@reviewpipe/core

The core of reviewpipe — the Pipeline engine, the pluggable interfaces, the NormalizedReview / AnalysisResult types, and the provider-agnostic aggregation logic.

Zero runtime dependencies. Model- and format-specific code lives in separate packages (adapters, providers, exporters); this package never pulls in an ML dependency.

npm install @reviewpipe/core

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

const result = await new Pipeline()
  .source(adapter) // a SourceAdapter — e.g. @reviewpipe/adapter-csv
  .provider(provider) // an AnalysisProvider — e.g. @reviewpipe/provider-lexicon
  .run(input);

result.overallSentiment; // { score, positive, neutral, negative, reviewCount }
result.trend; // average sentiment per month
result.flagged; // reviews needing attention (rating/text mismatch, strongly negative)
result.errors; // non-fatal per-review failures — one bad row never crashes the batch

What's here

• Pipeline — a builder wiring source → use(transform)* → provider → themes → export → run.
• Interfaces — SourceAdapter, Transformer, AnalysisProvider (with a capabilities flag), ThemeExtractor, and Exporter. Everything else in the ecosystem implements one of these.
• Types — NormalizedReview, AnalysisResult, SentimentResult, Theme, TrendPoint, FlaggedReview, PipelineError, and SCHEMA_VERSION.

Provider failures are isolated per review; adapter/transformer failures are fatal (nothing left to analyze). See the documentation for the full guide.

License

Apache-2.0