Optimization Algorithm Based On Collaborative Filtering Recommendation

With the rapid development of the Internet,information and data grow explosively.In face with massive data and information,how to quickly and accurately find the content that users are interested in has become an important issue to be solved urgently.The recommendation system was proposed in this context.The recommendation algorithm is the core of the recommendation system,it can construct the user interest model by mining users' behavior data,so as to actively recommend items to users.Collaborative filtering is the technology widely used in the recommendation system,the measurement of similarity between items and users is the key to collaborative filtering recommendation algorithm.It is impossible to accurately calculate similarity based on rare scoring data.Therefore,how to use the inner relevance between the items to improve the similarity between items and how to use the SimRank graph structure information to improve the similarity between users has become the focus of this paper.The main research work is as follows:(1)Aiming at the inaccuracy of the similarity calculation in the traditional collaborative filtering algorithms brought by the sparse scoring data.A collaborative filtering recommendation algorithm based on frequent itemsets mining is proposed.The algorithm uses Apriori algorithm to mine frequent itemsets on the transaction database,uses Jaccard similarity to measure the correlation between items in frequent itemsets,gets the similarity of items based on frequent itemsets,and then introduces weighting factors to synthesize the similarity with the similarity of the items calculated by the scoring data,and gets the overall similarity between items.The algorithm not only considers the inner correlation between items,but also considers the user's rating data to the items,and improves the recommendation quality of the recommendation algorithm.Experiments were carried out on three public datasets,MovieLens,FilmTrust and LastFM.The experimental results show that the proposed improved algorithm can achieve better recommendation results than the comparison algorithm on the public dataset.(2)A collaborative filtering recommendation algorithm based on improved SimRank is proposed.The algorithm optimizes the user-item bipartite graph,constructs the user node relationship graph based on the scoring data and user characteristics,solves the problem that the complex structure of the bipartite graph in the existing SimRank algorithm leads to high complexity of the algorithm,and considers different neighbors have different contributions to target user,the weighting factor is used to improve the SimRank method to more accurately measure the similarity between users.And uses the transitivity of SimRank algorithm,digs out more potential neighbors of target user,provides more kinds of recommended items,alleviates the problem of low accuracy and coverage in collaborative filtering algorithms caused by sparse data.Experiments were carried out on three public datasets,MovieLens,FilmTrust and LastFM,and the results demonstrate the effectiveness of the improved algorithm.