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...

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

I obtained the following key findings:• Some expected discription types with high inter-annotator agreement, thatare definition, order of time and procedure, were found.• In these description types, the possibility of accurate annotation by nonprofessionalswas shown.• Proposition of new techniques of answer categorization based on the descriptiontype. The experimental results showed a high accuracy of theproposed method with features of functional words, 0.8 in F-measure, forthe three description types: definition, order of time and procedure.I also pointed out some issues concerning the classification and tagging ofdescriptive answers as future work. The first task was to develop mechanisms tocontrol agreement and disagreement in discourse tagging. The second task wasQ&A corpus balanced in question type. I also needed to consider the mixture ofprofessional and non-professional tagging in discourse annotation.In Chapter 5, I present effective features that can be used to categorize listsin web pages by whether they explain a procedure. I showed that categorizationto extract texts including procedural expressions was different from traditionaltext categorization tasks with respect to the features and behaviors related toco-occurrences of words. I also showed the possibility of filtering to extract listsincluding procedural expressions in different domains by exploiting those featuresthat primarily consist of function words and patterns with mutual informationfiltering. Lists with procedural expressions in the Computer domain can be extractedwith higher accuracy.I obtained the following key findings:• When restricting the document structure of answers to a presentation oflists, a moderate accurate extraction of procedural expressions can be performedwith sequential pattern mining and support vector machines.• This method showed a high performance, more than 0.7 in F-measure, whenextracting lists of procedural expression.83

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

Saved successfully!

Ooh no, something went wrong!