Mining textual social media and crowdsourced questions / Iman Hassan Abdallah Mahmoud Saleh ; Supervised Osman Hegazy , Neamat Eltazi

Thesis (Ph.D.) - Cairo University - Faculty of Computers and Information - Department of Information Systems

Different types of social media platforms are getting more and more popular nowadays. Community Question Answering (cQA) websites are particularly getting increasingly popular. Such websites allow users to ask questions and find answers to them.Therefore, they can be considered as a valuable source of information. Numerous questions are posted to these websites on a daily basis, forming huge archives of questions. These archives offer a large body of text that can be analyzed and exploited in several text mining tasks. In this thesis, two different research problems related toUserGeneratedQuestions (UGQ{u2019}s), found in cQA websites, are studied. The first research problem is identifying semantic concepts in these questions. Identifying such concepts enables understanding questions and finding answers to them automatically.This is particularly useful in websites where a large number of questions is posted and it can be diffi- cult to answer all of them. In this thesis, we perform a comprehensive study on the use of reranking combined with linguistic information, in the form of syntactic and semantic structures, to find semantic concepts in UGQ{u2019}s. We evaluate our methods and show that techniques presented can outperform the state of the art for that task. The second research problem studied in the thesis is also related to UGQ{u2019}s. Our goal is to organize huge archives of UGQ{u2019}s by organizing tags assigned to questions into communities of semantically related tags. We investigate different ways that can be utilized to find semantic relationships between tags, and the best way to model tags in the vector space to find similarities between them

Issued also as CD