PubMed Health⌕ Search

SEARCH · PubMed Health

Results for “Natural Language Processing”

Explore indexed PubMed citations for clinical trials, systematic reviews and public health research. Read source abstracts and follow each citation to its original PubMed record.

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

At least 847 records · Page 47Linked to original sources
PubMed2006

A critical review of PASBio's argument structures for biomedical verbs.

BACKGROUND: Propositional representations of biomedical knowledge are a critical component of most aspects of semantic mining in biomedicine. However, the proper set of propositions has yet to be determined. Recently, the PASBio project proposed a set of propositions and argument structures for biomedical verbs. This initial set of representations presents an opportunity for evaluating the suitability of predicate-argument structures as a scheme for representing verbal semantics in the biomedical domain. Here, we quantitatively evaluate several dimensions of the initial PASBio propositional structure repository. RESULTS: We propose a number of metrics and heuristics related to arity, role labelling, argument realization, and corpus coverage for evaluating large-scale predicate-argument structure proposals. We evaluate the metrics and heuristics by applying them to PASBio 1.0. CONCLUSION: PASBio demonstrates the suitability of predicate-argument structures for representing aspects of the semantics of biomedical verbs. Metrics related to theta-criterion violations and to the distribution of arguments are able to detect flaws in semantic representations, given a set of predicate-argument structures and a relatively small corpus annotated with them.

Abstracting and Indexing↗
PubMed2006

Mapping data elements to terminological resources for integrating biomedical data sources.

BACKGROUND: Data integration is a crucial task in the biomedical domain and integrating data sources is one approach to integrating data. Data elements (DEs) in particular play an important role in data integration. We combine schema- and instance-based approaches to mapping DEs to terminological resources in order to facilitate data sources integration. METHODS: We extracted DEs from eleven disparate biomedical sources. We compared these DEs to concepts and/or terms in biomedical controlled vocabularies and to reference DEs. We also exploited DE values to disambiguate underspecified DEs and to identify additional mappings. RESULTS: 82.5% of the 474 DEs studied are mapped to entries of a terminological resource and 74.7% of the whole set can be associated with reference DEs. Only 6.6% of the DEs had values that could be semantically typed. CONCLUSION: Our study suggests that the integration of biomedical sources can be achieved automatically with limited precision and largely facilitated by mapping DEs to terminological resources.

Abstracting and Indexing↗
PubMed2005

GeneSeer: a sage for gene names and genomic resources.

BACKGROUND: Independent identification of genes in different organisms and assays has led to a multitude of names for each gene. This balkanization makes it difficult to use gene names to locate genomic resources, homologs in other species and relevant publications. METHODS: We solve the naming problem by collecting data from a variety of sources and building a name-translation database. We have also built a table of homologs across several model organisms: H. sapiens, M. musculus, R. norvegicus, D. melanogaster, C. elegans, S. cerevisiae, S. pombe and A. thaliana. This allows GeneSeer to draw phylogenetic trees and identify the closest homologs. This, in turn, allows the use of names from one species to identify homologous genes in another species. A website http://geneseer.cshl.org/ is connected to the database to allow user-friendly access to our tools and external genomic resources using familiar gene names. CONCLUSION: GeneSeer allows access to gene information through common names and can map sequences to names. GeneSeer also allows identification of homologs and paralogs for a given gene. A variety of genomic data such as sequences, SNPs, splice variants, expression patterns and others can be accessed through the GeneSeer interface. It is freely available over the web http://geneseer.cshl.org/ and can be incorporated in other tools through an http-based software interface described on the website. It is currently used as the search engine in the RNAi codex resource, which is a portal for short hairpin RNA (shRNA) gene-silencing constructs.

Alternative Splicing↗
PubMed2005

A controlled trial of automated classification of negation from clinical notes.

