Research On The Knowledge Aggregation In Network Community Based On Domain Conceptual Relations

This paper proposes a demand-oriented knowledge aggregation scheme in network community. The basis of knowledge organization in network community is the conceptual relations between topics of UGC (user-generated content), so a crucial part of it is to construct the domain conceptual relation system. The traditional knowledge organization systems can't satisfy the requirement of knowledge aggregation in network community. So our research focused on the construction of multi-component domain conceptual system which is built up from fusion of domain background knowledge and the topic relations mined from UGC in the given network community. Supported by the multi-component domain conceptual system, we discussed the implementation of knowledge aggregation in network community.In chapter 1, we discuss the aggregation patterns of knowledge in network communities. First, we discuss the knowledge organization requirement in network communities according to the characteristics of UGC. After reviewing the theories and development processes of knowledge aggregation, we emphatically analyze the forms of conceptual relations, based on which we figure out the aggregation patterns of knowledge in network communities.In chapter 2, we propose the knowledge aggregation model of network communities based on the domain conceptual relations. We first import the domain analysis method and discuss the characteristics of conceptual relations and its effect to knowledge organization. Then, the knowledge aggregation model of network communities based on the domain conceptual relations is proposed, which contains the construction process of multi-component domain conceptual system as well as the implementation of it in network community.In chapter 3, we discuss the construction of structured domain conceptual relation system. The existing universal knowledge organization system and professional knowledge organization system, such as taxonomy, thesaurus, ontology and topic map, can provide useful information about domain concept types and concept relations. Besides, it's necessary to gain domain concepts and extract concept relations from other resources, especially from the UGC in network communities. So we also discussed the method of gaining domain concepts and extracting concept relations from user queries and user tags, as well as from user discussion in network communities. Then, the integration of concept terms and concept relations gained from different resources are discussed. After comparing the SKOS (Simple Knowledge Organization System) and OWL 2 (Web Ontology Language), we choose OWL 2 as the formalized expression of the structured domain conceptual relation system. To test the proposed process and methods, we constructed a structured domain conceptual relation system of the cardiovascular field based on the CMesh (Chinese Medical subject headings) and the medical wiki in www.39.net.In chapter 4, we discuss the construction of co-occurrence conceptual relation system based on the concept co-occurrence relations in the UGC of network communities. The Graph-based Text Representation Model is utilized as the theoretical basis of the representation of concept co-occurrence relations in UGC. The process of constructing the co-occurrence conceptual relation system divided into two phases. The first phase is to construct co-occurrence matrix based on the co-occurrence relationships of concepts in UGC, including UGC corpus construction, Chinese word segmentation, synonyms combination, co-occurrence relationship extraction, and co-occurrence matrix generation. The second phase is to transform the co-occurrence matrix into concept network, including co-occurrence strength standardization, concept similarity computation, network link strength mapping, and concept network generation, In the experimental part, the cardiovascular forum in the DXY (Ding Xiang Yuan) medical BBS is taken as example. A co-occurrence conceptual relation system of cardiovascular is constructed based on the data mining from UGC.In chapter 5, we discussed the integration of the structured domain conceptual relation system and the co-occurrence conceptual relation system. The integration mechanism of them is firstly analyzed based on the comparison of their advantages and disadvantages. The principle of complementary advantage is set, based on which we propose the integration idea as mapping the concept type, the concept attribute information and the fine-grained conceptual relation from the structured domain conceptual relation system into the co-occurrence conceptual relation system. In the part of conceptual relation mapping, two important issues are specially discussed, including the similarity computation of concepts based on the combination of semantic distance and co-occurrence similarity, and the identification of fine-grained conceptual relations. Then, the formalized expression of the multi-component domain conceptual system is discussed based on the relational data model. In the experimental part, we integrate the two conceptual relation system constructed in chapter 3 and in chapter 4.In chapter 5, we discussed the implementation of knowledge aggregation in network community. With the support of the multi-component domain conceptual system, the granularity level of knowledge unit is more specific, and the form of knowledge aggregation is more variety. After analyzing the mechanism of multi-component domain conceptual system in knowledge aggregation, we propose a multidimensional knowledge aggregation model of network communities. The implementation of knowledge clustering as well as knowledge copolymerization is discussed both in topic level and in document level. The knowledge copolymerization is a new form of knowledge aggregation, which is quite different from the traditional form of knowledge clustering. Based on the multidimensional and fine-grained conceptual relation from the multi-component domain conceptual system, some new forms of knowledge aggregation can be implemented in network communities, typical forms including the faceted navigation and faceted search, the multidimensional topic recommendation, the knowledge cell linkage system, and the noninteractive literature-based discovery. In the experimental part, we implement those knowledge copolymerization methods in the cardiovascular forum of DXY medical BBS.