CN115130000A

CN115130000A - Information recommendation method and device, storage medium and electronic equipment

Info

Publication number: CN115130000A
Application number: CN202210856576.2A
Authority: CN
Inventors: 白琛; 罗灵锐; 杜小毅; 史鑫磊; 罗恒亮
Original assignee: Beijing Sankuai Online Technology Co Ltd
Current assignee: Xiamen Sankuai Online Technology Co ltd
Priority date: 2022-07-20
Filing date: 2022-07-20
Publication date: 2022-09-30

Abstract

The specification discloses an information recommendation method, an information recommendation device, a storage medium and electronic equipment. The information recommendation method comprises the following steps: acquiring user information and information to be recommended of a user, inputting the user information and the information to be recommended into a pre-trained information recommendation model, to determine the description information contained in the information to be recommended and used for describing the business object corresponding to the information to be recommended, and determines the user map characteristics corresponding to the user, the description map characteristics corresponding to the description information and the information map characteristics corresponding to the recommendation information through the pre-constructed feature map, the characteristic map is used for representing the historical interactive relation between each user and each historical recommendation information under the influence of different types of historical description information, determining a first click rate of a user for clicking the information to be recommended under the influence of the description information according to the user map characteristics, the description map characteristics and the information map characteristics, and recommending the information to the user according to the first click rate.

Description

Information recommendation method and device, storage medium and electronic equipment

Technical Field

The present disclosure relates to the field of information recommendation, and in particular, to a method, an apparatus, a storage medium, and an electronic device for information recommendation

Background

In recent years, with the development of science and technology, information recommendation models have been applied to more and more fields, and especially in the context of recommendation, search, advertisement and the like, the information recommendation models can predict click rates of different recommendation information by users through analysis of user information and historical behaviors, so that according to prediction results, corresponding recommendation information lists are generated to be shown to the users.

However, the current method usually only predicts the click rate of the user according to the characteristics of the user and the characteristics of the recommendation information, but ignores the guiding effect of other factors contained in the recommendation information on the user, which may result in that the recommendation information finally recommended to the user is not accurate enough.

Therefore, how to accurately predict the click condition of the user on different recommendation information so as to recommend the recommendation information meeting the preference of the user to the user, and further improve the use experience of the user is a problem to be solved urgently.

Disclosure of Invention

The present specification provides a method, an apparatus, a storage medium, and an electronic device for information recommendation, so as to partially solve the above problems in the prior art.

The technical scheme adopted by the specification is as follows:

the present specification provides an information recommendation method, including:

acquiring user information and information to be recommended of a user;

inputting the user information and information to be recommended into a pre-trained information recommendation model to determine description information contained in the information to be recommended and used for describing a service object corresponding to the information to be recommended, and determining user graph characteristics corresponding to the user, description graph characteristics corresponding to the description information and information graph characteristics corresponding to the information to be recommended through a pre-constructed feature graph, wherein the feature graph is used for representing historical interaction relations between the users and the historical recommendation information under the influence of different types of historical description information;

determining a first click rate of clicking the information to be recommended by the user under the influence of the description information according to the user map feature, the description map feature and the information map feature;

and recommending information to the user according to the first click rate.

Optionally, constructing the feature map specifically includes:

acquiring historical interaction records between each user and each historical recommendation information;

and for each user, according to the historical interaction records, connecting a user node corresponding to the user with an information node corresponding to the historical recommendation information clicked by the user by taking the determined description characteristics corresponding to the description information matched with the historical recommendation information as sides, and constructing the characteristic map.

Optionally, determining description information matched with the historical recommendation information specifically includes:

and if the historical recommendation information does not contain the description information, determining the description information matched with the historical recommendation information according to the historical interaction record.

Optionally, determining, according to the historical interaction record, description information matched with the historical recommendation information, specifically including:

determining other historical recommendation information with the same type as the information to which the historical recommendation information belongs;

determining reference recommendation information from the other historical recommendation information according to the historical interaction records corresponding to the other historical recommendation information;

and determining the description information matched with the historical recommendation information according to the description information contained in the reference recommendation information.

Optionally, before recommending information to the user according to the first click rate, the method further includes:

determining original user characteristics corresponding to the user and original information characteristics corresponding to the information to be recommended;

determining the click rate of the user for clicking the information to be recommended as a second click rate according to the original user characteristics, the user map characteristics, the original information characteristics and the information map characteristics;

according to the first click rate, recommending information to the user, specifically comprising:

and recommending information to the user according to the first click rate and the second click rate.

Optionally, recommending information to the user according to the first click rate and the second click rate, specifically including:

determining the weight corresponding to the first click rate as a first weight, and determining the weight corresponding to the second click rate as a second weight;

determining the click rate of the user for clicking the information to be recommended as a comprehensive click rate according to the first click rate and the first weight, and the second click rate and the second weight;

and recommending information to the user according to the comprehensive click rate.

