A Conversation Recommendation Method Based on Heterogeneous Graph of Multi-source Information

Abstract

The invention discloses a session recommendation method based on multi-source information heterogeneous graph. Based on the current session of the anonymous user, the method models the user's interest to recommend items that the user is most likely to be interested in in the next step. It mainly consists of four parts: the first part is to calculate the similarity between the current session and the sessions in the session set, recall similar neighbor session sets, and then sort and filter the transfer of items in the neighbor session set, and filter out the number of occurrences Less item transfer relationship, combined with the current session to construct a heterogeneous graph of two semantic edges; the second part is to use the multi-source information gate mechanism to select and fuse the two semantic neighbor node information of the node in the heterogeneous graph, Obtain the vector representation of the items in the current session; the third part is to characterize the item sequence in the current session to obtain the current user interest representation; the fourth part is to predict the most likely item that the user will be interested in in the next step based on the user interest representation.

Description

technical field [0001] The invention belongs to the technical field of Internet services, and in particular relates to a session recommendation method based on multi-source information heterogeneous graphs. Background technique [0002] Session (Session) is a user's interactive behavior within a period of time, and session-based recommendation is based on the current session to recommend the user's next click item. In actual scenarios, some users log in anonymously, and the user's historical interaction behavior data and user details cannot be obtained. Therefore, items of interest can only be recommended to the user based on the current session of the anonymous user. The traditional session recommendation method has an item-based collaborative filtering (Item-KNN) recommendation method, which recommends the most similar item to the user by calculating the similarity between the item in the candidate set and the item in the current session. In recent years, a session-based...

AI Technical Summary

Problems solved by technology

The number of edges between the two semantics in this heterogeneous graph is quite different, and the number of edges representing the general transfer relationship of items is often more than the number representing the transfer relationship of items in the current session, which brings difficulties to the transfer of node information in the graph.

Images