웹 상의 사실 정보를 이용한 지식 패턴 필터링 = Pattern Filtering for Knowledge Extraction Using Facts from Web

다국어 초록 (Multilingual Abstract)

The Web provides a large amount of knowledge that can be utilized for various purposes. Since most knowledge is expressed in natural language, a user can utilize the knowledge by reading a text expressing the knowledge. However, it is difficult to find knowledge appropriate to the user’s needs due to the large amount of texts. As a solution of this problem, there have been studies on automatically extracting knowledge from a text by using patterns. For this, in general, patterns are automatically generated and selected according to their statistics. However, there is no explicit way to consider whether a pattern is semantically right or not. Hence, semantically wrong patterns result in failure of knowledge extraction. This paper proposes a method for filtering semantically wrong patterns by using factual information obtained from the Web. The proposed method extracts sentences containing natural language expressions of already known knowledge and then these sentences are regarded as factual information. Confidence of a candidate pattern is measured by compared it with the factual sentences. If the confidence is lower than a threshold then the pattern is filtered out. In order to show superiority of proposed method, it is applied to a parse tree pattern-based knowledge extraction method for Korean texts. Our method achieved 0.826 accuracy and it outperformed the existing method by 0.191. The results imply that our proposed method is plausible for pattern filtering.

목차 (Table of Contents)

• I. 서 론 ············································································· 1
• II. 관련 연구 ······································································· 7
• III. 패턴 필터링 ································································· 13
• 3.1 관계에 대한 사실 정보 추출 ····································· 14
• 3.2 지식 패턴 필터링 ···················································· 15
• IV. 실 험 ········································································· 20
• 4.1 실험 방법 ······························································ 20
• 4.2 실험 데이터 ··························································· 21
• 4.3 비교 모델과 실험 설정 ············································ 25
• 4.4 실험 결과 ······························································ 27
• 4.5 실험 분석 ······························································ 29
• V. 결 론 ·········································································· 33
• 참고 문헌 ········································································· 35
• 영문초록 ··········································································· 39