Optionally, determining a weight corresponding to the first click rate as a first weight, and determining a weight corresponding to the second click rate as a second weight specifically includes:

determining the first weight according to the user profile feature, the description profile feature and the information profile feature;

and determining the second weight according to the first weight.

This specification provides an apparatus for information recommendation, including:

the acquisition module acquires user information of a user and information to be recommended;

the input module is used for inputting the user information and the information to be recommended into a pre-trained information recommendation model so as to determine description information which is contained in the information to be recommended and used for describing a service object corresponding to the information to be recommended, and determining user map features corresponding to the user, description map features corresponding to the description information and information map features corresponding to the information to be recommended through a pre-constructed feature map, wherein the feature map is used for representing historical interaction relations between the users and the historical recommendation information under the influence of different types of historical description information;

the determining module is used for determining a first click rate of clicking the information to be recommended under the influence of the description information by the user according to the user map feature, the description map feature and the information map feature;

and the recommending module is used for recommending information to the user according to the first click rate.

The present specification provides a computer-readable storage medium storing a computer program which, when executed by a processor, implements the above-described information recommendation method.

The present specification provides an electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the above information recommendation method when executing the program.

The technical scheme adopted by the specification can achieve the following beneficial effects:

in the information recommendation method provided in this specification, the acquired user information and the information to be recommended are input into a pre-trained information recommendation model, to determine the description information contained in the information to be recommended and used for describing the service object corresponding to the information to be recommended, and determines the user map characteristics corresponding to the user, the description map characteristics corresponding to the description information and the information map characteristics corresponding to the recommendation information through the pre-constructed feature map, the characteristic map is used for representing the historical interactive relation between each user and each historical recommendation information under the influence of different types of historical description information, and determining the click rate of the user for clicking the information to be recommended under the influence of the description information according to the user map characteristics, the description map characteristics and the information map characteristics, and further recommending the information to the user according to the click rate.

According to the method, when information is recommended to the user, the description information contained in the recommendation information is extracted, and the user map features corresponding to the user, the description map features corresponding to the description information and the information map features corresponding to the recommendation information are determined in the corresponding feature maps, so that the click rate of the user for clicking the information to be recommended under the influence of the description information can be accurately determined by an information recommendation model according to the features.

Drawings

The accompanying drawings, which are included to provide a further understanding of the specification and are incorporated in and constitute a part of this specification, illustrate embodiments of the specification and together with the description serve to explain the specification and not to limit the specification in a non-limiting sense. In the drawings:

FIG. 1 is a schematic flow chart of a method for information recommendation provided herein;

fig. 2 is a schematic diagram of a generation process of an interaction record provided in this specification;

FIG. 3 is a schematic diagram of a comprehensive click rate prediction process provided herein;

FIG. 4 is a schematic diagram of an information recommendation apparatus provided herein;

fig. 5 is a schematic diagram of an electronic device corresponding to fig. 1 provided in the present specification.

Detailed Description

In order to make the objects, technical solutions and advantages of the present disclosure more clear, the technical solutions of the present disclosure will be clearly and completely described below with reference to the specific embodiments of the present disclosure and the accompanying drawings. It is to be understood that the embodiments described are only a few embodiments of the present disclosure, and not all embodiments. All other embodiments obtained by a person of ordinary skill in the art based on the embodiments in the present specification without making any creative effort belong to the protection scope of the present specification.

The technical solutions provided by the embodiments of the present description are described in detail below with reference to the accompanying drawings.

Fig. 1 is a schematic flowchart of a method for recommending information provided in this specification, including the following steps:

s101: user information of a user is acquired.

When a user browses recommendation information such as take-out, hotels, restaurants and the like in a service client or a webpage, the preference of the user for the attribute of the recommendation information (such as the price, the position, the category and the like of a service object in the recommendation information) often determines whether the user clicks the recommendation information, but in practical application, description information which is contained in the recommendation information and used for describing the service object corresponding to information to be recommended also plays a certain guiding role for the user, some users often pay more attention to the description information recorded in the recommendation information, and if the user generates interest in the description information recorded in the recommendation information, the user clicks the recommendation information even if the attribute of the recommendation information does not accord with the preference of the user.

For example, when recommending hotel information to a user, the price, type and location of a hotel often affect the click condition of the user on the recommended information, the user tends to click on the hotel with the closest price and location that the user can receive when browsing the information, but when browsing a piece of hotel recommended information that does not meet the user's preference, if the descriptive information (such as hotel facilities, hotel environment, service quality, sound insulation effect, geographical location, etc.) contained in the recommended information can make the user interested, the user also can make the user interested, and then click on the recommended information.

Based on this, the present specification provides an information recommendation method, where user information of a user needs to be obtained, so as to describe a user image according to the user information, so as to determine a user characteristic corresponding to the user.

