CN113205377A - Information recommendation method and device - Google Patents

Information recommendation method and device Download PDF

Info

Publication number
CN113205377A
CN113205377A CN202110368069.XA CN202110368069A CN113205377A CN 113205377 A CN113205377 A CN 113205377A CN 202110368069 A CN202110368069 A CN 202110368069A CN 113205377 A CN113205377 A CN 113205377A
Authority
CN
China
Prior art keywords
service
node
feature vector
user
vector corresponding
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Pending
Application number
CN202110368069.XA
Other languages
Chinese (zh)
Inventor
崔强
张亚峰
张庆
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Beijing Sankuai Online Technology Co Ltd
Original Assignee
Beijing Sankuai Online Technology Co Ltd
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Beijing Sankuai Online Technology Co Ltd filed Critical Beijing Sankuai Online Technology Co Ltd
Priority to CN202110368069.XA priority Critical patent/CN113205377A/en
Publication of CN113205377A publication Critical patent/CN113205377A/en
Pending legal-status Critical Current

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
    • G06Q30/00—Commerce
    • G06Q30/06—Buying, selling or leasing transactions
    • G06Q30/0601—Electronic shopping [e-shopping]
    • G06Q30/0631—Item recommendations
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation

Landscapes

  • Engineering & Computer Science (AREA)
  • Business, Economics & Management (AREA)
  • Theoretical Computer Science (AREA)
  • Physics & Mathematics (AREA)
  • Data Mining & Analysis (AREA)
  • Finance (AREA)
  • Accounting & Taxation (AREA)
  • General Physics & Mathematics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • General Engineering & Computer Science (AREA)
  • Evolutionary Biology (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Computation (AREA)
  • Development Economics (AREA)
  • Economics (AREA)
  • Marketing (AREA)
  • Strategic Management (AREA)
  • General Business, Economics & Management (AREA)
  • Information Retrieval, Db Structures And Fs Structures Therefor (AREA)

Abstract

The specification discloses an information recommendation method and device, which respond to a service request of a user, obtain a preset service topological graph, determine candidate recommended service objects for the service request, then, for each candidate recommended service object, input the candidate recommended service object and user information of the user into a pre-trained feature extraction model so that the feature extraction model determines a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph, and input the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain a recommendation degree corresponding to the candidate recommended service object, and perform information recommendation for the user according to the recommendation degree corresponding to each candidate recommended service object, therefore, information recommendation can be more accurately carried out on the user.