BACKGROUND: Identification of negation in electronic health records is essential if we are to understand the computable meaning of the records: Our objective is to compare the accuracy of an automated mechanism for assignment of Negation to clinical concepts within a compositional expression with Human Assigned Negation. Also to perform a failure analysis to identify the causes of poorly identified negation (i.e. Missed Conceptual Representation, Inaccurate Conceptual Representation, Missed Negation, Inaccurate identification of Negation). METHODS: 41 Clinical Documents (Medical Evaluations; sometimes outside of Mayo these are referred to as History and Physical Examinations) were parsed using the Mayo Vocabulary Server Parsing Engine. SNOMED-C was used to provide concept coverage for the clinical concepts in the record. These records resulted in identification of Concepts and textual clues to Negation. These records were reviewed by an independent medical terminologist, and the results were tallied in a spreadsheet. Where questions on the review arose Internal Medicine Faculty were employed to make a final determination. RESULTS: SNOMED-CT was used to provide concept coverage of the 14,792 Concepts in 41 Health Records from John's Hopkins University. Of these, 1,823 Concepts were identified as negative by Human review. The sensitivity (Recall) of the assignment of negation was 97.2% (p < 0.001, Pearson Chi-Square test; when compared to a coin flip). The specificity of assignment of negation was 98.8%. The positive likelihood ratio of the negation was 81. The positive predictive value (Precision) was 91.2% CONCLUSION: Automated assignment of negation to concepts identified in health records based on review of the text is feasible and practical. Lexical assignment of negation is a good test of true Negativity as judged by the high sensitivity, specificity and positive likelihood ratio of the test. SNOMED-CT had overall coverage of 88.7% of the concepts being negated.

Abstracting and Indexing↗
PubMed2005

askMEDLINE: a free-text, natural language query tool for MEDLINE/PubMed.

BACKGROUND: Plain language search tools for MEDLINE/PubMed are few. We wanted to develop a search tool that would allow anyone using a free-text, natural language query and without knowing specialized vocabularies that an expert searcher might use, to find relevant citations in MEDLINE/PubMed. This tool would translate a question into an efficient search. RESULTS: The accuracy and relevance of retrieved citations were compared to references cited in BMJ POEMs and CATs (critically appraised topics) questions from the University of Michigan Department of Pediatrics. askMEDLINE correctly matched the cited references 75.8% in POEMs and 89.2 % in CATs questions on first pass. When articles that were deemed to be relevant to the clinical questions were included, the overall efficiency in retrieving journal articles was 96.8% (POEMs) and 96.3% (CATs.) CONCLUSION: askMEDLINE might be a useful search tool for clinicians, researchers, and other information seekers interested in finding current evidence in MEDLINE/PubMed. The text-only format could be convenient for users with wireless handheld devices and those with low-bandwidth connections in remote locations.

Academic Medical Centers↗
PubMed2006

The caCORE Software Development Kit: streamlining construction of interoperable biomedical information services.

BACKGROUND: Robust, programmatically accessible biomedical information services that syntactically and semantically interoperate with other resources are challenging to construct. Such systems require the adoption of common information models, data representations and terminology standards as well as documented application programming interfaces (APIs). The National Cancer Institute (NCI) developed the cancer common ontologic representation environment (caCORE) to provide the infrastructure necessary to achieve interoperability across the systems it develops or sponsors. The caCORE Software Development Kit (SDK) was designed to provide developers both within and outside the NCI with the tools needed to construct such interoperable software systems. RESULTS: The caCORE SDK requires a Unified Modeling Language (UML) tool to begin the development workflow with the construction of a domain information model in the form of a UML Class Diagram. Models are annotated with concepts and definitions from a description logic terminology source using the Semantic Connector component. The annotated model is registered in the Cancer Data Standards Repository (caDSR) using the UML Loader component. System software is automatically generated using the Codegen component, which produces middleware that runs on an application server. The caCORE SDK was initially tested and validated using a seven-class UML model, and has been used to generate the caCORE production system, which includes models with dozens of classes. The deployed system supports access through object-oriented APIs with consistent syntax for retrieval of any type of data object across all classes in the original UML model. The caCORE SDK is currently being used by several development teams, including by participants in the cancer biomedical informatics grid (caBIG) program, to create compatible data services. caBIG compatibility standards are based upon caCORE resources, and thus the caCORE SDK has emerged as a key enabling technology for caBIG. CONCLUSION: The caCORE SDK substantially lowers the barrier to implementing systems that are syntactically and semantically interoperable by providing workflow and automation tools that standardize and expedite modeling, development, and deployment. It has gained acceptance among developers in the caBIG program, and is expected to provide a common mechanism for creating data service nodes on the data grid that is under development.

