Natural Language Processing Techniques Research Articles

Computational linguistics and natural language processing techniques for semantic field extraction in Arabic online news

The research aimed to extract semantic fields from Arabic online news and advance Natural Language Processing (NLP) applications in understanding and managing news information effectively. It provides a comprehensive approach to processing and analyzing large volumes of Arabic news data by integrating semantic field analysis, NLP, and computational linguistics. Using quantitative methods, Arabic news articles were collected and processed with Python, a popular programming language in data analysis, and applied various NLP techniques and machine learning models to accurately extract semantic fields. The primary objective was to evaluate the effectiveness of different classification models in categorizing Arabic news and to identify the most suitable model for semantic field extraction. The research evaluated five classification models: Naive Bayes, Support Vector Machine (SVM), Logistic Regression, Random Forest, and Gradient Boosting. Among these, SVM achieves the highest overall accuracy of 90%. Specifically, SVM demonstrated exceptional performance in categorizing sports-related news, with a 99% probability and an F1-Score of 98%. However, it faced challenges in categorizing health and science news, achieving a lower F1-Score of 79%. Overall, the study demonstrated the effectiveness of computational methods, particularly SVM, in classifying Arabic news and extracting semantic fields, thereby advancing NLP and computational linguistics. The findings highlighted the potential of SVM for accurate news analysis and the need for further enhancement of NLP techniques to address multilingual and domain-specific challenges.

Mitigating Complex Cyber Threats: An Integrated Multimodal Deep Learning Framework for Enhanced Security

Abstract: The rapidly evolving landscape of cyber threats poses significant challenges to traditional cybersecurity methods, particularly in detecting and mitigating complex attacks that combine multiple data modalities. This article introduces a novel approach utilizing Multimodal Deep Learning (MDL) to enhance cybersecurity detection and prevention capabilities. By fusing Natural Language Processing (NLP) and computer vision techniques, our proposed MDL framework demonstrates superior performance in analyzing both textual and visual data associated with potential threats. Through a series of experiments and case studies, we show that this integrated approach significantly improves detection accuracy, reduces false positives, and exhibits remarkable adaptability to emerging attack vectors. Our results indicate a 37% increase in threat detection efficiency and a 42% reduction in false positives compared to traditional unimodal methods. Furthermore, we explore the potential applications of this technology in email security, social media monitoring, and advanced malware detection. This article not only contributes to the growing body of knowledge in AI-driven cybersecurity but also provides a roadmap for future developments, including the integration of audio analysis and the application of Explainable AI (XAI) to enhance trust and transparency in MDL-based security systems.

Artificial Intelligence-Enabled Intelligent Assistant for Personalized and Adaptive Learning in Higher Education

This paper presents a novel framework, artificial intelligence-enabled intelligent assistant (AIIA), for personalized and adaptive learning in higher education. The AIIA system leverages advanced AI and natural language processing (NLP) techniques to create an interactive and engaging learning platform. This platform is engineered to reduce cognitive load on learners by providing easy access to information, facilitating knowledge assessment, and delivering personalized learning support tailored to individual needs and learning styles. The AIIA’s capabilities include understanding and responding to student inquiries, generating quizzes and flashcards, and offering personalized learning pathways. The research findings have the potential to significantly impact the design, implementation, and evaluation of AI-enabled virtual teaching assistants (VTAs) in higher education, informing the development of innovative educational tools that can enhance student learning outcomes, engagement, and satisfaction. The paper presents the methodology, system architecture, intelligent services, and integration with learning management systems (LMSs) while discussing the challenges, limitations, and future directions for the development of AI-enabled intelligent assistants in education.

Explanatory argument extraction of correct answers in resident medical exams

Developing technology to assist medical experts in their everyday decision-making is currently a hot topic in the field of Artificial Intelligence (AI). This is specially true within the framework of Evidence-Based Medicine (EBM), where the aim is to facilitate the extraction of relevant information using natural language as a tool for mediating in human–AI interaction. In this context, AI techniques can be beneficial in finding arguments for past decisions in evolution notes or patient journeys, especially when different doctors are involved in a patient’s care. In those documents the decision-making process towards treating the patient is reported. Thus, applying Natural Language Processing (NLP) techniques has the potential to assist doctors in extracting arguments for a more comprehensive understanding of the decisions made. This work focuses on the explanatory argument identification step by setting up the task in a Question Answering (QA) scenario in which clinicians ask questions to the AI model to assist them in identifying those arguments. In order to explore the capabilities of current AI-based language models, we present a new dataset which, unlike previous work: (i) includes not only explanatory arguments for the correct hypothesis, but also arguments to reason on the incorrectness of other hypotheses; (ii) the explanations are written originally in Spanish by doctors to reason over cases from the Spanish Residency Medical Exams. Furthermore, this new benchmark allows us to set up a novel extractive task by identifying the explanation written by medical doctors that supports the correct answer within an argumentative text. An additional benefit of our approach lies in its ability to evaluate the extractive performance of language models using automatic metrics, which in the Antidote CasiMedicos dataset corresponds to a 74.47 F1 score. Comprehensive experimentation shows that our novel dataset and approach is an effective technique to help practitioners in identifying relevant evidence-based explanations for medical questions.

