Search NASA⌕ Search

SEARCH · Search NASA

Results for “Word Embeddings”

Search indexed NASA NTRS and DOE OSTI research on propulsion, heat transfer, battery materials and energy systems. Follow report and document links to the original sources.

Quote a phrase for an exact phrase match. Source license links do not imply unrestricted reuse.

At least 55 records · Page 3

Unstructured clinical notes within the 24 hours since admission predict short, mid & long-term mortality in adult ICU patients

Mortality prediction for intensive care unit (ICU) patients is crucial for improving outcomes and efficient utilization of resources. Accessibility of electronic health records (EHR) has enabled data-driven predictive modeling using machine learning. However, very few studies rely solely on unstructured clinical notes from the EHR for mortality prediction. In this work, we propose a framework to predict short, mid, and long-term mortality in adult ICU patients using unstructured clinical notes from the MIMIC III database, natural language processing (NLP), and machine learning (ML) models. Depending on the statistical description of the patients’ length of stay, we define the short-term as 48-hour and 4-day period, the mid-term as 7-day and 10-day period, and the long-term as 15-day and 30-day period after admission. We found that by only using clinical notes within the 24 hours of admission, our framework can achieve a high area under the receiver operating characteristics (AU-ROC) score for short, mid and long-term mortality prediction tasks. The test AU-ROC scores are 0.87, 0.83, 0.83, 0.82, 0.82, and 0.82 for 48-hour, 4-day, 7-day, 10-day, 15-day, and 30-day period mortality prediction, respectively. We also provide a comparative study among three types of feature extraction techniques from NLP: frequency-based technique, fixed embedding-based technique, and dynamic embedding-based technique. Lastly, we provide an interpretation of the NLP-based predictive models using feature-importance scores.

60 APPLIED LIFE SCIENCES↗

Exploring the Value of Learned Representations for Automated Syndromic Definitions

Objective: To better define and automate biosurveillance syndrome categorization using modern unsupervised vector embedding techniques. Introduction: Comprehensive medical syndrome definitions are critical for outbreak investigation, disease trend monitoring, and public health surveillance. However, because current definitions are based on keyword string-matching, they may miss important distributional information in free text and medical codes that could be used to build a more general classifier. Here, we explore the idea that individual ICD codes can be categorized by examining their contextual relationships across all other ICD codes. We extend previous work in representation learning with medical data [1] by generating dense vector embeddings of these ICD codes found in emergency department (ED) visit records. The resulting representations capture information about disease co-occurrence that would typically require SME involvement and support the development of more robust syndrome definitions.

60 APPLIED LIFE SCIENCES↗

Persuasive Influence Detection: The Role of Argument Sequencing

Automatic detection of persuasion in online discussion is key to understanding how social media is used. Predicting persuasiveness is difficult, however, due to the need to model world knowledge, dialogue, and sequential reasoning. Here, we focus on modeling the sequence of arguments in social media posts using neural models with embeddings for words, discourse relations, and semantic frames. We demonstrate significant improvement over prior work in detecting successful arguments. We also present an error analysis assessing novice human performance at predicting persuasiveness.

99 GENERAL AND MISCELLANEOUS↗

Machine Learning for Identifying Relevance to Biosurveillance in Multilingual Text

Objective: The objective is to develop an ensemble of machine learning algorithms to identify multilingual, online articles that are relevant to biosurveillance. Language morphology varies widely across languages and must be accounted for when designing algorithms. Here, we compare the performance of a word embedding-based approach and a topic modeling approach with machine learning algorithms to determine the best method for Chinese, Arabic, and French languages. Introduction: Global biosurveillance is an extremely important, yet challenging task. One form of global biosurveillance comes from harvesting open source online data (e.g. news, blogs, reports, RSS feeds). The information derived from this data can be used for timely detection and identification of biological threats all over the world. However, the more inclusive the data harvesting procedure is to ensure that all potentially relevant articles are collected, the more data that is irrelevant also gets harvested. This issue can become even more complex when the online data is in a non-native language. Foreign language articles not only create language-specific issues for Natural Language Processing (NLP), but also add a significant amount of translation costs. Previous work shows success in the use of combinatory monolingual classifiers in specific applications, e.g., legal domain [1]. A critical component for a comprehensive, online harvesting biosurveillance system is the capability to identify relevant foreign language articles from irrelevant ones based on the initial article information collected, without the additional cost of full text retrieval and translation.