It should be noted that, in this specification, before obtaining the user information of the user, authorization needs to be performed by the user, and after the user authorization, the user information of the user may be obtained. And if the user refuses the authorization, the user information of the user cannot be acquired. If the user firstly authorizes and then cancels the authorization, all user information of the user acquired after the user is authorized can be deleted.

In addition, the server may further obtain each piece of information to be recommended, so as to select a certain amount of information to be recommended from the pieces of information to be recommended and display the selected information to the user according to the determined order, where the recommended information may be recommended information corresponding to different types of business objects such as take-out, hotels, clothes, restaurants, articles for daily use, and certainly may also be recommended information corresponding to other business objects, and this specification does not specifically limit this.

In this specification, the execution subject for implementing the information recommendation method may refer to a designated device such as a server installed on the service platform, and for convenience of description, the following describes a method for information recommendation provided in this specification, by taking only the server as the execution subject.

S102: inputting the user information and the information to be recommended into a pre-trained information recommendation model to determine description information contained in the information to be recommended and used for describing a service object corresponding to the information to be recommended, and determining user map features corresponding to the user, description map features corresponding to the description information and information map features corresponding to the information to be recommended through a pre-constructed feature map, wherein the feature map is used for representing historical interaction relations between the users and the historical recommendation information under the influence of different types of historical description information.

After the user information corresponding to the user is acquired, the server can input the user information into a pre-trained information recommendation model, and for each piece of information to be recommended, the server can extract description information which is contained in the information to be recommended and used for describing a service object corresponding to the information to be recommended through the information recommendation model.

Of course, the description information may be a brief introduction and description of the information to be recommended, or may also be other information such as comments and labels in the information to be recommended, which is not specifically limited in this specification.

Further, the server may perform feature extraction on the user information of the user through a feature extraction layer set in the information recommendation model to obtain an original user feature corresponding to the user, perform feature extraction on the information to be recommended, perform feature extraction on the obtained original information feature corresponding to the information to be recommended and perform feature extraction on the extracted description information included in the recommendation information to obtain an original description feature corresponding to the description information.

In this specification, the original description feature corresponding to the description information may be an original description feature corresponding to a type to which the description information belongs. The type of the description information may be a type of different content recorded by the description information, and for example, in a hotel recommendation scene, the type of the description information may be an environment description of a hotel, a service quality description of the hotel, a geographic location description of the hotel, an equipment and facility description of the hotel, a sound insulation effect description of the hotel, and the like. Of course, the type of the description information may also be other content types corresponding to other service scenarios, and this is not specifically limited in this specification.

The server may map the user information to a one-hot coded (one-hot) vector through a feature extraction layer in the information recommendation model, and then multiply the one-hot vector corresponding to the user information by a corresponding embedding matrix, thereby obtaining the original user feature corresponding to the user (i.e., a vector obtained by multiplying the one-hot vector corresponding to the user information by the corresponding embedding matrix).

Similarly, the original information features corresponding to the recommendation information and the description features corresponding to the description information included in the recommendation information may also be extracted in the above manner, and this specification is not described herein in detail.

After obtaining the original user characteristics corresponding to the user, the original information characteristics corresponding to the recommendation information, and the original description characteristics corresponding to the description information included in the recommendation information, the server may input the original user characteristics, the original information characteristics, and the original description characteristics into a feature map that is pre-constructed in the information recommendation model, so that the feature map queries, according to the original user characteristics, the original information characteristics, and the original description characteristics, the user map characteristics corresponding to the user, the information map characteristics corresponding to the recommendation information, and the description map characteristics corresponding to the description information included in the recommendation information in the map.

Of course, the server may also directly input the user information corresponding to the user, the recommendation information, and the description information included in the recommendation information into the feature map, so that the feature map directly determines the user map feature corresponding to the user according to the user information, determines the information map feature corresponding to the recommendation information according to the recommendation information, and determines the description map feature corresponding to the description information included in the recommendation information according to the description information.

In practical applications, there may be some new users that do not appear in the constructed feature map, so that the user map features corresponding to the new users may not be found. Therefore, when information is recommended to the new users, some recommendation information meeting other conditions, such as recommendation information with the highest current click rate or recommendation information with the largest recommendation amount, can be recommended to the new users, and after interaction data of some users and different recommendation information under the influence of different description information are gradually generated along with the use of the users, the user graph characteristics of the users can be updated into the feature graph according to the interaction data.

In addition, the server may also perform feature matching on the original user features corresponding to the new users and original user features corresponding to other users in the feature map, so that the user map feature corresponding to the other user closest to the original user feature corresponding to the user is used as the user map feature corresponding to the user.

The feature map can be constructed based on historical interaction data of the user and other users in the sample data set under the influence of different types of description information and each piece of recommendation information, and is used for representing historical interaction relations between the users and the historical recommendation information under the influence of the different types of historical description information. That is to say, if there is data in the feature map, which is used by the user to click on the recommendation information under the description information, the user will click on the information to be recommended with a high probability under the guidance of the description information.

