Unsupervised Natural Language Processing Research Articles

Navigating the complex terrain of motivated behavior: a bibliometric and neuroscientific perspective.

Over several decades, motivated behavior has emerged as a crucial study area within neuroscience. Understanding the neural substrates and mechanisms driving behaviors related to reward, addiction, and other motivation forms is pivotal for novel therapeutic interventions. This review provides a bibliometric analysis of the literature, highlighting the main trends, influential authors, and the potential future direction of the field. Utilizing a dataset comprised by 3,150 publications from the Web of Science and Scopus databases ("motivated behavior as query), we delve into key metrics like publication trends, keyword prevalence, author collaborations, citation impacts, and employed an unsupervised natural language processing technique - Latent Dirichlet Allocation - for topic modeling. From early investigations focusing on basic neural mechanism and behaviors in animal models to more recent studies exploring the complex interplay of neurobiological, psychological, and social factors in humans, the field had undergone a remarkable transformation. The last century has seen a proliferation of research dedicated to uncovering the intricacies of motivation, significantly enriching our understanding of its myriad implications for human behavior and mental health. This bibliometric analysis aims to offer comprehensive insights into this dynamic research area, highlighting the field's key contributions and potential future directions, thereby serving as a valuable resource for researchers, and hopefully give a more thorough understanding of the research area.

A platform for connecting social media data to domain-specific topics using large language models: an application to student mental health.

To design a novel artificial intelligence-based software platform that allows users to analyze text data by identifying various coherent topics and parts of the data related to a specific research theme-of-interest (TOI). Our platform uses state-of-the-art unsupervised natural language processing methods, building on top of a large language model, to analyze social media text data. At the center of the platform's functionality is BERTopic, which clusters social media posts, forming collections of words representing distinct topics. A key feature of our platform is its ability to identify whole sentences corresponding to topic words, vastly improving the platform's ability to perform downstream similarity operations with respect to a user-defined TOI. Two case studies on mental health among university students are performed to demonstrate the utility of the platform, focusing on signals within social media (Reddit) data related to depression and their connection to various emergent themes within the data. Our platform provides researchers with a readily available and inexpensive tool to parse large quantities of unstructured, noisy data into coherent themes, as well as identifying portions of the data related to the research TOI. While the development process for the platform was focused on mental health themes, we believe it to be generalizable to other domains of research as well.

IS IT (MICRO)CHEATING? HOW SOCIAL MEDIA CONFOUND ASSUMPTIONS IN ROMANTIC RELATIONSHIPS

Social media have changed the ways we communicate, meet others, and form intimate relationships. However, technology can also mediate intimate partner surveillance and abuse (Muise, 2009; Tokunaga, 2010). One of the most explicit ways to understand these shifts is through the transgressing of relationship boundaries, defined and enforced by settler-colonial notions of compulsory monogamy (TallBear, 2020). Anxieties around cheating have evolved along with our technologies, as evidenced by ambiguous new terms like “microcheating” and “emotional cheating” (Lusinski, 2018). In this in-progress, mixed-methods study, we examine new definitions of cheating through analyzing discussions about potential transgressions on Reddit. Specifically, we investigate 1) which behaviors cause uncertainty in emerging forms of social media-enabled infidelity and 2) the degree to which relationship discourse online naturalizes the extension of compulsory monogamy into online space. For our pilot analysis, we used computational techniques to elicit common subjects within subreddit posts to then analyze qualitatively. We began with Latent Dirichlet Allocation (LDA), an unsupervised natural language processing tool, to organize Reddit posts and comments by topic (Blei, 2012). Then we qualitatively analyzed Reddit discourse by using critical discourse analysis. Our pilot analysis suggested a belief that proof of (in)fidelity can be found on a partner’s smartphone, such as by reading texts. This orientation toward evidence then justifies surveillance and hacking of a partner’s phone and computer presence, construing the invasion of privacy as the right to truth. This preliminary finding suggests that discourse around transgressive behaviors on social media likely reiterates compulsory monogamy and settler sexuality.

Natural language processing for knowledge discovery and information extraction from energetics corpora

AbstractWe present a demonstration of the utility of Natural Language Processing (NLP) for aiding research into energetic materials and associated systems. The NLP method enables machine understanding of textual data, offering an automated route to knowledge discovery and information extraction from energetics text. We apply three established unsupervised NLP models: Latent Dirichlet Allocation, Word2Vec, and the Transformer to a large curated dataset of energetics‐related scientific articles. We demonstrate that each NLP algorithm is capable of identifying energetic topics and concepts, generating a language model which aligns with Subject Matter Expert knowledge. Furthermore, we present a document classification pipeline for energetics text. Our classification pipeline achieves 59–76 % accuracy depending on the NLP model used, with the highest performing Transformer model rivaling inter‐annotator agreement metrics. The NLP approaches studied in this work can identify concepts germane to energetics and therefore hold promise as a tool for accelerating energetics research efforts and energetics material development.

