Topic Clustering of Student Complaints Based on Semantic Meaning Using the indoBERT and K-Means Models

This study applies Natural Language Processing (NLP) technology to extract and cluster information from student complaint text data. The model used is IndoBERT, a variant of BERT (Bidirectional Encoder Representations from Transformers) that has been adapted for the Indonesian language. The main objective of this research is to perform topic clustering based on semantic similarity. The process begins with data collection and cleaning, followed by tokenization and text normalization. Each complaint is transformed into a vector representation through IndoBERT embeddings, which are then used as input for the K-Means clustering algorithm. Evaluation is conducted using various metrics, and the results of the Silhouette Score and Elbow Method indicate that the optimal number of clusters is four. Cluster visualization using the t-distributed Stochastic Neighbor Embedding (t-SNE) method reinforces these findings by displaying four fairly distinct groups of complaints, although one cluster appears dispersed and less well-defined, indicating possible topic overlap. The quality of topics within each cluster is evaluated using the Topic Coherence (c_v) metric, where Cluster 3 achieved the highest score of 0.7084. The topics in this cluster highlight critical issues such as campus facilities, lecturer quality, and information delivery systems. Overall, the four resulting clusters reflect central themes: Facilities, Expectations or Impressions, Services, and Academic Lectures. These results are expected to serve as a reference for institutions in formulating service improvement policies based on student complaint analysis.