Research On Personalized Recommendation Method Based On Multi Similarity Measure Optimization

The rapid development of the Internet has brought about an exponential growth of information,which not only enriches people's knowledge and information,but also brings about the problem of information overload.In this context,recommender systems came into being.Personalized recommendation is to construct the user's interest preference by learning the user's historical behavior information,so as to actively recommend the information that the user is interested in.Collaborative filtering recommendation technology has become a relatively mature and widely used recommendation technology in current recommendation systems due to its simple construction and good interpretability.Collaborative filtering recommendation has made great progress in the field of research in recent years,and its application has become more and more extensive,but there are still some problems that similarity calculation is not sensitive to differences in user ratings,the determination of weight coefficients in comprehensive similarity calculation is relatively rough,and the selection of nearest neighbors is not enough.Accuracy and other issues have a great impact on the prediction accuracy of the recommendation system.On the basis of summarizing and analyzing the existing recommendation algorithms,this paper takes similarity calculation as the starting point,and proposes a personalized recommendation method based on optimization of multiple similarity measures.To optimize the recommendation process and improve the accuracy of the recommendation results,the main tasks are as follows:(1)In view of the problem that the traditional similarity calculation is not sensitive to the difference of user scores,by introducing the concept of information entropy,the similarity of user features is constructed by using the characteristic that information entropy can measure the degree of confusion of data.For the problem that the introduction of information entropy to calculate the similarity brings about offsetting differences and does not consider the proportion of common items,the original calculation formula is revised.First,the actual difference is used to weight the information entropy formula,and then Jaccard is used after normalization.The similarity is corrected to obtain the user feature similarity defined in this paper.User feature similarity measures the impact of user rating vector differences on user similarity,that is,the greater the user rating difference,the lower the user similarity.(2)Integrate user rating similarity and user feature similarity to construct comprehensive similarity,where user rating similarity is calculated by calculating Pearson similarity after filling the rating matrix with the SVD++ method.In the weighting of the two degrees of similarity,the particle swarm optimization algorithm is used to determine the weight ratio of the two degrees of similarity.The RMSE value is the objective function,and the weight coefficient is the particle to determine the weight coefficient with a higher degree of precision,so as to obtain a more accurate comprehensive similarity..(3)Aiming at the problem that the selection of nearest neighbors is not accurate enough,a method for secondary screening of nearest neighbor sets is proposed,which takes into account both the similarity of scores and the similarity of interests.Firstly,the user-category interest matrix is constructed by the attribute information of the item,and then the user's interest preference vector is calculated using the idea of TF-IDF,and then the similarity between vectors is calculated to obtain the user's interest similarity.In the secondary screening,the interest similarity threshold is firstly set to obtain the candidate nearest neighbor set,and then the strict nearest neighbor set is obtained through the secondary screening of comprehensive similarity.The simulation experiment is carried out on the data set.First,the comparison experiment between the user feature similarity defined in this paper and the traditional similarity in the collaborative filtering recommendation algorithm is set to verify the effectiveness of the optimized similarity calculation method in this paper,and then the algorithm in this paper and other algorithm models are set To verify the effectiveness of the proposed algorithm in this paper.It can be seen from the final experimental results that the algorithm proposed in this paper improves the accuracy of the recommendation system score prediction on the basis that the recommendation system coverage rate remains stable.