Adaptive Loss-based Curricula for Neural Team Recommendation
Bibliographic record
Abstract
Neural team recommendation models have brought state-of-the-art efficacy while enhancing efficiency at recommending collaborative teams of experts who, more likely than not, can solve complex tasks. Yet, they suffer from popularity bias and overfit to a few dominant popular experts and, hence, result in discrimination and reduced visibility for already disadvantaged non-popular experts. Such models are trained on randomly shuffled datasets with the disproportionate distribution of a few popular experts over many teams and a sparse long-tailed distribution of non-popular ones, overlooking the difficulty of recommending hard non-popular vs. easy popular experts. To bridge the gap, we propose three curriculum-based learning strategies to empower neural team recommenders sifting through easy popular and hard non-popular experts and to mitigate popularity bias and improve upon the existing neural models. We propose (1) a parametric curriculum that assigns a learnable parameter to each expert enabling the model to learn an expert's levels of difficulty (or conversely, levels of popularity) during training, (2) a parameter-free (non-parametric) curriculum that presumes the worst-case difficulty for each expert based on the model's loss, and (3) a static curriculum to provide a minimum base for comparison amongst curriculum-based learning strategies and lack thereof. Our experiments on two benchmark datasets with distinct distributions of teams over skills showed that our parameter-free curriculum improved the performance of non-variational models across different domains, outperforming its parametric counterpart, and the static curriculum was the poorest. Moreover, among neural models, variational models obtain little to no gain from our proposed curricula, urging further research on more effective curricula for them. The code to reproduce our experiments is publically available at https://github.com/fani-lab/OpeNTF/tree/cl-wsdm25
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.002 | 0.010 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.001 |
| Science and technology studies | 0.001 | 0.001 |
| Scholarly communication | 0.001 | 0.003 |
| Open science | 0.003 | 0.002 |
| Research integrity | 0.002 | 0.003 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".