Design And Implementation Of Intelligent Recommendation System Based On Big Data Of Paper And Information

The aim of literature recommendation system is to determine users' requirements in no time.It is relief of burden in literature retrieval for users to recommend them ones related to their research fields in plenty of papers.Literature recommendation is a branch of recommendation system.In this paper,both general features and many different aspects of literature information characteristics need to take into consideration.The multi aspects information contains literature properties that can decide whether users are interested or not,such as title,author,abstract,keyword,body,reference,etc.Natural language process algorithm will capture features of these information constituted by words.But for author information,it will be encoded directly into the whole.With analysis of application scenarios and technology,this literature recommendation system is built by Latent Factor Model with multiple information fusion.User latent vector and item latent vector are built within article title,abstract,content knowledge graph,reference knowledge graph,author knowledge graph.Generation of user latent vectors relies on historical information and input information,while the item one relies on literature attributes.In engineering,literature recommendation system has been realized in this paper.Some supporting functions are used to improve usability.In experiment,results of experiments on CiteULike dataset and user historical data provided by China Geography Library show that the recommendation system shares similar effect with state of the art.In addition,it avoids cold start by Latent Factor Model with multiple information fusion on a big data platform.The average recall of CiteULike dataset is 95.4%,while China Geography Library dataset is 64.6%.The NDCG value of CiteULike dataset is 0.628,while China Geography Library dataset is 0.498.Finally,the well-trained model is used to build literature recommendation web application,for recommending user papers.