, wherein 10 vectors in each row are mappings of 1 commodity on different categories.

506. And splicing the weighted weights of the characteristic information of each first resource under the multiple resource categories to obtain a weight vector corresponding to each first resource.

After the weighted vector matrix is obtained, 10 vectors of each row can be spliced to obtain 3 commodity vectors again, as shown in fig. 6 c. At this time, the dimension of each commodity vector becomes 10 times the original dimension.

Optionally, when the splicing processing is performed, the splicing processing may be performed according to the viewing sequence and/or the viewing time information, that is, according to the commodity browsing sequence, the browsing time point, the browsing time period, the browsing duration, and the like, so as to further improve the reliability of the determined push information and improve the recommendation effect. For example, the information processing apparatus may splice the weighted weights of the feature information of each first resource under the multiple resource categories according to a viewing order of the at least one first resource, such as a commodity browsing order, for example, a browsing order from first to last, so as to obtain a weight vector corresponding to each first resource. For another example, the information processing apparatus may splice the weighted weights of the feature information of each first resource under the multiple resource categories according to the viewing time information of the at least one first resource, such as the browsing time point, for example, according to the order from front to back of the browsing time point, so as to obtain a weight vector corresponding to each first resource; or according to the viewing time information of the at least one first resource, such as the browsing duration, for example, according to the sequence of the browsing duration from long to short, splicing the weighted weights of the feature information of each first resource under the multiple resource categories to obtain a weight vector corresponding to each first resource; or determining corresponding browsing duration according to the viewing time information of the at least one first resource, such as browsing time period, splicing the weighted weights of the feature information of each first resource under the multiple resource categories according to the browsing duration from long to short to obtain a weight vector corresponding to each first resource, or determining a corresponding browsing time point according to the browsing time period to determine a weight vector corresponding to each first resource, and so on. For another example, the information processing apparatus may splice the weighted weights of the feature information of each first resource under the multiple resource categories according to the viewing order and the viewing time information of the at least one first resource, so as to obtain a weight vector corresponding to each first resource.

507. And determining push information to the target user according to the weight vector corresponding to the at least one first resource and the weight vectors of the plurality of second resources corresponding to the target tower.

Optionally, the second resource may refer to a resource to be pushed, or may refer to a database, such as a resource in a designated database, and the like, which is not limited in this application. In some embodiments, the push information may include a push resource, and the determined push resource may be a portion of the second resource.

Wherein, the weight vectors of the plurality of second resources can be obtained and stored in advance. When the weight vector of the second resource is obtained, the feature information of the second resource may be obtained, and the weight of the feature information of the second resource under a plurality of resource categories is determined; weighting the weight of the characteristic information of the second resource under a plurality of resource categories to obtain the weighted weight of the characteristic information of the second resource under the plurality of resource categories; and splicing the weighted weights of the characteristic information of the second resource under the multiple resource categories to obtain a weight vector corresponding to the second resource. It can be understood that the calculation manner of the resource weight vector corresponding to the commodity side is similar to the calculation manner of the resource weight vector at the user side, and details are not described here. That is, the same attention calculation is performed also on the product side, and since the product side is a fixed product, the product side is multiplied by a weight, and therefore, the summation is not necessary.

Optionally, the plurality of resource categories corresponding to the first resource and the plurality of resource categories corresponding to the second resource may be the same, that is, the plurality of same resource categories are respectively virtualized on the user side and the commodity side, which is helpful for the information processing apparatus to better understand the relationship between the commodities, and further improves the reliability of the determined pushed commodities. The resource categories being the same may mean that the feature vectors corresponding to the resource categories are the same, for example, the vector of each category may be an average value of the vectors of the embedding layers of the commodities under the category.

For all commodities in the advertisement system, the embedding vector on the right side of the double tower, such as a 32-dimensional vector of the target tower, that is, a commodity side vector, can be obtained through model calculation. And then all the commodity embedding vectors can be packed into a set for the retrieval of an online system. It can be understood that the manner of combining the attribute mechanism during the two-tower model training is similar to that of not performing 503-507, for example, the relevant description of the first resource may be referred to for the feature processing of the third resource, and the relevant description of the second resource may be referred to for the feature processing of the push resource, and details are not repeated. Optionally, after the double-tower model is obtained through training, the embedding vector of each resource, that is, the commodity side vector, may be stored.

Therefore, the embedding feature of historical behavior data of any user can be obtained, the feature is spliced with other user portrait features to obtain a weight vector which is used as an input layer of a user side of the double-tower model, and a 32-dimensional vector, namely a user side vector, can be obtained through network (such as a full-connection network) calculation. And then, searching can be carried out in each commodity side vector of the set according to the user side vector so as to determine the pushed commodity. For example, offline evaluation may be implemented based on the stored commodity side vectors.

