The Concept Of Network Modeling, Implementation And Application

Information retrieval based on concept is one of the focal point of research in the filed of intelligent information retrieval. The representation and organization is the main question on concept-based retrieval. In this paper we give an object network modal on concept. We treat concept as a complex object with attribute, behavior, concept description, concept words. We construct a conceptual network by describing the structural and semantic relationships among concepts. Using this network we get "a grain of association of thought and a reliance of reasoning "Among all the knowledge.Aiming at the implement of conceptual network we present a new method based on the object type of ORDBMS. In this method conceptual network is realized by three data type : inner-object of concept, relation-object of concept, instance-object of concept. We get an well encapsulated concept by this way. At the same time we develop a model concept management system to helping expert to building this network.We then need to fill up the conceptual network after we have finished the data schema design. But how to get the related feature words of one concept is not an easy thing even for expertness. We spit it into two steps: first, we get a feature words from a sets of texts. Second, we build the concept network by feature words. In the course of implement, we not only realize the algorithm of Chinese word segmentation, but also we present a learning system of Chinese words segmentation aiming at the "new word learning", "ambiguous phrase segmentation" problem. Using the result of word segmentation we give a comprehensive feature weight calculation and we get the feature words set.At last we give a rough discussion on information retrieval based on conceptual network. Thinking of the fault of VSM based on term, which has a big dimension and little feature, we present a new way of convert the text feature from words space to concept space. After a statistic analysis is processed within categories and between categories. The feature of category is signified by two vectors: mean value and standard deviation. Furthermore, the category matching of text is implemented by using fuzz distance calculation. This new method eliminates the drawbacks of traditional IDF method and it can efficiently conduct text feature extraction thus to promote the accuracy of automation text categorization.