Domain-specific Resources Research Articles

MedNER: A Service-Oriented Framework for Chinese Medical Named-Entity Recognition with Real-World Application

Named-entity recognition (NER) is a crucial task in natural language processing, especially for extracting meaningful information from unstructured text data. In the healthcare domain, accurate NER can significantly enhance patient care by enabling efficient extraction and analysis of clinical information. This paper presents MedNER, a novel service-oriented framework designed specifically for medical NER in Chinese medical texts. MedNER leverages advanced deep learning techniques and domain-specific linguistic resources to achieve good performance in identifying diabetes-related entities such as symptoms, tests, and drugs. The framework integrates seamlessly with real-world healthcare systems, offering scalable and efficient solutions for processing large volumes of clinical data. This paper provides an in-depth discussion on the architecture and implementation of MedNER, featuring the concept of Deep Learning as a Service (DLaaS). A prototype has encapsulated BiLSTM-CRF and BERT-BiLSTM-CRF models into the core service, demonstrating its flexibility, usability, and effectiveness in addressing the unique challenges of Chinese medical text processing.

Automating literature screening and curation with applications to computational neuroscience.

ModelDB (https://modeldb.science) is a discovery platform for computational neuroscience, containing over 1850 published model codes with standardized metadata. These codes were mainly supplied from unsolicited model author submissions, but this approach is inherently limited. For example, we estimate we have captured only around one-third of NEURON models, the most common type of models in ModelDB. To more completely characterize the state of computational neuroscience modeling work, we aim to identify works containing results derived from computational neuroscience approaches and their standardized associated metadata (eg, cell types, research topics). Known computational neuroscience work from ModelDB and identified neuroscience work queried from PubMed were included in our study. After pre-screening with SPECTER2 (a free document embedding method), GPT-3.5, and GPT-4 were used to identify likely computational neuroscience work and relevant metadata. SPECTER2, GPT-4, and GPT-3.5 demonstrated varied but high abilities in identification of computational neuroscience work. GPT-4 achieved 96.9% accuracy and GPT-3.5 improved from 54.2% to 85.5% through instruction-tuning and Chain of Thought. GPT-4 also showed high potential in identifying relevant metadata annotations. Accuracy in identification and extraction might further be improved by dealing with ambiguity of what are computational elements, including more information from papers (eg, Methods section), improving prompts, etc. Natural language processing and large language model techniques can be added to ModelDB to facilitate further model discovery, and will contribute to a more standardized and comprehensive framework for establishing domain-specific resources.

Selective interference of liking and beauty judgements on affective working memory and visual working memory.

Aesthetic processing has profound implications for everyday life. Although liking and beauty judgements are outcomes of aesthetic processing and derive from a common hedonic value, there may be some differences in how they engage working memory. This study used maintenance and aesthetic judgement tasks to examine whether liking and beauty judgements make different demands on domain-specific working memory resources. Sixty participants (30 males) were instructed to rate picture for liking or beauty while maintaining the subjective affect or brightness of the presented pictures. Results indicated that liking judgements selectively impaired participants' performance in the affect maintenance task, and beauty judgements selectively impaired their performance in the brightness maintenance task. In addition, maintaining affect and brightness feelings in the mind increased image ratings on beauty but not on liking. Our findings provide evidence that liking judgements draw more on affective working memory resources than beauty judgements, and beauty judgements draw more on visual working memory resources than liking judgements.

Multilingual Language Diversity Protects Native Language Production under Different Control Demands.

The use of multiple languages has been found to influence individuals' cognitive abilities. Although some studies have also investigated the effect of multilingualism on non-native language proficiency, fewer studies have focused on how multilingual experience affects native language production. This study investigated the effect of multilingualism on native language production, specifically examining control demands through a semantic Go/No-Go picture naming task. The multilingual experience was quantified using language entropy, which measures the uncertainty and diversity of language use. Control demands were achieved by manipulating the proportion of Go (i.e., naming) trials in different conditions. Results showed that as control demands increased, multilingual individuals exhibited poorer behavioral performance and greater brain activation throughout the brain. Moreover, more diverse language use was associated with higher accuracy in naming and more interconnected brain networks with greater involvement of domain-general neural resources and less domain-specific neural resources. Notably, the varied and balanced use of multiple languages enabled multilingual individuals to respond more efficiently to increased task demands during native language production.

