A Study On The Extraction Of Non - Classified Relationships In Ontology Of Patent Domain

In the face of a broad array of information, traditional information retrieval methods keyword based on keyword information cannot meet user requirements. How to represent, manage, maintenance and reuse information resources have become industry and researchers’ common hotspot. As a new way to organize and describe information, ontology has good conceptual hierarchies and rich semantic information and supports. So it has been widely studied and applied in the field of information retrieval, digital library and so on.Concept is one of essential elements of ontology, the taxonomy relationships between the concepts are the skeleton of ontology and non-taxonomy relations between concepts are branches of ontology. The hierarchical model of ontology learning shows that non-taxonomic relations extraction is next step after concept extraction. And application requirements of patent ontology also need non-taxonomic relations between concepts. Meanwhile non-taxonomy relations are useful for patent analysis. In order to build a comprehensive patent ontology which is in new energy vehicles domain, non-taxonomy relations between concepts should be extracted from patent literature. By way of the relationship’s verb Automatic Clustering looking relationship type, and then the type of relation extraction are added to the patent ontology in the field of new energy vehicles. The major works in this paper are listed as follows:(1) Proposed a method which is based on conditional random fields to extract Chinese term in patent documents. In summarizing the previous term extraction research and characteristics of terminology in new energy domain, this paper select the word, word length, part of speech, dependencies, dictionary location, stop words and other characteristics as the feature templates. This paper explores the way to use dependencies between words to improve the term extraction problems precision. The experimental results show that dependencies between words as one of conditional random fields feature can improve it.(2) In order to solve the problem, relation structure is right but the semantic of relation is wrong, a new method is proposed. This method uses syntactic information and dictionary of relationship as new features and combine with traditional features in support vector machine. Combine traditional features such as lexical information, distance of concepts with new features to carry on the experiments. The experimental results show that this method can reduce the influence of semantic error and improve the results precision.(3) Designed and implemented ontology updating system in new energy vehicles domain. The purpose of building ontology is to build patent ontology knowledge base to use it for retrieving the patent document, building technology theme matrix, infringement detection and so on. With Patent Document explosive growth, the patent ontology also needs evolving. Implementation of this patent ontology updating system has three main functions. The first is that use method mentioned above to find new terms from patent documents and add them to patent ontology. The second is that non-taxonomy relations between the concepts was found and added to patent ontology. Before add relationship to ontology, the relation verb needs clustering to find relation types. Finally, the relation types are added to ontology. Through the above three functions the goal can be achieved.