Research On Recommendation Algorithms For Diversity Requirements And Service Resource Matching

There appears the problem of information overload due to the rapid growth of Internet scale and coverage: Too much information makes users can't obtain useful things to themselves,and the efficiency of information to be used is reduced.Also,it prevents merchants to displaying their products effectively to the group who are interested in them.Therefore,how to improve the accuracy,the diversity and the interpretability of the recommendation algorithm are the focus of the research.This paper first proposes a method to improve the diversity of recommendation systems based on joint features.Due to the deficiencies of the existing quadratic optimization algorithm which can not actively improve the recommendation diversity,this method improves idea of gaining the joint feature based on matrix similarity and applies to the recommended field using the MovieLens film recommendation dataset.By solving the joint features(tags)that have an effect on different categories,this method predicts products that may be of interest to users of other types of products that the user is not involved in to achieve the effect of actively increasing diversity of the recommendation.The method avoids the hypothesis defect caused by the quadratic optimization method,and the joint feature has good interpretability,which can explain the reason for recommending other products to the user more clearly,and let the user trust the system more.This paper also proposes a method about preference analysis and resource recommendation based on random forest.This method contains three stages:in the sampling stage,in order to sample accurately and rationaly we proposes a sampling method based on the combination principle;In the attribute analysis stage,we proposes a user preference analysis algorithm based on random forest to realize good interpretability and deep relationship analysis between attributes;In the recommendation stage,we proposes a user grouping method based on explicit scorea to solve the problem about super specialization.