TeCoFeS: Text Column Featurization using Semantic Analysis

Extracting insights from text columns can be challenging and time-intensive. Existing methods for topic modeling and feature extraction are based on syntactic features and often overlook the semantics. We introduce the semantic text column featurization problem, and present a scalable approach for automatically solving it. We extract a small sample smartly, use a large language model (LLM) to label only the sample, and then lift the labeling to the whole column using text embeddings. We evaluate our approach by turning existing text classification benchmarks into semantic categorization benchmarks. Our approach performs better than baselines and naive use of LLMs.