Deep Neural Network to Predict Answer Votes on Community Question Answering Sites

Abstract

Stack Exchange (SE) is a popular community question answering site where a large number of questions and answers are posted every day. Three tabs called active, oldest, and votes are used by SE to list the users’ answers. The default tab to display the answers is the votes’ tab. The problem with the current listing mechanism of the answer is that newly posted answers are always placed at the bottom of the list because of no votes. To receive the votes from the users, the answer needs to be seen by the user. However, the current mechanism favoring the oldest answers to gain the user’s votes. To resolve this bias, and provide an equal opportunity to all answers, this paper suggested a deep learning-based framework that predicts virtual votes for answers as soon as it is being posted on CQAs. The predicted votes may use to lists the answers on SE. The proposed model helps the users to know how many votes may be received their answer in the future. This also motivates the users to post high-quality answers to receive more votes. The prediction of votes required only the textual content of the answers and hence it is free from the handcrafted feature engineering. To validate the model, three different datasets of the SE are used and found a promising performance on each dataset.