Machine learning↗

Visualizing Temporal Topic Embeddings with a Compass

—Dynamic topic modeling is useful at discovering the development and change in latent topics over time. However, present methodology relies on algorithms that separate document and word representations. This prevents the creation of a meaningful embedding space where changes in word usage and documents can be directly analyzed in a temporal context. This paper proposes an expansion of the compass-aligned temporal Word2Vec methodology into dynamic topic modeling. Such a method allows for the direct comparison of word and document embeddings across time in dynamic topics. This enables the creation of visualizations that incorporate temporal word embeddings within the context of documents into topic visualizations. In experiments against the current state-of-the-art, our proposed method demonstrates overall competitive performance in topic relevancy and diversity across temporal datasets of varying size. Simultaneously, it provides insightful visualizations focused on temporal word embeddings while maintaining the insights provided by global topic evolution, advancing our understanding of how topics evolve over time.

Cluster analysis↗

ESTEEM: A Novel Framework for Qualitatively Evaluating and Visualizing Spatiotemporal Embeddings in Social Media

Analyzing and visualizing large amounts of social media communications and contrasting short-term conversation changes over time and geo-locations is extremely important for commercial and government applications. Earlier approaches for large-scale text stream summarization used dynamic topic models and trending words. Instead, we rely on text embeddings – low-dimensional word representations in a continuous vector space where similar words are embedded nearby each other. This paper presents ESTEEM,1 a novel tool for visualizing and evaluating spatiotemporal embeddings learned from streaming social media texts. Our tool allows users to monitor and analyze query words and their closest neighbors with an interactive interface. We used state-of- the-art techniques to learn embeddings and developed a visualization to represent dynamically changing relations between words in social media over time and other dimensions. This is the first interactive visualization of streaming text representations learned from social media texts that also allows users to contrast differences across multiple dimensions of the data.

social media dynamics↗

Root Cause Correlation Analysis of Software Failures via Orthogonal Defect Classification and Natural Language Processing

Systems theoretic process analysis (STPA) is becoming an increasingly popular technique to assess how complex digital software systems can fail. Rather than defining failures by their observable failure events, which may be sparse especially for safety rated nuclear digital instrumentation and control systems (DI&C), failures are defined as postulated unsafe actions under specific contextual conditions. This permits a top-down analysis of system hazards and identifies whether imposed constraints and requirements can sufficiently address undesirable hazards. However, STPA is a qualitative approach at identifying inadequacies in the development process and cannot currently be used to quantify unsafe action likelihoods for probabilistic risk assessment. Therefore, in this work, we examine the root causes of software failure and explore whether a consistent correlation can be linked to specific unsafe action classes. We implement Lbl2Vec, an unsupervised document classification and retrieval algorithm, on a database of 4,096 software defect reports acquired from various open-source software systems. By analyzing sentence structure, embedded labels, and word vectors, we show that certain defect types positively correlate to specific unsafe action classes over others. The correlations developed can be used to estimate the failure probability of safety intended DI&C systems which provides a licensing basis for nuclear plant modernization efforts.

22 GENERAL STUDIES OF NUCLEAR REACTORS↗

Root Cause Correlation Analysis of Software Failures via Orthogonal Defect Classification and Natural Language Processing

Systems theoretic process analysis (STPA) is becoming an increasingly popular technique to assess how complex digital software systems can fail. Rather than defining failures by their observable failure events, which may be sparse especially for safety rated nuclear digital instrumentation and control systems (DI&C), failures are defined as postulated unsafe actions under specific contextual conditions. This permits a top-down analysis of system hazards and identifies whether imposed constraints and requirements can sufficiently address undesirable hazards. However, STPA is a qualitative approach at identifying inadequacies in the development process and cannot currently be used to quantify unsafe action likelihoods for probabilistic risk assessment. Therefore, in this work, we examine the root causes of software failure and explore whether a consistent correlation can be linked to specific unsafe action classes. We implement Lbl2Vec, an unsupervised document classification and retrieval algorithm, on a database of 4,096 software defect reports acquired from various open-source software systems. By analyzing sentence structure, embedded labels, and word vectors, we show that certain defect types positively correlate to specific unsafe action classes over others. The correlations developed can be used to estimate the failure probability of safety intended DI&C systems which provides a licensing basis for nuclear plant modernization efforts.

