Tea Ontology Learning Methods Research Based On Text

Since"Semantic web"concept was put forward, ontology research is becoming increasingly widespread. Ontology is powerful knowledge representation and regulation tool, which are widely used in Semantic Web, search engine, knowledge acquisition, knowledge representation and other fields. Ontology learning is automatic or semi-automatic ontology construction, which use machine learning, nature language processing to solve the efficiency of ontology construction and the ontology quality problem. Ontology is significant to build large scale domain ontologies and Semantic Web.As tea pest and illness target, ontology methods and technologies are studied.especilly on concept and relation extracttion.The main contents on our research as followings:1. The paper mainly research ontology learning technologies based on Chinese text, especially focus on concept and relation extraction. Firstly, domain dictionary is built to guide preprocessing unstructured data, such as delete stop words, segmentation, part of speech tagging, and then multi-strategy approaches are used to extract ontology concept and relation.2. Two statistical methods and a method based on dictionary are applied to extract concept. WVSM extract concept by computing word weight with constructing of low-dimensional vector space model. Maximum word merge can extract concept directly and merge word that has been split based on domain dictionary.3. Formal concept analysis (FCA) concept lattice and association rules are used to study concept relation. The former method utilizes theory and method of FCA concept lattice, firstly constructing lattice, then learning relation according the same data structure between ontology and concept lattice. The later method constructs transaction database firstly, then extracts association rules with improved Apriori algorithm, relation can be got from association rules finally.At last, all the methods and technology are applied in the tea illness and pest field to test and verify the correctness and effectiveness of concept and relation extraction. Experiment shows correctness and effectiveness can be increased.