Search

CN-116257691-B - Recommendation method based on potential graph structure mining and user long-short-term interest fusion

CN116257691BCN 116257691 BCN116257691 BCN 116257691BCN-116257691-B

Abstract

The invention discloses a recommendation method based on potential diagram structure mining and user long-short-term interest fusion, which predicts the probability of clicking a target object by a user based on a historical behavior sequence of the user. The method mainly comprises seven parts, wherein a first part is used for dividing a user historical behavior sequence into a recent historical behavior sequence and a long-term historical behavior sequence, a second part is used for obtaining short-term interests of a user by utilizing a cyclic neural network based on the user recent historical behavior sequence, a third part is used for mining potential graph structures according to the similarity of objects in the user historical behavior sequence, a fourth part is used for updating object vectors by utilizing the graph neural network based on the mined graph structures, long-term interests of the user are obtained by utilizing the cyclic neural network based on the updated object vectors, a fifth part is used for fusing the short-term interests and the long-term interests of the user according to the interest diversity preference of the user, and a sixth part is used for predicting the click rate of the user on the objects.

Inventors

  • GU PAN
  • HU HAIYANG

Assignees

  • 杭州电子科技大学上虞科学与工程研究院有限公司

Dates

Publication Date
20260508
Application Date
20230412

Claims (9)

  1. 1. A recommendation method based on potential graph structure mining and user long-term interest fusion is characterized by comprising the following steps: S100, acquiring a user historical behavior sequence, and dividing the user historical behavior sequence into a recent historical behavior sequence and a distant historical behavior sequence according to the execution sequence of the user on the object; s200, modeling is conducted by using a cyclic neural network based on a recent historical behavior sequence of a user, so that short-term interests of the user are obtained; S300, mining a potential graph structure through a filtering operation formed by a plurality of circulating processes according to the similarity of the articles in the user history behavior sequence; S400, updating the object vector by using the graph neural network based on the mined graph structure, and obtaining the long-term interest of the user by using the long-term memory neural network based on the updated object vector, wherein the specific implementation method is as follows: s401, updating the article vector by using a graph neural network based on the mined graph structure, wherein the graph neural network is a multi-layer structure and is formed by total Layer messaging fuses user-remote behavioral information into which Nodes in the figure Is the first of (2) Layer article vector representation as The update formula is as follows: ; ; Wherein, the Representing pairs of adjacency matrices Regularizing; Is a graph structure Is calculated according to the degree matrix of (1) The degree matrix is a diagonal matrix; Is an adjacency matrix Regularized matrix, wherein the matrix elements of the ith row and the jth column are ; Representation of Is used to determine the neighbor set of a neighbor, Representation of Is the first of (2) Layer item vector representation; representing item nodes in a graph structure, nodes in the graph First, the Layer vector Initializing to Pass by After the information of the layers is transferred, the node is finally obtained The updated article vector is ; Representing the number of repeated cycles; s402, based on recent historical behavior sequence of user Is expressed as an updated item vector of (1) Obtaining the long-term interest of the user by using the long-term memory neural network Wherein ; S500, calculating the richness of the user behavior according to the diversity preference of the user interests, and combining the weights of the long-period interests obtained through learning to fuse the short-period interests and the long-period interests of the user so as to obtain the final user interest representation; S600, predicting the click rate of the user on the object by calculating the similarity between the user interest representation and the vector representation of the target object, and realizing object recommendation.
  2. 2. The recommendation method based on potential graph structure mining and user long-short term interest fusion according to claim 1, wherein the specific implementation method of S100 is as follows: acquiring user sets in a platform And an article collection Acquiring a user Historical behavior sequence of (a) Any of which Subscript of Representing a user Behavior sequence Is a length of (2); Is characterized by the vector of (a) Sequence of user historic behaviors Rear of (3) The recent historical behavior sequence of an item as a user is expressed as Front in the user history behavior sequence The long-term historical behavior sequence of the individual articles as users is expressed as 。
  3. 3. The recommendation method based on potential graph structure mining and user long-short term interest fusion according to claim 2, wherein the specific implementation method of S200 is as follows: For users A long-term and short-term memory neural network is adopted to carry out recent historical behavior sequence on users Modeling to obtain short-term interest of user : ; Wherein, the Is a sequence of recent historical behaviors of a user Is used in the vector representation of (a), The output of the last hidden state of the long-short-term memory neural network is the short-term interest of the user 。
  4. 4. The recommendation method based on potential graph structure mining and user long-short term interest fusion according to claim 3, wherein the specific implementation method of S300 is as follows: Calculating the similarity of cosine between the previous object and the next object in the user history behavior sequence to obtain an initial adjacency matrix of the graph structure Matrix of (a) Is a lower triangular matrix, when the ith row and the jth column in the matrix Subscript is satisfied In the time-course of which the first and second contact surfaces, ; Representing items in a user's historical behavior sequence And Is the similarity of the specific value of the articles And Vector characterization of (a) And Cosine similarity of (2); further to the initial adjacency matrix Filtering to obtain a sparse graph structure The filtering operation is a multiple cycle process: The first step, set the node set contained in the final graph structure as The set to be expanded is , And All initialized to ; Second, according to the initial adjacency matrix Obtaining Most similar item set for each item in a plurality Will be assembled The articles in the bag are put into the collection In, i.e And update Is that I.e. ; Third, repeating the second step K times to obtain a sparse graph structure And an adjacency matrix for the graph And corresponding adjacent matrix when repeating the second step each time in the multiple circulation processes The variation of (c) is expressed as: ; Wherein, the Representing a collection All of the articles in (a) , Representing found sum-of-items Most similar articles Then give the adjacency matrix Value assignment 。
  5. 5. The recommendation method based on potential graph structure mining and user long-short term interest fusion according to claim 1, wherein the specific implementation method of S500 is as follows: S501, counting the category number of category in the data set as The number of label types is The number of the behaviors of the user under a certain category or label is larger than As a criterion for determining that the user is interested in the category or label, counting the number of categories of interest to the user in the user history behavior sequence And number of tags Obtaining the richness of the user behavior The method comprises the following steps: ; Wherein, the And Is a super parameter, controls the importance degree of category and label information respectively, ; S502, learning the weight of the long-term interest by using a two-layer MLP model, and combining the richness of the user behavior The final long-term and short-term interest weight is obtained, and is specifically as follows: ; ; Wherein, the And Is the model parameter of the two-layer MLP model, is updated during training and is marked Is a transposed symbol; Representation of The function is activated and the function is activated, Representation of Activating a function; is the weight of the long-term interest of the user learned by the model itself via The function normalizes it to Within the range, then multiply by the user behavior richness Obtaining the long-term interest weight of the final user ; S503, based on long-term interest weight of user And user short-term interest weights Long-term interest to a user And short-term interest Fusion is carried out to obtain the final user interest representation 。
  6. 6. The recommendation method based on potential graph structure mining and user long-short term interest fusion according to claim 5, wherein the specific implementation method of S600 is as follows: According to the user Representation of user interest of (a) Representing user interests and target objects Vector representation of (a) Performing inner product calculation to predict target object of user Click rate of (2) : 。
  7. 7. The recommendation method based on potential graph structure mining and user long-short term interest fusion as set forth in claim 1, wherein the recommendation model framework formed by S100-S600 requires training in advance before actual reasoning, and the target object is obtained by a user during training Predicted click rate value of (2) Calculating click rate prediction value And click rate true value The cross entropy loss function between the two functions is used for guiding the updating process of the model parameters, and an Adam optimizer is used for updating the model parameters.
  8. 8. The recommendation method based on latent graph structure mining and user long-short term interest fusion according to claim 7, wherein the cross entropy loss function is calculated as: ; Wherein, the Is a true value representing whether the user clicked on the target item; is a sigmoid function.
  9. 9. The recommendation method based on latent image structure mining and user long-short term interest fusion according to claim 2, wherein the platform is a short video platform, the item is a short video, and Is initialized to a vector representation of the short video master graph.

