Research On Personalized Recommendation Algorithm Based On Collaborative Filtering

Collaborative filtering recommendation algorithm can help users find new things and provide personalized recommendation,attracting many scholars and experts to improve and expand their computing models and computing methods,so as to improve the accuracy of recommendation.With the improvement of people's demand quality and the increase of the amount of data,the collaborative filtering recommendation algorithm model is too rough and the near neighbor set is not accurate enough.On the basis of analyzing user behavior and related data,a series of improvement measures are put forward for its defects.After experimental verification,the recommended performance of the improved collaborative filtering recommendation algorithm has been greatly improved.The main work of this article is as follows:1.In view of the serious influence of traditional collaborative filtering recommendation algorithms which have too rough recommendation model,decreasing of the recommendation precision and data sparsity.In this paper,a multi factor collaborative filtering recommendation algorithm is proposed from the aspects of the user's evaluation of the project and the user attributes.The user rating factors and user attribute factors are fused by using parameters,and the superiority of the multi factor collaborative filtering recommendation algorithm and the algorithm performance under sparse data are verified by experiments.2.In the aspect of user scoring,the similarity between users is calculated by using the user score difference calculation model,and the calculation formula of the score difference is improved which solves the problem that the traditional formula is not applicable and unreasonable in the characteristics of the individualized user.In addition,the traditional formula are still unable to distinguish the fuzzy user and easily lost is very similar to the user,through processing formula of widening the fuzzy interval fuzzy user obviously separated into user score maximum difference variables,get rid of the influence factors,solve the problems such as the loss of extremely similar customers.3.In terms of user properties,the user's behavior attributes are excavated to increase the amount of users' information.Based on the analysis of the traditional user attribute similarity calculation formula neglects the different effects of different attributes on users,a quantitative attribute similarity computation method is proposed.The similar user is identified through user's score of the project,and then the calculation method of quantized user attribute is expressed by similar users.Finally,the quantitative value of the attribute is determined by the training data set,and the similarity is calculated by the quantitative attribute formula.4.In the end,the traditional collaborative filtering recommendation algorithm is not sufficient for user implicit feedback data mining,and a collaborative filtering recommendation algorithm with important label degree is proposed.User types and frequencies can reflect users' preferences.On this basis,a new user interest preference model is established to calculate the average importance of tags,for different users to use text inverse frequency files to calculate the importance of the tags.Finally,the target user's close neighbor set and the prediction score are obtained,and the effective recommendation for the target user is implemented.The experimental results show that the proposed algorithm greatly improves the accuracy of the recommendation and alleviates the cold start problem.