It should be noted that, in this specification, although the original feature extraction layer and the feature map all function to determine the corresponding features, however, the features extracted by the initial feature extraction layer are different from the features determined by the feature map in nature, the original user features, the original description features and the original information features extracted by the initial feature extraction layer are independent from each other and have no certain relevance, the target characteristics determined by the characteristic map are historical interactive data of each user and each recommendation information under the influence of different types of description information, the original features are obtained after iteration, that is, the user map features, the information map features and the description map features determined by the feature map have certain relevance, and the influence of different types of files on the interaction behavior between the user and the recommended information can be reflected.

Before the server constructs the feature map, historical interaction records between each user and each historical recommendation information are acquired, so that the feature map is constructed.

In practical applications, some pieces of recommendation information may include description information, some pieces of recommendation information may not include description information, and history interaction records of the same user and each piece of recommendation information are often sparse, for example, if a specified user does not click on one piece of recommendation information including description information, only an interaction record in which the user does not click on the piece of recommendation information under the description information is generated, and since it is not known whether the user clicks on the history recommendation information under the condition that the piece of recommendation information does not include description information, a corresponding interaction record under the condition that the piece of recommendation information does not include description information may not occur. Therefore, the sparse interaction record in the single direction is not beneficial to the construction of the feature map, and the precision of the constructed feature map is limited.

Therefore, the server can judge whether the historical recommendation information contains the description information or not according to each piece of historical recommendation information to obtain a judgment result, and then generate new sample data according to the judgment result and actual operation data of the user on the historical recommendation information.

Specifically, in the process of browsing information by the user, the description information included in the recommendation information often plays a positive role in guiding the clicking behavior of the user, so that for the history recommendation information including the history description information and clicked by the user, the server may generate a history interaction record that the user does not click on the history recommendation information under the condition that the history recommendation information does not include the history description information. In other words, if the history recommendation information does not include the history description information, the user does not click on the history recommendation information.

For the history recommendation information which does not include the history description information and is clicked by the user, the server may generate an interaction record that the user clicks the history recommendation information when the history recommendation information includes the history description information. In other words, if the history recommendation information is clicked by the user in the case where the history recommendation information does not include the history description information, the history recommendation information is clicked by the user in the case where the history recommendation information includes the history description information.

When the feature map is constructed, the user nodes and the information nodes are connected by edges, and for the historical recommendation information which is clicked by the user but has no description information, namely the historical recommendation information in the historical interaction record generated by the historical recommendation information which does not contain the historical description information and is clicked by the specified user, the server can judge whether the historical recommendation information contains the description information or not, and if the historical recommendation information does not contain the description information, the description information matched with the historical recommendation information is determined according to the historical interaction record.

Specifically, the server may determine other historical recommendation information of the same type as the information to which the historical recommendation information belongs, determine reference recommendation information from the other historical recommendation information according to historical interaction records corresponding to the other historical recommendation information, and then determine description information matched with the historical recommendation information according to description information included in the reference recommendation information.

For example, the server may determine, as reference recommendation information, historical recommendation information that has a largest number of clicks and includes description information of other users, and use the description information included in the reference recommendation information as description information that matches the historical recommendation information.

Of course, the server may also determine the description information with the most frequent historical recommendation information being presented to other users as the description information matched with the historical recommendation information.

For the historical recommendation information which is clicked by the user and contains the description information, the server does not need to reconfirm the description information of the historical recommendation information.

Of course, the description information in the generated history recommendation information may also be description information corresponding to other history recommendation information clicked by the user, and this is not particularly limited in this specification.

In this way, an actual interaction record is obtained, as well as a new generated interaction record generated based on the inference of the actual interaction record. For the convenience of understanding, the present specification also provides a schematic diagram of generating sample data, as shown in fig. 2.

Fig. 2 is a schematic diagram of a generation process of an interaction record provided in this specification.

The server can screen out an actual sample which contains the description information and is not clicked by the user and a historical interaction record which does not contain the description information and is clicked by the user, and then deduces the historical interaction record which contains the description information and is not clicked by the user, so that the historical recommendation information corresponding to the actual sample can be rejected by the user, and the historical recommendation information can not be clicked under the condition that the historical recommendation information does not contain the description information, and a corresponding newly generated interaction record which does not contain the description information and is not clicked by the user can be generated.

For the history interaction record which does not contain the description information and is clicked by the user, the user himself is interested in the history recommendation information, so that the user clicks the history recommendation information even when the history recommendation information contains the description information, and a corresponding generation interaction record which contains the description information and is clicked by the user is generated.