Investigating the Linguistic Fingerprint of GPT-4o in Arabic-to-English Translation Using Stylometry

This study explores the linguistic and stylistic characteristics of machine-generated texts, focusing on the output of GPT-4o. Using various natural language processing (NLP) techniques, including word frequency and stopword count analysis, readability and sentence structure metrics, lexical diversity measures, syntactic frequency analysis, and named entity recognition (NER), the research aims to uncover the stylometric fingerprints present in machine-generated content. The results reveal that GPT-4ogenerated texts exhibit moderate lexical diversity and syntactic complexity, with certain chapters reflecting higher readability and more varied sentence structures, while others lean toward simpler linguistic patterns. The findings also highlight thematic variation across chapters, as observed in the distribution of named entities, which contributes to understanding the model’s handling of different contextual content. The research suggests that while GPT-4o maintains a consistent style in its generated text, there are distinguishable characteristics that may serve as indicators of machine authorship. This provides valuable insights for stylometric analysis, authorship attribution, and the identification of machine-generated texts in various contexts. Future research could extend this work by exploring deeper stylometric features, conducting cross-model comparisons, and developing advanced authorship detection algorithms tailored for AI-generated content. Moreover, the ethical implications of stylometric analysis in the context of AI-generated texts warrant further investigation, particularly as machine-generated content becomes increasingly prevalent across different domains.

An Evaluation of the Maternal Patient Experience through Natural Language Processing Techniques: The Case of Twitter Data in the United States during COVID-19

The healthcare sector constantly investigates ways to improve patient outcomes and provide more patient-centered care. Delivering quality medical care involves ensuring that patients have a positive experience. Most healthcare organizations use patient survey feedback to measure patients’ experiences. However, the power of social media can be harnessed using artificial intelligence and machine learning techniques to provide researchers with valuable insights into understanding patient experience and care. Our primary research objective is to develop a social media analytics model to evaluate the maternal patient experience during the COVID-19 pandemic. We used the “COVID-19 Tweets” Dataset, which has over 28 million tweets, and extracted tweets from the US with words relevant to maternal patients. The maternal patient cohort was selected because the United States has the highest percentage of maternal mortality and morbidity rate among the developed countries in the world. We evaluated patient experience using natural language processing (NLP) techniques such as word clouds, word clustering, frequency analysis, and network analysis of words that relate to “pains” and “gains” regarding the maternal patient experience, which are expressed through social media. The pandemic showcased the worries of mothers and providers on the risks of COVID-19. However, many people also shared how they survived the pandemic. Both providers and maternal patients had concerns regarding the pregnancy risks due to COVID-19. This model will help process improvement experts without domain expertise to understand the various domain challenges efficiently. Such insights can help decision-makers improve the patient care system.

Generative AI as a New Paradigm for Personalized Tutoring in Modern Education

Conventional education often falls short when delivering personalized and immediate support to all students, resulting in disparities in understanding and unequal learning opportunities. Generative Artificial Intelligence (GenAI) provides a scalable and cost-effective solution for on-demand tutoring, offering personalized support 24/7. This paper delves into the potential application of GenAI as an on-demand tutoring system, addressing the critical need for immediate and personalized educational support. Implementing GenAI to establish an on-demand tutoring system that delivers personalized, real-time assistance is vital in meeting the demands of today’s academic landscape. Critical components of this approach encompass advanced natural language processing techniques to comprehend and respond to student inquiries, machine learning algorithms to tailor support to individual learning styles, and a scalable cloud-based infrastructure to ensure round-the-clock availability. This approach’s expected scientific surplus value lies in its capacity to significantly enhance educational outcomes by providing scalable, personalized learning experiences. This paper elucidates a pathway for future research and development in this area, highlighting the potential of GenAI to revolutionize education and improve learning outcomes for all students.

