KweriME

In today’s age, humans are encountering problems for which they need a best and accurate answer. This is where a Question and Answer based site such as Quora, Yahoo!Answers, Stack Overflow etc., comes into picture. They are important sources of content creation providing an entire knowledge base for general questions or domain specific questions.

Such growing contents are posing several challenges which open up new opportunities for researchers to analyze their field of interest. Mostly, there are two types of users involved in these sites one being the asker, who asks or posts questions on these sites and waits for the question to be answered and second being the answerer who answers these questions which is posted by the asker. Usually multiple answerers could provide the answer, thus it becomes difficult to select the best answer.

Since multiple answers are available to the asker on these sites, it is the need of the hour that a best answer must be selected amongst them to provide ease to the asker thus, ensuring a proper and optimized working environment for both the asker as well as the answerer.

Purpose

• KweriME is a Q&A based model which predicts accepted answers of various questions in a CQA sites. Seeking information on the internet has become a daily part of our lives. This inspired us to build an efficient system which reduces delay as well as gives us additional information on the content.
• Most of the Q&A systems experience a delay which we try to solve by forwarding the questions to experts thus, ensuring improved answer quality and also incorporate an anti-spammer control to filter out irrelevant and advertising verbiage to optimize user interaction.
• Instead of going through multiple questions, the asker will be satisfied with the best and optimal answer, thus, saving time and energy of the asker.

This study addresses the answer to unanswered research question - how are the best answers selected?

1. First, we use activity signatures, domain knowledge and topical similarity of the user to identify active answerers in the domain of the questions asked.
2. Second, we use topical interest, topical expertise for expertise computation using topic modeling and voting scores.
3. Third, we find topic relevancy to find the relationship between Q&A pairs.
4. Finally, we focus on features involved in posts for predicting whether the answer to the question will be accepted or not based on different classifiers.

The proposed approach examines StackOverflow Q&A posts with no less than five responses to extract features for pattern identification using which the best answer is selected for the asked questions based on topic modeling and classifier.

To assess rightness of the proposed model, a set of parameters are utilized, for example, Receiver Operating Characteristics Area Under Curve, Precision Recall Area Under Curve, F-score, and Accuracy. In our approach, a model is developed for choosing best answer for the question asked on the CQA site. Rather than taking data associated with question and answer alone into account as done in the manual process, this model takes both question-answer and answerers’ information into account, which gives an understanding perspective into the appropriate responses given by the experts that is more liable to be chosen as the best answer. Results demonstrate that the proposed model is compelling in anticipating the best answer.

Results

Dig deeper?

You can find out more about this project at Github.