Skip to content

sentiment-basic

Is this sentence positive, negative, or neutral? Split it into words, look each one up in a small polarity lexicon, sum the hits, and divide by the word count — a fast, transparent sentiment score with no model, no network, and no nondeterminism.

sentiment-basic is a pure-Rust, zero-dependency crate that does exactly that: a compact built-in word list of positive and negative terms, lowercase tokenization with basic stemming, and a single analyze call returning a polarity score, a label, and the counts behind both. It is the tone checker behind the BeLikeNative AI writing assistant, packaged as a small library for editors, CLIs, and feedback widgets.

How scoring works

  1. Tokenize — split on non-letter characters, lowercase each token.
  2. Score each word — +1 if in the positive lexicon, -1 if in the negative lexicon, 0 otherwise.
  3. Sum — total polarity across all tokens.
  4. Normalize — score = polarity / word_count, clamped to -1.0..=1.0.
  5. Label — score > 0 → Positive, < 0 → Negative, == 0 → Neutral.
field range meaning
positive usize number of positive words matched
negative usize number of negative words matched
polarity i64 positive - negative
score -1.0 .. 1.0 normalized polarity (0.0 if no words)
label enum Positive / Negative / Neutral

Install

[dependencies]
sentiment-basic = "0.1"

Quick start

use sentiment_basic::analyze;

let pos = analyze("Great job, this is wonderful and clean!");
let neg = analyze("This is terrible, awful, and broken.");

println!("pos score: {:.2} ({:?})", pos.score, pos.label); // Positive
println!("neg score: {:.2} ({:?})", neg.score, neg.label); // Negative

assert!(pos.score > 0.0);
assert!(neg.score < 0.0);

See it live

The crate gives you the score. For a live in-editor tone readout that flags harsh or lukewarm phrasing as you type, use the BeLikeNative writing assistant.

Features

  • Pure Rust, no deps — a single file, no features, no network.
  • Built-in lexicon — small curated positive/negative word lists; no data files to ship.
  • Deterministic — same input always yields identical output.
  • Normalized score — -1.0..=1.0 so results are comparable across inputs.
  • Three-way label — Sentiment enum for quick branching.
  • O(1) per word — hash-set lookups; linear in token count.
  • MIT licensed — embed anywhere.

When to use it (and when not)

sentiment-basic is a lexicon analyzer, not a machine-learning model. It shines for short, clearly opinionated text — reviews, status updates, headlines, and writing-tone feedback. It will not catch sarcasm, negation scope ("not bad"), or domain-specific idiom. For those you need a model; this crate is the fast, transparent baseline.