About Conftrace

Conftrace is an open browsing interface for 224,410 research papers from 36 top CS/AI venues: NeurIPS, ICML, ICLR, CVPR, ICCV, ECCV, ACL, EMNLP, NAACL, AAAI, IJCAI, MICCAI, Interspeech, and 23 more venues.

What you can do here

• Search 224k papers by title, narrowing by venue and year. Topic and keyword pages link into the same listing, so you can also arrive there from a topic or a keyword.
• Browse 36 venues with per-year paper counts, top authors, and DNA fingerprints.
• Explore a 752-topic taxonomy across the field, classified by an LLM.
• Walk through 215k keywords with co-occurrence graphs.
• Track emerging terms — year-over-year growth from paper titles.
• Visualise the graph — 8 interactive presets including Keyword Galaxy, Research Bridges, Time Machine, and Topic Landscape.

Who is behind this

Conftrace is built and maintained by Justyna Wojtczak (github.com/justi) as a personal research project. The data pipeline is public: github.com/justi/researchpooler holds the scrapers and the classification code. The Rails application itself is not published.

The data layer (scrapers, abstract extraction, taxonomy classification) is a continuation of Andrej Karpathy's 2011 researchpooler project (the scrapers here run from a fork). Karpathy outlined a "Stage 4 web interface" in the original README; Conftrace is that interface, finally built more than a decade later.

How is this different from Google Scholar / Semantic Scholar?

• Scoped, not universal. We only index 36 top CS/AI venues - no preprints, no random aggregators. The signal-to-noise is high.
• Taxonomy-first. Papers with an abstract are eligible for classification into a 3-level taxonomy of 752 topics by an LLM; the share done so far is below. You can browse the field by structure, not just by keyword.
• Co-occurrence graphs. 215k keywords linked through shared papers, queryable as a graph. Find research bridges between disjoint areas.
• Free to use. No paywall, no login, no ads. The data pipeline that builds the corpus is public on GitHub; the web application is not.

Limitations & caveats

• The sample is top-tier CS/AI venues, not all of research. Trends and achievements describe "among 36 venues", not "in the world".
• 12% of papers have no abstract (mostly pre-2015 publications without HTML pages). They are hidden from the default listing and from the search index, but one click away in /papers and reachable by direct URL.
• 96% of papers with an abstract have been classified into the taxonomy so far. Classification accuracy hasn't been rigorously validated against a human-labeled benchmark - treat topic/keyword assignments as LLM-generated best guesses, not verified ground truth (see Methodology).
• Each venue is indexed over a scrape window, not its full history - the year range on every card in /conferences is the real coverage. Some venues start decades after their first edition.
• Authors are identified by normalised name, so researchers sharing a name are merged into one profile. Paper counts, rankings and achievements for common names are inflated as a result (see Methodology).
• A paper presented at a co-located event (ACL-IJCNLP, AACL-IJCNLP, NAACL/SemEval and others) is published in one volume but appears in two venues' proceedings. Conftrace counts it once, under the venue the ACL Anthology filed it in, so the totals here are documents rather than listings - about 6,700 fewer than a naive count of both.
• Data refresh cadence: new proceedings are ingested in batches when I run the importer, not automatically on announcement, so a venue shows up here weeks to a couple of months after publishing.

See also: Methodology for technical details on data collection and classification.