@reviewpipe/clustering

Theme clustering for reviewpipe — groups reviews into labeled themes and fills result.themes.

npm install @reviewpipe/clustering

import { Pipeline } from '@reviewpipe/core';
import { ClusteringThemeExtractor } from '@reviewpipe/clustering';

const result = await new Pipeline()
  .source(adapter)
  .provider(provider)
  .themes(new ClusteringThemeExtractor({ k: 3 }))
  .run(input);

result.themes; // [{ label, keywords, mentions, sentiment, exampleReviewIds }]

Implements the core ThemeExtractor interface. When every review carries an embedding (from a provider that supports embed) it clusters with k-means; otherwise it falls back to keyword-frequency grouping — so it works with any provider. Clusters are labeled by cluster-vs-corpus TF-IDF, and each theme carries the mean sentiment of its reviews. k and the keyword/example counts are configurable.

License

Apache-2.0