SAPUTRA, GILAS ADI (2026) IMPLEMENTASI NAMED ENTITY RECOGNITION DAN ALGORITMA A-STAR PADA ASISTEN VIRTUAL UNIVERSITAS MERCU BUANA MERUYA. S1 thesis, Universitas Mercu Buana Jakarta.
|
Text
COVER.pdf Download (359kB) | Preview |
|
|
Text
BAB 1.pdf Download (52kB) | Preview |
|
|
Text
BAB 2.pdf Download (131kB) | Preview |
|
|
Text
BAB 3.pdf Download (460kB) | Preview |
|
|
Text
BAB 4.pdf Download (814kB) | Preview |
|
|
Text
BAB 5.pdf Download (31kB) | Preview |
|
|
Text
DAFTAR PUSTAKA.pdf Download (150kB) | Preview |
|
|
Text
LAMPIRAN.pdf Download (390kB) | Preview |
Abstract
In this increasingly sophisticated era, artificial intelligence technology enables more efficient human-computer interaction. In this study, the researchers designed and implemented a voice-based virtual assistant for the Faculty of Computer Science at Mercu Buana University Meruya using a hybrid NER architecture. This system combines two processing paths: a “fast lane” based on Named Entity Recognition (NER) with a fuzzy string matching method for deterministic recognition of faculty and staff entities, and a “slow lane” that utilizes ChromaDB semantic search and the Gemma 4 Large Language Model (LLM) as a fallback mechanism. The system is equipped with campus navigation features using the AStar algorithm, a Live2D character interface, and voice conversion via Retrievalbased Voice Conversion (RVC). NER performance evaluation was conducted on 1,000 test data points covering realistic input variations such as Speech-to-Text transcription errors, pronunciation variations, and partial name mentions across two data split scenarios (80:20 and 70:30) to measure the precision, recall, and F1-score for each entity label. Keywords: Virtual Assistants, Named Entity Recognition, Fuzzy String Matching, Large Language Model (LLM), A-Star Algorithm Pada perkembangan zaman yang semakin canggih, teknologi kecerdasan buatan memungkinan interaksi manusia dan komputer jadi lebih efisien, Pada penelitian ini peneliti merancang dan mengimplementasikan asisten virtual berbasis suara untuk Fakultas Ilmu Komputer Universitas Mercu Buana Meruya dengan meggunakan arsitektur Hybrid NER. Sistem ini menggabungkan dua jalur pemrosesan : fast lane berbasis Named Entity Recognition (NER) dengan metode fuzzy string matching untuk pengenalan entitas dosen dan tenaga kependidikan secara deterministik, lalu jalur slow lane yang memanfaatkan semantic search ChromaDB dan Large Language Model (LLM) Gemma 4 sebagai mekanisme fallback. Sistem dilengkapi fitur navigasi kampus menggunakan algoritma A-Star, antarmuka karakter Live2D, serta konversi suara melalui Retrieval-based Voice Conversion (RVC). Evaluasi Kinerja NER dilakukan terhadap 1.000 data uji yang mencangkup vairasi input realisitis berupa kesalahan transkripsi Speech-to-Text, variasi pelafalan, dan penyebutan nama parsial pada dua skenario pembagian data (80:20, dan 70:30) guna mengukur precision, recall, dan F1-score dari setiap label entitas. Kata kunci: Asisten virtual, Named Entity Recognition, Fuzzy String Matching, Large Language Model, Algoritma A-Star
Actions (login required)
 |
View Item |
