Text Summarizer

How does it decide which sentences to keep?

It counts how often each content word appears in the whole text, then gives each sentence the average of those counts for the words it contains, and multiplies the first sentence by 2, the second by 1.5 and the last by 1.3. The highest-scoring sentences are kept and shown in their original order. Content words are words of four or more characters that are not on a built-in English stop-word list. There is no model of meaning, topic or importance beyond repeated words and position.

Is this AI summarisation?

No. It is a short script that scores and selects sentences. Every sentence in the summary is copied word for word from your text, nothing is paraphrased, shortened or reworded, and nothing is added. That means it cannot invent facts, but it also cannot merge two ideas into one sentence, resolve a "this" or "they" that refers to an earlier sentence, or notice that the point of the article is never stated in a single sentence.

Why does the first sentence nearly always appear?

Because of the position bonus: its score is doubled before it competes with the others. In a ten-sentence test about solar panels, the opening sentence scored 5.00 against 2.31 for the next best. For articles that open with a hook, an anecdote or a greeting, that boost can put a weak sentence into the summary; for news-style writing with a strong lead it helps.

Why are some sentences cut off or split strangely?

Sentences are found by looking for full stops, exclamation marks and question marks, with no knowledge of abbreviations or decimals. "Dr. Lee measured 3.5 percent growth" is cut at "Dr." and again at "3.", and the summary can then contain the fragment "Lee measured 3." Text after the last punctuation mark is ignored when choosing sentences, and a text with no punctuation at all is treated as one long sentence. Tidy abbreviations and decimals, and end every line with punctuation, before summarising.

How many sentences will I get?

It depends on the format and length setting. Bullet points and executive summaries use 3, 5 or 7 sentences for brief, medium and detailed. The paragraph format uses 2, 3 or 4, and the one-sentence TL;DR always uses the single best-scoring sentence. The tool never returns more sentences than the text contains, so a four-sentence text asked for five points returns all four with 0% compression.

What do the compression percentage and the keywords mean?

Compression is one minus the number of selected sentences divided by the total number of sentences, so it compares sentence counts, not words. A 10-sentence text summarised to 3 bullets shows 70%. The keyword line lists the ten most frequent content words, including words that appear only once when the text is short, so use it as a rough topic hint. There is no word stemming, which means "panel" and "panels" are counted separately.

What is the minimum length, and is there a maximum?

The text must contain at least 30 words (separated by whitespace); a shorter paste returns a message asking for more. The code does not enforce an upper limit, although the placeholder suggests up to 5,000 words. Very long texts simply give more sentences competing for the same few slots, so expect the result to favour repeated, generic words.

Does it work for languages other than English?

It will run, but the stop-word list is English only. In another language, articles, prepositions and auxiliary verbs are not filtered, so they inflate sentence scores and the keyword list. Accented letters are handled as part of words, but nothing is stemmed or normalised, and sentence splitting still depends on . ! and ? alone. It is designed for ordinary English prose.