Description

Information recommendation method and device
Technical Field
The present specification relates to the field of machine learning technologies, and in particular, to a method and an apparatus for information recommendation.
Background
With the continuous development of information technology, a user can execute various services on a service platform, and accordingly, the service platform can recommend service objects under various services to the user, for example, in a take-out service, the service platform can recommend take-out food to the user, and in a consumption comment service, the service platform can recommend merchants in the area where the user is located and food under the merchants to the user.
In practical application, when information recommendation is performed on a service, it is often determined which service objects in the service are recommended to a user and the sequence of the service objects in the service recommended to the user only through the historical behavior of the user in the service, and information recommendation is performed in such a manner without considering the overall behavior of the user in a service platform, so that information recommendation performed in such a manner may be relatively inaccurate.
Therefore, how to accurately recommend information to a user is an urgent problem to be solved.
Disclosure of Invention
The present specification provides a method and an apparatus for information recommendation, which partially solve the above problems in the prior art.
The technical scheme adopted by the specification is as follows:
the present specification provides a method for information recommendation, including:
responding to a service request of a user, acquiring a preset service topological graph and determining candidate recommended service objects aiming at the service request, wherein the service topological graph is constructed according to historical service records of at least two services, each user node and each service object node are contained in the service topological graph, and if one user historically executes a service corresponding to one service object, the user node corresponding to the user in the service topological graph is connected with the service object node corresponding to the service object;
for each candidate recommended service object, inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph;
inputting the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object;
and recommending information to the user according to the recommendation degree corresponding to each candidate recommended service object.
Optionally, inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines, based on the service topology map, a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user, specifically including:
inputting the service attribute information corresponding to the candidate recommended service object and the user information of the user into the feature extraction model, so that the feature extraction model determines a service object node corresponding to the candidate recommended service object and a user node corresponding to the user from the service topological graph, and respectively uses the service object node corresponding to the candidate recommended service object and the user node corresponding to the user as target nodes;
and for each target node, determining a local topological graph of the target node in the service topological graph when the target node is used as the center of the service topological graph through the feature extraction model, and determining a corresponding node feature vector of the target node in the service topological graph according to the local topological graph.
Optionally, determining a node feature vector corresponding to the target node in the service topology according to the local topology, specifically including:
determining other nodes of the target node which do not exceed a set adjacency relation in the local topological graph;
determining an initial feature vector corresponding to the target node and initial feature vectors corresponding to other nodes through the feature extraction model;
aggregating the initial characteristic vectors corresponding to the other nodes to determine the aggregated characteristic vectors corresponding to the other nodes;
and determining a node characteristic vector corresponding to the target node in the service topological graph according to the initial characteristic vector corresponding to the target node and the aggregation characteristic vector.
Optionally, aggregating the initial feature vectors corresponding to the other nodes, and determining the aggregated feature vectors corresponding to the other nodes, specifically including:
determining a first-order adjacent node which has a first-order adjacent relation with the target node in the service topological graph from the other nodes;
determining attention weight between the target node and the first-order adjacent node according to the initial feature vector corresponding to the target node and the initial feature vector corresponding to the first-order adjacent node, wherein the initial feature vector corresponding to the first-order adjacent node is determined according to the initial feature vector corresponding to the node which has the first-order adjacent relationship with the first-order adjacent node in each other node;
and determining the aggregation feature vector according to the attention weight between the target node and the first-order adjacent node and the initial feature vector corresponding to the first-order adjacent node.
Optionally, determining an initial feature vector corresponding to the first-order adjacent node according to an initial feature vector corresponding to a node in the first-order adjacent relationship with the first-order adjacent node in the other nodes, specifically including:
determining a basic feature vector corresponding to the first-order adjacent node according to an initial feature vector corresponding to a node which has a first-order adjacent relationship with the first-order adjacent node in each other node;
and inputting the basic feature vector into a full connection layer contained in the feature extraction model to obtain an initial feature vector corresponding to the first-order adjacent node.
Optionally, determining the attention weight between the target node and the first-order neighboring node according to the initial feature vector corresponding to the target node and the initial feature vector corresponding to the first-order neighboring node, specifically including:
determining a number of iterations for the attention weight;
aiming at the Nth iteration, determining an initial feature vector corresponding to the target node in the Nth-1 th iteration according to the attention weight between the target node and the first-order adjacent node determined in the Nth-1 th iteration and the initial feature vector corresponding to the first-order adjacent node;
according to the initial feature vector corresponding to the target node in the N-1 th iteration and the initial feature vector corresponding to the first-order adjacent node, determining the attention weight between the target node and the first-order adjacent node in the N-th iteration, and according to the attention weight between the target node and the first-order adjacent node in the N-th iteration and the initial feature vector corresponding to the first-order adjacent node, determining the attention weight between the target node and the first-order adjacent node in the (N + 1) -th iteration until the iteration number is reached, wherein N is a positive integer.
Optionally, training the feature extraction model and the prediction model specifically includes:
obtaining a training sample, and determining at least one associated prediction model corresponding to the prediction model, wherein a service association exists between a first service corresponding to the prediction model and a second service corresponding to the at least one associated prediction model, the training sample comprises sample user information, a first historical service record and a second historical service record, the first historical service record is a historical service record of a user corresponding to the sample user information historically executing the first service aiming at the first historical service object, and the second historical service record is a historical service record of a user corresponding to the sample user information historically executing the second service aiming at the second historical service object;
inputting the sample user information, the first historical business object and the second historical business object into a feature extraction model to be trained, so that the feature extraction model determines a feature vector corresponding to the sample user information, a feature vector corresponding to the first historical business object and a feature vector corresponding to the second historical business object based on the business topological graph;
inputting the feature vector corresponding to the sample user information and the feature vector corresponding to the first historical service object into the prediction model to obtain a prediction recommendation degree corresponding to the first historical service object, and inputting the feature vector corresponding to the sample user information and the feature vector corresponding to the second historical service object into the at least one associated prediction model to obtain a prediction recommendation degree corresponding to the second historical service object;
