Research On Personalized Recommendation Algorithm For Community Group Buying Based On Big Data

At present,personalized recommendation system plays a very important role in both e-commerce and social network.Compared with the traditional e-commerce model,the community group buying model focuses on the time when the user purchases the commodities and the user’s social relationship.However,traditional community group buying platforms have not adopted effective methods to analyze the importance of time and social data for recommendation.The massive data brings storage and computing pressure to the traditional group buying platforms.In the context of big data,a research on personalized recommendation of community group buying based on deep learning is carried out.The research contents of the thesis are as follows:(1)A personalized recommendation algorithm RTSA in community group buying that integrates time-aware GRU and self-attention is proposed.In view of the fact that the recurrent neural network cannot consider the influence of time information,a time-aware GRU module is constructed.Further,irrelevant commodity data in the user-commodity interaction sequence always generate noise and other issues.Self-attention is adopted to realize the recommendation of a single user in the community group buying scenario.Comparative experiments are carried out using the Movie Lens-1M and Amazon Beauty datasets.The experimental results show that RTSA is superior than FPMC,GRU4 Rec and other recommendation algorithms.The effectiveness of the RTSA recommendation model is verified.(2)A personalized recommendation algorithm SSAGR in community group buying that integrates social network and hierarchical self-attention is proposed.First,a recurrent neural network is used to capture the complex potential interests of users in group buying over time.Then,considering that the predefined fusion strategy in group recommendation cannot dynamically obtain group user weights,a hierarchical self-attention network is used to realize a dynamic group preference aggregation strategy.Finally,for the sparsity problem of group-item interaction data,neural collaborative filtering is used to mine group-item interactions to complete the recommendation of group in the community group buying scenario.Comparative experiments are carried out using the Ma Feng Wo and Douban Book datasets.The results show that SSAGR is better than AGREE and other group recommendation algorithms.The effectiveness of the SSAGR recommendation model is verified.(3)The parallelization of RTSA and SSAGR recommendation algorithm based on Spark platform is realized.First,Spark big data cluster consisting of 1 Master node and 3 Worker nodes is built.Then,the parallelization of RTSA and SSAGR neural network models is realized by using the data parallelization strategy.Further,in different datasets and different nodes,comparative experiment are conducted using the speedup ratio and scalability as evaluation indicators.The experimental results show that for large-scale datasets,parallelizing neural network models is more efficient and has good scalability.(4)A prototype system for community group buying recommendation is built.Facing the application scenario of community group buying recommendation in the context of big data,this thesis designes and implementes a prototype system for community group buying recommendation based on the two implemented recommendation models RTSA and SSAGR,combined with the Pytorch deep learning framework.