After the history interaction record is obtained and the interaction record is generated, the server may construct a feature map, where the graph convolution network includes a plurality of convolution layers, and in each convolution layer, since one user clicks a plurality of recommendation information and one recommendation information is clicked by a plurality of users, a user node at an initial layer of the graph convolution network may be represented as:

the information node of this layer may be represented as:

for any one of the nodes f (which may be a user node or an information node), the combination with the neighbor node of the node f (i.e., the node connected to the node f) may be expressed as:

N _f ＝{(t,b)|(f,t,b)∈G}

wherein N is _f Is a neighbor of node fAnd b is a neighbor node of f, t is an edge between the two nodes, and G is the feature map.

Therefore, the feature corresponding to the node f can be represented by the neighbor nodes and the corresponding edges of the node f:

wherein the content of the first and second substances,

for the representation of the corresponding characteristic of the node f by the neighbor nodes of the node f, N _f Is a set of neighboring nodes, and is,

is the feature corresponding to the neighbor node b in the layer connected to node f, e _t Is the feature corresponding to the edge between node f and the neighbor node b.

After convolution propagation is performed on a plurality of convolution layers, the graph characteristics corresponding to the nodes after iteration can be expressed as:

wherein the content of the first and second substances,

for the map feature corresponding to the last layer (i.e. layer l) of the node f, LeakyReLU is the corresponding activation function, W ₁ And W ₂ Respectively corresponding transformation matrices.

By the method, the user map features corresponding to all users, the information map features corresponding to all historical recommendation information and the description map features corresponding to all description information can be obtained, and therefore the feature map is constructed through the map features.

Specifically, for each user, the server may associate the user profile characteristics with the user, and the userUsing the information map feature corresponding to the clicked historical recommendation information as a node, connecting the user map feature node with the information map feature node by using the description map feature corresponding to the historical description information contained in the historical recommendation information as a side, and thus, if the user u exists in the sample data _i In history description information t _k To history recommendation information i under guidance of _j When clicked, a triple (u) is formed _i ,t _k ,i _j ) By describing feature nodes t of the graph _k (edge) grouping user profile feature nodes u _i And information map feature node i _j The connection is made.

S103: and determining a first click rate of clicking the information to be recommended by the user under the influence of the description information according to the user map feature, the description map feature and the information map feature.

The information recommendation model can comprise two prediction networks, namely a first prediction network and a second prediction network, wherein the second prediction network is used for determining the click rate of the user on the information to be recommended as a second click rate based on the matching degree between the user information and the information to be recommended according to the input original user characteristics, original information characteristics, user map characteristics and information map characteristics;

the first prediction network is used for determining a first click rate of clicking on the information to be recommended under the influence of the description information contained in the information to be recommended by the user according to the input user map features, description map features and information map features.

Because the user spectrum features corresponding to the user and the information spectrum features corresponding to the information to be recommended are obtained after repeated iterative learning in the feature spectrum construction process, the user spectrum features, the information spectrum features and the description spectrum features are in different representation spaces, so that the user spectrum features and the information spectrum features need to be processed before being input into the first prediction network, and the processing process of the user spectrum features corresponding to the user can be expressed by the following formula:

e′ _u ＝e _u +σ(e _u ⊙e _t )⊙e _u

wherein e is _u Is the user map feature corresponding to the user, e' _u For the processed user profile features, σ is an activation function (e.g., sigmod function).

The processing procedure of the associated features corresponding to the information to be recommended can be represented by the following formula:

e′ _i ＝e _i +σ(e _i ⊙e _t )⊙e _i

wherein e is _i Is the information map characteristic corresponding to the information to be recommended, e' _i And sigma is an activation function for the processed information map features corresponding to the recommendation information.

In addition, the server can also perform corresponding processing on the description map characteristics.

After the user map features, the information map features and the description map features are processed, the server can input the processed user map features, the processed information map features and the processed description map features corresponding to the description information contained in the information to be recommended into a first prediction network, and therefore a first click rate of clicking on the information to be recommended by a user under the influence of the description information is determined.

The server can determine a weight corresponding to the first click rate as a first weight, and determine a weight corresponding to the second click rate as a second weight, and then the server can determine a comprehensive click rate corresponding to the information to be recommended according to the first weight corresponding to the first click rate and the first click rate, and the second weight corresponding to the second click rate and the second click rate.

Further, the server may determine, according to the map features, a first weight corresponding to a first click rate and a second weight corresponding to a second click rate through a multi-layer perceptron, where the weight corresponding to the first click rate may be represented by the following formula:

wherein, ω is a first weight corresponding to the first click rate, MLP is a multi-layer perceptron, σ is a Sigmoid function,

for user u _i The characteristics of the corresponding user's profile,

to recommend information i _j Description information t contained in _k The characteristics of the corresponding description map are described,

to recommend information i _j Corresponding information map features.

And then the server can determine a second weight corresponding to a second click rate according to the weight corresponding to the first click rate.