and performing joint training on the feature extraction model, the prediction model and the at least one associated prediction model by taking the minimization of the deviation between the prediction recommendation degree corresponding to the first historical service object and the first historical service record and the minimization of the deviation between the prediction recommendation degree corresponding to the second historical service object and the second historical service record as optimization targets.
This specification provides an apparatus for information recommendation, including:
the response module is used for responding to a service request of a user, acquiring a preset service topological graph and determining each candidate recommended service object aiming at the service request, wherein the service topological graph is constructed according to historical service records of at least two services, each user node and each service object node are contained in the service topological graph, and if one user historically executes a service corresponding to one service object, the user node corresponding to the user in the service topological graph is connected with the service object node corresponding to the service object;
the input module is used for inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model aiming at each candidate recommended service object so as to enable the feature extraction model to determine a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph;
the prediction module is used for inputting the characteristic vector corresponding to the user and the characteristic vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object;
and the recommending module is used for recommending information to the user according to the recommending degree corresponding to each candidate recommended service object.
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 and apparatus provided in this specification, in response to a service request of a user, a preset service topology map is obtained, and candidate recommended service objects for the service request are determined, then, aiming at each candidate recommended service object, inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines the feature vector corresponding to the candidate recommended service object and the feature vector corresponding to the user based on the service topological graph, inputting the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object, and recommending information to the user according to the recommendation degree corresponding to each candidate recommended service object.
It can be seen from the above method that the method can show the records of execution of each user in the service platform in various services by means of the topological graph, and when information recommendation needs to be performed to the user, determine the feature vectors showing the user and the candidate recommended service object by means of the constructed service topological graph, so that when information recommendation is performed, the method can refer to the relevant information in the service corresponding to the candidate recommended service object, and also refer to the relevant information in other services, thereby performing information recommendation to the user more accurately compared with the prior art.
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 in this specification;
fig. 2 is a schematic diagram of a service topology provided in the present specification;
FIG. 3 is a schematic diagram of determining an aggregate feature vector 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 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 flow chart of an information recommendation method in this specification, which specifically includes the following steps:
s101: responding to a service request of a user, acquiring a preset service topological graph and determining candidate recommended service objects aiming at the service request, wherein the service topological graph is constructed according to historical service records of at least two services, each user node and each service object node are contained in the service topological graph, and if one user historically executes a service corresponding to one service object, the user node corresponding to the user in the service topological graph is connected with the service object node corresponding to the service object.
In practical application, the service platform may provide a plurality of services to the user, and accordingly, in each service, the service platform may recommend a service object under the service to the user, for example, in a take-out service, the service platform may recommend a take-out meal to the user, and the take-out meal may refer to the service object in the take-out service, and further, for example, in a consumption comment service, the service platform may recommend a restaurant to the user, and the restaurant may refer to the service object in the consumption comment service.
In the two examples, information such as taste habits and living areas of one user can be reflected in the two service scenes, so that when information is recommended in one service, if the information can be combined with service behaviors of the user in other services, the information can be recommended in the service more accurately.
Based on this, the information recommendation method provided in this specification may refer to the service behaviors of users in multiple services to perform information recommendation for one service, and specifically, after responding to a service request of a user, the service platform may obtain a preset service topology map and determine candidate recommended service objects for the service request. The service request mentioned here may be a service request sent to the service platform through the terminal when the user needs to view a service object recommended under a certain service of the service platform. Therefore, each candidate recommended service object for the service request determined by the service platform may be a service object under one class of services, that is, a service object in a service corresponding to the service request.
The service topology graph mentioned here can be constructed according to historical service records of at least two services, and the service topology graph includes user nodes and service object nodes, where if a user has historically executed a service corresponding to a service object, the user node corresponding to the user in the service topology graph is connected to the service object node corresponding to the service object, and the service topology graph is exemplified below, as shown in fig. 2.
Fig. 2 is a schematic diagram of a service topology provided in this specification.
In the service topology diagram in fig. 2, the historical service records of the user 1 and the user 2 in the service a, the service B, and the service C can be shown, and it can be seen that the user 1 has executed the service corresponding to the service object E under the service B, the service object C under the service a, and the service object D under the service C, respectively, and the user 2 has executed the service corresponding to the service object a under the service a, the service object B under the service C, and the service object D. By the service topological graph, records executed by each user under a plurality of services can be clearly and directly represented.
S102: and for each candidate recommended service object, inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph.
After determining each candidate recommended service object, the service platform may input the candidate recommended service object and user information corresponding to the user into a pre-trained feature extraction model for each candidate recommended service object, so that the feature extraction model determines a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph.
That is to say, the feature extraction model may determine the feature vector corresponding to the candidate recommended service object according to the service object node corresponding to the candidate recommended service object in the service topological graph, and determine the feature vector corresponding to the user according to the user node corresponding to the user. The feature vector corresponding to the candidate recommended service object mentioned herein can characterize the features of the candidate recommended service object in the at least two services, and the feature vector corresponding to the user can also characterize the behavior features of the user in the at least two services.
Specifically, the service platform may input the service attribute information corresponding to the candidate recommended service object and the user information of the user into the feature extraction model, where the feature extraction model may determine, from the service topology map, a service object node corresponding to the candidate recommended service object and a user node corresponding to the user, and respectively use the service object node corresponding to the candidate recommended service object and the user node corresponding to the user as target nodes, and, for each target node, when it is determined that the target node is used as the center of the service topology map through the feature extraction model, the target node is in a local topology map in the service topology map, and a node feature vector corresponding to the target node in the service topology map is determined according to the local topology map.
