Collaborative Filtering Recommendation Algorithm Based on Improved KMEANS

The collaborative filtering recommendation algorithm looks for the nearest neighbor set with the target user or item according to the historical track of the user’s subscription or browsing of the item, and then predicts the score of the target object according to the score of the user in the nearest neighbor set of the item, and finally recommends the top several items to the user. This algorithm has a high degree of recognition in both academia and industry, but it also has some problems such as cold startup, data sparsity and poor scalability. For this reason, researchers at home and abroad have proposed many methods to improve the existing technology, among which how to accurately mine the user’s interest preference is the key to improve the collaborative filtering based recommendation method. Aiming at the defects and problems still existing in the existing collaborative filtering based personalized recommendation methods, through in-depth and systematic research, this paper first adopts semantic similarity computing technology to analyze the similarity of users’ comments on projects, and takes the similarity as the similarity of users’ preferences in user comments. The method of mixing the similarity with Pearson correlation coefficient and heuristic similarity is designed respectively. At the same time, matrix decomposition technology is used to fill the sparse score matrix, and then the improved KMEANS algorithm is integrated to build the project cluster model on the filled score matrix, and finally complete the project-based collaborative filtering recommendation on the candidate set determined by the cluster model. Experimental results show that the method proposed in this paper improves the precision of collaborative filtering.