| With volume of information generated every second increasing rapidly,how to pinpoint the information desired in large quantities of data presents itself as a major problem to the development of Internet.Besides traditional web portals which classify large quantities of data into various categories and search engines which generally requires users to provide specific query conditions,much importance is attached to recommender systems which enjoy the potential of relieving users of information burden.While existing recommender systems achieve some success in lighting the information load,they have some deficiencies too.Therefore,further research into recommender systems bears significance both in theory and in practice.Considering the fact that items constitute a central and fundamental part of recommender systems,this paper starts from discussion of properties of items,and then,noticing the similarity between items in recommendation systems and words in natural language processing and with power of word2 vec model in mind,proposes an item feature extraction method based on artificial neural network.Based on analysis of properties of users in recommender systems and the fact that the interaction between users and items constitutes the important basis upon which recommendation is made,this paper proposes a user feature extraction method which averages item feature vectors corresponding to items rated by users.Furnished with further extensions,the enhanced user feature extraction methods aim to characterize user properties from varied perspectives on the premise of not requesting users for any information.With the proposed item and user feature extraction methods,this paper proposes a refined recommendation method,contrasts it against existing mainstream recommendation methods on a public data set through experiments and evaluates its timeefficiency.Experiment results show that the proposed recommendation method outperforms existing ones and prove its feasibility. |
PDF Full Text Request