When the target node is used as the center of the service topological graph, the local topological graph of the target node in the service topological graph may refer to a local topological graph of a certain range defined by taking the target node as the center in the service topological graph, and the node feature vector corresponding to the target node determined by the local topological graph may represent the topological structure of the local topological graph taking the target node as the center.
For example, the service platform may determine that the target node does not exceed each other node in the local topological graph in a set adjacency relation, determine an initial feature vector corresponding to the target node and an initial feature vector corresponding to each other node through the feature extraction model, aggregate the initial feature vectors corresponding to each other node, determine an aggregated feature vector corresponding to each other node, and determine a node feature vector corresponding to the target node in the service topological graph according to the initial feature vector and the aggregated feature vector corresponding to the target node.
The setting of the adjacency relation mentioned here may be set according to actual needs. For example, if the set adjacency is set to a first-order adjacency, each other node is a node directly connected to the target node, and if the set adjacency is set to a second-order adjacency, each other node is a node having a first-order adjacency and a second-order adjacency with the node. If the set adjacency is set to be a larger adjacency, the number of other nodes may be larger.
The initial feature vector corresponding to each of the other nodes mentioned above is that each of the other nodes corresponds to an initial feature vector, and the aggregated feature vector corresponding to each of the other nodes aggregated by the initial feature vector corresponding to each of the other nodes is only one unique feature vector.
The initial feature vector corresponding to the target node may refer to a feature vector of the target node determined last time, that is, for the target node, the initial feature vector corresponding to the target node may be a last feature vector calculated by the service platform for the target node when information recommendation needs to be performed before. And when the feature extraction model is just trained, the feature vector of the target node can be initialized randomly.
There are various ways to determine the aggregated feature vector according to the initial feature vector corresponding to each other node. For example, the service platform may directly aggregate the initial feature vectors corresponding to the other nodes through a preset aggregation function to obtain an aggregated feature vector. For another example, the service platform may sequentially perform aggregation by the preset aggregation function from the node having the most adjacency with the target node from among other nodes that do not exceed the set adjacency, to obtain initial feature vectors of other nodes in each layer, so as to obtain the aggregated feature vector, where the other nodes in each layer are nodes having the same adjacency with the target node, as shown in fig. 3.
Fig. 3 is a schematic diagram of determining an aggregate feature vector provided in the present specification.
In fig. 3, a node B, C, D, E, F, G is another node in a local topological graph centered on a node a, which is not more than 2 th order adjacent relation to the node a, and assuming that the node a is a target node, a feature vector corresponding to the node a can be determined according to the local topological graph, first, initial feature vectors corresponding to the nodes F and G and initial feature vectors corresponding to the nodes D and E are determined, the initial feature vectors corresponding to the nodes F and G are aggregated to obtain an aggregated feature vector corresponding to the nodes F and G, then, an initial feature vector corresponding to the node B is determined according to the aggregated feature vector corresponding to the nodes F and G, and a manner of determining the initial feature vector of the node C is similar to that of the node B. The aggregation feature vector can be determined by the initial feature vectors corresponding to the node B and the node C, and the node feature vector corresponding to the node a can be determined by the aggregation feature vector and the initial feature vector corresponding to the node a. Of course, in this example, other nodes are merely used as examples of nodes that do not exceed the second-order adjacency with the target node, and in practical applications, other settings may be made to set the adjacency.
It can be seen that the initial feature vector corresponding to the first-order adjacent node of the target node is determined according to the initial feature vector corresponding to the node having the first-order adjacent relationship with the first-order adjacent node in each other node. The initial feature vector corresponding to the first-order adjacent node is determined according to the initial feature vector corresponding to the node having the first-order adjacent relationship with the first-order adjacent node in each other node, and the basic feature vector is input to the full connection layer included in the feature extraction model, so that the initial feature vector corresponding to the first-order adjacent node is obtained.
Specifically, the initial feature vectors corresponding to the nodes having the first-order adjacent relationship to the first-order adjacent node may be aggregated to obtain aggregated feature vectors, the basic feature vectors corresponding to the first-order adjacent node are determined according to the aggregated feature vectors and the initial feature vectors of the first-order adjacent node calculated last time, and the basic feature vectors are subjected to full-connection transformation to obtain the initial feature vectors corresponding to the first-order adjacent node. The reason why the initial feature vector is determined last time is that, in practical applications, for the same node, the node may be used multiple times to calculate the feature vector corresponding to the user or the business object when recommending information.
When the node feature vector corresponding to the node a is determined, the initial feature vectors of the node B and the node C may also be aggregated according to a preset aggregation function to obtain an aggregated feature vector corresponding to each other node, and then the node feature vector corresponding to the node a is obtained according to the aggregated feature vector and the initial feature vector of the node a. The node feature vector corresponding to the node a may be specifically calculated by the following formula:
OV＝max(V)
G＝ReLU(W×concat(q，OV)+w)
wherein V is the initial feature vector corresponding to each first-order adjacent node, OVIn order to aggregate the aggregation feature vector obtained after V, G is the node feature vector corresponding to the target node, W and W are model parameters in the feature extraction model, and max (V) means that the initial feature vectors corresponding to the first-order adjacent nodes are aggregated through the maximal pooling aggregation function, although other aggregation functions may be selected.
As can be seen from the above, when finally determining the aggregated feature vector, the initial feature vector corresponding to each first-order neighboring node of the target node is needed to determine, and the importance degree of each first-order neighboring node corresponding to the target node may be different from the target node.
Therefore, when determining the aggregated feature vector, the feature extraction model may determine each first-order neighboring node having a first-order neighboring relationship with the target node in the service topology map from the other nodes, determine an attention weight between the target node and each first-order neighboring node according to the initial feature vector corresponding to the target node and the initial feature vector corresponding to each first-order neighboring node, and finally determine the aggregated feature vector according to the attention weight between the target node and each first-order neighboring node and the initial feature vector corresponding to each first-order neighboring node.
The attention weight between the target node and the first order neighboring node can indicate the importance of the first order neighboring node relative to the target node. That is, when the first-order neighboring nodes are aggregated, the aggregation does not need to be performed by the aggregation function, but the initial feature vectors corresponding to the first-order neighboring nodes may be weighted and summed by the attention weight between the first-order neighboring nodes and the target node, so as to obtain the aggregated feature vector.
