import { KeyphraseEntry, SEEKTOPICOptions, SEEKTOPICResult } from './types';
/**
 * ### SEEKTOPIC \u2014 Keyphrase & Sentence Extraction
 *
 * Pulls the most important phrases and sentences out of any document.
 * Given raw text, it returns a ranked list of key concepts (e.g. "neural
 * network", "climate change") and, optionally, the sentences that best
 * summarise the document around those concepts.
 *
 * <img src="https://i.imgur.com/gZ4kI1V.png" width="360px" />
 *
 * **How it works \u2014 8-step pipeline:**
 *
 * 1. **Clean** \u2014 strip HTML tags and entities.
 * 2. **Split** \u2014 break text into sentences, respecting abbreviations and URLs.
 * 3. **Topic Extraction** *(LLM Path)* \u2014 sends the first 5000 words to an LLM
 *    to extract the most descriptive key topic phrases labeling the document.
 * 4. **Vector Search** *(LLM Path)* \u2014 calculates cosine similarity embeddings
 *    for all sentences vs topic phrases, finding the top relevant sentences per topic.
 * 5. **Tokenise & Extract** *(Fallback)* \u2014 if LLM fails, identify noun-anchored
 *    n-grams (1-N words).
 * 6. **Score & Fold** *(Fallback)* \u2014 weight phrases and merge subsumed phrases.
 * 7. **Re-weight** *(Fallback)* \u2014 boost Wikipedia entities and rare domain terms.
 * 8. **TextRank** *(Fallback)* \u2014 build sentence similarity graph and run random walk.
 *
 * <video src="https://github.com/user-attachments/assets/73348d63-7671-4e20-8df9-29a13d5b0768"
 *  width="550px" controls />
 *
 * @param docText - Plain text or HTML document to analyse.
 * @param options - Optional tuning parameters (word limits, thresholds, query bias).
 * @returns
 *   - `optionSkipRanking: true` (default) \u2192 `KeyphraseEntry[]` sorted by weight.
 *   - `optionSkipRanking: false` \u2192 `SEEKTOPICResult` with `topSentences`,
 *     `keyphrases`, and the full `sentences` array.
 *
 * @example
 * // Fast: just extract keyphrases
 * const keyphrases = extractSEEKTOPIC(articleText, { phrasesModel });
 * // \u2192 [{ keyphrase: "machine learning", weight: 84 }, ...]
 *
 * @example
 * // Full: keyphrases + summary sentences, biased toward a search query
 * const { topSentences, keyphrases } = extractSEEKTOPIC(articleText, {
 *   phrasesModel,
 *   optionSkipRanking: false,
 *   heavyWeightQuery: "transformer attention",
 *   limitTopSentences: 5,
 * }) as SEEKTOPICResult;
 *
 * @author [ai-research-agent (2024)](https://airesearch.js.org)
 * @category Topics
 */
export declare function extractSEEKTOPIC(docText: string, options?: SEEKTOPICOptions): Promise<KeyphraseEntry[] | SEEKTOPICResult>;