TextRank Summarization

TextRank is an unsupervised graph-based ranking algorithm that links sentences or words by similarity or co-occurrence and applies PageRank to select the most central sentences as a summary or the top keywords.

텍스트랭크는 문장이나 단어를 유사도나 동시출현으로 연결한 그래프에 페이지랭크를 적용하여 가장 중심적인 문장을 요약문으로, 또는 상위 키워드를 추출하는 비지도 그래프 기반 순위 알고리즘입니다.

How do I automatically summarize documents or extract their key sentences and keywords?

문서를 자동으로 요약하거나 핵심 문장과 키워드를 추출하려면 어떻게 하나요?

TextRank builds a graph in which sentences or words are nodes connected by similarity or co-occurrence, ranks them with the PageRank algorithm, and returns the top-ranked items as an extractive summary or keyword list. It fits when no labelled training data are available and the summary can consist of sentences taken directly from the original text.

When to use TextRank Summarization

• Long reports, articles, or transcripts that need short extractive summaries
• Extracting keywords from documents without training data
• Quick summaries of many documents in the same language

When to use something else

• A rewritten, fluent summary is required: use an abstractive transformer summarizer or a large language model
• Finding terms that distinguish documents across a corpus: use TF-IDF analysis

Assumptions and requirements

• Reasonably long, well-formed text
• Accurate sentence segmentation and tokenization for the language
• Preprocessing such as stopword removal for keyword extraction
• Summaries are extractive only

What the analysis reports

• Ranked sentences with scores
• Extractive summary of the top sentences
• Ranked keyword list

How to report it

TextRank was used to extract the top [k] sentences from each [document type] as an extractive summary.

Related methods

• TF-IDF Analysis
• LexRank
• Keyword Extraction (RAKE)

Run TextRank Summarization on Quorant