Results for “Entity Extraction”

DOE OSTI2014

Automatic Labeling for Entity Extraction in Cyber Security

Timely analysis of cyber-security information necessitates automated information extraction from unstructured text. While state-of-the-art extraction methods produce extremely accurate results, they require ample training data, which is generally unavailable for specialized applications, such as detecting security related entities; moreover, manual annotation of corpora is very costly and often not a viable solution. In response, we develop a very precise method to automatically label text from several data sources by leveraging related, domain-specific, structured data and provide public access to a corpus annotated with cyber-security entities. Next, we implement a Maximum Entropy Model trained with the average perceptron on a portion of our corpus (~750,000 words) and achieve near perfect precision, recall, and accuracy, with training times under 17 seconds.

Automatic Labeling↗

DOE OSTI2020

Survey of Current State of the Art Entity-Relation Extraction Tools

In the area of information extraction from text data, there exists a number of tools with the capability of extracting entities, topics, and their relationships with one another from both structured and unstructured text sources. Such information has endless uses in a number of domains, however, the solutions to getting this information are still in early stages and has room for improvement. The topic has been explored from a research perspective by academic institutions, as well as formal tool creation from corporations but has not made much advancement since the early 2000's. Overall, entity extraction, and the related topic of entity linking, is common among these tools, though with varying degrees of accuracy, while relationship extraction is more difficult to find and seems limited to same sentence analysis. In this report, we take a look at the top state of the art tools currently available and identify their capabilities, strengths, and weaknesses. We explore the common algorithms in the successful approaches to entity extraction and their ability to efficiently handle both structured and unstructured text data. Finally, we highlight some of the common issues among these tools and summarize the current ability to extract relationship information.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2019

Survey of Current State of the Art Entity-Relation Extraction Tools

In the realm of information extraction from text data, there exists a number of tools with the capability of extracting entities and their relationships with one another. Such information has endless uses in a number of domains, however, the solutions to getting this information is still in early stages and has room for improvement. The topic has been explored from a research perspective by academic institutions, as well as formal tool creation from corporations. Overall, entity extraction is common among these tools, though with varying degrees of accuracy, while relationship extraction is more difficult to find. In this report, we take a look at the top state of the art tools currently available and identify their capabilities, strengths, and weaknesses. We explore the common algorithms in the successful approaches and their ability to efficiently handle both structured and unstructured text data. Finally, we highlight some of the common issues among these tools and summarize the current status in the area of entity-relationship extraction.

97 MATHEMATICS AND COMPUTING↗

DOE OSTI2025

A Data Processing Pipeline To Extract A Knowledge Graph From Heterogeneous Data For Socio-technical Analysis Of Critical Infrastructure Influence

The code is written in Python and consists of the following pipeline that is implemented in Apache Airflow. This pipeline intends to understand the companies that are directly or indirectly involved with a type of critical infrastructure system at some point in that system's lifecycle. The pipeline takes a configuration file that specifies a list of initial companies to consider, a geographic region of interest, and a set of SEC form types as well as other data sources (e.g. CrunchBase) from which to extract entities and relations. There are four main components to this pipeline as currently implemented: Entity Extraction, Network Construction, Analysis, and Visualization. First, Entity Extraction, is implemented as the `topear-extract_organizations` Apache Airflow workflow. Given an initial query that specifies a geographic region of interest and a time interval, the software will extract CI facilities of interest and organizations that have a direct influence relationship to those facilities (e.g. ownership). During the course of the LDRD, we focused on Electric Vehicle charging stations and this information is available via the Department of Energy (DOE) database on fueling stations maintained by NREL. Within the context of the DOE CESER project, we have focused on Battery Energy Storage Systems (BESS). Second, the Network Extraction component will iteratively construct a social network graph given the set of organizations and people extracted in the previous step. Organizations (and eventually People if desired) are then fed as a query to the `topgear-construct_social_network` Apache Airflow workflow which given a set of initial companies and data sets (e.g. SEC EDGAR form types, OpenCorporates, Crunchbase). This Airflow workflow will iteratively query such data sources to discover relationships with new organizations and people. For example, this module can iteratively query SEC EDGAR for metadata that documents the number of each type of form for the given set of companies and their location. This forms metadata represents a catalog of data sources from SEC EDGAR for the extracted social network knowledge graph. The pipeline then downloads these forms from the website and saves them in a build directory for further processing. These documents are then parsed for entities and relations. Again, we note that in additional to SEC data sources, this step can also pull in information on organizations via API services such as CrunchBase and OpenCorporates or bulk data sources. At the end of this step, the resultant social network, the Critical Infrastructure network, and the edges that encode relationships between organizations and CI facilities, form the Adversarial Socio-Technical Network (ASTN) that informs the analysis. Third, the Analysis component processes these generated ASTN. Previously, that has included the ability to compare prevalence of different vendors for a given infrastructure component type across different regions as well as identify common public and private investors across those vendors. This was demonstrated for EV Charging Stations across several different metropolitan areas within an IEEE PES GridEdge publication. More recently, we have looked at ways to identify infrastructure owners and operators of BESS with the most nameplate capacity across different states as well as other indictors of risk resulting from changes in ownership over time. Finally, the Visualization component consists of an HTML/CSS/JS framework by which users can interact geospatial, operational, and organizational relationships across a given portfolio of Critical Infrastructure facilities. The objective is to provide a library of UI/UX modules that can be repurposed for stakeholder-specific dashboards. All of the modules are related via a common event model that enables UI actions in one view to percolate across the other views.

