Claim Missing Document
Check
Articles

Found 2 Documents
Search
Journal : Jurnal Algoritma

Evaluasi dan Implementasi Indobert Question Answering (QA) pada Domain Spesifik Menggunakan Mean Reciprocal Rank Ramadhan, Teguh Ikhlas; Supriatman, Agus; Kurniawan, Taufik Rahmat
Jurnal Algoritma Vol 21 No 1 (2024): Jurnal Algoritma
Publisher : Institut Teknologi Garut

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33364/algoritma/v.21-1.1542

Abstract

Applications of artificial intelligence, such as Chat-GPT and Bard, have become common in various aspects of life today. One of the main aspects is the use of the Question Answering (QA) model to meet specific domain needs. However, in some cases, models like Bard may not be able to provide specific information, such as registration procedures at a particular company or lecturer schedules at a university. To overcome this challenge, an Indonesian QA model called IndoBERT-QA has been developed. This research aims to evaluate the capabilities of IndoBERT-QA in a specific domain context using the Mean Reciprocal Rank (MRR) evaluation method. The evaluation results show that the IndoBERT-QA model is able to achieve an MRR of 0.91 with an approach to creating a customized context for each question. These results indicate that this model has good performance in providing relevant answers. The benefit of this research is that it serves as a valuable reference for the development of QA systems that will be created by other parties in the same environment. By utilizing a well-tested and evaluated approach, this research provides a strong foundation for the development of a high-performance QA system in Indonesian, so that it can meet domain-specific needs. Besides that, this research also provides insights for further development. One approach that can be explored is the use of Information Retrieval or Passage Retrieval as an initial step in the QA process. This can help the model in getting more precise and relevant context, thereby allowing further improvement in the quality of the answers provided by the model. Index Terms-BERT, Mean Reciprocal Rank, Question Answering However, this research also provides insights for further development. One approach that can be explored is the use of Information Retrieval or Passage Retrieval as an initial step in the QA process. This can help the model in getting more precise and relevant context, thereby allowing further improvement in the quality of the answers provided by the model.
Passage Retrieval untuk Question Answering Bahasa Indonesia Menggunakan BERT dan FAISS Ramadhan, Teguh Ikhlas; Supriatman, Agus; Kurniawan, Taufik Rahmat
Jurnal Algoritma Vol 21 No 2 (2024): Jurnal Algoritma
Publisher : Institut Teknologi Garut

Show Abstract | Download Original | Original Source | Check in Google Scholar | DOI: 10.33364/algoritma/v.21-2.2100

Abstract

This research develops a passage retrieval model for a Question Answering (QA) application in the Indonesian language, focusing on a specific domain. The model leverages BERT embedding techniques and the Faiss index to enhance the efficiency and accuracy of finding answers to user queries, with a particular focus on a text corpus related to Universitas Perjuangan Tasikmalaya. The evaluation was conducted on 80 questions, encompassing various informational aspects within the corpus. Results indicate an average execution time of 0.23 seconds per question and a total processing time of 18.8 seconds for all queries, achieving an accuracy rate of 43.75%. Accuracy was generally higher when the questions contained terms that exactly matched those in the corpus. While the initial findings are promising, the accuracy remains suboptimal and warrants further improvement. Potential areas for optimization include employing alternative embedding techniques, refining passage formation methods, and enhancing search performance using a cross-encoder. This research contributes to accelerating the retrieval process and improving the relevance of results for QA applications within specific domains.