Personalized Recommendations Of Web Resources Based On Topic Clustering

In recent years,with the rapid development of Web technology,network resources showing explosive growth trend,the number of resources increasingly large,the category of resources increasingly numerous.Such a rich resource has brought convenience for people learning and life,people can easily access the information they want through the network.At the same time,such a massive resource also bring the inconvenience and people fall into the"abundant resources,lack of information" dilemma.People are eager to solve the problem that how to find the valuable information that they really need from the massive network resources quickly and accurately.Therefore personalized recommendation technology attracted widespread attention.Personalized recommendations of Web resources based on Topic clustering technology will not only make use of clustering techniques to management Web resources,but also can recommend resource on the basis of user individual differences,which can better meet people quick access to information demand.Firstly,the research status of domestic and abroad about personalized recommendations of Web resources and clustering was introduced,and the existing clustering algorithms and personalized recommendations algorithms were brief introduced;then,the method of Web resources clustering based on theme was highlighted,traditional clustering methods or do not consider topical information of Web resources,or only consider the case of a single topic,and there is not a good clustering method for Web resources that contain multiple themes,the method Web resources clustering based on topic is proposed based on this.The methods consist of theme extraction,feature extraction,feature representation and clustering Web resources.Among them,the theme extraction is clustering a set of words that produced after Web resources pretreatment,which is extracting topics of Web resources;feature extraction is on the basis of theme weights,extract feature words from each topic to characterize Web resources;feature representation by means of VSM model,use features extracted words and the corresponding weight to represents Web resources;Web resource clustering is on the basis of feature representation,and calculates similarity of two Web resources,then clustering Web resources into one group if the similarity reaches a certain value.Then use the implicit tracking method for user interest mining,through observing user's browse behavior at service site,computing users interest degree of Web resources,and building Web resource queue that user interested in,then use TWTC clustering method mining user interest information from the Web resources queue.Lastly,the paper recommended Web resources based on the match of the Web resources and user interest,it also consider the case of user interest may offset over time,so it regular updates user preferences information and recommends Web resources dynamic according to the updated user preference information.In the experiment,three methods were selected to do two set of comparative test with the method proposed in this paper,and use precious and recall rate to evaluate the experiment results.Experimental results show that the method that personalized recommendation of Web resources based on theme clustering has higher precious and recall rate than the other three methods,even with the increase of the number of resources,Algorithm performance is stable.The innovation of this paper is that:firstly,proposed a Web resource topic-based clustering method,which fully considers the theme information of Web resources and clustering Web resource based on theme information,improve clustering precious and recall rate;then with this method,mining user interest,in order to improve the accuracy of user interest model,thus improving the accuracy of personalized recommendation,at the same time,consider the case of user interest may offset over time,use the time window for the user interest updating to achieve dynamic recommendation of Web resources.This paper uses content-based method to recommend Web resources,but this method is currently only available in Web document clustering,we will continue research on Web audio and video resources clustering and recommendation later.