GLARE: Guided LexRank for Advanced Retrieval in Legal Analysis

Abstract: The Brazilian Constitution, known as the Citizen's Charter, provides mechanisms for citizens to petition the Judiciary, including the so-called special appeal. This specific type of appeal aims to standardize the legal interpretation of Brazilian legislation in cases where the decision contradicts federal laws. The handling of special appeals is a daily task in the Judiciary, regularly presenting significant demands in its courts. We propose a new method called GLARE, based on unsupervised machine learning, to help the legal analyst classify a special appeal on a topic from a list made available by the National Court of Brazil (STJ). As part of this method, we propose a modification of the graph-based LexRank algorithm, which we call Guided LexRank. This algorithm generates the summary of a special appeal. The degree of similarity between the generated summary and different topics is evaluated using the BM25 algorithm. As a result, the method presents a ranking of themes most appropriate to the analyzed special appeal. The proposed method does not require prior labeling of the text to be evaluated and eliminates the need for large volumes of data to train a model. We evaluate the effectiveness of the method by applying it to a special appeal corpus previously classified by human experts.

• The paper introduces Guided LexRank, a modified algorithm that efficiently extracts and condenses essential information from legal appeals.
• It employs BM25 for similarity assessment, enabling a reliable ranking of relevant legal topics without extensive preprocessing.
• The evaluation on Brazilian special appeals shows that GLARE can match expert-level classification accuracy while reducing manual effort.

The paper "GLARE: Guided LexRank for Advanced Retrieval in Legal Analysis" introduces a novel approach tailored to support legal analysts in classifying special appeals in the Brazilian legal system. The Brazilian Constitution allows for special appeals to standardize interpretations of federal laws when there are discrepancies, a process demanding significant resources and expertise.

The authors propose a method called GLARE, which utilizes unsupervised machine learning to assist in this classification process. Central to GLARE is the modified version of the LexRank algorithm, known as Guided LexRank. This adaptation is designed to generate summaries of special appeals by operating on graph-based principles.

Key aspects of the GLARE method include:

1. Guided LexRank: This modification uses graph-based techniques to extract and summarize content from legal documents. The aim is to provide condensed representations of appeals that capture essential information.
2. Similarity Assessment with BM25: The summarized content is then compared to various legal topics using the BM25 algorithm, a state-of-the-art ranking function in information retrieval, which measures the similarity between text documents. This step enables the generation of a ranked list of topics relevant to the appeal being analyzed.
3. Efficiency and Effectiveness: The method operates without the need for preprocessing steps like labeling or extensive data training, which commonly hinder traditional machine learning approaches in legal contexts. This makes GLARE particularly suitable for the dynamic and data-limited environment of legal analysis.
4. Evaluation and Results: The authors test the effectiveness of the GLARE method on a dataset of special appeals previously classified by legal experts. Although specific quantitative outcomes aren't detailed, the approach is suggested to match or enhance the accuracy of human classification, providing timely and resource-efficient results.

Overall, GLARE exemplifies an innovative approach for automating and supporting legal analysis, reducing the workload on human experts while maintaining high standards of accuracy and relevance in complex legal interpretations.