Research On Big Data Recommendation Algorithm Based On Project Matching Degree

Nowadays,there are some problems in recommender systems,such as diverse data sources and complex data structure,cold start and poor recommendation diversity,etc.Among them,the problem of poor recommendation diversity has few solutions,which means the recommended results are often homogenized and less concerned about the "Long Tail" items.In this paper,the problems of recommender systems are deeply researched.In view of the poor recommendation diversity problem,a hybrid matching recommendation algorithm is proposed from the point of items collocation.Based on the user and items data generated by real application platforms and idea of hybrid recommendation algorithm,an items matching model for multi-source heterogeneous data is proposed to evaluate the relationship among the items.And then designed and implemented a Hybrid Recommendation for Items Matching algorithm(HRIM),so as to provide the recommendation of the matching items when the user concerned about a certain item.The main work is as follows:1.The mainstream algorithm in recommender systems are described,compared and analysed in detail.The algorithm flow of hybrid recommendation is proposed.By focusing on collocation between items,this paper improves the diversity of recommendations.2.Since data has complicated structure and diverse sources,this paper proposes an items matching model for multi-source heterogeneous data to evaluate degree of items' collocation,inspired by the ideas of content-based,knowledge-based and collaborative filtering algorithms.3.The problem of recommendation is transformed into ranking problem,and a hybrid recommendation for items matching algorithm based on ranking learning is proposed,which is concerned with the ranking accuracy of collocation recommendation,and improves the diversity of recommendation results.A fusion and filtering algorithm of matching is proposed before learning to rank,so that it reduces the computational range of learning to rank and improve the overall efficiency of the algorithm.4.Based on the data extracted from datasets of Tencent microblog and Taobao,this paper designs and implements a set of contrastive experiments.The experimental results show that the proposed algorithm can improve the diversity of recommendation results and have certain effect on forecasting matching items.