Assessing the significance of co-occurring terms in Goods and Services Tax in India using co-occurrence graphs and Attention based Deep Learning Models
Pankaj Dikshit, Bibhas Chandra · 2021 IEEE International Conference on Systems, Man, and Cybernetics (SMC) · 2021
Implementation of Goods and Services Tax (GST) in India in 2017 resulted in major tax reforms and the tax paying users tweeted their experience of the new system commenting on different aspects. Tweets of users commenting on the GST information system were extracted for the period June 2017 to May 2020 using pre-identified key terms. The paper presents a unique approach of combining co-occurrence graphs and Attention based deep learning technique for analyzing the importance of co-occurring terms in the tweets by taxpayers. Co-occurrence graphs between the terms in the tweets helped in identifying important terms related to GST, based on the frequency of their connections with the key terms. In order to assess how much the pair of key terms and the adjoining important term contribute, attention weights were found for this pair using Attention based bi-directional LSTM (Long Short Term Memory) model. A measure called "Attention factor" has been introduced to find the relevance of the pair in the tweets which will help in making improvements in the GST system.