Heterogeneous graph-based virtual article time sequence recommendation method and device, medium and equipment

Abstract

The invention discloses a heterogeneous graph-based virtual article time sequence recommendation method and device, a medium and equipment. The method comprises the following steps: firstly, using GCN on a heterogeneous graph comprising three nodes of users, virtual articles and article attributes to obtain user representations and article representations; and then, respectively extracting a social sub-graph and an article association sub-graph from the heterogeneous graph, respectively obtaining user representation and article representation through GCN, and fusing the user representation and the article representation with user representation and article representation obtained in the heterogeneous graph to obtain final user representation and article representation. Inputting the obtained article representations into a sequential network according to a purchase time sequence to obtain candidate virtual articles; and finally, performing dot product on the user representation and the candidate virtual article representation to obtain a recommendation probability of each candidate article. According to the virtual article recommendation method, the relationship between the user and the virtual article in the spatial dimension and the relationship of purchase of the virtual article in the time dimension can be fully considered in the virtual article recommendation task, so that more accurate virtual article recommendation is realized.

Description

technical field [0001] The invention relates to the field of personalized recommendation, in particular to a time-series recommendation method based on heterogeneous graphs on item recommendation tasks. Background technique [0002] In recent years, as a technology that can capture contextual information, temporal model has attracted more and more researchers' attention. The optimal architecture designed by the timing model has shown great advantages in various tasks, such as natural language processing tasks, machine translation and other tasks. [0003] In the field of virtual item recommendation, it is very effective to use temporal models to capture user preferences in the temporal dimension. Taking the time series model Transformer as an example, Positional Embedding is introduced to represent the sequence relationship of virtual items, and the user's historical behavior information is modeled through the self-attention mechanism to obtain hidden user behavior patterns...

AI Technical Summary

Problems solved by technology

[0004] In practice, due to the large number of users and virtual items, the historical behavior data of users is very sparse, and only time-series models are used to captu

Images