Weaver, Gabriel [Idaho National Laboratory (INL), ↗

NASA NTRSDate not supplied

What Went Wrong: A Survey of Wildfire UAS Mishaps through Named Entity Recognition

Increasingly, unmanned aircraft systems (UAS) are being applied to wildfire incidents for tasks such as mapping, aerial ignition, and delivery. As a result, aviation incident reporting systems for wildfires are beginning to accumulate data related to UAS mishaps in wildfire response. In this research, we apply state-of-the-art natural language processing (NLP) techniques to develop a custom Named Entity Recognition (NER) model which extracts entities relevant to safety analysts. The custom NER model is built by fine-tuning an existing Bidirectional Encoder Representations from Transformers (BERT) model, resulting in a generalizable NER model that can extract engineering relevant entities including failure modes, causes, effects, control processes, and recommendations from failure-relevant text. This model performs passably, with a weighted average f1 score of 0.33 across entity types, indicating more labeled training data is needed. Extracted entities are used to form a Failure Modes and Effects Analysis (FMEA)-style survey of wildfire UAS mishaps reported using the SAFECOM system. Similar mishaps are manually clustered and reported as single rows within an FMEA. Foreach cluster, we compute frequency, severity, and overall riskin accordance with FAA standards. This methodology can beapplied as part of a broader safety management system totrack trends in mishaps (e.g., likelihood, severity) and discoverknowledge (e.g., causes, effects) that can be utilized to improvesafety outcomes and system performance.

Machine Learning↗

DOE OSTI2025

Leveraging Large Language Models for Real-World Data Evidence: A Framework for Automated Treatment Extraction and Data Harmonization

Background: The ability to comprehensively collect treatment information from cancer patient medical records would enable studies to evaluate real-world benefits and risks tied to specific treatments. Currently, it is difficult to system- atically collect high-quality treatment information because it is often stored in unstructured text. Manually extracting and standardizing drug and regimen data is time-intensive. Recent advances in large language models (LLMs) offer a potential solution for automated extraction of structured treatment information from clinical text. Objective: This study systematically evaluates the utility of four LLMs from the Llama family for automated extraction of oncology treatment information from clinical text. This information can guide researchers using cancer registry data to provide insights into cancer care and outcomes beyond clinical trials. Methods: Four instruction-tuned Llama models with varying parameter counts (1B, 3B, 8B, and 70B) were evaluated for their ability to extract treatment information from clinical documents. A unified oncology knowledge base integrating seven major public data sources was developed to standardize and normalize extracted entities—a critical step for harmonizing data from diverse sources. Extracted treatment data were compared against expert-annotated ground truth. Model performance was assessed using accuracy metrics (Precision, Recall, F1-Score) and opera- tional feasibility metrics, including processing speed and structural compliance of the output. Results: A strong positive correlation was observed between model size and extraction accuracy. F1-score improved from 0.609 for the 1B model to 0.710 (3B), 0.807 (8B), and 0.828 (70B). While larger models demonstrated superior accuracy and compliance, they incurred higher computational costs. The modest performance difference between 8B and 70B suggests diminishing returns with increasing model size. Conclusions: LLMs represent a viable technology for automating oncology treatment extraction. The 8B-parameter model emerged as a highly effective option, balancing high accuracy and computational efficiency. Selecting an appropriate LLM for deployment in cancer registries involves a trade-off between desired accuracy and available operational resources. Harmonizing extracted entities with the oncology knowledge base facilitates standardized integration into common data models, enhancing data quality for real-world evidence analyses.

