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.

elements were usually eliminated. However, some particular areas of text categorization,for example, authorship identification, suggested the feasibility of textcategorization with functional expressions on a different axis of document topics[63, 147, 187].4.3 Annotating description types of answersAs stated at the beginning of this chapter, the classification of answers has oftenbeen discussed as it is integrated with the classification of questions. However,there are no established categories of descriptive answers, and the relationshipsbetween classification categories and question categories have not been clarifiedeither. Therefore, I conducted an experiment to classify answers using the classificationcategories based on the discursive features on general texts that wereproposed in leading studies. The classification was performed by tagging the answerarticles. I tried to clarify necessary conditions for categories of descriptiveanswers and those tagging methods.4.3.1 description typesTo further explore description types of answers, this thesis considered the frameworkto solve four problems comprising those described in last section. For thefirst problem of collection of corpus for annotating description type and the secondproblem of reduction of annotation cost for tagging, this thesis suppose a networkenvironment for anonymous annotators tagging descriptive types to articles.To realize such kind of annotation framework, at least, I have to know any descriptiontypes that can be stably assigned by non-professional annotators. Thisthesis supposed instructions of annotations and definitions of description types ina level of book of technical writing for general readers, and then investigated thefeasibility of annotation in such kind of discursive features of text. For the thirdproblems, this thesis stands on machine learning based approaches to automaticallyacquire rules to specify descriptive type from tagged corpus. Finally, forthe forth problem of feature analysis for answers in Japanese question-answering,I conducted annotation of description types to answers in a actual web Q&Aservice, and examine the features of description types.51

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

Saved successfully!

Ooh no, something went wrong!