Semantic Ontology Based Service Recommendation System For Social Network

The prosperity of social networks has produced large amounts of redundant data.How to obtain useful information from them as the basis of recommendation has becomea hot researching issue. Currently, traditional recommendation algorithms arecontent-based recommendation, collaborative filtering and hybrid recommendation.However, they always show shortcomings of sparse data, cold starting and highcomputational complexity in various degrees. With the proposed of Web3.0, theSemantic Web and ontology has guided a new direction for recommendation service.The Semantic Web is based on three core technologies of XML, Ontology and RDF.Ontology is the formalized expression of the concepts in specific domain andrelationships between them. It is to construct a structured conceptual model to provideuniversal and shared understanding of domain knowledge. In this paper, based on thedeeply analysis of the structure and characteristics of the data in social networks (SinaMicro blog), we have utilized ontology into the recommendation system.Firstly, the open API of Sina Micro blog is adopted to crawl and preprocess the rawdata of user profile. Secondly, as the existing keyword extraction algorithm does notfully consider the characteristics of social network, we propose the TF-MBF algorithmand combined it with the TextRank algorithm to extract keywords of user interests.Thirdly, the basic ontology of user interests is constructed based on the Chinese versionof WordNet. Finally, the results set to be recommended are obtained according to thespecific selecting rule. Then, we propose a novel algorithm to calculate the semanticsimilarity of sememes tree which takes the density into consideration. According to this,the final recommendation results can be selected.Experimental results show that combining the proposed TF-MBF algorithm withTextRank to extract keywords is more suitable for social network data. The proposedsemantic similarity algorithm can provide more accurate results. The recommendationresults can reflect the true interests of users well. Our system is suitable for practicalapplications and has good scalability.