Shiraz: A Proposed List Wise Approach to Answer Validation

Abstract

Answer Validation is an important step in Automatic Question Answering systems and nowadays by spreading Community Question Answering systems it is known as an important task by itself. Previous works just considered it as a binary classification problem in which they try to find the best answer among all the candidate answers for a question. Accordingly, they do not consider the possible unique information which may have been included other answers. This can be considered by having a multiclass label classification problem, it is not only able to find the best answer but also can find”potentially good”,”bad”, and etc. answers too. By doing so, it is fully expected to extract and rate all the necessary information from existing candidates to help questioner to find the best and general answer for his question. This work tries to consider some features which are gained from importance of comments of the questioner. Finally, by using a good classifier, we try to overcome this problem. The designed system participated in subtask A of the Semeval-2015 Task 3. The primary submission ranked at the 5th and 7th places in four class label and three class label evaluation, accordingly.