<![CDATA[This study discusses improving book search relevance in digital libraries by implementing semantic similarity. This study aims to improve search relevance and contextual understanding by integrating the TF-IDF method. This process includes text preprocessing, TF-IDF weighting, and semantic similarity calculation using the Wu-Palmer method with the aid of WordNet. Testing was carried out with 25 different queries on 1000 digital books in five scenarios. The test results show: (1) The 3-term TF-IDF scenario produces the highest average similarity, ranging from 71.83% (1-word query) to 40.08% (5-word query or more), with a standard deviation increasing from 15.82% to 28.76%; (2) The 5-term TF-IDF scenario shows an average similarity from 61.36% to 35.69% and a standard deviation from 19.91% to 28.64%; (3) The 10-term TF-IDF scenario provides an average similarity of 53.6% to 31.77%, with a relatively stable standard deviation in the range of 27-28%; (4) The scenario without TF-IDF has the lowest average similarity, 42.14% to 25.6%, with a standard deviation decreasing from 30.31% to 27.13%; (5) The non-semantic scenario yields the highest average similarity of 80% (1-word query) and decreases to 38.25% for long queries, with no standard deviation due to the absence of score variation. This study proves that the combination of semantic similarity with the top 3 TF-IDF terms is an optimal approach in increasing the relevance of search results compared to other scenarios, especially for short to medium queries. These results contribute to developing a more effective and contextual digital library search system.]]>
