Searching OWL Knowledge By Question-Answering

Searching information by natural language questions accords with behaviors of general users and is a more challenging problem than search engines. Database-driven question-answering system, Web-based question-answering system, and Ontology-driven question-answering are three cases. However, current researches on QA systems encounter a few urgent problems. Firstly, efficiency of building knowledge base by hand and quality of learning knowledge base from text are not satisfying. Secondly, most of mapping methods between natural language questions and Ontology queries are semiautomatic and require users to manually solve the ambiguity problems in semantic mapping. Thirdly, existing Ontology languages cannot represent fuzzy concepts and roles. OWL, the standard Web Ontology Language is recommended by W3C, has become the new standard for knowledge representation and exchange on the Internet. More and more Ontology knowledge of OWL version in different domains, such as Cyc general knowledge base and the time Ontology, is built by auto tools or by hand and released on the Web. How to utilize and query the OWL knowledge becomes an emergent research.Aiming at the above problems, the paper focused on three aspects of research on searching OWL knowledge by question-answering. They are acquiring right OWL knowledge, mapping natural language questions to OWL queries, and extending OWL representing fuzzy knowledge.In order to share and reuse existing OWL knowledge on the Web, the paper presents an idea of acquiring OWL knowledge indirectly. The key steps are searching OWL, clustering OWL, and perfect them. We propose an ordering method (called CWM) based on semantic analysis for coarse OWL constructs and a method of computing similarity between two OWL documents (named OWL-SSim) based on semantic analysis for refine OWL constructs. The CWM is used to filter OWL in a prototype search engine (WI OntoSearch). The OWL-SSim can be used to cluster OWL documents built by experts or OWL documents learned by automatic tools. We do some experiments on these methods and results are promising.Mapping a question into an equivalent OWL query is to map different syntactical constructs between them based on semantic equivalence. Fixing on words and elements as appropriate syntactical constructs, the paper proposes a three-phase semantic mapping framework (TPSM). In the framework, we research two algorithms for auto building semantic mapping between a set of words and a set of elements and a method for combining query language. They are FCSP-based semantic mapping (FCSP-SM), learning-based semantic mapping (LSM), and template-based combining method (TBCM). We have implemented a prototype system named Agile based on the framework and the above methods, and tested a series of experiments on three different topics of OWL knowledge bases and original sets of questions. The experimental results indicate that the precision of FCSP-SM, across all topics of questions is over 80%; natural language processing, priorities among constraints and refined methods have effect on the precision of the prototype system; J48 is a good method for all topic data; and the precision of LSM are between the most and worst priority orderings among constraints for FCSP-SM.Existing OWL only represents certain and complete concepts and roles. However, human being is used to represent and reason fuzzy knowledge. This paper extends OWL language by encoding fuzzy constructors, axioms and constraints, maps semantics of these new vocabularies to fuzzy description logic, and presents translation rules from OWL to extended OWL. In order to finish fuzzy query for extended OWL, we specify Precisiated Natural Language process based on extended OWL and gives an example of query film to show the system workflow.There are contributes in the paper as follow:1. In order to share and reuse existing OWL knowledge on the Web, the paper presents a method of computing similarity between two OWL documents (named OWL-SSim) based on semantic analysis. Results of experiments show that OWL-SSim is effective on clustering OWL documents built by experts or learned by automatic tools.2. The paper formalizes the task of semantic mapping into a Fuzzy Constraint Satisfaction Problem. The experimental results indicate that the model has three advantages. Firstly, the method is automatic. Secondly, domains and structures of OWL knowledge base have little influence on precision of the algorithm. Thirdly, priorities among constraints and refined methods have effect on the precision of the prototype system.3. In order to improve learning capability for semantic mapping from natural language questions to OWL query, we describe a mapping method based on learning in TPSM. The experimental results indicate that J48 is a good method for all topic data and the precision of the algorithm are between the most and worst priority orderings among constraints for FCSP-SM.4. The paper extends OWL as fuzzy OWL (FOWL) based on fuzzy description logic and discusses a fuzzy QA based on PNL process. FOWL can represent fuzzy Ontology knowledge and PNL process can deal with questions with fuzzy concepts and roles. In conclusion, these technologies and methods, special for OWL knowledge, proposed in the paper can be seen as cases and are used in other knowledge and information.