Design And Implementation Of Concept Lattice Based Text Filtering System

Text filtering is proposed to help people to get information from the Internet, which contains a large amount of informations and resources. Keyword search is a primary method of text filtering in traditional research works, which ignore the semanteme and relationships among the keywords. Traditional research on text filtering uses keyword search as primary, ignoring the semanteme and relations between keywords.And now it becomes a bottleneck to the further development of text filtering research. Therefore, it is inevitable to use some new theory to improve the ability of text filtering.Concept Lattice which formed a ripe theory has solid foundation in mathematics. Concept lattice is a useful tool for data analysis, and recently it draw more and more attention in information processing area.To solve the problem mentioned above, in this paper,we combine the traditional text filtering technique with concept lattice theory.Formal Context is used to organize texts for filtering and domain feature words in them, and a corresponding concept lattice is generated. By using the Hierarchy Structure of Concept Lattice and Object-Attribute association, the matching between text and user profile is transformed into the matching between nodes in concept lattice and user profile, so that the purpose of text filtering is achieved. Moreover, in order to measure the semantic relationships between concepts, the domain ontology is employed to compute concept similarity between domain feature words and user's interest words for improving the precision of text filtering.To achieve above purpose, a concept lattice based text filtering model is proposed, and several relevant algorithms are designed such as incremental formation algorithm of concept lattice and concept similarity computation based on domain ontology. Accordingly, a prototype system correspongding to the proposed model which is named as CL-TFS (Concept lattice based Text Filtering System) is designed and implemented.To verify the effectivity of proposed model and the usability of corresponding prototype system, the biological and biomedical ontologies supplied by the NCBO(stand for the National Center for Biomedical Ontologies) BioPortal is utilized in this paper. By using 106 user profiles supplied by TREC-9, the corpus named OHSUMED which is used in TREC-9 is filtered..The experimental results reveal that the design of the CL-TFS system has better recall and precision than those of traditional text filtering system based on keyword search.