Using Topic Modeling in Classification of Brazilian Lawsuits

Why Do Tenants Sue Their Landlords? Answers from a Topic Model

“…Overall, relevant CTs are more likely to be obtained with BERTopic rather than with LDA. One explanation is that unlike most works described in Section 2 that dealt with documents from different legal areas [11,12,13], our corpus of paragraphs is much more homogeneous as it is only related to housing law, hence making topic modeling more difficult. Consequently, the bag-of-words approach of LDA gives less relevant topics compared to BERTopic, which has access to word semantic information.…”

“…Such methods extract topics that can be described as collections of words clustered together based on their distribution across the documents. In the legal field, topic models were applied for instance to UK legislative documents [7], Latvian legal acts [8] and court decisions from Australia [9], the Netherlands [10], Brazil [11,12] and the United States [13]. We must emphasize that these topic modeling experiments usually yielded topics meant for experts in the field.…”

“…Despite readily available toolkits facilitating topic modeling, evaluating topics is still an open question, in part because the assessment differs depending on the data and the legal area. A possible strategy is an extrinsic evaluation of the topics, for instance measuring the improvement topics bring when carrying out text classification [11,12]. Such a protocol obviously requires the cases to be manually classified beforehand (this is not our case).…”