TRIVIR: A Visualization System to Support Document Retrieval with High Recall
Bibliographic record
Abstract
\n A high recall problem in document retrieval is described by scenarios in which one wants to ensure that, given one (or multiple) query document(s), (nearly) all relevant related documents are retrieved, with minimum human effort. The problem may be expressed as a document similarity search: a user picks an example document (or multiple ones), and an automatic system recovers similar ones from a collection. This problem is often handled with a so-called Continuous Active Learning strategy: given the initial query, which is a document described by a set of relevant terms, a learning method returns the most-likely relevant documents (e.g., the most similar) to the reviewer in batches, the reviewer labels each document as relevant/not relevant and this information is fed back into the learning algorithm, which uses it to refine its predictions. This iterative process goes on until some quality condition is satisfied, which might demand high human effort, since documents are displayed as ranked lists and need to be labeled individually, and impact negatively the convergence of the learning algorithm. Besides, the vocabulary mismatch issue, i.e., when distinct terminologies are employed to describe semantically related or equivalent concepts, can impair recall capability. We propose TRIVIR, a novel interactive visualization tool powered by an information retrieval (IR) engine that implements an active learning protocol to support IR with high recall. The system integrates multiple graphical views in order to assist the user identifying the relevant documents in a collection. Given representative documents as queries, users can interact with the views to label documents as relevant/not relevant, and this information is used to train a machine learning (ML) algorithm which suggests other potentially relevant documents. TRIVIR offers two major advantages over existing visualization systems for IR. First, it merges the ML algorithm output into the visualization, while supporting several user interactions in order to enhance and speed up its convergence. Second, it tackles the vocabulary mismatch problem, by providing terms synonyms and a view that conveys how the terms are used within the collection. Besides, TRIVIR has been developed as a flexible front-end interface that can be associated with distinct text representations and multidimensional projection techniques. We describe two use cases conducted with collaborators who are potential users of TRIVIR. Results show that the system simplified the search for relevant documents in large collections, based on the context in which the terms occur.\n
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 | 0.003 | 0.009 |
| Meta-epidemiology (narrow) | 0.002 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.005 | 0.003 |
| Science and technology studies | 0.001 | 0.000 |
| Scholarly communication | 0.004 | 0.004 |
| Open science | 0.003 | 0.003 |
| Research integrity | 0.002 | 0.002 |
| Insufficient payload (model declined to judge) | 0.031 | 0.021 |
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".