Party Based Sentiment Analysis of Legal Opinion Texts 
Advised by: Amal Shehan Perera, Nisansa de Silva, Gathika Ratnayaka
University of Moratuwa
Factual scenario analysis of previous court cases holds a significant importance to lawyers and legal officers whenever they are handling a new legal court case. Legal officials are required to analyse previous court cases and statutes to find arguments and evidence before they represent a client at a trial. As the number of legal cases increases, legal professionals typically endure heavy workloads on a daily basis, and they may be overwhelmed and may be unable to obtain quality analysis. In this analysis process, identifying advantageous and disadvantageous relevant to legal parties can be considered a critical and time consuming task. By automating this task, legal officers will be able to reduce their workload significantly. Our ultimate goal of this research is to introduce a system to predict sentiment value of sentences in legal documents in relation to its legal parties. To achieve this task, we use a fine-grained sentiment analysis technique called Aspect-Based Sentiment Analysis(ABSA). To the best of our knowledge, this would be the first study that brings the concepts of Aspect Based Sentiment Analysis to the legal domain. To the best of our knowledge, there is no publicly available dataset for the aspect (party)based sentiment analysis for legal opinion texts. Hence, we created a dataset(SigmaLaw-ABSA) which consists 2000 legal opinion text fetched from court cases in order to train the models and to conduct the experiments. Then we developed a rule-based model which is primarily built around a phrase-level sentiment annotator and using rationally built rules to perform Party-Based Sentiment Analysis. The complexity of the structure of the sentences in legal texts has prompted to develop first a rule-based approach over other approaches. Developing this model, we got a much understanding on the legal domain, complex structure of the legal texts and many other benefits for the future works. But in the evaluation process, we experienced some major limitations as it's significantly depends on the phrase level sentiment annotator and there's a huge amount of scenarios which can not cover all using manual rules. Then we developed a deep-learning based approach to mitigate those limitation and to perform the task efficiently. After going through the different existing architectures for other domains which have their own capabilities of achieving accurate output overcoming different limitations, we came up with a new architecture which can be introduced as a combination of all the capabilities of the above models. We evaluated this model and existing ABSA models on the SigmaLaw-ABSA dataset and experiments showed that our model outperforms the state-of-the-art models for the SigmaLaw-ABSA dataset.
Keywords: Natural Language Processing | Machine Learning / Deep Learning | Law | Aspect-based Sentiment Analysis | Legal Domain |