Recommendation Algorithm Based On Scoring Preference And Attribute Scoring

The rapid development of the Internet not only accelerates the pace of people's life,improves the quality of life,but also brings the Internet itself Huge data information.from the mountains of books,paper,calligraphy and painting in the old era to the dense file list,convenient storage also leads to the difficulty of data processing.Since its birth,recommendation system has brought great benefits and convenience to human beings.Although the recommendation system is mature but not perfect.this paper studies the similarity between users and projects from two perspectives,and integrates the current popular text processing and deep learning model to study how to improve and mine the deeper information of users or projects,so as to improve the performance of the system.The main contents of this paper are as follows:Similarity analysis.On the one hand,with the purpose to optimize similarity measurement formula,this paper improved on the existing basic similarity formula,reduced the influence brought by the preference and behavior differences of individual users by decreasing the user evaluation value differences among users,then optimized the user similarity,so as to obtain more accurate target user community.On the other hand,the cold start problem of the system was considered,and the user's attribute information was integrated into the improved similarity calculation formula,which improved the system's unfriendly recommendation to new users or users without historical behavior data.Finally,through the open data experiment,the superiority of the improved algorithm was verified and the recommendation level of the system was improved.Text Processing.The data analyzed by the traditional recommendation algorithm is too simple,which results in less effective information and affects the recommendation ability of the system.The text processing method makes good use of the data that the traditional algorithm in the system can't process and analyze,which greatly increases the recommendation performance of the recommendation algorithm.In terms of text processing,this paper also make use of the category characteristics of the project,and the correlation of the text information between them in the movie recommendation system to analyze the text processing method to obtain the category feature vector of the project,and take it as a calculation basis of similarity.Feature mining.the data of similarity calculation not only comes from the attribute data of users or projects themselves,and also contains the behavior relationship data between users and projects.The traditional recommendation algorithm processing these data is nothing more than direct calculation,matrix decomposition or text processing and other similar methods.Although the processing results are effective,the feature data obtained are superficial explicit features.From the perspective of mining deep feature data of information,this paper integrated deep learning ideas into collaborative filtering algorithm,analyzed and trained user or project behavior data through deep learning model to obtain deep feature vectors of the project.Finally,the similarity calculation method was obtained by combining the feature vectors of item category obtained via text processing method with deep feature vectors of the project.Through the comparative analysis of experiments,after the traditional collaborative filtering algorithm has the support of text processing and deep learning,the multi-dimensional recommendation algorithm provides users with more accurate and ideal recommendation results.