In this specification, there may be a plurality of ways to determine the attention weight, and in addition to the way to directly calculate the attention weight between the initial feature vector corresponding to the target node and the initial feature vector corresponding to each first-order neighboring node to obtain the attention weight between each first-order neighboring node and the target node, the feature extraction model may calculate the attention weight for the target node a plurality of times, and calculate the aggregated feature vector according to the attention weight determined at the last time, so that the determined attention weight corresponding to each first-order neighboring node can be more accurate.
Specifically, the feature extraction model may determine the number of iterations for the attention weight, determine, for the nth iteration, an initial feature vector corresponding to the target node in the nth-1 iteration according to the attention weight between the target node and each first-order adjacent node determined by the nth-1 iteration and the initial feature vector corresponding to the first-order adjacent node, determine, for the nth iteration, the attention weight between the target node and the first-order adjacent node in the nth iteration according to the initial feature vector corresponding to the target node in the nth-1 iteration and the initial feature vector corresponding to the first-order adjacent node, and determine, for the nth iteration, the attention weight between the target node and the first-order adjacent node in the nth +1 iteration according to the attention weight between the target node and the first-order adjacent node in the nth iteration and the initial feature vector corresponding to the first-order adjacent node, until the iteration number is reached, N is a positive integer.
That is to say, each time the attention weight between each first-order neighboring node and the target node is determined, each first-order neighboring node may be weighted and summed through the attention weight determined last time, and the initial feature vector corresponding to the target node is updated according to the result obtained by the weighted and summed, so as to obtain the initial feature vector corresponding to the target node in the current iteration, and the attention weight between each first-order neighboring node and the target node in the current iteration may be determined through the initial feature vector corresponding to the target node in the current iteration, and finally, the attention weight determined under the preset iteration number for the attention weight is taken, so as to calculate the aggregated feature vector.
If the attention weight between each first-order neighboring node and the target node in one iteration is calculated, the attention weight can be specifically calculated by the following formula:
OK＝KαN-1
GN＝R(GN-1+OK)
in the above formula, K may refer to an initial feature vector, α, corresponding to a first-order neighboring nodeN-1For attention weights, G, between the first-order neighboring nodes and the target node determined in the previous iterationN-1For the initial feature vector, G, corresponding to the target node determined in the previous iterationNFor the initial feature vector corresponding to the target node updated in the iteration, R is also a model parameter in the feature extraction model, and the initial feature vector is obtained through GNThe attention weight alpha in the iteration can be determined by KN. Of course, K may also be a basic feature vector corresponding to a first-order neighboring node before the full link layer, and G may be determined by the basic feature vectorNTo determine attention weights.
S103: and inputting the characteristic vector corresponding to the user and the characteristic vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object.
S104: and recommending information to the user according to the recommendation degree corresponding to each candidate recommended service object.
After determining the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object through the feature extraction model, the service platform may input the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object, and the service platform may recommend information to the user according to the recommendation degree corresponding to each candidate recommended service object.
It is mentioned here that the recommendation degree corresponding to the candidate recommended service object may be used to characterize the degree of interest of the user in recommending the candidate recommended service object to the user predicted by the prediction model. Therefore, if the recommendation degree corresponding to the candidate recommended service object is higher, the service platform is more likely to recommend the candidate recommended service object to the user.
It should be noted that the prediction model and the feature extraction model in this specification need to be trained in advance with supervision. Specifically, the service platform may obtain a training sample, and determine at least one associated prediction model corresponding to the prediction model, where a service association exists between a first service corresponding to the prediction model and a second service corresponding to the at least one associated prediction model, that is, each associated prediction model may correspond to a second service, and a certain relation may exist between the second service corresponding to the associated prediction model and the first service, for example, the second service corresponding to the associated prediction model may refer to a takeaway service, and the first service may refer to a consumption comment service, and both of the two services relate to food and drink, so that a certain service association is provided. Of course, the service association may not refer to similarity in service, and may be a second service associated with the first service presence service as long as the service is provided in the service platform.
The training sample comprises sample user information, a first historical service record and a second historical service record, wherein the first historical service record is a historical service record of a first service which is historically executed by a user corresponding to the sample user information and aims at the first historical service object, and the second historical service record is a historical service record of a second service which is historically executed by a user corresponding to the sample user information and aims at the second historical service object.
The service platform can input the sample user information, the first historical service object and the second historical service object into a feature extraction model to be trained, so that the feature extraction model determines a feature vector corresponding to the sample user information, a feature vector corresponding to the first historical service object and a feature vector corresponding to the second historical service object based on a service topological graph, and inputs the feature vector corresponding to the sample user information and the feature vector corresponding to the first historical service object into a prediction model to obtain the prediction recommendation degree corresponding to the first historical service object.
And finally, the service platform can take the minimized deviation between the predicted recommendation degree corresponding to the first historical service object and the first historical service record and the minimized deviation between the predicted recommendation degree corresponding to the second historical service object and the second historical service record as optimization targets, and perform joint training on the feature extraction model, the prediction model and the at least one associated prediction model.
It should be noted that the sample user information may include sample information of different users or only sample information of the same user, and if the sample user information includes sample information of different users, it is necessary that the prediction model and each associated prediction model correspond to sample information of one user, and the first historical service record corresponds to sample information of a user input into the prediction model, and the second historical service record corresponds to sample information of a user input into the associated prediction model. The first historical service record and the second historical record may indicate whether the user executes the service corresponding to the service object, may also indicate whether the user clicks on a link of the service object after recommending the service object to the user, and the like, and may specifically determine the required historical service record according to actual requirements.
The method can be seen that the service platform can show records of execution of each user in the service platform in various services in a topological graph mode, and when information recommendation needs to be performed on the user, the characteristic vectors showing the user and the candidate recommended service object are determined through the constructed service topological graph, so that when information recommendation is performed, related information in services corresponding to the candidate recommended service object can be referred to, related information under other services can also be referred to, and compared with the prior art, information recommendation can be performed on the user more accurately.
Based on the same idea, the present specification further provides a corresponding information recommendation apparatus, as shown in fig. 4.
Fig. 4 is a schematic diagram of an information recommendation apparatus provided in this specification, which specifically includes:
a response module 401, configured to obtain a preset service topology map in response to a service request of a user, and determine candidate recommended service objects for the service request, where the service topology map is constructed according to historical service records of at least two services, and the service topology map includes user nodes and service object nodes, where if a user has historically executed a service corresponding to a service object, a user node corresponding to the user in the service topology map is connected to a service object node corresponding to the service object;
an input module 402, configured to input, for each candidate recommended service object, the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines, based on the service topology map, a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user;
the prediction module 403 is configured to input the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain a recommendation degree corresponding to the candidate recommended service object;
and the recommending module 404 is configured to recommend information to the user according to the recommendation degree corresponding to each candidate recommended service object.
Optionally, the input module 402 is specifically configured to input the service attribute information corresponding to the candidate recommended service object and the user information of the user into the feature extraction model, so that the feature extraction model determines, from the service topology map, a service object node corresponding to the candidate recommended service object and a user node corresponding to the user, and uses the service object node corresponding to the candidate recommended service object and the user node corresponding to the user as target nodes, respectively; and for each target node, determining a local topological graph of the target node in the service topological graph when the target node is used as the center of the service topological graph through the feature extraction model, and determining a corresponding node feature vector of the target node in the service topological graph according to the local topological graph.
Optionally, the input module 402 is specifically configured to determine that the target node does not exceed each other node in the local topological graph, where an adjacency relationship is set; determining an initial feature vector corresponding to the target node and initial feature vectors corresponding to other nodes through the feature extraction model; aggregating the initial characteristic vectors corresponding to the other nodes to determine the aggregated characteristic vectors corresponding to the other nodes; and determining a node characteristic vector corresponding to the target node in the service topological graph according to the initial characteristic vector corresponding to the target node and the aggregation characteristic vector.
Optionally, the input module 402 is specifically configured to determine, from the other nodes, a first-order adjacent node that has a first-order adjacent relationship with the target node in the service topology map; determining attention weight between the target node and the first-order adjacent node according to the initial feature vector corresponding to the target node and the initial feature vector corresponding to the first-order adjacent node, wherein the initial feature vector corresponding to the first-order adjacent node is determined according to the initial feature vector corresponding to the node which has the first-order adjacent relationship with the first-order adjacent node in each other node; and determining the aggregation feature vector according to the attention weight between the target node and the first-order adjacent node and the initial feature vector corresponding to the first-order adjacent node.
Optionally, the input module 402 is specifically configured to determine, according to an initial feature vector corresponding to a node in the first-order adjacent node, which has a first-order adjacent relationship with the first-order adjacent node, a basic feature vector corresponding to the first-order adjacent node; and inputting the basic feature vector into a full connection layer contained in the feature extraction model to obtain an initial feature vector corresponding to the first-order adjacent node.
Optionally, the input module 402 is specifically configured to determine the number of iterations for the attention weight;
aiming at the Nth iteration, determining an initial feature vector corresponding to the target node in the Nth-1 th iteration according to the attention weight between the target node and the first-order adjacent node determined in the Nth-1 th iteration and the initial feature vector corresponding to the first-order adjacent node; according to the initial feature vector corresponding to the target node in the N-1 th iteration and the initial feature vector corresponding to the first-order adjacent node, determining the attention weight between the target node and the first-order adjacent node in the N-th iteration, and according to the attention weight between the target node and the first-order adjacent node in the N-th iteration and the initial feature vector corresponding to the first-order adjacent node, determining the attention weight between the target node and the first-order adjacent node in the (N + 1) -th iteration until the iteration number is reached, wherein N is a positive integer.
Optionally, the apparatus further comprises:
a training module 405, configured to obtain a training sample, and determine at least one associated prediction model corresponding to the prediction model, where a service association exists between a first service corresponding to the prediction model and a second service corresponding to the at least one associated prediction model, the training sample includes sample user information, a first historical service record, and a second historical service record, the first historical service record is a historical service record in which a user corresponding to the sample user information historically executes a first service for the first historical service object, and the second historical service record is a historical service record in which a user corresponding to the sample user information historically executes a second service for the second historical service object; inputting the sample user information, the first historical business object and the second historical business object into a feature extraction model to be trained, so that the feature extraction model determines a feature vector corresponding to the sample user information, a feature vector corresponding to the first historical business object and a feature vector corresponding to the second historical business object based on the business topological graph; inputting the feature vector corresponding to the sample user information and the feature vector corresponding to the first historical service object into the prediction model to obtain a prediction recommendation degree corresponding to the first historical service object, and inputting the feature vector corresponding to the sample user information and the feature vector corresponding to the second historical service object into the at least one associated prediction model to obtain a prediction recommendation degree corresponding to the second historical service object; and performing joint training on the feature extraction model, the prediction model and the at least one associated prediction model by taking the minimization of the deviation between the prediction recommendation degree corresponding to the first historical service object and the first historical service record and the minimization of the deviation between the prediction recommendation degree corresponding to the second historical service object and the second historical service record as optimization targets.
The present specification also provides a computer-readable storage medium storing a computer program, which can be used to execute the information recommendation method shown in fig. 1.
This specification also provides a schematic block diagram of the electronic device 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, but may also include hardware required for 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, the present specification does not exclude other implementations, such as logic devices or a combination of software and hardware, and the like, 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 for 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 invention may be provided as a method, system, or computer program product. Accordingly, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present invention 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 present invention is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. 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 non-transitory and non-transitory, removable and non-removable media, may implement 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 Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium that 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.
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.
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 specification, and is not intended to limit the present specification. 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 (10)