Identifying Common Topics in Patient Portal Messages with Unsupervised Natural Language Processing.

Patient portal messaging is an increasingly important form of communication between patients and medical providers. This has become particularly relevant in oncology, where patients undergo intense longitudinal treatments that require frequent communication regarding symptoms, appointments, and diagnostic results. The rise in the volume of these messages has significantly increased the workload of medical providers and consequent physician burn-out. Natural language processing (NLP), particularly transformer-based models, may offer an automated approach to characterize the content of patient messages and improve message triage and routing. In this study, we employed a state-of-the-art language model (Bidirectional Encoder Representations from Transformers; BERT) to identify data-derived categories of representative topics from real-world data thereby providing basic information to build an appropriate routing system. Patient-generated portal messages sent to a messaging pool for a single institution radiation oncology department from 2014 to 2023 were extracted. BERTopic, an NLP-based topic modeling technique based on BERT was optimized for topic modeling of patient messages. Uniform Manifold Approximation and Projection (UMAP) was used to reduce dimensionality and visualize topic relationships across messages. The BERTopic-identified topic categories were subsequently labeled manually by one of the physician investigators. Differences of number of messages over time were assessed using t-tests. A total of 47,492 messages were retrieved. The average number of messages per month from a single patient ranged from 1 to 18 (median 1.67, interquartile range 1.0-2.4). The total volume of patient messages showed a ten-fold increase over the study period, with 101 messages per month sent in 2014 and 999 messages per month in 2022 (p<0.001). BERTopic initially identified 35 topics whose relationships and degrees of overlap were visualized by UMAP. Due to physician-identified similarities, these topics were reduced into 13 categories. The most frequent topic category was messages about laboratory tests or imaging studies: 24.3%, followed by messages expressing appreciation: 18.9%, scheduling discussions: 15.6%, symptom-related messages: 11%, and treatment-related messages: 10.7%. Patient portal messages sent to a single institution radiation oncology department have increased dramatically in volume since implementation, corresponding to a broader national trend. NLP successfully identified common subject themes across patient messages, many of which are related to scheduling. This presents potential opportunities to apply NLP to automate message routing in the future.

S1189 Using Unsupervised Natural Language Processing to Automatically Identify Colonic Dysplasia in Pathology Reports From Patients With Chronic Colitis

S1189 Using Unsupervised Natural Language Processing to Automatically Identify Colonic Dysplasia in Pathology Reports From Patients With Chronic Colitis

S1188 Using Unsupervised Natural Language Processing to Automatically Identify Chronicity and Extent of Inflammation on Ileocolonoscopy From Pathology Reports

S1188 Using Unsupervised Natural Language Processing to Automatically Identify Chronicity and Extent of Inflammation on Ileocolonoscopy From Pathology Reports

Enhancing Knowledge Management in Healthcare: An Embedding Fusion Approach to Business Rule Representation

Enterprise Decision Management systems are vital for delivering efficient healthcare services. However, the ever-changing clinical terminology creates complex business rules, making healthcare IT systems difficult to maintain. In this study, we present an embedding fusion technique using unsupervised Natural Language Processing (NLP) to represent business rules as semantic vectors by incorporating multiple text data sources for each rule. We apply this method to a dental insurance administration case study and find that our approach is over 200 times more likely to identify redundant rule pairs compared to random pairs. This case study suggests that an embedding-based technique can significantly improve knowledge management efficiency in healthcare IT systems.

