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.