Exploration of Deep-Learning-Based Approaches for False Fact Identification in Social Judicial Systems

With the many applications of artificial intelligence (AI) in social judicial systems, false fact identification becomes a challenging issue when the system is expected to be more autonomous and intelligent in assisting a judicial review. In particular, private lending disputes often involve false facts that are intentionally concealed and manipulated due to unique and dynamic relationships and their nonconfrontational nature in the judicial system. In this article, we investigate deep learning techniques to identify false facts in loan cases for the purpose of reducing the judicial workload. Specifically, we adapt deep-learning-based natural language processing techniques to a dataset over 100 real-world judicial rules spanning four courts of different levels in China. The BERT (bidirectional encoder representations from transformers)-based classifier and T5 text generation models were trained to classify false litigation claims semantically. The experimental results demonstrate that T5 has a robust learning capability with a small number of legal text samples, outperforms BERT in identifying falsified facts, and provides explainable decisions to judges. This research shows that deep-learning-based false fact identification approaches provide promising solutions for addressing concealed information and manipulation in private lending lawsuits. This highlights the feasibility of deep learning to strengthen fact-finding and reduce labor costs in the judicial field.

Identifying ChatGPT-generated texts in EFL students’ writing: Through comparative analysis of linguistic fingerprints

The emergence of generative AI (GenAI) poses new challenges for L2 writing teachers. This study investigates the distinguishability of essays written by Japanese EFL learners from those generated by ChatGPT. Partially replicating Herbold et al. (2023), 140 first-year university students wrote essays and completed a survey on ChatGPT use. Among them, 125 wrote independently, 13 used ChatGPT for proofreading, and two asked ChatGPT to write the entire essay. To create a comparative dataset, 123 additional essays were generated by ChatGPT, imitating the two texts. The resulting 263 essays were then analyzed using the natural language processing (NLP) technique, including automated linguistic analysis and machine learning classification using random forest. The results reveal significant differences between human-written and ChatGPT-generated essays across all linguistic features, with the latter being easily identifiable. This study emphasizes the need for clear guidelines on the ethical use of AI in L2 writing, highlighting the potential risk of inappropriate AI use and the importance of fostering a mutual understanding of AI use with learners regarding responsible AI integration in academic work.

AI-Driven Drug Discovery: Accelerating the Development of Novel Therapeutics in Biopharmaceuticals

Artificial Intelligence (AI) has emerged as a transformative force in drug discovery, revolutionizing the biopharmaceutical industry's approach to developing novel therapeutics. This paper provides a comprehensive overview of AI-driven drug discovery, focusing on its applications in accelerating the development of innovative treatments. We examine the fundamental AI technologies employed in drug discovery, including machine learning algorithms, deep learning architectures, and natural language processing techniques. The paper analyzes the integration of AI across various stages of the drug discovery pipeline, from target identification to clinical trial design, highlighting significant improvements in efficiency and accuracy. We explore the impact of big data on AI-driven drug discovery, discussing the challenges and opportunities presented by multi-omics data integration, electronic health records mining, and the need for data standardization. The study also addresses ethical considerations and regulatory challenges associated with AI implementation in drug development. Finally, we present emerging trends and prospects for AI in biopharmaceuticals, emphasizing the importance of collaborative ecosystems and the potential for AI to revolutionize personalized medicine. This review synthesizes current research and industry practices, providing insights into the transformative potential of AI in drug discovery and the challenges that lie ahead in realizing its full potential.

Decision-making model for selecting products through online product reviews utilizing natural language processing techniques

Online product reviews are evaluations of products shared by customers on various online platforms, such as electronic commerce websites, social media, or dedicated review sites. These reviews offer valuable insights and opinions about products, potentially influencing the purchasing decisions of other prospective buyers. However, due to the vast volume of content provided by internet product evaluations, it can be challenging for new buyers to conduct a comprehensive qualitative assessment of competitive products. Utilizing natural language processing techniques, analyzing online product reviews posted on social media platforms can aid customers in making informed purchasing decisions. Hence, this article employs a 2-tuple linguistic q-rung orthopair fuzzy set for product selection based on online product reviews. In this study, we introduce a novel hybrid approach combining the Criteria Importance through Inter-Criteria Correlation (CRITIC) method with the Weighted Aggregated Sum Product Assessment (WASPAS) method. This hybrid method assists potential customers in evaluating alternative products by considering consumer opinions regarding product performance in the 2-tuple linguistic q-rung orthopair fuzzy environment. Moreover, for multi-attribute group decision-making problems, we develop a weighted average aggregation operator based on the 2-tuple linguistic q-rung orthopair fuzzy set. Finally, we apply the proposed approach to a real evaluation decision, validating its validity and practicality through parameter and comparison analyses.

