Depok, August 25th 2023. Faculty of Computer Science (Fasilkom) Universitas Indonesia (UI) student team won first place in the Ria Statistics and Data Science Festival (Satria Data) 2023 competition organized by the National Achievement Center, Ministry of Education, Culture, Research, and Technology of the Republic of Indonesia (Kemendikbudristek RI) in Malang on August 20th-24th 2023. The Fasilkom UI student team named Three Neuron V2 consisted of Bryan Tjandra, Nyoo Steven Christopher Handoko, and Oey Joshua Jodrian. The victory achieved in the Big Data Challenge category was obtained thanks to the innovation of the Three Neuron V2 Team in creating artificial intelligence (AI) that can automatically test conformity between laws and regulations.
Legislation has a central role in regulating the rights and obligations of every citizen. However, disharmony sometimes occurs between laws and regulations, such as conflicting contents of regulations that can threaten legal certainty for the community. Efforts to detect and test the conformity of laws and regulations is a complex task, and so far, it has been done manually. The innovation by the Three Neuron V2 Team can assist legislators and policy makers in keeping these laws and regulations in harmony and not colliding with each other.
This proposed AI-assisted solution takes advantage of Legal Textual Entailment (LTE), a subfield method in Natural Language Processing (NLP) and AI that focuses on determining a logical relationship between two legal texts. This solution is implemented using the LawID dataset which was also created by the Three Neuron V2 team for LTE purposes. The dataset consists of tens of thousands of data which includes primary data such as laws and regulations, starting from the level of the constitution, laws, to derivative regulations, such as government regulations and ministerial regulations, and includes secondary data such as court decisions from The Constitutional Court
In processing and summarizing regulatory documents, the Three Neuron V2 Team uses various language models to produce subtopics using the BERTopic technique. One of the models proposed in the experiment, namely BigBird-NLI, has been applied to the primary LTE section of the LawID dataset and managed to achieve an accuracy of 99% in the validation data after going through a fine-tuning process using 800 paragraph pairs from the document. These verse pairs are labeled using semi-supervised labeling and Large Language Models (LLM).
From this process it was found that the majority of pairs of paragraphs from regulatory documents showed harmony. While the percentage of incongruity is only around 0.6% of all verse pairs. Pairs of verses that are not in harmony are generally semantically incompatible without the need for in-depth context.
Dean of Fasilkom UI, Dr. Ir. Petrus Mursanto, M.Sc., appreciated the achievements of the students in the Satria Data 2023 event. “This achievement adds to a series of excellence track records of Fasilkom UI students in the field of information and communication technology at the national level. This is also a form of real contribution from the Fasilkom UI academic community to the nation and state,” said Dr. Ir. Peter. Meanwhile, the success of this team cannot be separated from the support of supervisor Dr. Panca Hadi Putra, B.Sc., S.H., M.Bus., who is a legal technology expert from Fasilkom UI .
The Satria Data 2023 competition was attended by 600 students consisting of 209 teams from 50 universities in Indonesia. Some of them are IPB University, Institut Teknologi Sepuluh November, Institut Teknologi Bandung, Universitas Gadjah Mada, Universitas Airlangga, Universitas Diponegoro, Universitas Brawijaya, and others. There are four categories in competition, namely the National Statistics Competition (NSC), Statistics Essay Competition (SEC), Statistics Infographic Competition (SIC), and Big Data Challenge (BDC).