Reinforcement learning based distantly supervised biomedical named entity recognition

Data used by current Biomedical named entity recognition (BioNER) systems has mostly been manually labelled for supervision. However, it might be difficult to find large amounts of annotated data, especially in fields with a high level of specialization, such as biomedical, bioinformatics, and so on. When dictionaries and ontologies are available, which are domain-specific knowledge resources, automatically tagged distantly supervised biomedical training data can be developed. However, any such distantly supervised NER result is normally noisy. The prevalence of false positives and false negatives with this type of autonomously generated data is the main problem that directly affects efficiency. This research investigates distant supervision to detect false positive occurrences in BioNER task. A reinforcement learning technique is employed that is modelled as a graphical processing unit (GPU) accelerated Markov decision process (MDP) with a neural network policy. To deal with false negative cases, we employ a partial annotation conditional random field (CRF) technique. Results on two benchmark datasets show a cutting-edge methodology that can enhance the functionality of the neural NER system. It goes on to show how the proposed approach cuts down on human annotated data for BioNER tasks in Natural Language Processing (NLP).

Predicting Challenge and Threat Appraisal of Job Demands among Nurses: The Role of Matching Job Resources.

(1) Background: Empirical studies have started to examine employees' subjective appraisals of job demands and their relations to employees' health. However, knowledge of working conditions, which might contribute to how employees appraise specific job demands, is scarce. The present study aimed to examine predictors of nurses' appraisals of job demands (i.e., time pressure, emotional demands, physical demands, and role ambiguity) as challenges and/or threats among corresponding job resources (i.e., autonomy, social support, physical resources, participation in decision-making). It also examined moderating effects of these predictors. (2) Methods: Cross-sectional data were collected via an online survey in a sample of 425 nurses working in Luxembourg. (3) Results: Multiple regression analyses indicated that matching job resources predicted nurses' appraisal of job demands as challenging. Threat appraisal was predicted by three out of four kinds of job resources (i.e., autonomy, physical resources, participation in decision-making). However, the current study did not find any moderating effects between job demands and job resources on challenge/threat appraisals. (4) Conclusions: The present study identified domain-specific job resources that contribute to how employees perceive selected job demands. Accordingly, we encourage scholars and practitioners to align job demands with matching job resources to prevent nurses' threat appraisal of job demands, and to promote their challenge appraisals.

How To Teach Domain Ontology-based Knowledge Graph Construction? An Irish Experiment

Domains represent concepts which belong to specific parts of the world. The particularized meaning of words linguistically encoding such domain concepts are provided by domain specific resources. The explicit meaning of such words are increasingly captured computationally using domain-specific ontologies, which, even for the same reference domain, are most often than not semantically incompatible. As information systems that rely on domain ontologies expand, there is a growing need to not only design domain ontologies and domain ontology-grounded Knowl­edge Graphs (KGs) but also to align them to general standards and conventions for interoperability. This often presents an insurmountable challenge to domain experts who have to additionally learn the construction of domain ontologies and KGs. Until now, several research methodologies have been proposed by different research groups using different technical approaches and based on scenarios of different domains of application. However, no methodology has been proposed which not only facilitates designing conceptually well-founded ontologies, but is also, equally, grounded in the general pedagogical principles of knowl­edge organization and, thereby, flexible enough to teach, and reproduce vis-à-vis domain experts. The purpose of this paper is to provide such a general, pedagogically flexible semantic knowl­edge modelling methodology. We exemplify the methodology by examples and illustrations from a professional-level digital healthcare course, and conclude with an evaluation grounded in technological parameters as well as user experience design principles.

Cognitive Resources in Working Memory: Domain-Specific or General?

An experiment in the dual-task paradigm was carried out to explore the nature of domain-specific and domain-general resource distribution in working memory. The subjects (N = 32) performed symmetry span and letter reading span tasks under visuospatial (tapping) and verbal (articulatory suppression) cognitive load. The effects of task type and cognitive load modality were analyzed. The results are described within the concentric model framework: significant distinctions in relative accuracy under visuospatial and verbal cognitive load in visuospatial and verbal tasks were observed when N elements in the set exceeded the region of direct access capacity, while no such effect was observed for 2-3 element sets. This is attributed to domain-general resources in the region of direct access, and domain-specific resources in the activated long-term memory. We also found evidence for the asymmetric distribution of visuospatial and verbal working memory resources in that the verbal component is more susceptible to cognitive load.

