Domain Knowledge Mining In Network Environments

The massive, heterogeneous and fast-growing data resources in current web environments have brought the contradictions between'data rich'and'knowledge poor', which increases difficulty to acquire potential and valuable domain knowledge.The thesis uses heterogeneous web data sources as object of study to explore the feasibility of knowledge discovery from them. Following the train of thought of 'model proposition, algorithm implementation and data verification', the thesis conducts the practice study of mining potential and valuable domain knowledge using web data resources.1. It proposes a domain knowledge mining model in the network environments. The model is a three-layer one, which are data layer, knowledge layer and application layer from bottom to top. It guides to mining multidimensional knowledge. Based on the model, the thesis conducts the practice study of domain knowledge mining with three data types of scientific literatures, blog posts and social annotations.2. The thesis proposes a new probabilistic topic model:Topic-Author model, which jointly model information of literature content and authors. Based on the model, a domain knowledge mining framework is proposed to perform the multi-dimension analysis for domain knowledge discovery, including concept discovery, expert finding, articles recommendation, trends analysis, and correlations identification.3. The thesis proposes a blog knowledge mining framework to study topic mining, opinion analysis and information diffusion. Potential concepts are discovered based on text clustering and topic modeling and the opinions of the concepts are analyzed. Based on the study of social network diffusion models, the thesis summarizes the methods for maximization of information diffusion and proposes an improved threshold model.4. The thesis analyzes the collective intelligence in social annotations and different knowledge classification methods in web environment. A lightweight ontology construction method is proposed to discover semantic knowledge of social annotations. The method uses a clustering algorithm of social tags based on weighed network division to perform semantic cluster and semantic layering.The results show that the study of domain knowledge mining proposed in this thesis can discover vastly valuable and potential knowledge, provide multiple knowledge services and support knowledge acquisition and learning.