1. A method for information recommendation, comprising:
responding to a service request of a user, acquiring a preset service topological graph and determining candidate recommended service objects aiming at the service request, wherein the service topological graph is constructed according to historical service records of at least two services, each user node and each service object node are contained in the service topological graph, and if one user historically executes a service corresponding to one service object, the user node corresponding to the user in the service topological graph is connected with the service object node corresponding to the service object;
for each candidate recommended service object, inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph;
inputting the feature vector corresponding to the user and the feature vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object;
and recommending information to the user according to the recommendation degree corresponding to each candidate recommended service object.
2. The method of claim 1, wherein inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model, so that the feature extraction model determines a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topology map, specifically comprises:
inputting the service attribute information corresponding to the candidate recommended service object and the user information of the user into the feature extraction model, so that the feature extraction model determines a service object node corresponding to the candidate recommended service object and a user node corresponding to the user from the service topological graph, and respectively uses the service object node corresponding to the candidate recommended service object and the user node corresponding to the user as target nodes;
and for each target node, determining a local topological graph of the target node in the service topological graph when the target node is used as the center of the service topological graph through the feature extraction model, and determining a corresponding node feature vector of the target node in the service topological graph according to the local topological graph.
3. The method according to claim 2, wherein determining a node feature vector corresponding to the target node in the service topology according to the local topology specifically includes:
determining other nodes of the target node which do not exceed a set adjacency relation in the local topological graph;
determining an initial feature vector corresponding to the target node and initial feature vectors corresponding to other nodes through the feature extraction model;
aggregating the initial characteristic vectors corresponding to the other nodes to determine the aggregated characteristic vectors corresponding to the other nodes;
and determining a node characteristic vector corresponding to the target node in the service topological graph according to the initial characteristic vector corresponding to the target node and the aggregation characteristic vector.
4. The method according to claim 3, wherein aggregating the initial feature vectors corresponding to the other nodes and determining the aggregated feature vectors corresponding to the other nodes specifically comprises:
determining a first-order adjacent node which has a first-order adjacent relation with the target node in the service topological graph from the other nodes;
determining attention weight between the target node and the first-order adjacent node according to the initial feature vector corresponding to the target node and the initial feature vector corresponding to the first-order adjacent node, wherein the initial feature vector corresponding to the first-order adjacent node is determined according to the initial feature vector corresponding to the node which has the first-order adjacent relationship with the first-order adjacent node in each other node;
and determining the aggregation feature vector according to the attention weight between the target node and the first-order adjacent node and the initial feature vector corresponding to the first-order adjacent node.
5. The method according to claim 4, wherein determining the initial feature vector corresponding to the first-order neighboring node according to the initial feature vector corresponding to the node having the first-order neighboring relationship with the first-order neighboring node in the other nodes specifically includes:
determining a basic feature vector corresponding to the first-order adjacent node according to an initial feature vector corresponding to a node which has a first-order adjacent relationship with the first-order adjacent node in each other node;
and inputting the basic feature vector into a full connection layer contained in the feature extraction model to obtain an initial feature vector corresponding to the first-order adjacent node.
6. The method of claim 4, wherein determining the attention weight between the target node and the first-order neighboring node according to the initial feature vector corresponding to the target node and the initial feature vector corresponding to the first-order neighboring node comprises:
determining a number of iterations for the attention weight;
aiming at the Nth iteration, determining an initial feature vector corresponding to the target node in the Nth-1 th iteration according to the attention weight between the target node and the first-order adjacent node determined in the Nth-1 th iteration and the initial feature vector corresponding to the first-order adjacent node;
according to the initial feature vector corresponding to the target node in the N-1 th iteration and the initial feature vector corresponding to the first-order adjacent node, determining the attention weight between the target node and the first-order adjacent node in the N-th iteration, and according to the attention weight between the target node and the first-order adjacent node in the N-th iteration and the initial feature vector corresponding to the first-order adjacent node, determining the attention weight between the target node and the first-order adjacent node in the (N + 1) -th iteration until the iteration number is reached, wherein N is a positive integer.
7. The method of claim 1, wherein training the feature extraction model and the prediction model comprises:
obtaining a training sample, and determining at least one associated prediction model corresponding to the prediction model, wherein a service association exists between a first service corresponding to the prediction model and a second service corresponding to the at least one associated prediction model, the training sample comprises sample user information, a first historical service record and a second historical service record, the first historical service record is a historical service record of a user corresponding to the sample user information historically executing the first service aiming at the first historical service object, and the second historical service record is a historical service record of a user corresponding to the sample user information historically executing the second service aiming at the second historical service object;
inputting the sample user information, the first historical business object and the second historical business object into a feature extraction model to be trained, so that the feature extraction model determines a feature vector corresponding to the sample user information, a feature vector corresponding to the first historical business object and a feature vector corresponding to the second historical business object based on the business topological graph;
inputting the feature vector corresponding to the sample user information and the feature vector corresponding to the first historical service object into the prediction model to obtain a prediction recommendation degree corresponding to the first historical service object, and inputting the feature vector corresponding to the sample user information and the feature vector corresponding to the second historical service object into the at least one associated prediction model to obtain a prediction recommendation degree corresponding to the second historical service object;
and performing joint training on the feature extraction model, the prediction model and the at least one associated prediction model by taking the minimization of the deviation between the prediction recommendation degree corresponding to the first historical service object and the first historical service record and the minimization of the deviation between the prediction recommendation degree corresponding to the second historical service object and the second historical service record as optimization targets.
8. An apparatus for model training, comprising:
the response module is used for responding to a service request of a user, acquiring a preset service topological graph and determining each candidate recommended service object aiming at the service request, wherein the service topological graph is constructed according to historical service records of at least two services, each user node and each service object node are contained in the service topological graph, and if one user historically executes a service corresponding to one service object, the user node corresponding to the user in the service topological graph is connected with the service object node corresponding to the service object;
the input module is used for inputting the candidate recommended service object and the user information of the user into a pre-trained feature extraction model aiming at each candidate recommended service object so as to enable the feature extraction model to determine a feature vector corresponding to the candidate recommended service object and a feature vector corresponding to the user based on the service topological graph;
the prediction module is used for inputting the characteristic vector corresponding to the user and the characteristic vector corresponding to the candidate recommended service object into a pre-trained prediction model to obtain the recommendation degree corresponding to the candidate recommended service object;
and the recommending module is used for recommending information to the user according to the recommending degree corresponding to each candidate recommended service object.
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.
CN202110368069.XA 2021-04-06 2021-04-06 Information recommendation method and device Pending CN113205377A (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN202110368069.XA CN113205377A (en) 2021-04-06 2021-04-06 Information recommendation method and device

