Semantic Clustering of News Headlines Using K-Means with Contextual Embeddings

Clustering short textual content like news headlines poses unique challenges due to their brevity and sparse context. Traditional techniques such as TF-IDF often fail to capture semantic relationships within such data. This study investigates the application of K-Means clustering on news headlines represented using contextual embeddings from state-of-the-art transformer models including BERT and GPT-۲. By leveraging the semantic richness provided by these models, we evaluate the effectiveness of clustering through various metrics. Our experiments demonstrate that contextual embeddings significantly enhance clustering quality over traditional methods.

کلیدواژه ها:

K-Means ، Contextual Embeddings ، News Headlines ، Clustering ، BERT ، GPT ، TF-IDF