In practical applications, if a user pays more attention to the description information included in the recommendation information, attributes of the recommendation information (such as a commodity category and a price interval corresponding to the recommendation information) may have a smaller influence on the click behavior of the user, and if the user pays more attention to the attributes of the recommendation information, the description information included in the recommendation information may have a smaller influence on the click behavior of the user. Thus. The integrated click rate can be expressed by the following formula:

wherein the content of the first and second substances,

for the information i to be recommended _j The corresponding comprehensive click rate is obtained by the method,

is to be recommended toA first click rate corresponding to the information,

and a second click rate corresponding to the information to be recommended, wherein ω is a first weight corresponding to the first click rate, and (1- ω) is a second weight corresponding to the second click rate.

As can be seen from the above formula, the larger the first weight corresponding to the first click rate is, the smaller the second weight corresponding to the second click rate is, and conversely, the smaller the first weight corresponding to the first click rate is, the larger the second weight corresponding to the second click rate is, but the sum of the first weight and the second weight is a fixed value (i.e., the sum is 1).

For easy understanding, the present specification further provides a schematic diagram of a comprehensive click rate prediction process, as shown in fig. 3.

FIG. 3 is a schematic diagram of a comprehensive click rate prediction process provided herein.

The server can input the processed user map features, the processed information map features and the processed description map features into a first prediction network, determine a first click rate corresponding to the recommendation information, input the user map features, the information map features, the original user features and the original information features into a second prediction network, and determine a second click rate corresponding to the recommendation information.

S104: and recommending information to the user according to the first click rate.

After the first click rate and the comprehensive click rate corresponding to each piece of information to be recommended are determined, the server can recommend the information to the user according to the comprehensive click rate corresponding to each piece of information to be recommended, for example, the server can sort the pieces of information to be recommended according to the descending order of the comprehensive click rate to obtain the sorting results corresponding to the pieces of information to be recommended, then select the pieces of information to be recommended before the designated sorting order, generate the recommendation lists corresponding to the pieces of recommendation information, and recommend the information to the user.

Of course, the server may also recommend information to the user only according to the first click rate corresponding to each piece of information to be recommended, or recommend information to the user only according to the second click rate corresponding to the piece of information to be recommended.

Before the information recommendation model is used, the information recommendation model needs to be trained in advance, and it should be noted that the historical interaction records between each user and each historical recommendation information included in the sample data set used in the training process of the information recommendation model may be the same historical interaction records between each user and each historical recommendation information in the process of constructing the feature map, or may be different historical interaction records, but the actual interaction records may be inferred in the same manner, so that a new generated interaction record is generated.

That is, for the history interactive record which includes the description information and is not clicked by the user, the user himself may reject the history recommendation information corresponding to the actual sample, so that the history recommendation information is not clicked even when the history recommendation information does not include the description information, and therefore, a newly generated interactive record which corresponds to the history recommendation information and is not clicked by the user may be generated. For the history interaction record which does not contain the description information and is clicked by the user, the user himself can generate interest for the history recommendation information, so that under the condition that the history recommendation information contains the description information, the user can click on the history recommendation information, and therefore the corresponding generation interaction record which contains the description information and is clicked by the user can be generated.

In the present specification, the training subject for training the information recommendation model may refer to a server or a specific device such as a desktop computer or a notebook computer, and for convenience of description, the training of the information recommendation model will be described below by taking the server as the training subject for training the information recommendation model as an example.

The server can obtain historical user information and historical recommendation information of each user, input the historical user information and the historical recommendation information into an information recommendation model to be trained to determine that the historical recommendation information contains description information, determine user map features corresponding to the user, description map features corresponding to the description information and information map features corresponding to the historical recommendation information through a pre-constructed feature map, determine a first click rate of the user for clicking the historical recommendation information under the influence of the description information according to the user map features, the map features and the information map features, and perform information recommendation on the user according to the first click rate to obtain a recommendation result.

The server can train the information recommendation model by taking the deviation between the minimum recommendation result and the actual operation condition of the information recommended to each history by the specified user as an optimization target until the training target is met, and deploy the information recommendation model. Wherein, the training target can be: the information recommendation model converges to a preset threshold range, or reaches a preset training frequency to ensure that the information recommendation model accurately recommends recommendation information conforming to the preference of the user to the user, and the preset threshold range and the preset training frequency can be set according to actual conditions, which is not specifically limited in the specification.

The loss function of the information recommendation model can be determined according to the actual interaction records of the appointed user on the historical data and the comprehensive click rate corresponding to the historical data, and the information recommendation model is trained by using the minimization of the loss function as an optimization target. The loss function can be calculated by the following formula:

wherein L is a loss function corresponding to the information recommendation model, y _j Is sample data j and userThe actual interaction scenario of (c). For example, when a user clicks on sample data j, y _j When the user does not click on sample data j, y _j ＝0。

And lambda is a hyper-parameter for controlling regularization, and theta represents all trainable parameters in the model, wherein lambda is the comprehensive click rate corresponding to the sample data j predicted by the information recommendation model.

