CN-121998738-A - Method, device, computer equipment and storage medium for determining item recommendation strategy

CN121998738ACN 121998738 ACN121998738 ACN 121998738ACN-121998738-A

Abstract

The application relates to a method, a device, computer equipment and a storage medium for determining an item recommendation strategy. The method comprises the steps of determining four-dimensional transaction behavior characteristics of a target user according to historical transaction data of the target user initiating a transaction article recommendation request, carrying out standardization processing on the four-dimensional transaction behavior characteristics, determining standardization characteristics, constructing a standardization matrix according to the standardization characteristics, training an initialized neural network based on the standardization matrix, determining a target neural network, determining an initial clustering centroid set according to a weight vector of each grid node of an output layer of the target neural network, taking the initial clustering centroid set as a starting point of a K-means algorithm, carrying out K-means clustering on the standardization matrix through the K-means algorithm, determining a clustering result, determining a user characteristic label based on the clustering result, and determining an article recommendation strategy corresponding to the target user according to the user characteristic label. By the aid of the scheme, individuation and reliability of the article recommendation strategy are improved.

Inventors

WANG JIANFEI
GUO ZHENGJUN
YAO JIAMING
CAI XIAOHONG
CHEN WEN

Assignees

浙江中烟工业有限责任公司

Dates

Publication Date: 20260508
Application Date: 20260123

Claims (10)

1. A method for determining an item recommendation policy, comprising: Determining four-dimensional transaction behavior characteristics of a target user according to historical transaction data of the target user initiating a transaction object recommendation request, wherein the four-dimensional transaction behavior characteristics comprise a latest transaction time ratio, a transaction frequency ratio, an accumulated transaction credential ratio and a transaction category ratio of the target user; carrying out standardization processing on the four-dimensional transaction behavior characteristics, determining standardization characteristics, and constructing a standardization matrix according to the standardization characteristics; Training the initialized neural network based on the standardized matrix to determine a target neural network; Determining an initial cluster centroid set according to the weight vector of each grid node of the target neural network output layer; Taking the initial clustering centroid set as a starting point of a K-means algorithm, and carrying out K-means clustering on the standardized matrix through the K-means algorithm to determine a clustering result, wherein the clustering result comprises cluster labels of the standardized matrix and final centroid coordinates of each cluster; and determining a user characteristic label of the target user based on the clustering result, and determining an article recommendation strategy corresponding to the target user according to the user characteristic label.
2. The method as recited in claim 1, further comprising: Evaluating the standardized matrix by adopting a contour coefficient method, and determining an optimal cluster number; and determining an output layer grid structure of the self-organizing map neural network according to the optimal cluster number, so as to construct an initialization neural network.
3. The method of claim 2, wherein evaluating the normalized matrix using a contour coefficient method to determine the optimal number of clusters comprises: Determining an intra-cluster average distance and an inter-cluster minimum average distance of the standardized matrix; determining a contour coefficient corresponding to the clustering number of the standardized matrix according to the average distance in the clusters and the minimum average distance among the clusters; And determining the optimal cluster number according to the average coefficient value of the contour coefficient.
4. The method of claim 2, wherein the training the initializing neural network based on the normalization matrix to determine a target neural network comprises: Constructing a node weight matrix of an initialized neural network output layer based on the characteristic dimension of the standardized matrix; determining Euclidean distance between each standardized feature in the standardized matrix and each node weight vector in the node weight matrix, and determining a target node from the node weight vectors according to the Euclidean distance; Updating the target node and node weight vectors in the target node neighborhood range based on a preset Gaussian neighborhood function and the self-adaptive learning rate of the initialized neural network, and determining candidate neural networks; Determining whether the maximum variation of the node weight vector in the node weight matrix of the output layer of the candidate neural network is smaller than or equal to a preset variation threshold; if yes, determining the candidate neural network as a target neural network.
5. The method of claim 1, wherein the determining the clustering result by K-means clustering the standardized matrix by the K-means algorithm using the initial cluster centroid set as a starting point of the K-means algorithm comprises: Taking the initial cluster centroid set as a starting point of a K-means algorithm, and determining the nearest centroid of each standardized feature according to the distance between each standardized feature in the standardized matrix and the initial cluster centroid in the initial cluster centroid set through the K-means algorithm; And respectively distributing the standardized features to clusters to which the nearest centroid belongs according to the nearest centroid of the standardized features, and determining a clustering result.
6. The method of claim 1, wherein the determining the user feature label of the target user based on the clustering result comprises: based on a standardized rule and service definitions corresponding to feature dimensions of the four-dimensional transaction behavior feature, analyzing a centroid value of a final centroid coordinate of each cluster on each feature dimension; And determining the user characteristic label of the target user according to the centroid value, the absolute value of the centroid value and a service definition preset standardized mapping rule based on each characteristic dimension of the four-dimensional transaction behavior characteristic.
7. The method of claim 1, wherein determining four-dimensional transaction behavioral characteristics of the target user based on historical transaction data of the target user initiating the transaction item recommendation request comprises: Collecting historical transaction data of a target user initiating a transaction article recommendation request, and performing data cleaning on the historical transaction data to determine effective transaction data, wherein the historical transaction data comprises a user identification code of the target user, a transaction timestamp, a transaction article name and a user transaction credential; Based on the valid transaction data, four-dimensional transaction behavioral characteristics of the target user are determined.
8. An apparatus for determining an item recommendation policy, wherein the apparatus for determining an item recommendation policy includes: The four-dimensional transaction behavior feature comprises a latest transaction time ratio, a transaction frequency ratio, an accumulated transaction credential ratio and a transaction category ratio of the target user; The standardized matrix determining module is used for carrying out standardized processing on the four-dimensional transaction behavior characteristics, determining standardized characteristics and constructing a standardized matrix according to the standardized characteristics; The neural network training module is used for training the initialized neural network based on the standardized matrix and determining a target neural network; The initial cluster centroid determining module is used for determining an initial cluster centroid set according to the weight vector of each grid node of the target neural network output layer; the centroid coordinate determining module is used for taking the initial clustering centroid set as a starting point of a K-means algorithm, and carrying out K-means clustering on the standardized matrix through the K-means algorithm to determine a clustering result, wherein the clustering result comprises cluster labels of the standardized matrix and final centroid coordinates of each cluster; and the article recommendation strategy determining module is used for determining the user characteristic label of the target user based on the clustering result and determining the article recommendation strategy corresponding to the target user according to the user characteristic label.
9. A computer device comprising a memory and a processor, the memory storing a computer program, characterized in that the processor implements the steps of the method of any one of claims 1 to 7 when the computer program is executed.
10. A computer readable storage medium, on which a computer program is stored, characterized in that the computer program, when being executed by a processor, implements the steps of the method of any one of claims 1 to 7.

