Trending Topic Detection during Pandemic (Covid- 19) In Indonesian Tweets Using the Document Pivot (Doc-p) Method and BN-grams

People are providing a variety of information about the coronavirus outbreak, including the needs of affected communities and the location of the outbreak. Such information can be used as one of the resources to map the coronavirus outbreak events and the needs of affected people in Indonesia. However, the information obtained from social media has an informal structure and has low reliability as an information provider. Unstructured social media data makes it difficult to identify information related to trending topics, especially related to the coronavirus outbreak. Therefore, in this study, we used Document Pivot (Doc-p) and BN-grams (January to May 2020) methods to detect trending topics in Indonesian tweets. In our experiments, we investigate the impact of different topic numbers and master data on the quality of the resulting trending topics. We measure the accuracy of detecting trending topics by comparing both methods to trending topics found in local news and Twitter. Our experimental results show that using 10 topics yields the highest topic recall. Trending topics generated by BN-grams have the highest topic recall values. Stemming also reduces the quality of the resulting trending topics. The topic recall values of Doc-p and BN-gram from the four datasets are 75% and 50%, respectively. Overall, Doc-p has higher topic recall compared to his BN-gram because, unlike BN-gram, the dataset is used without stemming.