With the commodity recommendation based on the attention mechanism, the AUC (area Under cutter) effect is improved, and the consumption of DPA is promoted. For example, the AUC effect of model off-line evaluation can be improved by 2%, and the consumption of DPA per day is improved by 5%.

In the embodiment of the application, an attention mechanism can be modified on the basis of a double-tower model, a plurality of resource types are virtualized on a commodity side and a user side respectively, in the process of calculating two towers, attention operation is carried out on the characteristics needing attention operation and 10 virtual types respectively, the result of attention is weight, original multi-valued characteristic processing is carried out according to the weight, and subsequent operations such as full connection and the like are carried out continuously, so that the attention mechanism can be approximately applied to the double-tower model, the estimation of million-level commodities can be returned by the attention mechanism within millisecond-level time, the reliability of commodity pushing is improved, the time consumption of retrieval is reduced, and the operation complexity is reduced.

It is to be understood that the above method embodiments are all illustrations of the information processing method of the present application, and descriptions of various embodiments have respective emphasis, and reference may be made to relevant descriptions of other embodiments for parts that are not described in detail in a certain embodiment.

Based on the description of the information processing method embodiment, the embodiment of the invention also discloses an information processing device. Alternatively, the information processing apparatus may be a computer program (including program code/program instructions) running in the information processing device. For example, the information processing apparatus may execute the methods shown in fig. 4 and 5. Referring to fig. 7, the information processing apparatus 700 may operate as follows:

an obtaining unit 701, configured to obtain feature information of a target user, where the feature information includes feature information of at least one first resource associated with the target user;

a processing unit 702, configured to determine weights of the feature information of the at least one first resource under multiple resource categories, respectively;

the processing unit 702 is further configured to perform weighting processing on the weight corresponding to the at least one first resource to obtain a weighted weight corresponding to the at least one first resource;

the processing unit 702 is further configured to determine, according to the weighting corresponding to the at least one first resource, push information for the target user.

In an optional implementation manner, when determining, according to the weighting corresponding to the at least one first resource, the push information for the target user, the processing unit 702 is specifically configured to:

splicing the weighted weights of the characteristic information of each first resource under the multiple resource categories to obtain a weight vector corresponding to each first resource;

and determining push information of the target user according to the weight vector corresponding to the at least one first resource.

In an optional embodiment, the feature information of the first resource includes a feature vector after vectorization of the feature of the first resource by a user tower of a double-tower model, the double-tower model includes the user tower and a target tower, and the push information includes a push resource;

when determining the push information for the target user according to the weight vector corresponding to the at least one first resource, the processing unit 702 is specifically configured to:

and determining the push resources to the target user according to the weight vector corresponding to the at least one first resource and the weight vectors of the plurality of second resources corresponding to the target tower.

In an optional embodiment, the characteristic information further includes viewing information of the at least one first resource, and the viewing information includes viewing sequence and/or viewing time information of the at least one first resource;

the processing unit 702 is specifically configured to, when the processing unit splices the weighted weights of the feature information of each first resource in the multiple resource categories to obtain a weight vector corresponding to each first resource:

and according to the viewing sequence and/or viewing time information of the at least one first resource, splicing the weighted weights of the characteristic information of each first resource under the multiple resource categories to obtain a weight vector corresponding to each first resource.

In an optional implementation manner, the obtaining unit 701 may be further configured to obtain feature information of the second resource;

the processing unit 702 is further configured to determine a weight of the feature information of the second resource under multiple resource categories;

the processing unit 702 is further configured to perform weighting processing on the weights of the feature information of the second resource in multiple resource categories, so as to obtain weighted weights of the feature information of the second resource in multiple resource categories;

the processing unit 702 is further configured to perform splicing processing on the weighted weights of the feature information of the second resource in the multiple resource categories to obtain a weight vector corresponding to the second resource.

In an optional implementation manner, the obtaining unit 701 may be further configured to obtain feature information of a plurality of users, where the feature information of each user includes a resource feature of at least one third resource associated with the user;

the processing unit 702 is further configured to train the double-tower model by taking a resource feature of a fourth resource in the multiple pushed resources as a positive sample, taking a resource feature of a pushed resource other than the fourth resource in the multiple pushed resources as a negative sample, taking a resource feature of the at least one third resource as a feature of the user tower, and taking a resource feature of the multiple pushed resources as a feature of the target tower; the fourth resource is a pushed resource clicked by a user in the plurality of pushed resources.