Humans↗
PubMed2003

Generating hypotheses by discovering implicit associations in the literature: a case report of a search for new potential therapeutic uses for thalidomide.

The availability of scientific bibliographies through online databases provides a rich source of information for scientists to support their research. However, the risk of this pervasive availability is that an individual researcher may fail to find relevant information that is outside the direct scope of interest. Following Swanson's ABC model of disjoint but complementary structures in the biomedical literature, we have developed a discovery support tool to systematically analyze the scientific literature in order to generate novel and plausible hypotheses. In this case report, we employ the system to find potentially new target diseases for the drug thalidomide. We find solid bibliographic evidence suggesting that thalidomide might be useful for treating acute pancreatitis, chronic hepatitis C, Helicobacter pylori-induced gastritis, and myasthenia gravis. However, experimental and clinical evaluation is needed to validate these hypotheses and to assess the trade-off between therapeutic benefits and toxicities.

Databases, Bibliographic↗
PubMed2004

A multi-aspect comparison study of supervised word sense disambiguation.

OBJECTIVE: The aim of this study was to investigate relations among different aspects in supervised word sense disambiguation (WSD; supervised machine learning for disambiguating the sense of a term in a context) and compare supervised WSD in the biomedical domain with that in the general English domain. METHODS: The study involves three data sets (a biomedical abbreviation data set, a general biomedical term data set, and a general English data set). The authors implemented three machine-learning algorithms, including (1) naïve Bayes (NBL) and decision lists (TDLL), (2) their adaptation of decision lists (ODLL), and (3) their mixed supervised learning (MSL). There were six feature representations (various combinations of collocations, bag of words, oriented bag of words, etc.) and five window sizes (2, 4, 6, 8, and 10). RESULTS: Supervised WSD is suitable only when there are enough sense-tagged instances with at least a few dozens of instances for each sense. Collocations combined with neighboring words are appropriate selections for the context. For terms with unrelated biomedical senses, a large window size such as the whole paragraph should be used, while for general English words a moderate window size between 4 and 10 should be used. The performance of the authors' implementation of decision list classifiers for abbreviations was better than that of traditional decision list classifiers. However, the opposite held for the other two sets. Also, the authors' mixed supervised learning was stable and generally better than others for all sets. CONCLUSION: From this study, it was found that different aspects of supervised WSD depend on each other. The experiment method presented in the study can be used to select the best supervised WSD classifier for each ambiguous term.

Abbreviations as Topic↗
PubMed2001

Speech recognition software.

This article discusses the use of speech recognition software by means of reviewing two leading packages. Both programs require considerable training before they can be used effectively, but are then able to convert continuous speech into text with varying degrees of success.

CD-ROM↗
PubMed2026

Benchmarking large language models for extracting biobank-derived insights into health and disease.

Biobank-scale datasets such as the UK Biobank have become foundational resources for advancing biomedical discovery. Yet the complexity and heterogeneity of these resources, spanning genomics, imaging, clinical records, and metadata, pose substantial barriers to access and interpretation. Large Language Models (LLMs) offer a promising avenue for making such datasets more navigable through natural language interfaces. However, the extent to which current general-purpose LLMs can retrieve and synthesize biobank-specific insights has not yet been systematically evaluated. In this study, we present a reproducible, multi-metric evaluation framework to benchmark the capabilities of leading LLMs. We evaluated six leading large language models: Gemini 3 Pro, Claude Opus 4.5, Claude Sonnet 4.5, GPT-5.2, Mistral Large 2, and DeepSeek V3, on four benchmark tasks designed to assess biobank-related knowledge retrieval. We evaluate model performance across six dimensions (semantic accuracy, factual correctness, domain knowledge, reasoning quality, response depth, and biobank specificity) and assessed output consistency using curated UK Biobank references and a robust random baseline. All models outperformed the baseline by 2&#xd7; to 3&#xd7;&#x2009;, with strong statistical separation (p&#x2009;<&#x2009;0.001), confirming meaningful biobank-specific knowledge retrieval. Gemini 3 Pro achieved the highest overall accuracy across tasks such as keyword synthesis, institution recognition, and topic inference, while Claude Sonnet 4.5 demonstrated the most uniform performance across evaluation dimensions. Our benchmark provides a rigorous framework for evaluating LLMs in biomedical settings. Using the UK Biobank as a real-world testbed, we highlight both the capabilities and limitations of current models, measuring their capacity to recall structured biomedical knowledge consistent with authoritative biobank metadata.

