Research On Sequence Recommendation Method Based On Hybrid Neural Network

Based on the idea of sequence recommendation,aiming at the low accuracy of recommendation system,this paper improves the recommendation model in the project embedding stage and sequence modeling stage to improve the performance of recommendation.The specific problems and improvement methods are described as follows.In the process of user interaction sequence,the existing recommendation model only cares about the order of user interaction items when embedding,ignoring the correlation between user interaction sequence item attributes and content.This paper introduces the knowledge graph of related fields to obtain auxiliary information in the project embedding stage,and improves the conventional embedding method by enriching the form of user interaction information.The experimental results show that the performance of the embedding method is improved compared with the previous mainstream method and the method without data set related domain knowledge graph.In the sequence modeling stage,most of the existing models use recurrent neural network(RNN)to model the user interaction sequence.Compared with the recurrent neural network,the temporal convolutional network(TCN)has better parallelism,and has structural advantages,And in the field of natural language processing has proved to have a good effect.In the user's recent interaction stage,preferences change frequently.Because of its structural advantages,temporal convolution network is more conducive to user's recent interaction modeling.Therefore,this paper introduces temporal convolution network to model user recent interaction sequence to improve the accuracy of recent preference acquisition.In addition to the problem of modeling in the near future,in view of the problem that the accuracy of recommendation is low due to the lack of full consideration of user interaction sequence,this paper also adds long-term user interaction sequence in the modeling process,and fully considers the interaction sequence of each stage of the user to improve the performance of the recommendation model.Considering that long-term and short-term preferences of users are beneficial,but simple combination of user preferences in two stages will not have a good recommendation effect.Based on the consideration of users' short-term preference and long-term preference,this paper dynamically integrates the user's preferences of two interaction sequences,namely,the short-term and long-term,based on the attention mechanism,Thus,the accuracy of recommendation is improved.To sum up,in view of the shortcomings of the current recommendation methods mentioned above,this paper introduces the knowledge graph of related fields to enrich the interactive information in the embedding stage,introduces the temporal convolution network to improve the accuracy of user recent preference modeling,and adds the attention mechanism at the end of the model to dynamically fuse user preferences in each stage,so as to improve the performance of recommendation.Finally,in the experiment,experiments are carried out on the public data sets movielens and Last Fm in the field of film and music.The results show that the proposed model plays a role and has a good effect.