Design And Implementation Of A Personalized Social Search Engine Based On The Modeling Of Multidimensional User Characteristics

At present, search engines have become an important way for the users of internet to getinformation. However, with the rapid development of Web2.0and the increasing rise ofvarious social networking sites, some of the flaws of the traditional search engine exposed.On the one hand, a variety of Web2.0websites and social networking sites have vast amountsof user data, but traditional search engines are unable to support its retrieval. Different users,on the other hand, have different occupation, education background, interests, preferences andsocial relationships, the personalization of search results also have certain requirements.In view of this, this paper aims to design and implement a personalized social searchengine, whose data sources are from the social networking sites, by analyzing thecharacteristics of users and their social relationships to improve results of the traditionalsearch engine.User modeling is the prerequisite and basis for personalized social search. In this paper,we propose a multidimensional and multi-level user model according to the characteristics ofthe actual data from the social networking sites called Sina microblog. This model includesthree sub models, named user influence model, user social relation model and user interestmodel. It is a comprehensive description of the characteristics of users in social networkingsites. In order not to affect normal use of social networks, we design an implicit user dataacquisition algorithm according to the characteristics of Sina microblog API. Furthermore, wedesign a set of user characteristics generation algorithm, including the user influence featuregeneration algorithm named PersonRank used to calculate the global influence of a user in thesocial network, the user social relation feature generation algorithm used to calculate thesimilarity and familiarity between users, and the user interest feature generation algorithmwhich uses Naive Bayes text classification method to get users’ interest vectors. Besides, inthis paper we built the user model update mechanism using the implicit feedback strategy andthe active feedback strategy.On the basis of the user modeling, we design the personalized score and sortingmechanisms. The score and the sorting algorithms work on the basis of the traditional full-textsearch engine Lucene. The algorithms combine the user influence score, the social relationsscore, and user interest score to generate the final score of a document. Then the documentsare reordered according to the final scores.Finally, we design and implement this personalized social search engine according to thetheories above, and let some users use it for evaluation. According to the statistics of click logs, users are more satisfied with the personalized social search engine than the full-textsearch engine Lucene.