Short sequence recommendation method based on user repeated behavior pattern mining

Abstract

The invention discloses a short sequence recommendation method based on user repeated behavior pattern mining, and the method comprises the following steps: 1), carrying out the article representationlearning of a given user behavior sequence, and obtaining the feature representation of an article; 2) converting the behavior sequence of the user into a repeated behavior pattern sequence, and predicting the item selection probability of repeated behaviors: predicting the probability that each item in the sequence is re-clicked next time according to the repeated behavior pattern sequence; 3) article selection probability prediction of exploration behaviors: predicting the probability that each article which does not appear in the behavior sequence of the user is clicked in the next behavior; 4) calculating probability distribution of repeated behaviors and exploration behaviors of the user; and 5) according to the results of the steps 2) to 4), obtaining the probability that each article is clicked next time. According to the invention, the repeated behaviors of the user are modeled, and the long-distance dependency relationship in the sequence is captured, so that the accuracy ofrecommending articles to the user can be effectively improved.

Description

technical field [0001] The invention relates to the technical field of data mining, in particular to a short-sequence recommendation method based on user repetitive behavior pattern mining. Background technique [0002] Traditional short-sequence recommendation methods are usually based on sequential pattern mining or first-order Markov chains, but the former usually only focuses on items with high frequency of occurrence, while the latter only models short-term dependencies and ignores long-term dependencies, making it difficult to obtain accurate prediction results. With the advancement of deep learning, many deep neural network-based models are devoted to enhancing the effect of short sequence recommendation tasks, most of which are mainly based on recurrent neural network (Recurrent Neural Network) and self-attention mechanism (Self-Attention) to build architectures , although these models have shown some improvement in short sequence recommendation tasks, it is difficu...

AI Technical Summary

Problems solved by technology

[0002] Traditional short-sequence recommendation methods are usually based on sequential pattern mining or first-order Markov chains, but the former usually only focuses on items with high frequency of occurrence, while the latter only models short-term dependencies and ignores long-term dependencies, making it difficult to obtain accurate The predicted results of

With the advancement of deep learning, many deep neural network-based models are devoted to enhancing the effect of short sequence recommendation tasks, most of which are mainly based on recurrent neural network (Recurrent Neural Network) and self-attention mechanism (Self-Attention) to build architectures , although these models have shown some improvement in short sequence recommendation tasks, it is difficult for recurrent neural networks to pay attention to the dependencies between contexts, and the self-attention mechanism ignores the modeling of the time order of items, which still has obvious shortcomings

In addition, the current work ignores the mining of user behavior patterns, making it difficult to model users' repetitive behaviors, so it has great limitations.

Images