Grouping bug reports into clusters can assist in verifying and validating bugs in the software development cycle. One of the clustering methods is Agglomerative Hierarchical Clustering (AHC). It relies on distance calculations to determine the degree of similarity between clusters. One of the distance calculations is the Jaccard coefficient. The Jaccard Coefficient method has the disadvantage that it only considers the same set of words between two documents but does not consider their importance. Previous research added Inverse Document Frequency (IDF) algorithm to the Jaccard coefficient to calculate the importance of word groups and in this research is referred to the weighted Jaccard coefficient. Clustering is carried out using a combination of AHC and that coefficient. The silhouette score is then compared with the silhouette score of AHC with the Jaccard coefficient. Results indicate that increasing term complexity reduces cluster quality, with silhouette scores dropping from 13.13% (bigram) to 0.45% (4-gram). Furthermore, many clusters exhibited negative silhouette scores, highlighting the difficulty of separating high-dimensional bug data using unsupervised methods. In contrast, the supervised classification baseline achieved significantly higher accuracy. This paper contributes a critical analysis demonstrating that while Weighted Jaccard captures semantic nuance, unsupervised clustering remains insufficient for this domain compared to supervised approaches.