22 - GENERAL STUDIES OF NUCLEAR REACTORS↗

Automating the Analysis of Large Language Models Responses through Zero-Shot Question Answering

Recent advancements in Large Language Models (LLMs) have shown significant potential in various applications, yet their evaluation, particularly in zero-shot question answering scenarios, remains a challenging task. In this study, our objective was to explore precision metrics for Large Language Models (LLM) and design and implement a software pipeline to automatically evaluate LLMs' outputs under zero-shot question answering. Zero-shot question answering involves a model providing answers to questions about topics it hasn't seen during training. It leverages the principles of zero-shot learning by relying on semantic understanding and generalization from related knowledge. The data used was metadata from medical databases on congenital heart disease. We explored eleven LLM metrics and selected three for our evaluation: BLEU, BERTScore, and MoverScore. BLEU calculates a score based on the overlap of n-grams (contiguous sequences of n items, typically words) between the machine-generated translation and the reference translations. Higher BLEU scores indicate better correspondence between the machine-generated and human-generated translations. BERTScore is a metric used to evaluate the quality of machine-generated text by measuring the similarity of token embeddings produced by BERT (Bidirectional Encoder Representations from Transformers) between the generated text and reference text. MoverScore is a metric that quantifies the dissimilarity between the distributions of word embeddings from machine-generated text and reference text, emphasizing semantic similarity over exact token overlap. We also introduced HBKI, a composite metric summarizing these approaches. We tested five models —GPT-3, Llama-2, Gemini 1.5 Pro, Solar 10.7B, and Mixtral-8x7b. Our software pipeline, designed and implemented using Object-Oriented Programming principles, allows users to customize the selection and extraction of features for topics of interest in their own research. Our results show that MoverScore delivered the most precise evaluation of the LLM's outputs, while Mixtral-8x7b achieved the best overall performance in extracting metadata from the databases.

97 MATHEMATICS AND COMPUTING↗

Replacing non-biomedical concepts improves embedding of biomedical concepts

Embeddings are semantically meaningful representations of words in a vector space, commonly used to enhance downstream machine learning applications. Traditional biomedical embedding techniques often replace all synonymous words representing biological or medical concepts with a unique token, ensuring consistent representation and improving embedding quality. However, the potential impact of replacing non-biomedical concept synonyms has received less attention. Embedding approaches often employ concept replacement to replace concepts that span multiple words, such as non-small-cell lung carcinoma, with a single concept identifier (e.g., D002289). Also, all synonyms of each concept are merged into the same identifier. Here, we additionally leveraged WordNet to identify and replace sets of non-biomedical synonyms with their most common representatives. This combined approach aimed to reduce embedding noise from non-biomedical terms while preserving the integrity of biomedical concept representations. We applied this method to 1,055 biomedical concept sets representing molecular signatures or medical categories and assessed the mean pairwise distance of embeddings with and without non-biomedical synonym replacement. A smaller mean pairwise distance was interpreted as greater intra-cluster coherence and higher embedding quality. Embeddings were generated using the Word2Vec algorithm applied to a corpus of 10 million PubMed abstracts. Our results demonstrate that the addition of non-biomedical synonym replacement reduced the mean intra-cluster distance by an average of 8%, suggesting that this complementary approach enhances embedding quality. Future work will assess its applicability to other embedding techniques and downstream tasks. Python code implementing this method is provided under an open-source license.

algorithms↗

Data entry and error embedding system

A data entry and error embedding system in which, first, a document is bitmapped and recorded in a first memory. Then, it is displayed, and portions of it to be replicated by data entry are underlayed by a window, into which window replicated data is entered in location and size such that it is juxtaposed just below that which is replicated, enhancing the accuracy of replication. Second, with this format in place, selected portions of the replicated data are altered by the insertion of character or word substitutions, thus the embedding of errors. Finally, a proofreader would endeavor to correct the error embedded data and a record of his or her changes recorded. In this manner, the skill level of the proofreader and accuracy of the data are computed.

