IMPLEMENTASI PENCARIAN SEMANTIK DALAM TAFSIR AL-QURAN DENGAN ALGORITMA COSINE SIMILARITY DAN LARGE LANGUAGE MODELS

The search for relevant and accurate information is a crucial aspect of understanding the interpretation of the Quran. Implementing semantic search can help users find interpretations that align with their context and needs. The cosine similarity algorithm can be used to measure the similarity between a query and interpretation data. After using the cosine similarity algorithm, Large Language Models (LLMs) can be employed to deeply understand the meaning of queries and interpretations, resulting in more relevant and accurate searches. Therefore, this research aims to implement cosine similarity and LLMs for semantic search. The proposed semantic search system is built following the ML-SDLC method, which consists of the stages: planning, data collection, data preprocessing, model development, evaluation, and deployment. Users can perform searches by entering inputs into the search engine, the input will be converted into vectors, then the system will search the Quran interpretation dataset that has been converted into vectors and stored in a vector database. Cosine similarity is used to measure the semantic relevance between the query representation and the vector representation of the Quran interpretations. The GPT-4 model will summarize the semantic search results and present them to the user. Based on the model evaluation results using F1-Score vs. Threshold and ROC-AUC Score, the F1-Score yielded a value of 0.819 with a threshold of 0.410, while the ROC-AUC Score result was 0.587. The model testing results showed an accuracy of 96%, and for relevance testing, the results showed an accuracy of 89%.

Keywords: Semantic Search, Cosine Similarity, LLM, Quran Interpretation, Vector