Governance in multi-system transitions: A new methodological approach for actor involvement in policy making processes

Multi-system interactions associated with the decarbonisation of energy and mobility systems represent a complex phenomenon in the acceleration phase of net-zero transitions. In this paper, we present a novel methodological approach to examine actor involvement in the governance of multi-system transitions, with a focus on the UK's net-zero energy-mobility transitions from 2008 to 2021. Utilising Named Entity Recognition (NER), a natural language processing technique, we systematically map actors and their interactions within policy consultations and how these have changed over time. Our analysis differentiates between single-system and multi-system policy making processes; identifies weak and strong links among actors as two types of multi-system interactions; categorises actors into business, policy, academia, and society groups; and examines the evolution of engagement across multiple governance levels. Our findings indicate an increasing trend of multi-system interactions, suggesting the UK's progression towards the acceleration phase of net-zero transitions. Our analysis further reveals the predominance of policy actors, particularly from the national level, in governing such multi-system transitions processes, followed by business actors. Despite some limitations, our approach offers a scalable method for analysing large volumes of text, providing valuable insights into the governance dynamics of multi-system transitions. We conclude with implications for policy making and offer suggestions for future research, emphasising the importance of understanding actor involvement and political contestations around net-zero trajectories for ensuring the achievement of sustainability goals.

Automated System to Capture Patient Symptoms From Multitype Japanese Clinical Texts: Retrospective Study.

Natural language processing (NLP) techniques can be used to analyze large amounts of electronic health record texts, which encompasses various types of patient information such as quality of life, effectiveness of treatments, and adverse drug event (ADE) signals. As different aspects of a patient's status are stored in different types of documents, we propose an NLP system capable of processing 6 types of documents: physician progress notes, discharge summaries, radiology reports, radioisotope reports, nursing records, and pharmacist progress notes. This study aimed to investigate the system's performance in detecting ADEs by evaluating the results from multitype texts. The main objective is to detect adverse events accurately using an NLP system. We used data written in Japanese from 2289 patients with breast cancer, including medication data, physician progress notes, discharge summaries, radiology reports, radioisotope reports, nursing records, and pharmacist progress notes. Our system performs 3 processes: named entity recognition, normalization of symptoms, and aggregation of multiple types of documents from multiple patients. Among all patients with breast cancer, 103 and 112 with peripheral neuropathy (PN) received paclitaxel or docetaxel, respectively. We evaluate the utility of using multiple types of documents by correlation coefficient and regression analysis to compare their performance with each single type of document. All evaluations of detection rates with our system are performed 30 days after drug administration. Our system underestimates by 13.3 percentage points (74.0%-60.7%), as the incidence of paclitaxel-induced PN was 60.7%, compared with 74.0% in the previous research based on manual extraction. The Pearson correlation coefficient between the manual extraction and system results was 0.87 Although the pharmacist progress notes had the highest detection rate among each type of document, the rate did not match the performance using all documents. The estimated median duration of PN with paclitaxel was 92 days, whereas the previously reported median duration of PN with paclitaxel was 727 days. The number of events detected in each document was highest in the physician's progress notes, followed by the pharmacist's and nursing records. Considering the inherent cost that requires constant monitoring of the patient's condition, such as the treatment of PN, our system has a significant advantage in that it can immediately estimate the treatment duration without fine-tuning a new NLP model. Leveraging multitype documents is better than using single-type documents to improve detection performance. Although the onset time estimation was relatively accurate, the duration might have been influenced by the length of the data follow-up period. The results suggest that our method using various types of data can detect more ADEs from clinical documents.

Corporate environmental, social, and governance information disclosure and audit governance in the context of green development

The purpose of this paper is to investigate the relationship between Environmental, Social, and Governance (ESG) disclosure and audit governance. The objective is to assess the impact of auditor team independence, auditor professional quality, and transparency of audit procedures on the quality of information disclosure, and to elucidate the role of these factors in advancing sustainable development. Research methodologies include the optimization of information disclosure text analysis using natural language processing (NLP) and text generation techniques, complemented by empirical analysis. Key findings indicate that the maximum error between NLP and manual annotation in the proportion of positive emotions in information disclosure is 0.03, and 0.06 in audit governance. The Rouge index score surpasses the 0.5 threshold, while the Perplexity value varies across three experiments, though remaining generally low. Deep learning model results reveal that auditor team independence, professional quality, and procedural transparency significantly enhance the quality of information disclosure. These findings offer crucial insights for enterprises, regulators, and investors, affirming the rationality and stability of the model, and providing valuable input for future green development strategies and decision-making processes.

