// the find
liuhuanyong/TextGrapher
Text Content Grapher based on keyinfo extraction by NLP method。输入一篇文档，将文档进行关键信息提取，进行结构化，并最终组织成图谱组织形式，形成对文章语义信息的图谱化展示。
TextGrapher takes a Chinese-language document and extracts key entities, keywords, and subject-verb-object triples, then renders them as an interactive graph in an HTML file using vis.js. It's aimed at Chinese NLP practitioners who want a quick visualization of semantic structure in news or event-driven text. The examples are all crime/controversy news stories from Chinese media circa 2017-2021.
The pipeline is simple and self-contained — one call to `handler.main(content)` produces a browsable HTML graph with no server required. It combines three complementary extraction methods (TextRank keywords, NER, SVO triples) rather than relying on just one. The vis.js output is genuinely interactive, which makes it useful for demos or quick exploration. The code is short enough that you can actually read and modify it in an afternoon.
Hard-wired for Chinese text only — the NLP pipeline depends on Chinese-specific tokenization and models (likely jieba/HanLP underneath), so it won't work on English without a complete rewrite of the extraction layer. Last commit was October 2021 and the README links to a personal blog for support — this is effectively abandoned. No requirements.txt or setup.py is visible in the tree, which means you're on your own figuring out which dependency versions actually work together. The SVO extraction quality is acknowledged as weak even by the author, so the graphs for anything beyond simple news sentences will be noisy and require manual cleanup.