An official website of the United States government
Neural entity linking models are very powerful, but run the risk of overfitting to the domain they are trained in. For this problem, a "domain" is characterized not just by genre of text but even by factors as specific as the particular distribution of entities, as neural models tend to overfit by memorizing properties of frequent entities in a dataset. We tackle the problem of building robust entity linking models that generalize effectively and do not rely on labeled entity linking data with a specific entity distribution. Rather than predicting entities directly, our approach models fine-grained entity properties, which can help disambiguate between even closely related entities. We derive a large inventory of types (tens of thousands) from Wikipedia categories, and use hyperlinked mentions in Wikipedia to distantly label data and train an entity typing model. At test time, we classify a mention with this typing model and use soft type predictions to link the mention to the most similar candidate entity. We evaluate our entity linking system on the CoNLL-YAGO dataset and show that our approach outperforms prior domain-independent entity linking systems. We also test our approach in a harder setting derived from the WikilinksNED dataset where all the mention-entity pairs are unseen during test time. Results indicate that our approach generalizes better than a state-of-the-art neural model on the dataset.
more »
« less
Knowledge-Driven Drug-Use NamedEntity Recognition with Distant Supervision
d public health. For such high-impact areas, accurately capturing relevant entities at a more granular level is critical, as this information influences real-world processes. On the other hand, training NER models for a specific domain without handcrafted features requires an extensive amount of labeled data, which is expensive in human effort and time. In this study, we employ distant supervision utilizing a domain-specific ontology to reduce the need for human labor and train models incorporating domain-specific (e.g., drug use) external knowledge to recognize domain specific entities. We capture entities related the drug use and their trends in government epidemiology reports, with an improvement of 8% in F1-score.
more »
« less
- PAR ID:
- 10403183
- Date Published:
- Journal Name:
- Studies in health technology and informatics
- ISSN:
- 1879-8365
- Page Range / eLocation ID:
- 140-144
- Format(s):
- Medium: X
- Sponsoring Org:
- National Science Foundation
More Like this
-
-
Neural entity linking models are very powerful, but run the risk of overfitting to the domain they are trained in. For this problem, a “domain” is characterized not just by genre of text but even by factors as specific as the particular distribution of entities, as neural models tend to overfit by memorizing properties of frequent entities in a dataset. We tackle the problem of building robust entity linking models that generalize effectively and do not rely on labeled entity linking data with a specific entity distribution. Rather than predicting entities directly, our approach models fine-grained entity properties, which can help disambiguate between even closely related entities. We derive a large inventory of types (tens of thousands) from Wikipedia categories, and use hyperlinked mentions in Wikipedia to distantly label data and train an entity typing model. At test time, we classify a mention with this typing model and use soft type predictions to link the mention to the most similar candidate entity. We evaluate our entity linking system on the CoNLL-YAGO dataset (Hoffart et al. 2011) and show that our approach outperforms prior domain-independent entity linking systems. We also test our approach in a harder setting derived from the WikilinksNED dataset (Eshel et al. 2017) where all the mention-entity pairs are unseen during test time. Results indicate that our approach generalizes better than a state-of-the-art neural model on the dataset.more » « less
-
We present CyBERT, a domain-specific Bidirectional Encoder Representations from Transformers (BERT) model, fine-tuned with a large corpus of textual cybersecurity data. State-of-the-art natural language models that can process dense, fine-grained textual threat, attack, and vulnerability information can provide numerous benefits to the cybersecurity community. The primary contribution of this paper is providing the security community with an initial fine-tuned BERT model that can perform a variety of cybersecurity-specific downstream tasks with high accuracy and efficient use of resources. We create a cybersecurity corpus from open-source unstructured and semi-unstructured Cyber Threat Intelligence (CTI) data and use it to fine-tune a base BERT model with Masked Language Modeling (MLM) to recognize specialized cybersecurity entities. We evaluate the model using various downstream tasks that can benefit modern Security Operations Centers (SOCs). The finetuned CyBERT model outperforms the base BERT model in the domain-specific MLM evaluation. We also provide use-cases of CyBERT application in cybersecurity based downstream tasks.more » « less
-
An ontology is a structured framework that categorizes entities, concepts, and relationships within a domain to facilitate shared understanding, and it is important in computational linguistics and knowledge representation. In this paper, we propose a novel framework to automatically extend an existing ontology from streaming data in a zero-shot manner. Specifically, the zero-shot ontology extension framework uses online and hierarchical clustering to integrate new knowledge into existing ontologies without substantial annotated data or domain-specific expertise. Focusing on the medical field, this approach leverages Large Language Models (LLMs) for two key tasks: Symptom Typing and Symptom Taxonomy among breast and bladder cancer survivors. Symptom Typing involves identifying and classifying medical symptoms from unstructured online patient forum data, while Symptom Taxonomy organizes and integrates these symptoms into an existing ontology. The combined use of online and hierarchical clustering enables real-time and structured categorization and integration of symptoms. The dual-phase model employs multiple LLMs to ensure accurate classification and seamless integration of new symptoms with minimal human oversight. The paper details the framework's development, experiments, quantitative analyses, and data visualizations, demonstrating its effectiveness in enhancing medical ontologies and advancing knowledge-based systems in healthcare.more » « less
-
This study builds a coronavirus knowledge graph (KG) by merging two information sources. The first source is Analytical Graph (AG), which integrates more than 20 different public datasets related to drug discovery. The second source is CORD-19, a collection of published scientific articles related to COVID-19. We combined both chemo genomic entities in AG with entities extracted from CORD-19 to expand knowledge in the COVID-19 domain. Before populating KG with those entities, we perform entity disambiguation on CORD-19 collections using Wikidata. Our newly built KG contains at least 21,700 genes, 2500 diseases, 94,000 phenotypes, and other biological entities (e.g., compound, species, and cell lines). We define 27 relationship types and use them to label each edge in our KG. This research presents two cases to evaluate the KG’s usability: analyzing a subgraph (ego-centered network) from the angiotensin-converting enzyme (ACE) and revealing paths between biological entities (hydroxychloroquine and IL-6 receptor; chloroquine and STAT1). The ego-centered network captured information related to COVID-19. We also found significant COVID-19-related information in top-ranked paths with a depth of three based on our path evaluation.more » « less
- Free Publicly Accessible Full Text
-
Accepted Manuscript1.0
- Conference Paper:
- The DOI is not currently available.
