Recommending Batch of Experts to Questions on Stackoverflow

Abstract

order to assist the process of questions answering on CQA (Community Question Answering) websites, this paper proposes an improved methodology of batch recommendation of answerers (experts) to questions called BESF (BERT Expert Recommendation using Multi-Objective Sailfish Algorithm with Genetic Algorithm). First, experts and questions modeling is done using BERT Topic modeling technique, which creates clusters on the base of topics. Using TF-IDF values calculated by BERT, Question-Expert similarity, Question-Topic similarity and New-Old questions similarity are calculated, which helps in classification of new questions. Using the calculated similarities in each cluster, experts are ranked on the base of four basic parameters, i.e. reputation, past performance, recent activity and activeness. Keeping in view the bounded number of experts and avoiding duplicate answers to repeated or similar questions, this methodology optimizes three parameters i.e. increased question coverage, increased answerability and decreased expert resources utilization. This becomes a multiobjective optimization problem and MOSFO-GA (Multi-Objective Sailfish Optimization with Genetic Algorithm) is used to address this problem. The proposed approach is evaluated on StackOverflow dataset which shows that using BERT for topic modeling and clustering, gives better clustering results as well as increases the performance as a whole, in comparison with using MOSFO-GA for clustering. This approach can be helpful in time conservation of users and providing better answers to questions by recommending batch of experts to answer the questions.