From the above formula, it can be seen that when the sample data j corresponds to the actual click condition y _j When the value is 1, namely the user clicks on the sample data j, the result is that

The larger the comprehensive click rate corresponding to the predicted sample data j is, the smaller the value of L is. And when the actual click condition y corresponding to the sample data j _j 0, i.e. the user has not clicked on sample data j,

the larger the predicted comprehensive click rate corresponding to the sample data j is, the larger the value of L is.

According to the method, when information recommendation is performed on the user, the description information contained in the recommendation information is extracted, and the user map features corresponding to the user, the description map features corresponding to the description information and the information map features corresponding to the recommendation information are determined in the corresponding feature map, so that the click rate of the user for clicking on the information to be recommended under the influence of the description information can be accurately determined by an information recommendation model according to the features.

It should be noted that all the actions of acquiring signals, information or data in this specification are performed under the premise of complying with the corresponding data protection regulation policy of the country of the location and obtaining the authorization given by the owner of the corresponding device.

Based on the same idea, the present specification also provides a corresponding information recommendation device, as shown in fig. 4.

Fig. 4 is a schematic diagram of an information recommendation apparatus provided in this specification, including:

the obtaining module 401 obtains user information of a user and information to be recommended;

an input module 402, configured to input the user information and information to be recommended into a pre-trained information recommendation model, so as to determine description information included in the information to be recommended and used for describing a service object corresponding to the information to be recommended, and determine, through a pre-constructed feature map, a user map feature corresponding to the user, a description map feature corresponding to the description information, and an information map feature corresponding to the information to be recommended, where the feature map is used to represent a historical interaction relationship between each user and each piece of historical recommendation information under the influence of different types of historical description information;

the determining module 403 is configured to determine, according to the user graph feature, the description graph feature and the information graph feature, a first click rate at which the user clicks the information to be recommended under the influence of the description information;

and the recommending module 404 is used for recommending information to the user according to the first click rate.

Optionally, the apparatus further comprises: a build module 405;

the building module 405 is specifically configured to obtain a history interaction record between each user and each history recommendation information; and for each user, according to the historical interaction records, connecting a user node corresponding to the user with an information node corresponding to the historical recommendation information clicked by the user by taking the determined description characteristics corresponding to the description information matched with the historical recommendation information as sides, and constructing the characteristic map.

The building module 405 is specifically configured to, if the historical recommendation information does not include description information, determine, according to the historical interaction record, description information that matches the historical recommendation information.

The building module 405 is specifically configured to determine other history recommendation information of the same type as the information to which the history recommendation information belongs; determining reference recommendation information from the other historical recommendation information according to the historical interaction records corresponding to the other historical recommendation information; and determining the description information matched with the historical recommendation information according to the description information contained in the reference recommendation information.

Optionally, before information recommendation is performed on the user according to the first click rate, the recommendation module 404 is configured to determine an original user characteristic corresponding to the user and an original information characteristic corresponding to the information to be recommended; determining the click rate of the user for clicking the information to be recommended as a second click rate according to the original user characteristics, the user map characteristics, the original information characteristics and the information map characteristics;

the recommending module 404 is specifically configured to recommend information to the user according to the first click rate and the second click rate.

Optionally, the recommending module 404 is specifically configured to determine a weight corresponding to the first click rate as a first weight, and a weight corresponding to the second click rate as a second weight; determining the click rate of the user for clicking the information to be recommended as a comprehensive click rate according to the first click rate and the first weight, and the second click rate and the second weight; and recommending information to the user according to the comprehensive click rate.

Optionally, the recommending module 404 is specifically configured to determine the first weight according to the user graph feature, the description graph feature, and the information graph feature; and determining the second weight according to the first weight.

The present specification also provides a computer-readable storage medium storing a computer program, which can be used to execute a method of information recommendation provided in fig. 1 above.

The present specification also provides a schematic block diagram of an electronic device corresponding to fig. 1 shown in fig. 5. As shown in fig. 5, at the hardware level, the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and may also include hardware required by other services. The processor reads a corresponding computer program from the non-volatile memory into the memory and then runs the computer program to implement the information recommendation method described in fig. 1. Of course, besides the software implementation, this specification does not exclude other implementations, such as logic devices or combination of software and hardware, and so on, that is, the execution subject of the following processing flow is not limited to each logic unit, and may be hardware or logic devices.

