Σεμινάριο για μεταπτυχιακούς φοιτητές "Deep Learning for Legal Texts" 25/11 16:00, Αίθουσα Α56

Ημερομηνία: Δευτέρα 25/11, Αίθουσα: Α56

Ομιλητής: Ηλίας Χαλκίδης, Υποψήφιος Διδάκτορας, Οικονομικό Πανεπιστήμιο Αθηνών

Τίτλος: Deep Neural Networks for Information Mining from Legal Texts

Το σεμινάριο θα προσμετρηθεί στις υποχρεώσεις για λήψη διπλώματος των μεταπτυχιακών φοιτητών του ΠΜΣ "Πληροφορική και Τηλεπικοινωνίες".

Abstract:

Legal text processing [Ashley, 2017] is a growing research area where Natural Language Processing (NLP) techniques are applied in the legal domain. There are three main legal sub-domains where such techniques are applied: legislation, court cases and legal agreements (contracts). A large number of companies, including hundreds of start-ups, operate in the emerging legal-tech and reg-tech industries in order to provide text analytics. They target a wide variety of use cases (i.e., Due diligence/Contract Negotiations, Contract Management/Vendor Management, Legislation/Regulation Drafting, Regulatory Compliance, and Judicial Assistance) that are currently poorly handled due to the excessive amount of data (documents), which is difficult to be analyzed by humans.
Recently, Deep Learning [Goodfellow et al., 2016; Goldberg, 2017] has gained significant attention in the Natural Language Processing (NLP) research community, as in many areas of Artificial Intelligence. Deep Neural Networks (DNNs) have been rapidly replacing rule-based approaches, dictionary-based models and traditional machine learning techniques (i.e., linear models, decision trees), which in their majority require intensive manual feature engineering. DNNs and machine learning in general have also been gradually introduced in the legal domain, where researchers traditionally employed manually crafted knowledge bases and patterns to capture legal concepts, terms of interest and synonyms that were defined beforehand. The goal of this PhD project is to to explore and advance deep learning methods for legal tasks, such as legal judgment prediction, topic classification and identification of relevant court cases, that have already been discussed in the literature or may arise in this project in order to assist the aforementioned use cases. Further on, we plan to investigate how can legal NLP systems be efficiently transferred from one language to another [Artetxe and Schwenk, 2018;  Lample and Conneau, 2019] and how can we provide explanation for the decisions (predictions) of such systems, which is of crucial importance in the legal domain (e.g., when predicting the outcome of a trial).