Find ML papers worth your time. Five algorithms, side by side.

Pick any arXiv paper from the ML / AI / NLP / CV corpus on OpenAlex. Four algorithms each return their own top-10 most-similar papers: a hybrid blend, a sentence-transformer content tower, classic TF-IDF, and an ALS model fit on the citation graph. Latency badges are live calls against the FastAPI service. The leaderboard at the bottom is the offline evaluation with bootstrap 95% confidence intervals on every metric.

Best MAP@10 (hybrid)

0.167

16x the popularity baseline

Citation graph

46.4kedges

in-subset, citing to cited

Best p95 latency

75.4ms

top-10 from a 28,000-paper catalogue

Catalogue

28,424

CS arXiv papers since 2019 via OpenAlex

Seed selection

Type any title or author. The FastAPI service does the lookup.

top-k

Seed paper

Recommendations, side by side

Pick a seed paper above

Hybrid

Neural + ALS + TF-IDF + popularity blend, cold-paper aware

Neural

MiniLM sentence-transformer over title + abstract, cosine

Content (TF-IDF)

Title + abstract + authors + topic, sparse cosine

Citation ALS

Implicit ALS over citing -> cited edges, 96 factors

Leaderboard

Held-out evaluation on 2,000 citing-paper seeds, top-10, bootstrap 95% confidence intervals

Algorithm	MAP@10	NDCG@10	Recall@10	Precision@10	Coverage	p95 lat (ms)
Popularity	0.0101	0.0155	0.0320	0.0033	0.000	0.1
TF-IDF	0.1546	0.1925	0.3110	0.0317	0.387	66.4
Neural	0.1279	0.1623	0.2709	0.0277	0.366	3.7
Citation ALS	0.0392	0.0588	0.1225	0.0127	0.184	0.7
Hybridbest	0.1665	0.2101	0.3465	0.0353	0.333	75.4

MAP@10

NDCG@10

Recall@10