Woo, Jr., John↗

Walker Ranch 3D seismic images

Amplitude images (both vertical and depth slices) extracted from 3D seismic reflection survey over area of Walker Ranch area (adjacent to Raft River). Crossline spacing of 660 feet and inline of 165 feet using a Vibroseis source. Processing included depth migration. Micro-earthquake hypocenters on images. Stratigraphic information and nearby well tracks added to images. Images are embedded in a Microsoft Word document with additional information. Exact location and depth restricted for proprietary reasons. Data collection and processing funded by Agua Caliente. Original data remains property of Agua Caliente.

15 GEOTHERMAL ENERGY↗

A Cliffed Cape Does Not Trap Water Waves in the Sea of Constant Depth

The water wave problem is considered for a class of semi-infinite domains each having its shore shaped as a cliffed cape. In particular, the free surface of a water domain is supposed to be an infinite sector whose vertex angle is greater than π, whereas the water layer lying under the free surface is of constant depth. Under these assumptions it is shown that there are no trapped mode solutions of the problem for all values of a non-dimensional spectral parameter; in other words, no point eigenvalues are embedded in the continuous spectrum of the problem.

71 CLASSICAL AND QUANTUM MECHANICS, GENERAL PHYSIC↗

Search Enhancements using Natural Language Processing Techniques

NASA Goddard Earth Sciences Data and Information Services Center (GESDISC) is one of the 12 NASA Science Mission Directorate Data Centers. The main goal of GESDISC is to provide earth science data, information, and services to the earth science data community. Consequently, data discovery is at the center of our mission and our search engine is the primary tool for our users to interact, find, and access our data. Existing search approaches are largely focused on hard-matching of keywords in the search query with dataset metadata. Here we propose to expand the search by introducing a complementary natural language processing (NLP) search. At the heart of our proposed NLP search, we trained a joint embedding using scientific text corpus and a curated set of dataset metadata. The embedding learns the association between words in our dataset metadata and those of the scientific text corpus. This enables us to go beyond simple hard-matching of a query and data set metadata and have a notion of “similarity” between the search query and the datasets. We further integrated our NLP search into the Elastic Search (ES) framework leveraging similarity search capabilities offered through the “dense_vector” field type. Our preliminary evaluations show that our proposed NLP search has the potential to be utilized to complement the existing search engine and serve as a base for a dataset recommendation system.

Armin Mehrabian↗

TX$^2$: Transformer eXplainability and eXploration

The Transformer eXplainability and eXploration (Martindale & Stewart, 2021), or TX 2 software package, is a library designed for artificial intelligence researchers to better understand the performance of transformer models (Vaswani et al., 2017) used for sequence classification. The tool is capable of integrating with a trained transformer model and a dataset split into training and testing populations to produce an ipywidget (Project Jupyter Contributors, 2021) dashboard with a number of visualizations to understand model performance with an emphasis on explainability and interpretability. The TX 2 package is primarily intended to integrate into a workflow centered around Jupyter Notebooks (Kluyver et al., 2016), and currently assumes the use of PyTorch (Paszke et al., 2019) and Hugging Face transformers library (Wolf et al., 2020). The dashboard includes visualization and data exploration features to aid researchers, including an interactive UMAP embedding graph (McInnes et al., 2018) to understand classification clusters, a word salience map that can be updated as researchers alter textual entries in near real time, a set of tools to understand word frequency and importance based on the clusters in the UMAP embedding graph, and a set of traditional confusion matrix analysis tools.

