<![CDATA[Selecting an appropriate method for an information retrieval system is a critical factor in achieving accurate and efficient search performance. This study aims to compare the performance of the Vector Space Model (VSM) and Latent Semantic Indexing (LSI) in the context of book information retrieval, with particular emphasis on semantic capability and computational efficiency. The dataset was constructed by merging two book datasets obtained from the Kaggle platform, which were originally sourced from Amazon's book catalog. After data normalization and duplicate removal, the final dataset consisted of 133,491 book records. The analysis focused on two primary attributes: book titles and book descriptions. The evaluation was conducted through two experimental scenarios: polysemy and synonymity testing, assessed using the Mean Absolute Percentage Error (MAPE) across 20 documents and five search queries, and retrieval speed testing, measured by response time on Google Colab using 20 dataset size variations. The experimental results indicate that LSI outperformed VSM in three of the five search queries, achieving the best MAPE score of 33.20%, whereas VSM recorded its lowest MAPE of 36.35% but deteriorated to 72.01% for queries with high semantic ambiguity. In contrast, VSM demonstrated superior computational efficiency in the retrieval speed evaluation, with response times ranging from 0.4583 ms to 7.2597 ms, substantially faster than LSI, which required between 2.5290 ms and 27.6777 ms. Both algorithms exhibited a linear increase in response time as the dataset size increased, with coefficients of determination of R² = 0.999 for VSM and R² = 0.997 for LSI. The findings reveal a significant trade-off between semantic accuracy and computational efficiency: LSI provides superior semantic understanding for information retrieval, whereas VSM offers substantially faster retrieval performance.]]>
Copyrights © 2026
