Improving the Performance of Answers Ranking in Q&A Communities: A Long-Text Matching Technique and Pre-Trained Model

Abstract

This paper introduces TR-BERT, a novel method to answer ranking in Question & Answer (Q&A) communities, designed to tackle the widespread challenges of irrelevant popular answers and the neglect of new questions. TR-BERT integrates a long-text matching technique with a pre-trained language model. This ranking method effectively filters the noise and extracts textual features of questions and answers in QA communities. The experimental results on the Zhihu Q&A community dataset and the SemEval-2017 dataset showed the effectiveness and superiority of the TR-BERT. The contributions are as follows: Designing a new framework to process long-text data by filtering the noise and developing the TR-BERT to optimize the issue of answer ranking in the Q&A community. The experiment also showed that the TR-BERT model has the advantages of faster speed and requires less computational resources, which makes the TR-BERT valuable for practical applications. Meanwhile, TR-BERT offers an insight: By removing noise from the input text to shorten the length of the input sequence, we can decrease the time and computational resources required for model training and computation. This leads to the potential for smaller models, faster speeds, reduced computational resource demands, and improved efficiency.