Bibliographic record
Abstract
Recommender systems help users find their items of interest from large data collections with little effort. Collaborative filtering (CF) is one of the most popular approaches for making recommendations. While significant work has been done on improving accuracy of CF methods, some of the most popular CF approaches are limited in terms of scalability and efficiency. The size of data in modern recommender systems is growing rapidly in terms of both new users and items and new ratings. Item-based recommendation is one of the CF approaches used widely in practice. It computes and uses an item-item similarity matrix in order to predict unknown ratings. Previous works on item-based CF method confirm its usefulness in providing high quality top-k results. In this paper, we design a scalable algorithm for top-k recommendations using this method. We achieve this by probabilistic modeling of the similarity matrix. A unique challenge here is that the ratings that are aggregated to produce the aggregate predicted score for a user should be obtained from different lists for different candidate items and the aggregate function is non-monotone. We propose a layered architecture for CF systems that facilitates computation of the most relevant items for a given user. We design efficient top-k algorithms and data structures in order to achieve high scalability. Our algorithm is based on abstracting the key computation of a CF algorithm in terms of two operations -- probe and explore. The algorithm uses a cost-based optimization whereby we express the overall cost as a function of a similarity threshold and determine its optimal value for minimizing the cost. We empirically evaluate our theoretical results on a large real world dataset. Our experiments show our exact top-k algorithm achieves better scalability compared to solid baseline algorithms.
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 | 0.008 |
| Meta-epidemiology (narrow) | 0.003 | 0.002 |
| Meta-epidemiology (broad) | 0.002 | 0.003 |
| Bibliometrics | 0.003 | 0.005 |
| Science and technology studies | 0.003 | 0.001 |
| Scholarly communication | 0.007 | 0.007 |
| Open science | 0.008 | 0.006 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.248 | 0.275 |
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".