Leveraging lexical resources as external knowledge for entity reasoning using deep learning frameworks
Bibliographic record
Abstract
Entity reasoning is an important aspect of natural language understanding.Most existing approaches towards entity reasoning tend to rely on the information extracted from training data.Since humans interpret texts with respect to some prior knowledge, we hypothesize that leveraging external knowledge, specifically in the form of lexical resources, could also be beneficial to entity reasoning.Lexical resources have been studied extensively in natural language processing, with a wide range of applications from word sense disambiguation to measuring semantic relatedness.However, their effects on entity reasoning have rarely been examined.In this thesis, we develop novel methods and model architectures that are able to leverage lexical resources in two different problem settings: knowledge base completion and rare entity prediction.Empirical results show that leveraging lexical resources as external knowledge in deep learning-based models can improve the performances of entity reasoning in both problem settings, indicating the potential of lexical resources in other similar applications, such as common sense reasoning and dialogue agents.iii Rsum Le raisonnement sur les entits est un aspect important de la comprhension du langage naturel.La plupart des approches utilises dans le raisonnement sur les entits se concentrent sur l'information extraite de l'ensemble de donnes d'entrainement.Puisque les gens interprtent les textes en fonction de leurs connaissances, nous mettons l'hypothse que le recours aux connaissances externes, en particulier, sous forme de ressources lexicales, pourrait galement tre bnfique pour le raisonnement sur les entits.Les ressources lexicales ont t tudies de manire approfondie dans le traitement du langage naturel, et jouissent d'une large gamme d'applications telles que la dsambigusation lexicale et la mesure de la simialrit smantique.Cependant, leurs effets sur le raisonnement sur les entits sont rarement examins.Dans cette thse, nous dveloppons de nouvelles mthodes et architectures de modles capables d'utiliser les ressources lexicales dans le cadre de deux problmes diffrents: l'achvement de base de connaissances et la prdiction d'entits rares.Nos rsultats dmontrent que l'utilisation des ressources lexicales comme connaissance externe aux rseaux neuronaux profonds peut amliorer considrablement la performance d'apprentissage dans les deux problmes considrs.Cela indique le potentiel de l'utilisation des ressources lexicales dans d'autres applications similaires, telles que le raisonnement de sens commun et les agents conversationnels.
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 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 | 0.002 | 0.003 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.001 |
| Bibliometrics | 0.001 | 0.000 |
| Science and technology studies | 0.005 | 0.000 |
| Scholarly communication | 0.001 | 0.002 |
| Open science | 0.003 | 0.001 |
| Research integrity | 0.002 | 0.005 |
| 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 itClassification
machine, unvalidatedMachine predicted; both teacher heads agree on what is shown here.
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".