X Tang Q Zeng/ The Journal of Systems and Software 85(2012)87-101 日 per Management Module Pa User 1 te rface 日 Subject User Behavior Monitoring E Subject2.1.1 Module User n User ProfilIng Module Behavi Datab ser Profiles Fig. 1. Framework for generating user interest profiles. stored in the paper database. Each paper in the paper database In the paper database storing the research paper da is classified according to the subject ontology and can readily each papers information contains its corresponding category be retrieved by users. The paper management module plays the mark based on the subject ontology. In this paper, vector role of fundamental component in the whole framework. space model is used to represent the research papers. Key (3)User behavior monitoring module This module is responsible for words are provided by the papers authors, representing the the background collection of the behavior data of each user. key content of each corresponding paper. In the vector space The user behavior data include searching keywords, browsing, model, a paper is represented by a keyword vector, i.e., downloading and commenting on papers, etc. The monitoring paper=(keywordl,., keywor keyword)(1≤i≤n).The nd collecting processes are totally unobtrusive. weight of the ith keyword keyword in paper is (4)User profiling module. The user profiling module makes use of as WKP(keyword, paper), which is computed by the tE the user behavior data recorded by the user behavior moni- oring module the paper database and the subject to create user profiles. The user profiles obtained ca to recommend papers to these users. We will elaborate our COMPUTER SICENCI profiling approach in Section 6. 4. Automatic ontology extension through clustering Communications and Theory of computation Algorithms and data structures Databases In order to solve the problems in the user profiles based on the traditional ontologies, we pi ontology extension algorithm Programming languages and to refine the user profiles. Before presenting it, we introduce an Artificial intelligence com pliers riginal subject ontology, which is defined by the Science Paper Concurrent, paralleL, and Online website(Sciencepaper Online, 2010). It is a taxonomy of distributed systems Computer graphics research subjects and has been in use on the Internet for many ears. This simple ontology consists of two levels of classification, Software engineering Scientific computing primary subjects and secondary subjects, and it holds is-a relation- ships between the subjects in different levels. In the first level, there Computer architecture Collaborative Networks are 43 primary subjects. Each primary subject has secondary sub- jects as its subordinate classifications. Fig. 2 shows the section of Fig. 2. The classification of"computer science"in the research paper subject onto- the primary subject"computer science "in this subject ontologyX. Tang, Q. Zeng / The Journal of Systems and Software 85 (2012) 87–101 89 Fig. 1. Framework for generating user interest profiles. stored in the paper database. Each paper in the paper database is classified according to the subject ontology and can readily be retrieved by users. The paper management module plays the role of fundamental component in the whole framework. (3) User behavior monitoring module. This module is responsible for the background collection of the behavior data of each user. The user behavior data include searching keywords, browsing, downloading and commenting on papers, etc. The monitoring and collecting processes are totally unobtrusive. (4) User profiling module. The user profiling module makes use of the user behavior data recorded by the user behavior moni￾toring module, the paper database and the subject ontology, to create user profiles. The user profiles obtained can be used to recommend papers to these users. We will elaborate our profiling approach in Section 6. 4. Automatic ontology extension through clustering weighted keyword graphs In order to solve the problems in the user profiles based on the traditional ontologies, we propose an ontology extension algorithm to refine the user profiles. Before presenting it, we introduce an original subject ontology, which is defined by the Science Paper Online website (Sciencepaper Online, 2010). It is a taxonomy of research subjects and has been in use on the Internet for many years. This simple ontology consists of two levels of classification, primary subjects and secondary subjects, and it holds is–a relation￾ships between the subjects in different levels. In the first level, there are 43 primary subjects. Each primary subject has secondary sub￾jects as its subordinate classifications. Fig. 2 shows the section of the primary subject “computer science” in this subject ontology. In the paper database storing the research paper data, each paper’s information contains its corresponding category mark based on the subject ontology. In this paper, vector space model is used to represent the research papers. Key￾words are provided by the paper’s authors, representing the key content of each corresponding paper. In the vector space model, a paper is represented by a keyword vector, i.e., paper = (keyword1,..., keywordi,..., keywordn) (1 ≤ i ≤ n). The weight of the ith keyword keywordi in paperp is denoted as WKP(keywordi, paperp), which is computed by the TF-IDF Fig. 2. The classification of “computer science” in the research paper subject ontol￾ogy