Large Language Models↗
PubMed2004

Evaluation of a computerized aid for creating human behavioral representations of human-computer interaction.

The research reported herein presents the results of an empirical evaluation that focused on the accuracy and reliability of cognitive models created using a computerized tool: the cognitive analysis tool for human-computer interaction (CAT-HCI). A sample of participants, expert in interacting with a newly developed tactical display for the U.S. Army's Bradley Fighting Vehicle, individually modeled their knowledge of 4 specific tasks employing the CAT-HCI tool. Measures of the accuracy and consistency of task models created by these task domain experts using the tool were compared with task models created by a double expert. The findings indicated a high degree of consistency and accuracy between the different "single experts" in the task domain in terms of the resultant models generated using the tool. Actual or potential applications of this research include assessing human-computer interaction complexity, determining the productivity of human-computer interfaces, and analyzing an interface design to determine whether methods can be automated.

Auditory Perception↗
PubMed2004

Improving the precision of the keyword-matching pornographic text filtering method using a hybrid model.

With the flooding of pornographic information on the Internet, how to keep people away from that offensive information is becoming one of the most important research areas in network information security. Some applications which can block or filter such information are used. Approaches in those systems can be roughly classified into two kinds: metadata based and content based. With the development of distributed technologies, content based filtering technologies will play a more and more important role in filtering systems. Keyword matching is a content based method used widely in harmful text filtering. Experiments to evaluate the recall and precision of the method showed that the precision of the method is not satisfactory, though the recall of the method is rather high. According to the results, a new pornographic text filtering model based on reconfirming is put forward. Experiments showed that the model is practical, has less loss of recall than the single keyword matching method, and has higher precision.

Algorithms↗
PubMed2004

Biosphere: the interoperation of web services in microarray cluster analysis.

UNLABELLED: The growing use of DNA microarrays in biomedical research has led to the proliferation of analysis tools. These software programs address different aspects of analysis (e.g. normalisation and clustering within and across individual arrays) as well as extended analysis methods (e.g. clustering, annotation and mining of multiple datasets). Therefore, microarray data analysis typically requires the interoperability of multiple software programs involving different analysis types and methods. Such interoperation is often hampered by the heterogeneity inherent in the software tools (which may function by implementing different interfaces and using different programming languages). To address this problem, we employed the simple object access protocol (SOAP)-based web service approach that provides a uniform programmatic interface to these heterogeneous software components. To demonstrate this approach in the microarray context, we created a web server application, Biosphere, which interoperates a number of web services that are geographically widely distributed. These web services include a clustering web service, which is a suite of different clustering algorithms for analysing microarray data; XEMBL, developed at the European Bioinformatics Institute (EBI) for retrieving EMBL Nucleotide Sequence Database sequence data; and three gene annotation web services: GetGO, GetHAPI and GetUMLS. GetGO allows retrieval of Gene Ontology (GO) annotation, and the other two web services retrieve annotation from the biomedical literature that is indexed based on the Medical Subject Headings (MeSH) terms. With these web services, Biosphere allows the users to do the following: (i) cluster gene expression data using seven different algorithms; (ii) visualise the clustering results that are grouped statistically in colour; and (iii) retrieve sequence, annotation and citation data for the genes of interest. AVAILABILITY: Biosphere and its web services described in Web Service Description Language (WSDL) can be accessed at http://rook.cecid.hku.hk:8280/BiosphereServer.

