Studi Literatur Information Retrieval System Semantik Untuk Pencarian Produk E-Commerce

Abstrak

This research aims to analyze the development of the Semantic Information Retrieval System (Semantic IRS) approach in e-commerce product search based on descriptions through a literature study of 15 scientific articles consisting of national and international publications. The results of the analysis show that 33% of articles use the semantic IR and dense retrieval approaches as the basis for semantic mapping between queries and product documents. The late interaction and multimodal semantic retrieval approaches were each applied in 27% of articles, indicating an increasing research focus on token-level semantic interaction modeling and the integration of textual and visual information. Additionally, 13% of articles utilized query expansion and semantic relation modeling as supporting methods to improve search relevance. In terms of methodology, 80% of article used a quantitative experimental approach with information retrieval system metric-based evaluation, and 67% of articles adopted neural models. Overall, these quantitative findings indicate that neural model-based Semantic IR, late interaction, and multimodal approaches are the dominant and most relevant directions for handling long and unstructured description-based product searches in modern e-commerce systems.