Semantic Indexing of Open Scientific Literature to Help Users Discover and Navigate through Publications Networks

In recent years, several evolutions have drastically transformed the way researchers as well as scientific and technical information (STI) services interact with scientific literature. The amount and pace of publications are skyrocketing, whether in journals and conferences or through pre-publication repositories (e.g., arxiv.org), such that it is increasingly difficult to keep up, find and make sense of relevant articles. Furthermore, the specialization of research communities makes it difficult to discover cross-disciplinary knowledge, which is essential to meet the growing demand of funding agencies for interdisciplinary projects. Scientific open archives are central in this landscape, however the keyword-based search services that they usually provide fail to grasp the semantic relationships between articles. Therefore, it is necessary to develop new tools that allow users to find their way in this mass of knowledge. In this talk, we wish to present the methods, tools and services implemented in the ISSA*2 project to address these needs, and discuss how they could fit and be deployed in the biodiversity area. Guided by the open science goals and embracing the FAIR*1 principles, the project aims to: provide a generic, transferable and extensible pipeline for the analysis and processing of the articles of an open scientific archive; turn the processing results into a semantic index stored and published as a public RDF knowledge graph; develop innovative search and visualization services that leverage this semantic index to allow researchers, decision makers or STI professionals to explore thematic association rules, networks of co-publications, articles with co-occurring topics, etc. provide a generic, transferable and extensible pipeline for the analysis and processing of the articles of an open scientific archive; turn the processing results into a semantic index stored and published as a public RDF knowledge graph; develop innovative search and visualization services that leverage this semantic index to allow researchers, decision makers or STI professionals to explore thematic association rules, networks of co-publications, articles with co-occurring topics, etc. The semantic index construction process involves several artificial intelligence techniques: natural language processing, knowledge engineering and Semantic Web. These techniques are used to process the publications’ metadata and text to automatically extract thematic descriptors and named entities. These descriptors and named entities are linked to knowledge bases such as Wikidata, DBpedia and GeoNames, or domain-specific terminological resources suited to the archive's domain. The semantic index linked with the third-party resources serves as a keystone to support the development of rich search and visualization tools aimed at researchers and/or STI professionals. We demonstrated the effectiveness of this solution in the use case of Agritrop, an institutional archive of 110,000+ resources among which are 12,000 open access articles, specialized in the fields of agronomy, biodiversity and sustainable development. In this context, the Agrovoc multilingual thesaurus was used as a domain-specific reference vocabulary. Fig. 1 illustrates how the concepts mentioned in the articles of the archive can be used to discover and visualize association rules. In this example, articles mentioning concepts COVID-19 and food security (a) frequently mention concept pandemics (b). Fig. 2 shows how other visualization techniques can help users search articles mentioning concept health or any of its sub-concepts (a and b), discover that it is often co-mentioned with climate change (c), and get the list of related publications (d) and their time distribution (e). Being designed as a generic, transferable solution, the pipeline and visualization tools delivered by ISSA could be easily adapted to open archives of biodiversity literature. Typically, terminological references such as Darwin Core Terms, Access to Biological Collection Data (ABCD), open Digital Specimens (openDS), Audubon Core Metadata Schema as well as various taxonomic registries, could be considered for the description of an article's metadata or the linking of thematic descriptors and named entities. From there, the proposed visualization techniques could easily be reconfigured to explore the articles from a biodiversity open archive to answer various competency questions, for instance: what are the articles that mention a taxon or any of its child taxa? What are the museums/institutions that are more frequently mentioned together with certain taxonomic groups? What are the research topics that frequently co-occur with climate change, and how do these topics evolve through the years? What public policies frequently occur in articles that mention endangered species? Furthermore, the pipeline could be extended by including existing third-party tools to carry out e.g., the extraction of relationships between entities or the reconciliation of authors' names.

Characterizing domain-specific open educational resources by linking ISCB Communities of Special Interest to Wikipedia.