Abstract P2-13-05: Triple Negative Breast Cancer (TNBC) patients are more likely to digitally explore clinical trial options and prior to receiving treatment for advanced disease compared to non-TNBC patients

Abstract Introduction: Triple-negative breast cancer (TNBC), an aggressive form of breast cancer (BC) that is associated with poor prognosis, accounts for 10-15% of all BCs. Chemotherapy remains the standard of care (SOC) for advanced disease, with limited clinical benefit. Oncology clinical trials (CTs) are globally recommended and encouraged as the preferred treatment (tx) option for any cancer patient (pt). ‘TrialJectory’ (TJ) is an artificial intelligence (AI)-based technology that matches pts to oncology CTs. Here, we identified distinct characteristics of TNBC pts who signed up to the TJ platform compared to non-TNBC pts. Methods: Using AI and an unsupervised natural language processing approach, the TJ platform clinically matches pts to CTs. Matching is achieved by pt response to an online dynamic questionnaire (www.trialjectory.com) that collects detailed clinical data including clinico-pathologic characteristics, tx history, general health, and comorbidities. Those are compared to the eligibility criteria of available CTs to yield a high-quality actionable matched-trial list. Results: Between 1/2020 and 12/2021, out of 9796 BC pts that signed up, 2688 were TNBC pts (27%). There was no significant difference in age at sign-up between TNBC and non-TNBC patients (median age of 57 years vs 58 years, respectively). Consistently with Non-hispanic black (NHB) race prevalence in the different molecular subtypes in the general US population, NHB race had higher signup rate in TNBC compared to non-TNBC (9.95% vs 5.76%, respectively). TNBC pts signed up at a later disease stage compared to non-TNBC pts (19% of TNBC reported having a stage 1 disease compared to 27% of non-TNBC pts, p&amp;lt; 0.001).A significantly higher percentage of pts with advanced/metastatic TNBC signed up to the TJ platform before starting tx compared to non-TNBC patients (34% vs 22%, respectively, p&amp;lt; 0.001). Furthermore, there was a significant difference in the willingness to travel any distance within the US for a matched clinical trial between TNBC and non-TNBC pts (39% vs 34%, respectively, p&amp;lt; 0.001). Conclusions: In this study, we found significant differences in the characteristics of TNBC vs non-TNBC pts that have signed up to the TJ platform. There was an up to 2-fold enrichment of TNBC on the TJ platform pts compared to their frequency in the general population. While previous studies do not show a difference in stage distribution between different subtypes, TNBC patients initiated their search for CTs at a higher stage. In addition, advanced TNBC patients started their search earlier in their journey, before starting chemotherapy. This may reflect the lack of effective SOC and possibly, the motivation to avoid the use of chemotherapy. This is also supported by the willingness of TNBC patients to travel farther in order to identify and enroll in a CT compared to non-TNBC pts. Importantly, the motivation of TNBC pts to travel any distance has not been reduced despite the COVID-19 pandemic, reflecting a strong drive of this pt population to enroll in CTs. It also demonstrates that with the right access, diverse patient populations are willing to participate in clinical trials. In sum, TNBC pts are more likely to explore CT options, in the advanced stage setting, earlier in their journey. This study demonstrates the power of TJ platform for clinico-pathologic characterization and diverse pt groups, including their drivers and behavioral choices during their battle with cancer. Citation Format: Michal Safran, Yelena Lapidot, Tzvia Bader, Avital Gaziel. Triple Negative Breast Cancer (TNBC) patients are more likely to digitally explore clinical trial options and prior to receiving treatment for advanced disease compared to non-TNBC patients [abstract]. In: Proceedings of the 2022 San Antonio Breast Cancer Symposium; 2022 Dec 6-10; San Antonio, TX. Philadelphia (PA): AACR; Cancer Res 2023;83(5 Suppl):Abstract nr P2-13-05.

Scalable Feature Engineering from Electronic Free Text Notes to Supplement Confounding Adjustment of Claims-Based Pharmacoepidemiologic Studies.

Natural language processing (NLP) tools turn free-text notes (FTNs) from electronic health records (EHRs) into data features that can supplement confounding adjustment in pharmacoepidemiologic studies. However, current applications are difficult to scale. We used unsupervised NLP to generate high-dimensional feature spaces from FTNs to improve prediction of drug exposure and outcomes compared with claims-based analyses. We linked Medicare claims with EHR data to generate three cohort studies comparing different classes of medications on the risk of various clinical outcomes. We used "bag-of-words" to generate features for the top 20,000 most prevalent terms from FTNs. We compared machine learning (ML) prediction algorithms using different sets of candidate predictors: Set1 (39 researcher-specified variables), Set2 (Set1 + ML-selected claims codes), and Set3 (Set1 + ML-selected NLP-generated features), vs. Set4 (Set1 + 2 + 3). When modeling treatment choice, we observed a consistent pattern across the examples: ML models utilizing Set4 performed best followed by Set2, Set3, then Set1. When modeling the outcome risk, there was little to no improvement beyond models based on Set1. Supplementing claims data with NLP-generated features from free text notes improved prediction of prescribing choices but had little or no improvement on clinical risk prediction. These findings have implications for strategies to improve confounding using EHR data in pharmacoepidemiologic studies.