artificial intelligence↗

DOE OSTI2024

A Data Processing Pipeline To Extract A Knowledge Graph From Sec Documents For Socio-technical Analysis Of Critical Infrastructure Influence

The code is written in Python and consists of the following pipeline that is implemented in Apache Airflow. This pipeline intends to understand the companies that are directly or indirectly involved with a type of critical infrastructure system at some point in that system's lifecycle. The pipeline takes a configuration file that specifies a list of initial companies to consider, a geographic region of interest (disk) expressed as a latitude/longitude point and distance, and a set of SEC form types from which to extract entities and relations. There are three main components to this pipeline as currently implemented: Social Network Extraction, Critical Infrastructure Network Extraction, and Inference and Fusion. First, Social Network Extraction, implemented as the `organizations_sec` component of the workflow graph queries the SEC EDGAR webservice using the list of initial companies from the configuration file. Given this, it extracts metadata that documents the number of each type of form for the given set of companies and their location. This forms metadata represents a catalog of data sources for the extracted social network knowledge graph. The pipeline then downloads these forms from the website and saves them in a build directory for further processing. These documents are then parsed for entities and relations. Second, the Critical Network Extraction component extracts entities and relations for a critical infrastructure sector. Currently, we focus on Electric Vehicle charging stations and this information is available via the Department of Energy (DOE) database on fueling stations maintained by NREL. Third, the Inference and Fusion component relates the social network graph to the critical infrastructure graph in order to understand the impact of a company within a geographic region. Relations include ownership of the EV Charging Station asset as well as maintenance/ownership of the EV payment networks. The fused network can be represented in many ways and currently we emit a knowledge graph.

Weaver, GabrielA.↗

NASA NTRS2022

Natural Language Understanding and Extraction of Flight Constraints Recorded in Letters of Agreement

This paper presents an automated information extraction and inference technique using natural language processing for extracting flight operational procedures and constraints embedded in heritage air traffic management documents. The extracted flight constraints can be digitized and fit into existing airspace information exchange models such as the Aeronautical Information Exchange Model (AIXM). This approach offers a digitized solution to disseminate airspace operating conditions to diverse air users and stakeholders in the National Airspace System (NAS). Furthermore, the digitized flight procedures can provide operational flexibility for emerging advanced air mobility providers and reduce traffic controller workload while maintaining current safety standards. To demonstrate this process, 1,972 Letters of Agreement (LOAs) have been selected for processing, named entity extraction, constraint identification and extraction. This dataset is derived from a subset of documents related to Air Route Traffic Control Centers (ARTCC) operations. We experimented with various traditional information extraction techniques, state-of-the-art machine learning and deep learning models to perform named entity recognition and pattern recognition on our dataset. We present the results from our experiments and demonstrate 99.0% F-1 score for named entity recognition, and a 96.6% accuracy for our entire workflow up to named entity recognition. We also discuss constraint definitions using generic patterned templates and extensions to this work in applying entity linking to digitally extracting relevant constraints.

Natural Language Processing↗

NASA NTRSDate not supplied

Natural Language Understanding and Extraction of Flight Constraints Recorded in Letters of Agreement

This paper presents an automated information extraction and inference technique using natural language processing for extracting flight operational procedures and constraints embedded in heritage air traffic management documents. The extracted flight constraints can be digitized and fit into existing airspace information exchange models such as the Aeronautical Information Exchange Model (AIXM). This approach offers a digitized solution to disseminate airspace operating conditions to diverse air users and stakeholders in the National Airspace System (NAS). Furthermore, the digitized flight procedures can provide operational flexibility for emerging advanced air mobility providers and reduce traffic controller workload while maintaining current safety standards. To demonstrate this process, 1,972 Letters of Agreement (LOAs) have been selected for processing, named entity extraction, constraint identification and extraction. This dataset is derived from a subset of documents related to Air Route Traffic Control Centers (ARTCC) operations. We experimented with various traditional information extraction techniques, state-of-the-art machine learning and deep learning models to perform named entity recognition and pattern recognition on our dataset. We present the results from our experiments and demonstrate 99.0% F-1 score for named entity recognition, and a 96.6% accuracy for our entire workflow up to named entity recognition. We also discuss constraint definitions using generic patterned templates and extensions to this work in applying entity linking to digitally extracting relevant constraints.