PageRank Algorithm-Based Recommendation System for Construction Safety Guidelines

The construction industry faces significant challenges with frequent accidents, largely due to the inefficient use of safety guidelines. These guidelines, which are often text and figure heavy, demand substantial human effort to identify the most relevant items for specific tasks and conditions. Additionally, the guidelines contain both central and peripheral elements, and central items are critical yet difficult to identify without extensive domain knowledge. This study proposes a novel recommendation framework to enhance the usability of these safety guidelines. By leveraging natural language processing (NLP) and knowledge graph (KG) modeling techniques, unstructured safety texts are transformed into a structured, interconnected KG. The PageRank and Louvain Clustering algorithm is then employed to rank guidelines by their relevance and importance. A case study on “High-rise Building Construction (General) Safety and Health Guidelines”, using ‘scaffolding’ as the keyword, demonstrates the framework’s effectiveness in improving retrieval efficiency and practical application. The analysis highlighted key clusters such as ‘fall’, ‘drop’, and ‘scaffolding’, with critical safety measures identified through their interconnections. This research not only overcomes the fragmentation of safety management documents but also contributes to advancing hazard analysis and risk prevention practices in construction management.

Vulnerability detection based on transformer and high‐quality number embedding

SummarySoftware vulnerability detection is an important problem in software security. In recent years, deep learning offers a novel approach for source code vulnerability detection. Due to the similarities between programming languages and natural languages, many natural language processing techniques have been applied to vulnerability detection tasks. However, specific problems within vulnerability detection tasks, such as buffer overflow, involve numerical reasoning. For these problems, the model needs to not only consider long dependencies and multiple relationships between statements of code but also capture the magnitude property of numerical literals in the program through high‐quality number embeddings. Therefore, we propose VDTransformer, a Transformer‐based method that improves source code embedding by integrating word and number embeddings. Furthermore, we employ Transformer encoders to construct a hierarchical neural network that extracts semantic features from the code and enables line‐level vulnerability detection. To evaluate the effectiveness of the method, we construct a dataset named OverflowGen based on templates for buffer overflow. Experimental comparisons on OverflowGen with a well‐known static vulnerability detector and two state‐of‐the‐art deep learning‐based methods confirm the effectiveness of VDTransformer and the importance of high‐quality number embeddings in vulnerability detection tasks involving numerical features.

Bringing Artificial Intelligence to Research Analytics: Research Highlighter-MatchMaker

It is more significant to identify accurate grant opportunities given the increasing challenges in securing grants these days. Combining Artificial Intelligence (AI) and Natural Language Processing (NLP) techniques, we have developed Research Highlighter-MatchMaker to facilitate the grant application process. The tool aims to provide an overview of research strength and to recommend matched funding opportunities as well as potential collaborators for researchers based on their research track records. It has the potential to expand to provide matching for on-campus researchers and industrial partners. It will also include a chatbot to facilitate the research administration infrastructure and process on campus.

Uncovering the Complexity of Perinatal Polysubstance Use Disclosure Patterns on X: Mixed Methods Study.

