Clustering Algorithms for Queries: A Comparative Analysis of Farmer Call Center Data

Extracting insights from queries and feedback helps identify trends, enhance products and services, personalize customer interactions, and ‎craft effective marketing strategies. Data clustering, a powerful method, organizes unstructured data and refines queries by offering ‎suggestions based on similar or related inputs, ultimately enhancing the search experience. This study compares the performance of several ‎clustering algorithms, including Agglomerative Clustering, K-Means (KM), Hierarchical Density-Based Spatial Clustering of Applications ‎with Noise (HDBSCAN), and Density-Based Spatial Clustering of Applications with Noise (DBSCAN), as well as various embeddings, ‎such as Term Frequency-Inverse Document Frequency (TF-IDF)—Sentence-Bidirectional Encoder Representations from Transformers ‎‎(SBERT), Word2Vec, and GloVe. The Calinski-Harabasz Index, Davies-Bouldin Index, and Silhouette Score are used to measure the ‎effectiveness of these algorithms. Results indicated that HDBSCAN outperformed other clustering algorithms within the farmer helpline ‎dataset. The conclusions were derived from the medium-level performance of clustering algorithms. The findings showed that HDBSCAN, ‎combined with different embeddings, achieved a Silhouette Score of 0.85, a Davies-Bouldin Index of 0.66, and a Calinski-Harabasz Index ‎of 4,239.9‎.