fast-topic-analysis

A tool for analyzing text against predefined topics using average weight embeddings and cosine similarity. It creates several weighted average embeddings for each topic instead of a single representation, capturing different semantic variations, and groups similar phrases within topics into coherent clusters using agglomerative or HDBSCAN algorithms. Cluster quality is measured via per-cluster cohesion and a global silhouette score, and preset configurations are offered for high precision, balanced, and performance use cases. The project centers on a generator that creates topic embeddings from training data and an interactive demo that analyzes text against them, powered by the embedding-utils library.

Setup

Install dependencies:

npm install