טכניון מכון טכנולוגי לישראל
הטכניון מכון טכנולוגי לישראל - בית הספר ללימודי מוסמכים  
M.Sc Thesis
M.Sc StudentBraunstain Liora
SubjectSupporting Human Answers for Advice-Seeking Questions
in Community Question Answering Sites
DepartmentDepartment of Industrial Engineering and Management
Supervisor Professor Oren Kurland
Full Thesis textFull thesis text - English Version


Abstract

Community-based Question Answering (CQA) sites have become very       popular in the last few years.

 In these sites, users post questions of a wide range of topics (e.g., travel, music, health, etc.) and receive answers from other users.

In many questions posted on CQA sites users look for the advice or opinion of other users who might offer a diverse perspective on a topic at hand; the answers can be based on factual knowledge or subjective opinions.

Many times answers provided for these questions do not include additional information that can corroborate the answer with respect to the question. Thus, it may not be trivial for the asker to choose which answer, from all those posted, suits her information need as expressed in the question.

The task we address in this work  is providing supportive information for human answers to advice-seeking questions, which will potentially help the asker in choosing  answers that fit her needs. Our hypothesis is that additional supporting information will increase the effective utilization of CQA sites by users who look for answers.

We present a support-based  retrieval model that ranks sentences from the Web by their presumed support  for a specific human answer that was posted with respect to an advice-seeking question.

We use a learning-to-rank approach that integrates various features, and present a novel dataset for the task.

Empirical evaluation shows that our model is highly effective for ranking sentences by support. Furthermore, we demonstrate the merits of integrating relevance-oriented and support-oriented features for support-based ranking. In addition, the model outperforms a state-of-the-art textual entailment system designed to infer factual claims from texts.