MotivationWikipedia is one of the most important channels for the public communication of science and is frequently accessed as an educational resource in computational biology. Joint efforts between the International Society for Computational Biology (ISCB) and the Computational Biology taskforce of WikiProject Molecular Biology (a group of expert Wikipedia editors) have considerably improved computational biology representation on Wikipedia in recent years. However, there is still an urgent need for further improvement in quality, especially when compared to related scientific fields such as genetics and medicine. Facilitating involvement of members from ISCB Communities of Special Interest (COSIs) would improve a vital open education resource in computational biology, additionally allowing COSIs to provide a quality educational resource highly specific to their subfield.ResultsWe generate a list of around 1500 English Wikipedia articles relating to computational biology and describe the development of a binary COSI-Article matrix, linking COSIs to relevant articles and thereby defining domain-specific open educational resources. Our analysis of the COSI-Article matrix data provides a quantitative assessment of computational biology representation on Wikipedia against other fields and at a COSI-specific level. Furthermore, we conducted similarity analysis and subsequent clustering of COSI-Article data to provide insight into potential relationships between COSIs. Finally, based on our analysis, we suggest courses of action to improve the quality of computational biology representation on Wikipedia.

Complex Portal 2022: new curation frontiers.

The Complex Portal (www.ebi.ac.uk/complexportal) is a manually curated, encyclopaedic database of macromolecular complexes with known function from a range of model organisms. It summarizes complex composition, topology and function along with links to a large range of domain-specific resources (i.e. wwPDB, EMDB and Reactome). Since the last update in 2019, we have produced a first draft complexome for Escherichia coli, maintained and updated that of Saccharomyces cerevisiae, added over 40 coronavirus complexes and increased the human complexome to over 1100 complexes that include approximately 200 complexes that act as targets for viral proteins or are part of the immune system. The display of protein features in ComplexViewer has been improved and the participant table is now colour-coordinated with the nodes in ComplexViewer. Community collaboration has expanded, for example by contributing to an analysis of putative transcription cofactors and providing data accessible to semantic web tools through Wikidata which is now populated with manually curated Complex Portal content through a new bot. Our data license is now CC0 to encourage data reuse. Users are encouraged to get in touch, provide us with feedback and send curation requests through the ‘Support’ link.

Detecting Multi Word Terms in patents the same way as entities

In English patent document information retrieval, Multi Word Terms (MWTs) are an important factor in determining how relevant a patent document is for a particular search query. Detecting the correct boundaries for these MWTs is no trivial task and often complicated by the special writing style of the patent domain. In this paper we describe a method for detecting MWTs in patent sentences based on a method for detecting technical entities using deep learning. On our annotated dataset of 22 patents, our method achieved an average precision of 0.75, an average recall of 0.74 and an average F1 score of 0.74. Further, we argue for the use of domain specific word embedding resources and suggest that our model mostly learns whether individual words should be included in MWTs or not.

Technical language processing: Unlocking maintenance knowledge

Out-of-the-box natural-language processing (NLP) pipelines need re-imagining to understand and meet the requirements of engineering data. Text-based documents account for a significant portion of data collected during the life cycle of asset management and the valuable information these documents contain is underutilized in analysis. Meanwhile, researchers historically design NLP pipelines with non-technical language in mind. This means industrial implementations are built on tools intended for non-technical use cases, suffering from a lack of verification, validation, and ultimately, personnel trust. To mitigate these sources of risk, we encourage a holistic, domain-driven approach to using NLP in a technical engineering setting, a paradigm we refer to as Technical Language Processing (TLP). Toward this end, the industrial asset management community must collectively redouble efforts toward production of and consensus around key domain-specific resources, including: (1) goal-driven data representations, (2) flexible entity type definitions and dictionaries, and (3) improved access to data-sets – raw and annotated. This collective action allows the maintenance community to follow in the path of other scientific communities, e.g., medicine, to develop and utilize these public resources to make TLP a key contributor to Industry 4.0.

