The Research Of Combine Learning To Rank And Topic Model

With the rapid popularization and emergence of the Internet,more and more people obtain information from online instead of offline.Due to the influx of data,the data size is getting larger and larger,presenting the situation of "information explosion".How to mine the information from the massive data that people may need is becoming more and more important.In order to solve this problem,the recommendation system came into being.The recommendation system plays an increasingly important role in almost all internet recommendation scenes,like e-commerce,social networks,online media and so on.Among them,personalized learning to rank algorithm in recommendation system is also praised by many scholars for its excellent effect and interpretability.Whether an item will be recommended to users,is mainly according to the historical information of the user and the inherent property of the item to determine whether the user will accept the item or not.Therefore,as well as the information of users and items is better mined and applied,the item can be better recommended to the user.Most of the previous recommendation systems are only use rating information of users to items,this thesis studies how to effectively apply the text information of users and items for recommendation.This thesis is mainly divided into two research points:1.Research on the recommendation algorithm for fusion of learning to rank and topic modelThe result of a recommended task is mostly displayed in the form of a recommended list in practical applications,so that the recommended task should have more optimization on the final recommended sequence.The recommended sequence generated by the traditional recommendation algorithm does not consider the order of the items relationship,and most of the algorithms only use the rating information of items and users.However,the text information contains a lot of valid information but is not well applied.The purpose of this thesis is to solve these two problems and propose to use the better fusion strategy to add the text information into the learning to rank recommendation algorithm.The method is validated experimentally and has obvious improvement compared with the benchmark algorithm.Moreover,it can solve part of cold start problem.2.Research and design of recommender system based on the fusion algorithm of learning to rank and topic modelAt the end of this thesis,we present a recommendation system based on the fusion algorithm we proposed.The design of the recommendation system mainly includes the following aspects:the design of the architecture of the system,the online and offline module design,the user interface of fore-end page design and the design and the design of Database for the recommendation system.In the end,the offline experiment is designed and experimented.