Natural Language Processing↗

DOE OSTI2018

Evaluation of a bis-oxo-dibenzothiacrown ether as an efficient Hg(II)-extracting macrocyclic platform

In this work, we discuss the evaluation of an oxygen-containing dibenzothiacrown ether as an effective macrocyclic scaffold for the efficient capture of Hg(II) and Pb(II) ions in acidic aqueous solutions. The affinity of the macrocyclic ether towards Hg(II) ions was measured in nitric acid solutions and found to be strong under short time scales, while no affinity towards Pb(II) was observed. Experiments demonstrated that the macrocycle decomposed over longer contact times with nitric acid. As this is a new macrocyclic platform, current studies are focusing in on its potential as an efficient Hg(II) extraction entity as well as the design of more acid-resistant versions of it.

37 INORGANIC, ORGANIC, PHYSICAL, AND ANALYTICAL CH↗

DOE OSTI2025

Text Mining for Process–Structure–Properties Relationships in Metals

With the advent of large language models (LLMs), the vast unstructured text within millions of academic papers is increasingly accessible for materials discovery—although significant challenges remain. While LLMs offer promising few- and zero-shot learning capabilities, particularly valuable in the materials domain where expert annotations are scarce, general-purpose LLMs often fail to address key materials-specific queries without further adaptation. To bridge this gap, fine-tuning LLMs on human-labeled data is essential for effective structured knowledge extraction (Liu in The Importance of Human-Labeled Data in the Era of LLMs, 2023). Here, in this study, we introduce a novel annotation schema designed to extract generic process–structure–properties relationships from scientific literature. We demonstrate the utility of this approach using a dataset of 128 abstracts, with annotations drawn from two distinct domains: high-temperature materials (Domain I) and uncertainty quantification in simulating materials microstructure (Domain II). Initially, we developed a conditional random field (CRF) model based on MatBERT—a domain-specific BERT variant—and evaluated its performance on Domain I. Subsequently, we compared this model with a fine-tuned LLM (GPT-4o from OpenAI) under identical conditions. Our results indicate that fine-tuning LLMs can significantly improve entity extraction performance over the BERT-CRF baseline on Domain I. However, when additional examples from Domain II were incorporated, the performance of the BERT-CRF model became comparable to that of the GPT-4o model. These findings underscore the potential of our schema for structured knowledge extraction and highlight the complementary strengths of both modeling approaches.

Materials science↗

NASA NTRS2018

Earth Science Deep Learning: Applications and Lessons Learned

Deep Learning: A subfield of machine learning; Algorithms inspired by function of the brain; Scales with amount of training data; Powerful tool without the need for feature engineering; Suitable for Earth Science applications. Deep Learning for Earth science at MSFC (Marshall Space Flight Center): Phenomena identification; Hurricane intensity (wind speed) estimation; Severe storm (hailstorm) detection; Transverse bands detection; Entity extraction for knowledge graph creation; Ephemeral water detection.

Labeled Data↗

NASA NTRSDate not supplied

BERT-E: An Earth Science Specific Language Model for Domain-Specific Downstream Tasks

Language models are fast approaching human-like understanding of natural language. They have been shown to perform equally, if not better than humans in a myriad of language tasks such as next sentence prediction, question answering, entity extraction etc. Part of the success of the models are owed to the fact that they have been trained on varied natural language text over the internet. By virtue of this, the models do not contain the semantic information present in Earth science literature. Hence, there is a lot of room for improvement when using these models for earth science specific tasks. In this work, we showcase our approach on developing Earth science specific language models. Furthermore, we justify the need for such a model by using the embeddings generated by the model to perform a domain specific downstream task that performs better than a generic model.

Prasanna Koirala↗

NASA NTRSDate not supplied

Natural Language Processing Methods for Air Traffic Management Text and Speech Data

This presentation discusses two efforts of the NARI AI/ML Intern team during the Fall 2021 OSTEM Internship term. For Letters of Agreement (LoA), we have studied how LoAs are structured and explored the question ‘What is an LoA constraint?’ To do this, our approach is data-driven, iterative, and assisted by machine learning when available. In this presentation, we will walk through our tasks of manually scanning through documents, performing a preliminary entity labelling task, and our unsupervised analysis on LoA procedures sections. After this research phase, we define the smallest constraint unit in an LoA, and start to perform entity extraction. Looking towards constraint extraction, we are also exploring the use of a one-class support vector machine (OneClassSVM) model to identify patterns within the data. The second effort of our team this term is focused on Air Traffic Control System Command Center (ATCSCC) advisory meetings, and the subsequent advisory documents that get published from their content. These advisory documents are important to give readily accessible summaries of daily operations, so that data centers, airline officials, and other stakeholders can easily understand the context of these meetings in real time. In applying machine learning to this scenario, two natural language processing tasks are used. First is developing machine learning models to convert the meeting speech data into text. With this text, use of extractive and abstractive text summarization models are used to automatically generate preliminary versions of the advisory documents.

