Personalized document clustering with dual supervision
Why this work is in the frame
A frame that forgets how it found something cannot be audited. These are the routes that admitted this work.
Bibliographic record
Abstract
The potential for semi-supervised techniques to produce personalized clusters has not been explored. This is due to the fact that semi-supervised clustering algorithms used to be evaluated using oracles based on underlying class labels. Although using oracles allows clustering algorithms to be evaluated quickly and without labor intensive labeling, it has the key disadvantage that oracles always give the same answer for an assignment of a document or a feature. However, different human users might give different assignments of the same document and/or feature because of different but equally valid points of view. In this paper, we conduct a user study in which we ask participants (users) to group the same document collection into clusters according to their own understanding, which are then used to evaluate semi-supervised clustering algorithms for user personalization. Through our user study, we observe that different users have their own personalized organizations of the same collection and a user's organization changes over time. Therefore, we propose that document clustering algorithms should be able to incorporate user input and produce personalized clusters based on the user input. We also confirm that semi-supervised algorithms with noisy user input can still produce better organizations matching user's expectation (personalization) than traditional unsupervised ones. Finally, we demonstrate that labeling keywords for clusters at the same time as labeling documents can improve clustering performance further compared to labeling only documents with respect to user personalization.
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.
Full frame distilled prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. Learned from the 10,348 direct Codex labels and 10,348 direct Gemma labels. Candidate is the union of thresholded teacher heads; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels or direct frontier model labels.
Codex and Gemma teacher scores by category
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.000 | 0.000 |
| Meta-epidemiology (narrow) | 0.000 | 0.000 |
| Meta-epidemiology (broad) | 0.000 | 0.000 |
| Bibliometrics | 0.000 | 0.000 |
| Science and technology studies | 0.000 | 0.000 |
| Scholarly communication | 0.000 | 0.001 |
| Open science | 0.000 | 0.000 |
| Research integrity | 0.000 | 0.000 |
| Insufficient payload (model declined to judge) | 0.000 | 0.000 |
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 it