Description

Method, device, computer equipment and storage medium for determining item recommendation strategy Technical Field The present application relates to the field of computer technologies, and in particular, to a method and apparatus for determining an item recommendation policy, a computer device, and a storage medium. Background Along with the gradual trend of market competition and the increasing diversification of user demands on personalized recommendation, the traditional empirical recommendation mode is difficult to meet the demands of accurate pushing and differentiated recommendation strategies, the current object recommendation field based on user group subdivision mainly faces a plurality of technical bottlenecks, the traditional K-means algorithm is sensitive to initial centroids and easy to fall into local optimal solutions, clustering precision is obviously reduced when high-dimensional user behavior data are processed, accuracy of recommending user group division is directly affected, SOM neural networks can effectively maintain a topological structure of the user behavior data, but clustering boundaries are fuzzy and difficult to directly use for refined recommendation group division, an existing recommendation subdivision model is difficult to support accurate object recommendation, most clustering methods need to preset clustering number K values, and the traditional method depending on experience or trial-error lacks objective quantification basis, so that stability and interpretability of subdivision results are poor, consistency and reliability of recommendation strategies are affected. Therefore, how to improve the recommendation accuracy and reliability of the article is a problem to be solved. Disclosure of Invention In view of the foregoing, it is desirable to provide a method, an apparatus, a computer device, and a storage medium for determining an item recommendation policy that can improve the accuracy and reliability of item recommendation. In a first aspect, the present application provides a method for determining an item recommendation policy, the method comprising: Determining four-dimensional transaction behavior characteristics of a target user according to historical transaction data of the target user initiating a transaction object recommendation request, wherein the four-dimensional transaction behavior characteristics comprise a latest transaction time ratio, a transaction frequency ratio, an accumulated transaction credential ratio and a transaction category ratio of the target user; carrying out standardization processing on the four-dimensional transaction behavior characteristics, determining standardization characteristics, and constructing a standardization matrix according to the standardization characteristics; Training the initialized neural network based on the standardized matrix to determine a target neural network; Determining an initial cluster centroid set according to the weight vector of each grid node of the target neural network output layer; Taking the initial clustering centroid set as a starting point of a K-means algorithm, and carrying out K-means clustering on the standardized matrix through the K-means algorithm to determine a clustering result, wherein the clustering result comprises cluster labels of the standardized matrix and final centroid coordinates of each cluster; and determining a user characteristic label of the target user based on the clustering result, and determining an article recommendation strategy corresponding to the target user according to the user characteristic label. In one embodiment, the method for determining an item recommendation policy further includes: Evaluating the standardized matrix by adopting a contour coefficient method, and determining an optimal cluster number; and determining an output layer grid structure of the self-organizing map neural network according to the optimal cluster number, so as to construct an initialization neural network. In one embodiment, evaluating the normalized matrix by using a contour coefficient method to determine an optimal cluster number includes: Determining an intra-cluster average distance and an inter-cluster minimum average distance of the standardized matrix; determining a contour coefficient corresponding to the clustering number of the standardized matrix according to the average distance in the clusters and the minimum average distance among the clusters; And determining the optimal cluster number according to the average coefficient value of the contour coefficient. In one embodiment, training the initializing neural network based on the normalization matrix, determining the target neural network includes: Constructing a node weight matrix of an initialized neural network output layer based on the characteristic dimension of the standardized matrix; determining Euclidean distance between each standardized feature in the standardized matrix and each node weight vecto