In an optional embodiment, the feature information of the at least one first resource is a feature vector; the processing unit 702, when determining the weights of the feature information of the at least one first resource under the multiple resource categories, may specifically be configured to:

acquiring feature vectors corresponding to the plurality of resource categories;

and determining the weight of the characteristic information of the at least one first resource under the multiple resource categories according to the characteristic vector of the at least one first resource and the characteristic vectors corresponding to the multiple resource categories.

According to an embodiment of the present application, each step involved in the methods shown in fig. 4 and 5 may be executed by each unit in the information processing apparatus shown in fig. 7. For example, step 401 shown in fig. 4 may be performed by the obtaining unit 701 shown in fig. 7, and step 402 and step 404 may be performed by the processing unit 702 shown in fig. 7; for another example, the

steps

501 and 503 shown in fig. 5 may be performed by the obtaining unit 701 shown in fig. 7, the

steps

502 and 504 and 507 may be performed by the processing unit 702 shown in fig. 7, and so on, which are not described herein again.

According to another embodiment of the present application, the units in the information processing apparatus shown in fig. 7 may be respectively or entirely combined into one or several other units to form a structure, for example, the obtaining unit and the processing unit may be combined into a processing unit, or some unit(s) therein may be further split into multiple functionally smaller units to form a structure, which may achieve the same operation without affecting the achievement of the technical effect of the embodiment of the present application. The units are divided based on logic functions, and in practical application, the functions of one unit can be realized by a plurality of units, or the functions of a plurality of units can be realized by one unit. In other embodiments of the present application, the information processing apparatus may also include other units, and in practical applications, these functions may also be implemented by being assisted by other units, and may be implemented by cooperation of multiple units, which is not limited in this application.

According to another embodiment of the present application, the information processing apparatus shown in fig. 7 may be constructed by running a computer program (including program codes/program instructions) capable of executing the steps involved in the respective methods shown in fig. 4 and 5 on a general-purpose computing device such as a computer including a processing element such as a Central Processing Unit (CPU), a random access storage medium (RAM), a read-only storage medium (ROM), and a storage element, and implementing the information processing method of the embodiment of the present application. The computer program may be recorded on a computer-readable recording medium, for example, and loaded and executed in the above-described computing apparatus via the computer-readable recording medium.

Based on the description of the method embodiment and the device embodiment, the embodiment of the application also provides an information processing device. Referring to fig. 8, the information processing apparatus includes at least a processor 801 and a computer (readable) storage medium 802. Optionally, the information processing apparatus may further include an input device 803 and an output device 804. The processor 801, the input device 803, the output device 804, and the computer storage medium 802 in the information processing apparatus may be connected by a bus or other means.

A computer storage medium 802 may be stored in the memory of the information processing apparatus, the computer storage medium 802 being used to store a computer program comprising program instructions, the processor 801 being used to execute the program instructions stored by the computer storage medium 802. The processor 801 (or CPU) is a computing core and a control core of the information Processing apparatus, and is adapted to implement one or more instructions, and specifically, adapted to load and execute the one or more instructions so as to implement a corresponding method flow or a corresponding function; in one embodiment, the processor 801 according to the embodiment of the present application may be configured to perform a series of information processing processes, including the following steps: acquiring feature information of a target user, wherein the feature information comprises feature information of at least one first resource associated with the target user; respectively determining the weight of the characteristic information of the at least one first resource under a plurality of resource categories; weighting the weight corresponding to the at least one first resource to obtain the weighted weight corresponding to the at least one first resource; and determining the push information of the target user according to the weighted weight corresponding to the at least one first resource, and the like.

Wherein the input device 803 may include one or more of a keyboard, a touch screen, a radio frequency receiver, or other input device; the output devices 804 may include one or more of a speaker, a display, a radio frequency transmitter, or other output devices. Optionally, the information processing apparatus may further include a memory module, a power supply module, an application client, and the like.

Embodiments of the present application also provide a computer (readable) storage medium (Memory), which may be a Memory device in an information processing device, and is used to store programs and data. It is understood that the computer storage medium herein may include both a built-in storage medium in the information processing apparatus and, of course, an extended storage medium supported by the information processing apparatus. The computer storage medium provides a storage space that stores an operating system of the information processing apparatus. Also stored in this memory space are one or more instructions, which may be one or more computer programs (including program code), suitable for loading and execution by processor 801. It should be noted that the computer storage medium may be a high-speed RAM memory, or may be a non-volatile memory (non-volatile memory), such as at least one disk memory; and optionally at least one computer storage medium located remotely from the processor.

