Benchmarks for Open Relation Extraction
Bibliographic record
Abstract
* Introduction* Benchmarks for Open Relation Extraction was developed by the University of Alberta and contains annotations for approximately 14,000 sentences from The New York Times Annotated Corpus (LDC2008T19) and Treebank-3 (LDC99T42). This corpus was designed to contain benchmarks for the task of open relation extraction (ORE), along with sample extractions from ORE methods and evaluation scripts for computing a method's precision and recall. ORE attempts to extract as many relations as described in a corpus without relying on relation-specific training data. The traditional approach to relation extraction requires substantial training effort for each relation of interest. That can be unpractical for massive collections such as found on the web. Open relation extraction offers an alternative by extracting unseen relations as they come. It does not require training data for any particular relation, making it suitable for applications that require a large (or even unknown) number of relations. Results published in ORE literature are often not comparable due to the lack of reusable annotations and differences in evaluation methodology. The goal of this benchmark data set is to provide annotations that are flexible and can be used to evaluate a wide range of methods. *Data* Binary and n-ary relations were extracted from the text sources. Sentences were annotated for binary relations manually and automatically. In the manual sentence annotation, two entities and a trigger (a single token indicating a relation) were identified for the relation between them, if one existed. A window of tokens allowed to be in a relation was specified; that included modifiers of the trigger and prepositions connecting triggers to their arguments. For each sentence annotated with two entities, a system must extract a string representing the relation between them. The evaluation method deemed an extraction as correct if it contained the trigger and allowed tokens only. The automatic annotator identified pairs of entities and a trigger of the relation between them; the evaluation script for that experiment deemed an extraction correct if it contained the annotated trigger. For n-ary relations, sentences were annotated with one relation trigger and all of its arguments. An extracted argument was deemed correct if it was annotated in the sentence. This release also includes extractions from the following ORE methods: ReVerb, SONEX, OLLIE, PATTY, TreeKernel, SwiRL, Lund and EXEMPLAR. Evaluation scripts are also provided for computing a method's precision and recall. *Samples* Please view this sample. *Updates* None at this time.
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 | 0.012 | 0.049 |
| Meta-epidemiology (narrow) | 0.003 | 0.001 |
| Meta-epidemiology (broad) | 0.002 | 0.002 |
| Bibliometrics | 0.011 | 0.012 |
| Science and technology studies | 0.004 | 0.002 |
| Scholarly communication | 0.006 | 0.007 |
| Open science | 0.005 | 0.006 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.023 | 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".