This thesis describes a concept-based question-answering system. User's natural language queries that ask the same types of answer have common semantic elements. According to this fact, we define common semantic elements as concepts and use them for q...
This thesis describes a concept-based question-answering system. User's natural language queries that ask the same types of answer have common semantic elements. According to this fact, we define common semantic elements as concepts and use them for question-answering.
By analyzing the natural language query corpus, we define concepts in each type of answers and construct concept-based rules for question analysis. A concept-based question-answering system has three steps like a typical question-answering system. The first step is question analysis, the second step is document and/or passage extraction, and the third step is answer extraction.
In the question analysis, if a user's natural language query is inputted, the concept-based rules for question analysis are applied to the user's query. If one of concept-based rules does apply to user's query, types of answer are classified and concepts in user's query are extracted by the rule.
In the document and/or passage extraction, the user's query is reconstructed and/or expanded by applying concepts extracted from the user's query and synonyms to the rules for query expansion. And after documents are extracted through the common web search engine, passages are chosen by the reconstructed and/or expanded queries.
In the answer extraction, answer candidates are extracted by applying the concept-based rules for answer extraction to passages extracted in the document and/or passage extraction step. If the concept-based rules for question analysis do apply to user's query, question-answering is performed by using the properties of query's form.
The experimental results show that the concept-based question-answering system has good performance even if it is performed under the restricted person domain. It also shows that the result of the question-answering using purely concept-based rules is accurate remarkably.
In order to enhance the total performance, a concept-based question-answering system may be a subsystem of other legacy question-answering system.
The concept-based rules used in a concept-based question-answering system may be dependent on language, but the concept-based approach, which gives meanings to words of user's query, is independent of language.
，免费韩语论文，韩语论文题目 |
