The transformer paper-recommendation advantage is real at the head of the impact distribution and decays to null across the broad literature: a 4-checkpoint, all-26-field convergence study of SPECTER vs. TF-IDF
Bucket Foundation research-atlas working group
Does a transformer-based recommender (SPECTER) beat a classical TF-IDF baseline at recommending related papers, and does that advantage hold across the whole literature or only at the top?
Directed the study as part of the research-atlas working group; did not take sole authorship credit for this paper.
Compares SPECTER and TF-IDF paper recommendation across all 26 OpenAlex fields at four training checkpoints.
Research direction and review within the working group.
AI-assisted implementation and analysis under working-group direction.
Preprint. Peer review has not yet occurred.