2021/01/26 by Ilias Chalkidis, Manos Fergadiotis, Chalkidis, Ilias +7 · 2 citations
Computer Science · #Computation and Language (cs.CL) #Cybercrime and Law Enforcement Studies #FOS: Computer and information sciences #Information Retrieval (cs.IR) #Sentiment Analysis and Opinion Mining #Software Engineering Research
paper · pdf · doi:10.48550/arxiv.2101.10726
openalex publication_date 2021/01/26 · openalex created_date 2025/10/10 · openalex updated_date 2026/07/28
Major scandals in corporate history have urged the need for regulatory\ncompliance, where organizations need to ensure that their controls (processes)\ncomply with relevant laws, regulations, and policies. However, keeping track of\nthe constantly changing legislation is difficult, thus organizations are\nincreasingly adopting Regulatory Technology (RegTech) to facilitate the\nprocess. To this end, we introduce regulatory information retrieval (REG-IR),\nan application of document-to-document information retrieval (DOC2DOC IR),\nwhere the query is an entire document making the task more challenging than\ntraditional IR where the queries are short. Furthermore, we compile and release\ntwo datasets based on the relationships between EU directives and UK\nlegislation. We experiment on these datasets using a typical two-step pipeline\napproach comprising a pre-fetcher and a neural re-ranker. Experimenting with\nvarious pre-fetchers from BM25 to k nearest neighbors over representations from\nseveral BERT models, we show that fine-tuning a BERT model on an in-domain\nclassification task produces the best representations for IR. We also show that\nneural re-rankers under-perform due to contradicting supervision, i.e., similar\nquery-document pairs with opposite labels. Thus, they are biased towards the\npre-fetcher's score. Interestingly, applying a date filter further improves the\nperformance, showcasing the importance of the time dimension.\n