Description

Recommendation method based on potential graph structure mining and user long-short-term interest fusion Technical Field The invention belongs to the technical field of internet service, and particularly relates to a recommendation method based on potential diagram structure mining and user long-term interest fusion. Background In recent years, domestic mobile netizens spend longer and longer on short video platforms. The duration of the short video is short, so that the user can watch many short videos in one day. In general, a user may browse different types of short videos in a short video platform, i.e., the user has a strong diversity preference in the short video platform. The historical sequence of the user is input into the short-term interest modeling model of the user by research, and the fact that the latest behaviors of the user are removed is found, the effect of the short-term interest modeling model is improved, and the strong diversity preference of the user is reflected to a certain extent. The conventional recommendation method generally inputs the latest behavior sequence of the user into the model to obtain the short-term interest of the user, and further recommends the short video of interest to the user. However, the behaviors of the user in the short video platform are very rich and dense, if the historical behavior sequence of the user in the long term is ignored, only the recent historical behavior sequence of the user is considered, rich information contained in the historical behavior sequence of the user in the long term can be lost, the recent interests of the user can be possibly over-fitted, and the recommendation result is single. If all the historical behavior sequences of the user are directly input into the interest modeling model, a large amount of noise contained in the historical sequence of the user long-term behavior can damage the recommending effect of the model, and the model has high complexity and long training and running time. Disclosure of Invention The problem of the method is defined as predicting the probability of a user clicking on a target item based on the user's behavior sequence. The mathematical symbols involved are represented by the user set in the platform asThe collection of items is represented as. User' sIs the historical behavior sequence of (1)Any of whichSubscript ofRepresenting a userBehavior sequenceIs a length of (c). The existing recommendation method generally inputs the latest behavior sequence of the user into the model to obtain the short-term interest of the user, and then recommends the short video of interest to the user. However, the behaviors of the user in the short video platform are very rich and dense, if the historical behavior sequence of the user in the long term is ignored, only the recent historical behavior sequence of the user is considered, rich information contained in the historical behavior sequence of the user in the long term can be lost, the recent interests of the user can be possibly over-fitted, and the recommendation result is single. If all the historical behavior sequences of the user are directly input into the interest modeling model, a large amount of noise contained in the historical sequence of the user long-term behavior can damage the recommending effect of the model, and the model has high complexity and long training and running time. For this purpose, the invention adopts the following technical scheme: A recommendation method based on potential graph structure mining and user long-term interest fusion comprises the following specific steps: S1, acquiring a user historical behavior sequence, and dividing the user historical behavior sequence into a recent historical behavior sequence and a distant historical behavior sequence according to the execution sequence of the user on the object; s2, modeling is conducted by using a cyclic neural network based on a recent historical behavior sequence of the user, so that short-term interests of the user are obtained; s3, mining a potential graph structure through a filtering operation formed by a plurality of circulating processes according to the similarity of the articles in the user history behavior sequence; s4, updating the object vector by using the graph neural network based on the mined graph structure, and obtaining the long-term interest of the user through the long-term memory neural network based on the updated object vector; S5, calculating the richness of the user behavior according to the diversity preference of the user interests, and combining the weights of the long-period interests obtained through learning to fuse the short-period interests and the long-period interests of the user so as to obtain the final user interest representation; s6, predicting the click rate of the user on the object by calculating the similarity between the user interest representation and the vector representation of the target object, and realizing object r