Research On Knowledge Service Oriented Intelligent Recommendation System

"Research on Knowledge Service oriented Intelligent Recommendation System"belongs to relative research domains about library and information resource management,information retrieval and personal service. The research task comes from the MOE(Ministry of Education) Project "Research on Comparison and Synthesis of Bibliometricsand Content Analysis" (Project No: 02JAZJD870003) of Key Research Institute ofHumanities and Social Sciences in Universities, and the foundational research project"Research on Technology Scheme and Application of Personalized Information Servicebased on Web Portal" (Project No: 2004QB1505) supported by General EquipmentMinistry of CPLA (Chinese People's Liberation Army).Knowledge service stands for the direction and extension of traditional library andinformation service patterns, which emphasizes on the character of personalization andspecialization of daily service work, concerns deeply about the knowledge value-addingand service value-adding in service procedures. The knowledge service oriented intelligentrecommendation system, namely IKRS, is a retrieval service system with ability ofsemantic process and user oriented personalized recommendation, which can overcome theproblems about "lack of service personalization" and "lack of information semantics" thattraditional information retrieval system has once been proved to be. Our research hassignificant theoretic and realistic sense. Firstly, the research is according with thedeveloping direction of personalization, knowledgelization of information service, and thesemantic retrieval trend of information retrieval system under the requirement ofknowledge service; Secondly, the research can provide technology support for thedevolopment and application of knowledge service, and meet knowledge and informationneeds of users better; Thirdly, the research can enrich and develop relative theory, methodsand technology about Chinese information resource organizaton, information retrieval andpersonalized service, and provide some reference for relative research and application.Under the direction of knowledge service, aiming at the target of high efficientretrieval service of digital information resources in network environment, we have paidgreat attention to the theories, system framework, several critical techniques and theirapplication methods about IKRS, and the main work and content of the research includesthe following five aspects:(1) Research on the formal model and architecture of knowledge service orientedintelligent recommendation system. The research status of three aspects such as knowledgeservice, personalized service and recommendation system, semantic retrieval is(?) summarized and reviewed. According to the connotation and requirement of knowledgeservice, a five tripled formal model, domain ontology and data mining based framework ofthe IKRS is proposed. By promoting and extending the four tripled formal model oftraditional information retrieval, a dynamic cycled interactive process mechanism amongthe three core elements such as user model, domain knowledge model and document modelof the proposed system framework is constructed, so that the concept level resourceretrieval and personalized recommendation feedback can be realized.(2) Exploration of the construction and visualized navagation of domain ontology.Ontology is the formal, explicit specification of a shared conceptualization, expressesconcepts and the relations between concepts, which is the basis of semantic process ofknowledge content in IKRS. The theories, methods and tools about ontology based domainknowledge modeling is introduced and disscussed. By referencing serval ontologyengineering methods, an experimental and OntoAvion named domain ontology aboutwarplane is constructed in five steps, mainly on the basis of reference book such asChinese Library Taxonomy, Chinese Defense Scientific and Technical Thesaurus. Furthermore, a visualized concept navigation scheme of OntoAvion is designed and realized forthe aid of user query, by appling software components of ontology reasoning andvisualization.(3) Research and experiment of concept semantic based Chinese text similaritymeasure. In IKRS, the text similarity measure can be applied in several functional modulessuch as resource retrieval for user query, content based resource recommendation etc. Theconcept comprehension of text similarity and relativity, and commonly used computationformulas for text similarity measure is discussed, the procedure and problems of currentlytypical text similarity method is also analyzed. Based on the discussion and analysis above,a simple and efficient, domain ontology and concept semantic concerned text similaritymeasure method called SCSM is promoted, which measures similarity between twoChinese text concept vectors by filling each vector with near neighbored superior conceptsaccording to their connection path and semantic relation in the network structure of domainontology. By taking the advantages of domain knowledge background, the measuringmethod SCSM showed better effects than traditional word vector based measuring method,which means that the SCSM method can be applied in cocept level retrieval and result listordering for content based retrieval and recommendation.(4) The personalized recommendation techniques of IKRS have been deeplyconcerned and explored. The concept and main tasks of user preference modeling is introduced, and the fine-grained user preference modeling and preference resourcerecommendation is focused. By applying domain ontology, a new fine-grained userpreference modeling method is proposed, which congregates near neighbored conceptsfrom coarse-grained user preference vector to find and describe user's multi-aspectinterested contents. Experiment showed that our fine-grained preference modeling cancatch the composition and variation of one user's multi-aspect preferences in a reasonablemanner. Furthermore, collaborative resource recommendation techniques of IKRS areapproached. After analyzing the pricipal and typical techniques of collaborativerecommendation, two effective collaborative recommendation methods is proposed andapplied. One method is on the basis of fine-grained preference similar users finding, theother is on the basis of weighted association rules among documents.(5) An IKRS prototype system is designed and developed on the basis of Javaplatform and techniques of open source during the research period of the project. TheIKRS prototype system has four primary functional modules such as ontology process andtransition, document resource annotation and process, user interaction, recommendationpre-process. The prototype system can automatically record users' visit behavior, finduser's fine-grained preference, provide multiple resource recommendation modes, andintegrated several important elements such as user oriented personalized recommendation,concept level retrieval match, keywords retrieval, and visualized navigation etc. duringuser-system interaction.