How do students become experts? An in-depth study on the development of domain-specific awareness in a materials chemistry course

ABSTRACT The need to develop clean-energy storage systems, harvesting technologies, and new functional materials are but a few of the many contemporary challenges in the domain of science today. To address these challenges, problem solving has become a key skill of the 21st century. It is generally documented, however, that experts and students differ in terms of their problem-solving skills. Experts in a domain are fluent in the use of different resources to solve domain-specific problems, whereas students stumble when using formulas, devices, or measurement methods. In most cases, the students are not aware if and why the resources are appropriate to solve the given problem. To acquire expertise in a domain, it is necessary to become fluent in the domain-specific resources. This means being aware of various resources and how they can be used reflectively in problem solving. In this study, we investigated how students’ awareness of various domain-specific resources developed within minimally guided, authentic research projects in an undergraduate materials chemistry course that focuses on crystal structures and its characterisation. By monitoring participants’ problem solving through audio-diaries and in-depth qualitative interview data, we examined to what extent their awareness of domain-specific resources developed. We further characterised productive moments of failure and uncertainty during students’ problem-solving processes that influenced their awareness. Thick descriptions of these moments facilitate understanding of how students’ awareness develops in a minimally guided, authentic learning environment and how students can become fluent in a given domain.

When and how stereotype threat influences older adults' arithmetic performance: Insight from a strategy approach.

In 3 experiments, we investigated how age-related differences in cognitive performance are exacerbated by age-based stereotype threat. We adopted a strategy approach and investigated a domain, namely arithmetic, where age-related differences in participants' performance are either nonexistent or very small and where effects of age-based stereotype threat have never been investigated. In 2 types of tasks (problem verification in Experiment 1 and computational estimation in Experiments 2 and 3), we found that age-based stereotype threat led older adults to obtain poorer performance, to adopt less systematically and less often the better strategy on each problem, to repeat the same strategy across trials even when it was inappropriate, and to execute available strategies more poorly. We also found that poorer strategy use mediated threat effects and that individual differences in processing resources moderated individuals' sensitivity to effects of stereotype threat. Our results establish that age-based stereotype threat effects occur in a wide variety of cognitive domains and tasks and are independent of pre-experimental differences in young and older adults' performance. They deepen our knowledge of the mechanisms underlying age-based, and other stereotype threat effects. They also document how domain-general and domain-specific processing resources moderate individual differences in age-based stereotype threat effects. Our findings have important implications to improve our understanding of how and when age-based (and other) stereotype threat effects occur, and, more generally, how psychosocial factors modulate age-related changes in human cognition. (PsycINFO Database Record (c) 2020 APA, all rights reserved).

Transfer learning for Turkish named entity recognition on noisy text

AbstractIn this article, we investigate using deep neural networks with different word representation techniques for named entity recognition (NER) on Turkish noisy text. We argue that valuable latent features for NER can, in fact, be learned without using any hand-crafted features and/or domain-specific resources such as gazetteers and lexicons. In this regard, we utilize character-level, character n-gram-level, morpheme-level, and orthographic character-level word representations. Since noisy data with NER annotation are scarce for Turkish, we introduce a transfer learning model in order to learn infrequent entity types as an extension to the Bi-LSTM-CRF architecture by incorporating an additional conditional random field (CRF) layer that is trained on a larger (but formal) text and a noisy text simultaneously. This allows us to learn from both formal and informal/noisy text, thus improving the performance of our model further for rarely seen entity types. We experimented on Turkish as a morphologically rich language and English as a relatively morphologically poor language. We obtained an entity-level F1 score of 67.39% on Turkish noisy data and 45.30% on English noisy data, which outperforms the current state-of-art models on noisy text. The English scores are lower compared to Turkish scores because of the intense sparsity in the data introduced by the user writing styles. The results prove that using subword information significantly contributes to learning latent features for morphologically rich languages.

