University of Twente Student Theses

This website will be unavailable due to maintenance December 1st between 8:00 and 12:00 CET.

Towards transfer learning in e-discovery : finding the optimal classifier and evaluating domain adaptation methods

Pebesma, J.L. (2020) Towards transfer learning in e-discovery : finding the optimal classifier and evaluating domain adaptation methods.

[img] PDF
Abstract:Once fraud is suspected, an e-discovery process can be started. Currently, this process starts with the scoping the case, whereby filters and search terms are identified to limit the search results. These results are then manually reviewed on their relevance to the case, which is expensive due to the number of hours people needed for review. Machine learning has shown potential for this task, whereby both the content and metadata of communication seem to help the classification. However, not much research has been done on the use of machine learning in e-discovery. Ideally, a classification model is trained, in a way that requires no further input, except the emails themselves. Apart from limited available research into the most optimal machine learning model, there are two other major limitations to the use of machine learning in e-discovery. One is that the datasets are relatively small, thus some patterns may not be picked up on. Knowledge about finding relevant documents may be transferred between different fraud cases. By using the knowledge learnt from each case and applying transfer learning, the knowledge can be strengthened and findings can be solidified. Another limitation is that the machine learning model may act as a black box, which is inconvenient for tasks that focus on creating transparency, such as fraud investigations. This limitation is further discussed and possible solutions are considered, such as the implications of using XAI in e-discovery. This aim of this thesis is to optimise a classification model and create more insight into the transferability between cases while considering the implications of applying explainable AI in the legal domain.
Item Type:Essay (Master)
Faculty:EEMCS: Electrical Engineering, Mathematics and Computer Science
Subject:54 computer science
Programme:Computer Science MSc (60300)
Link to this item:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page