University of Amsterdam at the TREC 2019 Complex Answer Retrieval Track.
Bibliographic record
Abstract
This paper documents the University of Amsterdam’s participation in the TREC 2019 Complex Answer Retrieval Track. This is the first year we actively participate in TREC CAR, attracted by the introduction to the limited “budget” of 20 passages per heading in the outline. We conducted initial exploratory experiments on making each heading contain a unique set of passages within the outline, and even do this hierarchical for each subtree and main title/article level, hence remove any redundancy between passages for different “queries” within the same title. We experimented with top-down and bottom-up filtering approaches. At the time of writing we are still in the process of analyzing the results. Some initial observations are the following. First, the restriction makes the task very challenging, as assigning any passage to the right heading in the outline is highly non-trivial. Qualitative analysis shows that our simple heuristics often make a different decision than the editorial judges on the heading under which a passage relevant to the title’s topic is assigned. Second, the fraction of judged and relevant passages per individual query or leave node is very small, making it hard to draw any definite conclusions on our experiments, and also resulting in a too small recall base to evaluate our non-pooled runs in a meaningful way. Third, when aggregating all qrels and runs to the title level, there is reasonable effectiveness of the underlying BM25 rankings, showing that the underlying passage ranking is not unreasonable, and that the hard and interesting problem is in the exact assignment of passages to the “right” headings.
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 | 0.007 | 0.013 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.001 |
| Bibliometrics | 0.004 | 0.004 |
| Science and technology studies | 0.002 | 0.001 |
| Scholarly communication | 0.009 | 0.006 |
| Open science | 0.002 | 0.003 |
| Research integrity | 0.003 | 0.002 |
| Insufficient payload (model declined to judge) | 0.235 | 0.146 |
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".