Head Concepts Selection for Verbose Medical Queries Expansion

Semantic concepts and relations encoded in domain-specific ontologies and other medical semantic resources play a crucial role in deciphering terms in medical queries and documents. The exploitation of these resources for tackling the semantic gap issue has been widely studied in the literature. However, there are challenges that hinder their widespread use in real-world applications. Among these challenges is the insufficient knowledge individually encoded in existing medical ontologies, which is magnified when users express their information needs using long-winded natural language queries. In this context, many of the users query terms are either unrecognized by the used ontologies, or cause retrieving false positives that degrade the quality of current medical information search approaches. In this article, we explore the combination of multiple extrinsic semantic resources in the development of a full-fledged medical information search framework to: i) highlight and expand head medical concepts in verbose medical queries (i.e. concepts among query terms that significantly contribute to the informativeness and intent of a given query), ii) build semantically enhanced inverted index documents, iii) contribute to a heuristical weighting technique in the query document matching process. To demonstrate the effectiveness of the proposed approach, we conducted several experiments over the CLEF eHealth 2014 dataset. Findings indicate that the proposed method combining several extrinsic semantic resources proved to be more effective than related approaches in terms of precision measure.

Modeling Islamist Extremist Communications on Social Media using Contextual Dimensions

Terror attacks have been linked in part to online extremist content. Online conversations are cloaked in religious ambiguity, with deceptive intentions, often twisted from mainstream meaning to serve a malevolent ideology. Although tens of thousands of Islamist extremism supporters consume such content, they are a small fraction relative to peaceful Muslims. The efforts to contain the ever-evolving extremism on social media platforms have remained inadequate and mostly ineffective. Divergent extremist and mainstream contexts challenge machine interpretation, with a particular threat to the precision of classification algorithms. Radicalization is a subtle long-running persuasive process that occurs over time. Our context-aware computational approach to the analysis of extremist content on Twitter breaks down this persuasion process into building blocks that acknowledge inherent ambiguity and sparsity that likely challenge both manual and automated classification. Based on prior empirical and qualitative research in social sciences, particularly political science, we model this process using a combination of three contextual dimensions -- religion, ideology, and hate -- each elucidating a degree of radicalization and highlighting independent features to render them computationally accessible. We utilize domain-specific knowledge resources for each of these contextual dimensions such as Qur'an for religion, the books of extremist ideologues and preachers for political ideology and a social media hate speech corpus for hate. The significant sensitivity of the Islamist extremist ideology and its local and global security implications require reliable algorithms for modelling such communications on Twitter. Our study makes three contributions to reliable analysis: (i) Development of a computational approach rooted in the contextual dimensions of religion, ideology, and hate, which reflects strategies employed by online Islamist extremist groups, (ii) An in-depth analysis of relevant tweet datasets with respect to these dimensions to exclude likely mislabeled users, and (iii) A framework for understanding online radicalization as a process to assist counter-programming. Given the potentially significant social impact, we evaluate the performance of our algorithms to minimize mislabeling, where our context-aware approach outperforms a competitive baseline by 10.2% in precision, thereby enhancing the potential of such tools for use in human review.

BioWordVec,\xa0improving biomedical word embeddings with subword information and MeSH

Distributed word representations have become an essential foundation for biomedical natural language processing (BioNLP), text mining and information retrieval. Word embeddings are traditionally computed at the word level from a large corpus of unlabeled text, ignoring the information present in the internal structure of words or any information available in domain specific structured resources such as ontologies. However, such information holds potentials for greatly improving the quality of the word representation, as suggested in some recent studies in the general domain. Here we present BioWordVec: an open set of biomedical word vectors/embeddings that combines subword information from unlabeled biomedical text with a widely-used biomedical controlled vocabulary called Medical Subject Headings (MeSH). We assess both the validity and utility of our generated word embeddings over multiple NLP tasks in the biomedical domain. Our benchmarking results demonstrate that our word embeddings can result in significantly improved performance over the previous state of the art in those challenging tasks.

Natural Language Processing in OTF Computing: Challenges and the Need for Interactive Approaches

The vision of On-the-Fly (OTF) Computing is to compose and provide software services ad hoc, based on requirement descriptions in natural language. Since non-technical users write their software requirements themselves and in unrestricted natural language, deficits occur such as inaccuracy and incompleteness. These deficits are usually met by natural language processing methods, which have to face special challenges in OTF Computing because maximum automation is the goal. In this paper, we present current automatic approaches for solving inaccuracies and incompletenesses in natural language requirement descriptions and elaborate open challenges. In particular, we will discuss the necessity of domain-specific resources and show why, despite far-reaching automation, an intelligent and guided integration of end users into the compensation process is required. In this context, we present our idea of a chat bot that integrates users into the compensation process depending on the given circumstances.