97 MATHEMATICS AND COMPUTING↗

Machine Learning for Predicting Team Functioning in HERA Missions

Team functioning is integral to success in future long term space exploration missions. Proactively detecting declines in team functioning can mitigate conflict and ensure mission success. This project developed a speech-based artificial intelligence (AI) system that unobtrusively predicts degradation in team functioning, including performance and cohesion, in the Human Exploration Research Analog (HERA) Campaigns 4 and 5. The AI system conducted automated analysis of the prosodic (tone of voice) and linguistic (language content) components of speech, modeling interpersonal dynamics at both the turn-taking and day-wide levels. We investigated team functioning via observing structured interactions (i.e., multi-mission space exploration vehicle-extra vehicular activity [MMSEV-EVA], team interaction battery [TIB]) and unstructured interactions before the MMSEV-EVA task. We developed machine learning models to predict team functioning (objective task accuracy, self reported team efficacy and self reported team cohesion) by analyzing OpenSmile acoustic features, linguistic descriptors extracted via the linguistic inquiry and word count (LIWC) dictionary, and semantic embeddings. In the TIB, static models using logistic regression and random forests were not able to predict task accuracy, but predicted team efficacy and cohesion during both the decision making and relational tasks to a moderate level (60-70%). Majority voting on the individual turns to predict day long team efficacy further increased accuracies (70-80%). Finally, long short-term memory (LSTM) models showed the best performance across all variables (80-91%), including task performance. In the MMSEV-EVA, static models achieved an accuracy of 60% with majority voting, which increased to 80% through the incorporation of mission day as a variable, accounting for the learning effect. A key finding across both tasks was the "team-dependent" nature of these interactions; models achieved much higher accuracy when trained on prior days of the same team's data rather than attempting to generalize across entirely different teams, with even 1-2 days of prior data per team achieving 5-15% improvement over team-independent models. In addition, the incorporation of pre-task data from the same team also improves model performance, e.g., incorporating data from the decision-making task of the TIB, which preceded the relational task, improved the prediction of team efficacy and cohesion during the latter. We compared model performance when trained on machine-generated data compared to data that had been further corrected by human annotators. Overall, models trained on human-corrected data exhibited a modest improvement in performance, particularly when acoustic features were used. We found no significant correlation between word error rate (WER) and model accuracy (r(55) = -0.08, p = 0.51), but model’s accuracy was significantly higher for medium/high quality transcription (0.74 (SD = 0.48)) compared to the low-quality group (0.64 (SD = 0.36)) (t(63)=2.82, p = 0.006). Based on these, several design recommendation emerge, that could inform Standards at NASA. Models predicting team functioning should incorporate at least one to two days of historical interaction data, include brief pre-task discussions, and explicitly model temporal learning effects, especially for longer operational tasks. Minimum quality standards for automated speech-processing pipelines are needed, given the performance gains observed with manually corrected acoustic data. Finally, systems should leverage both acoustic features and language embeddings in complementary ways, with modality choices and fusion strategies tailored to mission context, task demands, and data quality requirements.

Shrivatsa Mishra↗

RTSJ Memory Areas and Their Affects on the Performance of a Flight-Like Attitude Control System

The two most important factors in improving performance in any software system, but especially a real-time, embedded system, are knowing which components are the low performers and knowing what can be done to improve their performance. The word performance with respect to a real-time, embedded system does not necessarily mean fast execution, which is the common definition when discussing non real-time systems. It also includes meeting all of the specified execution dead-lines and executing at the correct time without sacrificing non real-time performance. Using a Java prototype of an existing control system used on Deep Space 1[1], the effects from adding memory areas are measured and evaluated with respect to improving performance.

Real-time Specification for Java2 (RTSJ)↗

Using Topic Modeling and Text Embeddings to Predict Deleted Tweets

Predictive models for tweet deletion have been a relatively unexplored area of Twitter-related computational research. We first approach the deletion of tweets as a spam detection problem, applying a small set of handcrafted features to improve upon the current state-of-the- art in predicting deleted tweets. Next, we apply our approach to a dataset of deleted tweets that better reflects the current deletion rate. Since tweets are deleted for reasons beyond just the presence of spam, we apply topic modeling and text embeddings in order to capture the semantic content of tweets that can lead to tweet deletion. Our goal is to create an effective model that has a low-dimensional feature space and is also language-independent. A lean model would be computationally advantageous processing high-volumes of Twitter data, which can reach 9,885 tweets per second. Our results show that a small set of spam-related features combined with word topics and character-level text embeddings provide the best f1 when trained with a random forest model. The highest precision of the deleted tweet class is achieved by a modification of paragraph2vec to capture author identity.

Potash, Peter J.↗