Sequence recommendation method based on edge enhanced global decoupling graph neural network

Abstract

The invention discloses a sequence recommendation method based on an edge enhanced global decoupling graph neural network, and the method comprises the steps: an input layer: taking all items in training data as nodes, taking an interaction sequence between a user and the items as edges, constructing a global link graph, and inputting the global link graph to a graph neural network; the decoupling learning layer is used for aggregating influence probabilities of a preset number of factors from all neighbor projects of the project and influence probabilities of a preset number of factors from a plurality of previous projects of the project of the target user according to an interaction time sequence; according to the prediction layer, decoupling item representations of the global level and the local level are accumulated and multiplied by the initial embedding representations of the candidate items, and the probability that the target user appears as the next interaction item after the item vi is obtained; iteratively training to complete the graph neural network; the item with the maximum probability output by prediction of the trained graph neural network is the item recommended to the target user. The method has the beneficial effects of perceiving transfer of user intentions and recommending products to users more accurately.

Description

technical field [0001] The invention relates to the technical field of sequence recommendation. More specifically, the present invention relates to a sequential recommendation method based on edge-enhanced globally decoupled graph neural networks. Background technique [0002] Recommender systems play a vital role in the rapidly developing Internet era, and sequence recommendation is one of the important components. Sequence recommendation models user behavior as a sequence of items rather than a collection of items. Markov chains are a classic approach that models short-term item switching and predicts the next item a user is likely to like. With the development of deep learning networks, recurrent neural networks have achieved success in sequence recommendation. For example, long short-term memory networks, a common variant of recurrent neural networks, are used to enhance a model's ability to retain sequence information through memory cells. GRU4REC applies gated recu...

AI Technical Summary

Problems solved by technology

However, these studies did not consider the link relationship patterns of the items and failed to distinguish different user intentions behind the sequences

Therefore, the learned sequence model will be sensitive to noisy data and lack interpretability

Images