According to the Morbidity and Mortality Weekly Report, polysubstance use among pregnant women is prevalent, with 38.2% of those who consume alcohol also engaging in the use of one or more additional substances. However, the underlying mechanisms, contexts, and experiences of polysubstance use are unclear. Organic information is abundant on social media such as X (formerly Twitter). Traditional quantitative and qualitative methods, as well as natural language processing techniques, can be jointly used to derive insights into public opinions, sentiments, and clinical and public health policy implications. Based on perinatal polysubstance use (PPU) data that we extracted on X from May 1, 2019, to October 31, 2021, we proposed two primary research questions: (1) What is the overall trend and sentiment of PPU discussions on X? (2) Are there any distinct patterns in the discussion trends of PPU-related tweets? If so, what are the implications for perinatal care and associated public health policies? We used X's application programming interface to extract >6 million raw tweets worldwide containing ≥2 prenatal health- and substance-related keywords provided by our clinical team. After removing all non-English-language tweets, non-US tweets, and US tweets without disclosed geolocations, we obtained 4848 PPU-related US tweets. We then evaluated them using a mixed methods approach. The quantitative analysis applied frequency, trend analysis, and several natural language processing techniques such as sentiment analysis to derive statistics to preview the corpus. To further understand semantics and clinical insights among these tweets, we conducted an in-depth thematic content analysis with a random sample of 500 PPU-related tweets with a satisfying κ score of 0.7748 for intercoder reliability. Our quantitative analysis indicates the overall trends, bigram and trigram patterns, and negative sentiments were more dominant in PPU tweets (2490/4848, 51.36%) than in the non-PPU sample (1323/4848, 27.29%). Paired polysubstance use (4134/4848, 85.27%) was the most common, with the combination alcohol and drugs identified as the most mentioned. From the qualitative analysis, we identified 3 main themes: nonsubstance, single substance, and polysubstance, and 4 subthemes to contextualize the rationale of underlying PPU behaviors: lifestyle, perceptions of others' drug use, legal implications, and public health. This study identified underexplored, emerging, and important topics related to perinatal PPU, with significant stigmas and legal ramifications discussed on X. Overall, public sentiments on PPU were mixed, encompassing negative (2490/4848, 51.36%), positive (1884/4848, 38.86%), and neutral (474/4848, 9.78%) sentiments. The leading substances in PPU were alcohol and drugs, and the normalization of PPU discussed on X is becoming more prevalent. Thus, this study provides valuable insights to further understand the complexity of PPU and its implications for public health practitioners and policy makers to provide proper access and support to individuals with PPU.

Using natural language processing approaches to characterize professional experiences of child welfare workers

High turnover rates among child welfare workers, including caseworkers and supervisors, are associated with negative attitudes and perceptions about work (NAPW). Our study aimed to (a) explore how child welfare workers’ demographic characteristics, professional attributes, and their work narratives are related to NAPW and (b) extract themes from these narratives to understand their professional needs and concerns. In 2022, a statewide online survey was administered to active child welfare caseworkers and supervisors in a large Midwestern state. This survey, combining standardized scales and an open-ended question about work experiences, yielded numeric (N = 551 survey respondents) and text data (N = 2,193 words). We analyzed the numeric data using general linear modeling and the text data using natural language processing (NLP) techniques. Results showed that workers associated with higher NAPW were female, younger, from racial and ethnic minority backgrounds, less tenured, and more likely to be employed in public agencies and to work longer hours. They are predominantly caseworkers rather than supervisors. Worker role and agency affiliation had the largest effect sizes. NLP analysis showed that workers with more NAPW frequently used third-person pronouns and emotionally negative language, suggesting a disconnection from colleagues and job dissatisfaction. Caseworkers often used first-person singular pronouns, indicating feelings of isolation. Two key themes emerged: a) the need for better agency support during home visits and b) concerns over traumatic experiences and ineffective trauma-informed practices. The findings suggest that future interventions should prioritize the well-being of caseworkers in public agencies and consider language patterns in workers’ narratives to understand their vulnerability to NAPW.

Development of a Natural Language Processing (NLP) model to automatically extract clinical data from electronic health records: results from an Italian comprehensive stroke center

IntroductionData collection often relies on time-consuming manual inputs, with a vast amount of information embedded in unstructured texts such as patients’ medical records and clinical notes. Our study aims to develop a pipeline that combines active learning (AL) and NLP techniques to enhance data extraction in an acute ischemic stroke cohort. Materials and methodsConsecutive acute ischemic stroke patients who received reperfusion therapies at IRCCS Humanitas Research Hospital were included. The Italian NLP Bidirectional Encoder Representations from Transformers (BERT) model was trained with AL to automatically extract clinical variables from electronic health text. Simulated active learning performances were evaluated on a set of labels representing patients’ comorbidities, comparing Bayesian Uncertainty Sampling by Disagreement (BALD) and random text selection. Prognostic models predicting patients’ functional outcomes using Gradient Boosting were trained on manually labelled and semi-automatically extracted data and their performance was compared. ResultsThe active learning process initially showed null performance until around 20% of texts were labelled, possibly due to root layers freezing in the BERT model, yet overall, active learning improves model learning efficiency across most comorbidities. Prognostic modelling showed no significant difference in performance between models trained on manually labelled versus semi-automatically extracted data, indicating effective prediction capabilities in both settings. ConclusionsWe developed an efficient language model to automate the extraction of clinical data from Italian unstructured health texts in a cohort of ischemic stroke patients. In a preliminary analysis, we demonstrated its potential applicability for enhancing prediction model accuracy.