12.07.2015 Views

file - ChaSen - 奈良先端科学技術大学院大学

file - ChaSen - 奈良先端科学技術大学院大学

file - ChaSen - 奈良先端科学技術大学院大学

SHOW MORE
SHOW LESS
  • No tags were found...

Create successful ePaper yourself

Turn your PDF publications into a flip-book with our unique Google optimized e-Paper software.

answers. Although different distributions of description style regarding differentdomains are predictable, some style can be considered to appear in various domains.Thus we can expect the feature of style in one domain to be also effectivein other domains. What styles are frequently used in descriptive answers? Howshould a style of description, that is description type, be defined? Because we aimat extraction of answers from articles in documents, do we have to take accountof linguistic expressions to define types of description style? Description type isnot equal to general document style or format but are not individual writing styleeither. We intend to find description types that can be used to accurately extracteach type of answer.As another solution to the difficulty of DQA, we could take account of exploitinghuman annotated semantic meta-data in the case of difficulty in extractingthe answer only using NLP, such as the example of a list and the procedurementioned above. What style in a Q&A corpus can be annotated as semanticmeta-data with high inter-annotator agreement? As the first step toward solvingthis problem, we performed description style annotation for Q&A articlesand studied the annotation results, clarifying features of the description style ofthe answer. Using the features of style, we conducted experiments of extractingarticles of a descriptive answer type, that is procedural expression from theWeb pages. Additionally, we explored the effective features of the extraction ofprocedural expressions.1.3 Guide to remaining chaptersWe overview previous studies of question-answering and related researches inChapter 2. Chapter 3 looks at multiple sentence query processing, and focuseson question segment extraction and question type identification for multi-sentencequeries. Chapter 4 and Chapter 5 are devoted to answer extraction. We discussannotation of description type to Q&A corpus in Chapter 4, and explore someexpected description type resulted in annotation experiment. In Chapter 5, wepropose the methodology of extraction of procedural expression from the Webpages using description type and machine learning, and show the effectiveness ofthe approach. Finally the thesis is concluded in Chapter 6.5

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!