Estimating Twitter user's interest for inviting potential participants into Web-based debate platforms based on BERT

As the growing usage of online debate system, more and more new platforms or that of local community encounter the barrier of inviting participants before the debate starts or when debate in the stagnation situation. The traditional way to promote debates and invite participants for the platforms can be expensive and low efficient due to the feature of web debate, time-free and local-free. We propose a system to tackle the problem by applying Bidirectional Encoder Representations from Transformers (BERT) in this paper. We adapt BERT to two folds of our system to get feature vectors. First fold output aspect-like feature by using hashtags of Twitter and second fold output contextual representation of text to help obtain contextual feature. Finally, the system will generate a score to estimate how much the SNS user is interested in certain debate topic for supporting the administrator of the debate to make decision on whether to make the invitation. We conduct an estimation experiment with questionnaire and obtain over 0.61 of the correlation coefficient which shows that the goal of our system is feasible.