Cluster Analysis↗
PubMed2003

Searching for cancer information on the internet: analyzing natural language search queries.

BACKGROUND: Searching for health information is one of the most-common tasks performed by Internet users. Many users begin searching on popular search engines rather than on prominent health information sites. We know that many visitors to our (National Cancer Institute) Web site, cancer.gov, arrive via links in search engine result. OBJECTIVE: To learn more about the specific needs of our general-public users, we wanted to understand what lay users really wanted to know about cancer, how they phrased their questions, and how much detail they used. METHODS: The National Cancer Institute partnered with AskJeeves, Inc to develop a methodology to capture, sample, and analyze 3 months of cancer-related queries on the Ask.com Web site, a prominent United States consumer search engine, which receives over 35 million queries per week. Using a benchmark set of 500 terms and word roots supplied by the National Cancer Institute, AskJeeves identified a test sample of cancer queries for 1 week in August 2001. From these 500 terms only 37 appeared >or= 5 times/day over the trial test week in 17208 queries. Using these 37 terms, 204165 instances of cancer queries were found in the Ask.com query logs for the actual test period of June-August 2001. Of these, 7500 individual user questions were randomly selected for detailed analysis and assigned to appropriate categories. The exact language of sample queries is presented. RESULTS: Considering multiples of the same questions, the sample of 7500 individual user queries represented 76077 queries (37% of the total 3-month pool). Overall 78.37% of sampled Cancer queries asked about 14 specific cancer types. Within each cancer type, queries were sorted into appropriate subcategories including at least the following: General Information, Symptoms, Diagnosis and Testing, Treatment, Statistics, Definition, and Cause/Risk/Link. The most-common specific cancer types mentioned in queries were Digestive/Gastrointestinal/Bowel (15.0%), Breast (11.7%), Skin (11.3%), and Genitourinary (10.5%). Additional subcategories of queries about specific cancer types varied, depending on user input. Queries that were not specific to a cancer type were also tracked and categorized. CONCLUSIONS: Natural-language searching affords users the opportunity to fully express their information needs and can aid users naïve to the content and vocabulary. The specific queries analyzed for this study reflect news and research studies reported during the study dates and would surely change with different study dates. Analyzing queries from search engines represents one way of knowing what kinds of content to provide to users of a given Web site. Users ask questions using whole sentences and keywords, often misspelling words. Providing the option for natural-language searching does not obviate the need for good information architecture, usability engineering, and user testing in order to optimize user experience.

Benchmarking↗
PubMed2000

Continuous speech recognition in MR imaging reporting: advantages, disadvantages, and impact.

OBJECTIVE: Our objective was to describe our experience with a commercially available continuous speech recognition system, highlighting the advantages, disadvantages, and costs compared with those of conventional transcription for MR imaging reports. MATERIALS AND METHODS: Data from 5072 reports generated in our MR imaging section during a 9-month period after the implementation of a commercial continuous speech recognition system were compared with 4552 reports produced during the same period 1 year earlier. Information pertaining to the use of continuous speech recognition, report turnaround time, word recognition rate, report appearance, and equipment costs was collected. RESULTS: After its system installation, continuous speech recognition was used to dictate 81.8% of all reports. The mean report turnaround time decreased from 87.8 to 43.6 hr, and report availability at 24 hr increased from 10.5% to 62.5%. The system was found to have an average word recognition accuracy of 92.7% for spontaneous dictation. Mean report length declined from 95 to 60 words, with an increase in spacing errors from 0.3 to 8.0 per 1000 words and a decrease in spelling errors from 3.0 to 0.8 per 1000 words. Initial hardware and software costs were approximately $10,000, compared with a yearly cost of $12,000 for human transcription. CONCLUSION: Although the technology is still evolving and was evaluated in its earliest implementation stages, continuous speech recognition nonetheless markedly improved report turnaround time and proved cost-effective.

Cost-Benefit Analysis↗