Article Preview
Top1. Introduction
Research in artificial intelligence and its implementation in the legal domain have been ongoing processes. The law has always been a challenge, involving the extraction of meaningful and relevant insights from judgments and law-related complex and voluminous texts. Modern innovations like deep learning techniques have opened up a world of possibilities and made research exciting. This work aims to predict case types issued by the Supreme Court of India. This is done by training and building a deep learning model that offers lawyers solutions in terms of ascertaining the success rate of a particular case with considerable accuracy.
Text processing is becoming an important research area of practical applicability in this period of modern computing. Now a days, most legal documents and jobs are digitized for unchallenging processing and retrieval of legal information. Digitalization has made a critical effect on the economy, particularly within the legitimate system. Legitimate advisors and Attorneys frequently access long lengthy documents for referring to the previous cases. Our approach provides high-quality legal reference information from texts of past judgments, and helps lawyers predict the probability of a case receiving a favorable judgment in a cost-effective and time-saving mode.
Do et al. (2017) stated that Deep Neural Network has been used steadily in the legal domain. In recent times, deep learning techniques have started to play a significant role in solving challenging applications that require natural language processing. The legal domain consisting of massive volumes of text data has significant potential for accommodating deep learning techniques, particularly deep neural networks. Deep neural networks have been speedily returning rule-based methods, standard dictionary-based methods, and traditional machine-learning algorithms in their most essential in-depth manual feature- engineering.
We found that most researchers used deep representations for generic purposes, or in some cases, domain-specific word embedding, which was pre-trained over small legal datasets. We focus on the semantic feature representations, a key instrument for the successful application of deep learning in natural language processing. Chalkidis, I., &Kampas, D. (2019) explained the pre-trained legal word embeddings using the word2vec model over large corpora, comprising legislations from the countries, Australia, United Kingdom of Britain, European or French, America, Canada, and Japan among others.
Here, we present a brief discussion about multitasking applications that have been implemented by deep learning algorithms to the legal analytics which can be divided into 3 categories; 1. Legal search document, 2. Legal document analysis, 3. Legal perspective interfaces as shown in Fig.1. The first list defines developed systems used to retrieve and classify the applicable document. The second list defines the Natural Language Processing (NLP) analysis such as legal information retrieval, classification, case prediction, text extraction from the legal case judgments. In our paper we have focused on the text classification, information extraction and prediction of judgment text.
Figure 1. Legal Analytics-Multitask
To the best of our knowledge, there is no publicly available legal subdomain of deep learning applications. Our work spotlights the feature representations, the deep neural networks architecture, and important observations derived from the results.
We prefer to address following three questions about the use of multi-task deep learning in the legal domain:
- (1)
Is multi-task deep learning useful for legal tasks?
- (2)
What are the outcomes of multiple issues training jointly and separately?
- (3)
In the legal domain, can the multi-task method outperform the state of the art?
In our work we have used Indian Supreme Court data. Our proposed work represents,