Institutional Repository of Vilnius University
Teisinių dokumentų skaitmeninė teksto analizė /
Abstract
dc:descriptionThe majority of the research in the field of the natural language processing are conducted for the most widely spread world languages, therefore the languages with difficult word structure, such as lithuanian, do not get much of attention. In this paper sequence autoencoder model is created for encoding legal documents. The model can later be used as a baseline model for an initialization of the other models for supervised learning in order to be more accurate and achieve faster training because of the learned features. For this purpose, a research is conducted about recurrent neural networks adaptation for solving problems with different lengths of input and output sequences. Also are reviewed: Vanishing gradient problem; Improvements on encoder-decoder model for processing sequence with bidirectional network; Stacking multiple layers; Providing reversed input sequences or using attention. During the research different models’ capability of restoring legal texts was compared, which showed that the best-performing model was LSTM model with attention. In addition, possible applications of the language model were specified such as text classification by case type or by judge. It was established, that the features of the created language model can be successfully applied for legal texts classification by case type.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2020
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Žaliauskas, Tadas,
- Contributors dc:contributor
-
- Petkevičius, Linas
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD81705840&prefLang=en_US