daviidli / summary

Extractive text summarization using TextRank and RAKE

summary

Extractive text summarization using TextRank and RAKE.

Demo

API

Hosted for demo purposes only. Please host your own server if you require consistent performance.

Selection types:

enum TextRankSelections {
	Text = 'text',
	Sentences = 'sentences',
	Ranks = 'ranks',
}

enum RakeSelections {
	Text = 'text',
	Sentences = 'sentences',
	Ranks = 'ranks',
	Keywords = 'keywords',
}

Getting the summary for a URL:

Endpoint: https://summsumm.herokuapp.com/summary/url/ POST Request body:

interface UrlRequest {
	url: string;
	selections: {
		textRank?: TextRankSelections[];
		rake?: RakeSelections[];
	}
}

• API will return the requested selections only

Getting the summary of inputted text:

Endpoint: https://summsumm.herokuapp.com/summary/text/ POST Request body:

interface TextRequest {
	text: string;
	selections: {
		textRank?: TextRankSelections[];
		rake?: RakeSelections[];
	}
}

• API will return the requested selections only

Response

interface Response {
	textRank?: TextRankResponse;
	rake?: RakeResponse;
}

// response is based on selections made in request

interface TextRankResponse {
	text?: string;
	sentences?: string[];
	ranks?: Rank[];	// shown below
}

interface RakeResponse {
	text?: string;
	sentences?: string[];
	ranks?: Rank[]; // shown below
	keywords?: KeywordScore[]; // shown below
}

interface Rank {
	sentenceIndex: number;
	sentence: string;
	rank: number;
	keywords?: string[];
	score?: number;
}

interface KeywordScore {
	keyword: string;
	degree: number;
	frequency: number;
	degFreq: number;
}

Algorithms

TextRank

Uses Google's PageRank algorithm but instead of webpages, it ranks sentences. Cosine similarity is used to compute the similarity of sentences.

RAKE (Rapid Automatic Keyword Extraction)

Method for summarization that extracts keywords from text and then computes a score for each word. Each sentence has a score of the sum of its keyword's scores. The sentences are then ranked by the sentence score. Tends to give preference to longer keywords and sentences

Original v1 code