In some embodiments, one or more instructions stored in a computer storage medium may be loaded and executed by processor 801 to perform the corresponding steps described above with respect to the method in the information processing embodiments; for example, in a particular implementation, one or more instructions in a computer storage medium are loaded and executed by the processor 801 to perform the steps of:

In an alternative embodiment, when determining push information for the target user according to the weighted weight corresponding to the at least one first resource, the one or more instructions may be loaded and executed by the processor 801:

when determining push information for the target user according to the weight vector corresponding to the at least one first resource, the one or more instructions may be loaded and executed by the processor 801:

when the splicing processing is performed on the weighted weights of the feature information of each first resource under the multiple resource categories to obtain the weight vector corresponding to each first resource, the one or more instructions may be loaded and executed by the processor 801:

In an alternative embodiment, the one or more instructions may also be loaded and executed by the processor 801 to:

acquiring characteristic information of the second resource, and determining the weight of the characteristic information of the second resource under a plurality of resource categories;

weighting the weight of the characteristic information of the second resource under a plurality of resource categories to obtain the weighted weight of the characteristic information of the second resource under the plurality of resource categories;

and splicing the weighted weights of the characteristic information of the second resource under the multiple resource categories to obtain a weight vector corresponding to the second resource.

acquiring characteristic information of a plurality of users, wherein the characteristic information of each user comprises resource characteristics of at least one third resource associated with the user;

taking the resource characteristics of a fourth resource in a plurality of pushed resources as a positive sample, taking the resource characteristics of the pushed resources except the fourth resource in the plurality of pushed resources as a negative sample, taking the resource characteristics of at least one third resource as the characteristics of a user tower, taking the resource characteristics of the plurality of pushed resources as the characteristics of a target tower, and training the double-tower model; the fourth resource is a pushed resource clicked by a user in the plurality of pushed resources.

In an optional embodiment, the feature information of the at least one first resource is a feature vector; in the determining the weight of the characteristic information of the at least one first resource under the plurality of resource categories, respectively, the one or more instructions may be loaded and executed by the processor 801:

It is understood that in the present application, "and/or", such as, a and/or B, is used to describe the association relationship of the associated object, such as can represent: a exists alone, A and B exist simultaneously, and B exists alone. The sequence numbers of the above processes do not mean the execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the present application, and the contents of the embodiments may be referred to each other.

The present application also provides a blockchain system including the information processing apparatus (or information processing device) described above and a blockchain link point. Optionally, the system may further include other devices that interact with the block link points or information processing devices (information processing apparatuses). The information processing apparatus (information processing device) may perform part or all of the steps in the methods in the embodiments shown in fig. 4 to 5, which are not described herein again.

Embodiments of the present application also provide a computer program product including a non-transitory computer-readable storage medium storing a computer program operable to cause a computer to perform some or all of the steps described in the information processing apparatus (information processing apparatus) in the above method embodiment.

The above disclosure is only for the purpose of illustrating the preferred embodiments of the present application and is not to be construed as limiting the scope of the present application, so that the present application is not limited thereto, and all equivalent variations and modifications can be made to the present application.

Claims

1. An information processing method, characterized in that the method comprises:

2. The method according to claim 1, wherein the determining the push information for the target user according to the weighted weight corresponding to the at least one first resource comprises:

3. The method of claim 2, wherein the feature information of the first resource comprises a feature vector after vectorization of features of the first resource by a user tower of a dual tower model, the dual tower model comprising the user tower and a target tower, the push information comprising a push resource;

the determining, according to the weight vector corresponding to the at least one first resource, push information for the target user includes:

4. The method of claim 2, wherein the characteristic information further comprises viewing information of the at least one first resource, the viewing information comprising viewing order and/or viewing time information of the at least one first resource;

the splicing processing of the weighted weights of the feature information of each first resource under the multiple resource categories to obtain the weight vector corresponding to each first resource includes:

5. The method of claim 3, further comprising:

6. The method of claim 3, further comprising:

taking the resource characteristics of a fourth resource in a plurality of pushed resources as a positive sample, taking the resource characteristics of the pushed resources except the fourth resource in the plurality of pushed resources as a negative sample, taking the resource characteristics of at least one third resource as the characteristics of a user tower, taking the resource characteristics of the plurality of pushed resources as the characteristics of a target tower, and training the double-tower model; and the fourth resource is a pushed resource clicked by a user in the plurality of pushed resources.

7. The method according to any of claims 1-6, wherein the feature information of the at least one first resource is a feature vector; the determining the weight of the characteristic information of the at least one first resource under a plurality of resource categories respectively comprises:

8. An information processing apparatus characterized by comprising:

9. An information processing apparatus characterized by comprising:

a computer storage medium having one or more instructions stored thereon, the one or more instructions adapted to be loaded by the processor and to perform the method of any of claims 1-7.

10. A computer-readable storage medium, characterized in that the computer-readable storage medium stores a computer program comprising program instructions that, when executed by a processor, cause the processor to carry out the method according to any one of claims 1-7.