Natural Language Processing↗

DOE OSTI2019

Creating Training Data for Scientific Named Entity Recognition with Minimal Human Effort

Scientific Named Entity Referent Extraction is often more complicated than traditional Named Entity Recognition (NER). For example, in polymer science, chemical structure may be encoded in a variety of nonstandard naming conventions, and authors may refer to polymers with conventional names, commonly used names, labels (in lieu of longer names), synonyms, and acronyms. As a result, accurate scientific NER methods are often based on task-specific rules, which are difficult to develop and maintain, and are not easily generalized to other tasks and fields. Machine learning models require substantial expert-annotated data for training. Here we propose polyNER: a semi-automated system for efficient identification of scientific entities in text. PolyNER applies word embedding models to generate entity-rich corpora for productive expert labeling, and then uses the resulting labeled data to bootstrap a context-based word vector classifier. Evaluation on materials science publications shows that the polyNER approach enables improved precision or recall relative to a state-of-the-art chemical entity extraction system at a dramatically lower cost: it required just two hours of expert time, rather than extensive and expensive rule engineering, to achieve that result. This result highlights the potential for human-computer partnership for constructing domain-specific scientific NER systems.

Crowdsourcing↗

DOE OSTI2025

Coreii - Scout

COREII Scout employs React, Vite, TypeScript, Tailwind, and Daisy UI for its graphical user interface (GUI), offering both dark and light modes. The code is modular, with components and reusable wrappers to enhance efficiency. The primary goal of COREII Scout is to aid analysts in collecting and analyzing various sources related to cyber attacks, utilizing models to automate the report writing process. It uses Named Entity Recognition (NER), a type of Natural Language Processing (NLP), to extract key entities from each source. Analysts review and classify these entities using the COREII Attack Chain Estimator (ACE), adding their comments. Ultimately, a Large Language Model (LLM) generates a detailed report with user guidance. This setup ensures a streamlined and effective approach to cyber attack analysis and reporting.

Pluth, Adam [Idaho National Laboratory (INL), Idah↗

NASA NTRSDate not supplied

Exploring Semantic Search Capability of Graph Convolutions Over a Knowledge Graph Built Using Earth Science Corpora

Traditional knowledge graphs tend to be too generic, and often perform poorly on complex scientific queries. Often times, precedence is given to pop culture over scientific knowledge for queries. This is predominantly due to the use of internet sources for building the knowledge graph. With this work, we aim to explore the effectiveness of combining a knowledge graph generated from earth science corpora with a language model and graph convolutions for the purpose of surfacing latent and related sentences given a natural language query. In this model, sentences are conceptualized in the graph as nodes which are connected through entities—words and phrases of interest found in the text—extracted using Google Cloud’s entity extraction model. The language model we used for this is Bidirectional Encoder Representations from Transformers (BERT).The sentences are given a numeric representation by the BERT model. Graph convolutions are then applied to sentence embeddings in order to obtain a vector representation of the sentence as well as the surrounding graph structure, thereby leveraging the power of adjacency inherently encoded in graph structures. With this presentation, we demonstrate the ability of graph convolutions and their improved ability to surface relevant, latent information based on the subject of the input query.

Muthukumaran Ramasubramanian↗

NASA NTRSDate not supplied

What Went Wrong: A Survey of Wildfire UAS Mishaps through Named Entity Recognition

Increasingly, unmanned aircraft systems (UAS) are being applied to wildfire incidents for tasks such as mapping, aerial ignition, and delivery. As a result, incident reporting systems for wildfires are beginning to accumulate data related to UAS mishaps in wildfire response. In this research, we apply state-of-the-art natural language processing (NLP) techniques to develop a custom Named Entity Recognition (NER) model which extracts a Failure Modes and Effects Analysis (FMEA)-style survey of wildfire UAS mishaps reported in SAFECOM. The custom NER model is built by fine-tuning an existing (BERT) model, resulting in a generalizable NER model that can extract engineering relevant entities including failure modes, causes, effects, control processes, and recommendations from any failure-relevant text. Similar mishaps are clustered and reported as single rows within the FMEA. For each cluster, frequency, severity, and overall risk are computed. The methodology can be applied as part of a broader safety management system to track trends in mishaps and discover knowledge that can be utilized to improve safety outcomes and system performance.

Machine Learning↗