In the 90 s of the 20 th century, improvements in a technology could clearly distinguish between improvements in hardware (e.g., improvements in circuit structures such as diodes, transistors, switches, etc.) and improvements in software (improvements in process flow). However, as technology advances, many of today's process flow improvements have been seen as direct improvements in hardware circuit architecture. Designers almost always obtain the corresponding hardware circuit structure by programming an improved method flow into the hardware circuit. Thus, it cannot be said that an improvement in the process flow cannot be realized by hardware physical modules. For example, a Programmable Logic Device (PLD), such as a Field Programmable Gate Array (FPGA), is an integrated circuit whose Logic functions are determined by programming the Device by a user. A digital system is "integrated" on a PLD by the designer's own programming without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Furthermore, nowadays, instead of manually making an Integrated Circuit chip, such Programming is often implemented by "logic compiler" software, which is similar to a software compiler used in program development and writing, but the original code before compiling is also written by a specific Programming Language, which is called Hardware Description Language (HDL), and HDL is not only one but many, such as abel (advanced Boolean Expression Language), ahdl (alternate Hardware Description Language), traffic, pl (core universal Programming Language), HDCal (jhdware Description Language), lang, Lola, HDL, laspam, hardward Description Language (vhr Description Language), vhal (Hardware Description Language), and vhigh-Language, which are currently used in most common. It will also be apparent to those skilled in the art that hardware circuitry that implements the logical method flows can be readily obtained by merely slightly programming the method flows into an integrated circuit using the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer-readable medium storing computer-readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, an Application Specific Integrated Circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller as pure computer readable program code, the same functionality can be implemented by logically programming method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers and the like. Such a controller may thus be considered a hardware component, and the means included therein for performing the various functions may also be considered as a structure within the hardware component. Or even means for performing the functions may be regarded as being both a software module for performing the method and a structure within a hardware component.

The systems, devices, modules or units illustrated in the above embodiments may be implemented by a computer chip or an entity, or by a product with certain functions. One typical implementation device is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smartphone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being divided into various units by function, and are described separately. Of course, the functions of the various elements may be implemented in the same one or more software and/or hardware implementations of the present description.

As will be appreciated by one skilled in the art, embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the description may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the description may take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, and the like) having computer-usable program code embodied therein.

The description has been presented with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the description. It will be understood that each flow and/or block of the flow diagrams and/or block diagrams, and combinations of flows and/or blocks in the flow diagrams and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In a typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include forms of volatile memory in a computer readable medium, Random Access Memory (RAM) and/or non-volatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.

Computer-readable media, including both permanent and non-permanent, removable and non-removable media, may implement the information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of computer storage media include, but are not limited to, phase change memory (PRAM), Static Random Access Memory (SRAM), Dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), Read Only Memory (ROM), Electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), Digital Versatile Disks (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. As defined herein, a computer readable medium does not include a transitory computer readable medium such as a modulated data signal and a carrier wave.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising an … …" does not exclude the presence of other like elements in a process, method, article, or apparatus that comprises the element.

This description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

The embodiments in the present specification are described in a progressive manner, and the same and similar parts among the embodiments are referred to each other, and each embodiment focuses on the differences from the other embodiments. In particular, for the system embodiment, since it is substantially similar to the method embodiment, the description is simple, and for the relevant points, reference may be made to the partial description of the method embodiment.

The above description is only an example of the present disclosure, and is not intended to limit the present disclosure. Various modifications and alterations to this description will become apparent to those skilled in the art. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present specification should be included in the scope of the claims of the present specification.

Claims

1. A method for information recommendation, comprising:

acquiring user information and information to be recommended of a user;

inputting the user information and information to be recommended into a pre-trained information recommendation model to determine description information contained in the information to be recommended and used for describing a service object corresponding to the information to be recommended, and determining user map features corresponding to the user, description map features corresponding to the description information and information map features corresponding to the information to be recommended through a pre-constructed feature map, wherein the feature map is used for representing historical interaction relations between the users and the historical recommendation information under the influence of different types of historical description information;

and recommending information to the user according to the first click rate.

2. The method according to claim 1, wherein constructing the feature profile specifically comprises:

3. The method of claim 2, wherein determining descriptive information that matches the historical recommendation information comprises:

4. The method of claim 3, wherein determining the descriptive information that matches the historical recommendation information based on the historical interaction records comprises:

determining other historical recommendation information of which the type is the same as that of the information to which the historical recommendation information belongs;

5. The method of claim 1, wherein prior to making information recommendations to the user based on the first click-through rate, the method further comprises:

6. The method of claim 5, wherein recommending information to the user according to the first click rate and the second click rate specifically comprises:

7. The method of claim 6, wherein determining the weight corresponding to the first click-through rate as a first weight and the weight corresponding to the second click-through rate as a second weight comprises:

and determining the second weight according to the first weight.

8. An apparatus for information recommendation, comprising:

the input module is used for inputting the user information and the information to be recommended into a pre-trained information recommendation model so as to determine description information which is contained in the information to be recommended and is used for describing a service object corresponding to the information to be recommended, and determining user map features corresponding to the user, description map features corresponding to the description information and information map features corresponding to the information to be recommended through a pre-constructed feature map, wherein the feature map is used for representing historical interaction relations between the users and the historical recommendation information under the influence of different types of historical description information;

9. A computer-readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1 to 7.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method of any of claims 1 to 7 when executing the program.