Design And Implementation Of Commodity Recommendation Algorithm Based On Sequence Data

In real life,the recommendation system often faces such a problem,which is only based on short-term session data(such as small news websites),rather than long-term user records(such as Taobao,Jingdong,etc.).In this case,the common matrix decomposition method is not accurate.Recently,Recurrent Neural Network(RNN)has been widely applied in recommendation based on sequence data.However,RNN also has its shortcomings in the recommendation based on sequence data,because it only considers the sequence information,but does not consider other information(such as user preference information,association between items,time length information).In order to make up for some shortcomings of RNN,this paper adopts a hybrid recommendation method,which takes the correlation between items into account.Based on the idea of Item Based Collaborative Filtering(Item-based CF)recommendation algorithm,the similarity in sequence data is calculated,and the similarity information is integrated into the existing RNN network,so as to improve the overall performance of RNN.Through analysis,the similarity of item-based CF can't be embedded in RNN cell very well.And in the existing hybrid strategy,only the parallel hybrid strategy can mix the recommended results,the other hybrid strategies need a certain coupling between the algorithms.The parallel hybrid strategy includes mixed,switch,weighted and dynamic weighted.Because RNN performs far better than item-based CF in the sequence data recommendation,the mixed strategy is abandoned and the rest strategies are adopted in this paper.The experiment on Recsys Challenge 2015 choses Recall and Mean Reciprocal Rank(MRR)as evaluation indexes.Item-based CF is introduced to get the similarity between items,and RNN is trained to get the fitted network model.The hybrid recommendation method is adopted to get the end result.The result shows that the proposed method improves the performance of the network.