Research On Web Search Based On Social Relationship

At present,the search engine has become a main way for Internet users to get information.However,with the rapid expansion of WEB 2.0,as well as the prosperity of various types of social networking sites,the traditional search engine has exposed the great defect.Many social networking sites and web2.0 sites have a mass of user data,but seriously the traditional search engine can not support its retrieval.For users of different occupation,education,interest and social relationship,the search results are expected to vary,so they urgently need personalized search engine.For entrepreneurs,in order to build better relationships with customers and render personalized service for customers,they are dying to popularize the new search methods that users participate in the decision-making,so they could get more user information.Given this,this paper unveils a new search engine,named PERSO.And it will realize the target of personalized search by using the searcher's social behavioral data.Through the analysis of rich user characteristics and social relations captured from open social networks,It will return the most important results in the most prominent position to searchers.That is to say,it can Improvement traditional search results.User modeling is the premise and foundation of personalized social search.Using real social network data from Sina Weibo,this paper proposes the new relevance models called persocial relevance models of three levels.In the level one it demonstrates the searchers own social behavior.In the level two,it increases the demonstration of the searchers' friends social behavior.In the level three,it increase the demonstration of the relationships between documents.So it is a comprehensive depiction of the users' feature in social networks.On the basis of the user modeling,this paper puts forward three methods to integrate the persocial relevance models into the processing procedure of web search,namely three different approaches to rank the documents: two-step TP,in which it will firstly use the text feature to filter pages,and then use social features to rank the page.two-step PT,in which it will firstly use social features to filter the page,and then use the text feature to rank pages.one-step HB,in which it will use both social features and the text feature to rank pages.Finally,capturing data from 10 million Baidu Wikipedia documents and 20 real Sina Weibo users,using F1 and n DCG@K as the evaluation indexes,this paper designs and implements four groups of experiments: the experiment to test the three levels of the persocial relevance model;the experiment to test the three approaches to rank the documents;the experiment to test the influential power contrast users and friends information on search performance;the experiment to test the influential power of the number of friends on search performance.These experiments show the effectiveness of the three ranking approaches and the varying improvements of the search performance by each level of the persocial relevance model.