The Research And Implementation Of Tag-based Recommender System

The last few years have witnessed an explosion of information that theexponential growth of the Internet and World Wide Web confronts us with aninformation overload: there are too much data and sources, that make it not easy tofind out those most relevant for us.The recommender technology has been proved tobe a powerful and valuable tool of handling information overload problem,playing anindispensible role in many areas.With the dramatic development of the Internet, Web2.0has emerged andbecome popular, which transforms users from passive consumers to active producersof content. Along with tagging behaviors, a great deal of valuable informationemerged, which strongly suggests the need to make use of such information toprovide personalized services.Firstly, this thesis introduces the background,meaning and current study situationat home and abroad.The classification,application, experimental method and cold startproblem of recommender system is presented.Secondly,this thesis states the study work that makes an improvement,includingthe following two aspects:(1)Modifing the calculation method of the user-item ratingsmatrix.This thesis modifies the traditional content-based recommender algorithm andimproves the traditional user-based collaborative recommender algorithm.Therepresentation of item model is modified through using the tags as the attributes ofitem.Calculating user’s rating toward a item through the the tags assigned on the itemto modify the user-item ratings matrix calculation method.A simple example and anexperiment is done to prove the modified user-based collaborative recommenderalgorithm outperforms the traditional user-based collaborative recommenderalgorithm in terms of Hit-rate and Hit-rank.(2)A recommender engine isdesigned,including data preprocessing module,related matrix generation module,recommendation list generation module and ranking postprocessing module.At last,the design of the functional module architecture is done,including7functional modules that are user tagging history query module,tag usage tendency query module,similar user network construction module,user resource recommendingmodule,tag cleaning module,tag recommendation module and tag extensionmodule.This thesis describle the procedure of training data and testing datapartition,user interest model construction and user similarity calculation,and describlehow to make the transformation of feature and data format.The implementation ofthe resource recommender module is done and present the user interface of therecommendation list.