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN202110368069.XA CN113205377A (en) 2021-04-06 2021-04-06 Information recommendation method and device

Publications (1)

Publication Number Publication Date
CN113205377A true CN113205377A (en) 2021-08-03

Family

ID=77026193

Family Applications (1)

Application Number Title Priority Date Filing Date
CN202110368069.XA Pending CN113205377A (en) 2021-04-06 2021-04-06 Information recommendation method and device

Country Status (1)

Country Link
CN (1) CN113205377A (en)

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115018009A (en) * 2022-07-07 2022-09-06 北京百度网讯科技有限公司 Object description method, and network model training method and device

Cited By (1)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN115018009A (en) * 2022-07-07 2022-09-06 北京百度网讯科技有限公司 Object description method, and network model training method and device

Similar Documents

Publication Publication Date Title
CN108537568B (en) Information recommendation method and device
CN113011483B (en) Method and device for model training and business processing
CN110020427B (en) Policy determination method and device
CN110378400B (en) Model training method and device for image recognition
CN113688313A (en) Training method of prediction model, information pushing method and device
CN113344098A (en) Model training method and device
CN115187311A (en) Shop site selection method and device suitable for multiple industries
CN113325855B (en) Model training method for predicting obstacle trajectory based on migration scene
CN113641894A (en) Information recommendation method and device
CN113205377A (en) Information recommendation method and device
CN113343085B (en) Information recommendation method and device, storage medium and electronic equipment
CN116308738A (en) Model training method, business wind control method and device
CN115618748A (en) Model optimization method, device, equipment and storage medium
CN114996570A (en) Information recommendation method and device
CN115168726A (en) Model training method and information recommendation method
CN114860967A (en) Model training method, information recommendation method and device
CN114331602A (en) Model training method based on transfer learning, information recommendation method and device
CN114119139A (en) Information recommendation method and device, storage medium and electronic equipment
CN114120273A (en) Model training method and device
CN112417275A (en) Information providing method, device storage medium and electronic equipment
CN114116816A (en) Recommendation method and device
CN113010564B (en) Model training and information recommendation method and device
CN113641892A (en) Information recommendation method and device
CN113010563B (en) Model training and information recommendation method and device
CN114297491A (en) Model training method, information recommendation method and device

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication