marcpinet / citracer

What is citracer?

Citracer is a Python CLI tool that traces citation chains across research papers for any keyword or concept you specify, starting from a PDF, arXiv ID, DOI, or URL. Feed it a paper and term like "self-attention," and it recursively downloads cited papers mentioning the concept nearby, rendering an interactive graph in your browser with nodes colored by status (analyzed, unavailable) and edges for keyword-associated citations. Supports semantic matching for synonyms, reverse traces to see who cites your paper on the concept, and exports to JSON/GraphML.

Why is it gaining traction?

Unlike basic citation scrapers, citracer delivers polished interactive graphs with layout controls, centrality metrics (PageRank, betweenness), pivot detection, and timelines—turning raw lit review drudgery into explorable maps of idea spread. Semantic search catches "channel-independent" in "univariate processing" passages regex misses, while diff mode and `--since` highlight new papers for ongoing monitoring. Docker GROBID integration handles PDF parsing reliably, with caching for fast re-runs.

Who should use this?

ML researchers mapping concept evolution (e.g., "patching" across transformers), academics automating citation trace de pas or laisser une trace citation workflows, or anyone surveying chains across bioRxiv/arXiv/OpenReview papers. Ideal for spotting origin papers or tracking forward citations without manual Semantic Scholar dives.

Verdict

Worth a spin for lit-heavy workflows—solid docs, CLI, and viz punch above 15 stars, though 1.0% credibility signals early maturity (test it on your cache first). Pair with a Semantic Scholar key to avoid throttles.

(198 words)