Research On Hybrid Recommendation Algorithm Based On Multi-feature Fusion

In recent years,with the vigorous development of network technology,human beings are gradually changing from a period of information shortage to a period of information overload.In order to dig out the information that meets the needs of users from the massive data more accurately and efficiently,the recommendation algorithm came into being,and gradually replaced the search engine as a more popular solution.Through a large number of research and experimental verification,although the recommendation algorithms are very mature,there is still room for improvement.As one of the most widely used algorithms,collaborative filtering recommendation algorithm usually has the following problems:(1)Improvement and optimization of the algorithm in the scenario of data sparsity;(2)User similarity calculation is not ideal,and it is easy to cause deviations in the selection of neighboring users;(3)The user's cold start problem.Aiming at these problems,this subject makes further improvements to the recommendation algorithm after in-depth research.The main research contents are as follows:1.In view of data sparsity,since the current matrix filling methods only consider the score and do not pay attention to the own attributes of the item,the recommendation accuracy is not high.This subject improves on this problem.,firstly,the user's ratings and preferences for item attributes are calculated based on the ratings and the item's own attribute characteristics,and then the results are used to complete the missing items in the original matrix,which alleviates the problem of low accuracy due to missing data.2.Aiming at the problem of low accuracy caused by unsatisfactory similarity calculation results in CFRA,this subject considers a variety of features that affect the similarity calculation,and proposes the similarity of multi-feature fusion.First,according to the user's preference for attributes,the similarity based on item attributes is obtained.Then,according to the impact of user activity,common scores and item popularity on user similarity,the score-based Pearson similarity calculation is performed.Finally,the improved Pearson similarity based on score and the similarity based on item attribute features are fused by weighting factors,and the CFRA of multifeature fusion is improved.3.Aiming at the problem of user cold start,this subject considers the use of user attribute features to improve the recommendation algorithm,and uses the user's age,gender and occupation information to calculate the new user's neighbors and old user sets to achieve recommendation.Finally,the multi-feature fusion hybrid recommendation algorithm(MFFHRA)is obtained by combining the multi-feature fusion CFRA and the user attribute-based recommendation algorithm according to whether the user is a new user,and the effectiveness of the algorithm is verified by experiments.It is shown that compared with other algorithms,MFFHRA can further alleviate the data sparsity,improve the accuracy,and solve the problem of user cold start.