Similarity Calculation based on Spectral Clustering in Bipartite Graph

With the rapid development of Internet applications, Internet information has also exploded, and the problem of information overload has become increasingly prominent on users’ online experience. The recommendation system was born to solve this problem. The system uses the user's historical information to recommend items that are of interest to the user. Memory-based collaborative filtering is an important recommendation method in the recommendation field. The key is to accurately measure the similarity between users or products. In order to achieve better measurement results, a collaborative filtering similarity algorithm based on spectral clustering in bipartite network (Similarity Calculation based on Spectral Clustering in Bipartite Graph, SC-SCBG) is now proposed. Perform spectral clustering in a bipartite network composed of items; then divide users into categories, and calculate the similarity between user categories; finally, integrate the similarity between user categories into the traditional correlation coefficient in the form of weights, The penalty function is introduced to construct a brand new similarity. This method measures both linear and non-linear relationships among users, and in terms of non-linear relationships, more common heuristic similarities are considered more comprehensively. The verification on real data shows that the collaborative filtering similarity algorithm based on the spectral clustering in the bipartite network can improve the accuracy of the target user recommendation, thereby improving the recommendation quality of the collaborative filtering recommendation system and improving the diversity of collaborative filtering.