Automatic Identification of Self-Reported COVID-19 Vaccine Information from Vaccine Adverse Events Reporting System.

The short time frame between the coronavirus disease 2019 (COVID-19) pandemic declaration and the vaccines authorization led to concerns among public regarding the safety and efficacy of the vaccines. The Food and Drug Administration uses the Vaccine Adverse Events Reporting System (VAERS) where general population can report their vaccine side effects in the text box. This information could be utilized to determine self-reported vaccine side effects. To develop a supervised and unsupervised natural language processing (NLP) pipeline to extract self-reported COVID-19 vaccination side effects, location of the side effects, medications, and possibly false/misinformation seeking further investigation in a structured format for analysis and reporting. We utilized the VAERS dataset of COVID-19 vaccine reports from November 2020 to August 2022 of 725,246 individuals. We first developed a gold-standard (GS) dataset of randomly selected 1,500 records. Second, the GS was split into training, testing, and validation sets. The training dataset was used to develop the NLP applications (supervised and unsupervised) and testing and validation datasets were used to test the performances of the NLP application. The NLP application automatically extracted vaccine side effects, body locations of the side effects, medication, and possibly misinformation with moderate to high accuracy (84% sensitivity, 82% specificity, and 83% F-1 measure). We found that 23% people (386,270) faced arm soreness, 31% body swelling (226,208), 23% fatigue/body weakness (168,160), and 22% (159,873) cold/flue-like symptoms. Most of the complications occurred in the body locations such as the arm, back, chest, neck, face, and head. Over-the-counter pain medications such as Tylenol and Ibuprofen and allergy medication like Benadryl were most reported self-reported medications. Death due to COVID-19, changes in the DNA, and infertility were possible false/misinformation reported by people. Some self-reported side effects such as syncope, arthralgia, and blood clotting need further clinical investigations. Our NLP application may help in extracting information from big free-text electronic datasets to help policy makers and other researchers with decision making.

Unsupervised natural language processing in the identification of patients with suspected COVID-19 infection.

Patients with post-COVID-19 syndrome benefit from health promotion programs. Their rapid identification is important for the cost-effective use of these programs. Traditional identification techniques perform poorly especially in pandemics. A descriptive observational study was carried out using 105,008 prior authorizations paid by a private health care provider with the application of an unsupervised natural language processing method by topic modeling to identify patients suspected of being infected by COVID-19. A total of 6 models were generated: 3 using the BERTopic algorithm and 3 Word2Vec models. The BERTopic model automatically creates disease groups. In the Word2Vec model, manual analysis of the first 100 cases of each topic was necessary to define the topics related to COVID-19. The BERTopic model with more than 1,000 authorizations per topic without word treatment selected more severe patients - average cost per prior authorizations paid of BRL 10,206 and total expenditure of BRL 20.3 million (5.4%) in 1,987 prior authorizations (1.9%). It had 70% accuracy compared to human analysis and 20% of cases with potential interest, all subject to analysis for inclusion in a health promotion program. It had an important loss of cases when compared to the traditional research model with structured language and identified other groups of diseases - orthopedic, mental and cancer. The BERTopic model served as an exploratory method to be used in case labeling and subsequent application in supervised models. The automatic identification of other diseases raises ethical questions about the treatment of health information by machine learning.

Processamento de linguagem natural não supervisionado na identificação de pacientes suspeitos de infecção por COVID-19

Patients with post-COVID-19 syndrome benefit from health promotion programs. Their rapid identification is important for the cost-effective use of these programs. Traditional identification techniques perform poorly especially in pandemics. A descriptive observational study was carried out using 105,008 prior authorizations paid by a private health care provider with the application of an unsupervised natural language processing method by topic modeling to identify patients suspected of being infected by COVID-19. A total of 6 models were generated: 3 using the BERTopic algorithm and 3 Word2Vec models. The BERTopic model automatically creates disease groups. In the Word2Vec model, manual analysis of the first 100 cases of each topic was necessary to define the topics related to COVID-19. The BERTopic model with more than 1,000 authorizations per topic without word treatment selected more severe patients - average cost per prior authorizations paid of BRL 10,206 and total expenditure of BRL 20.3 million (5.4%) in 1,987 prior authorizations (1.9%). It had 70% accuracy compared to human analysis and 20% of cases with potential interest, all subject to analysis for inclusion in a health promotion program. It had an important loss of cases when compared to the traditional research model with structured language and identified other groups of diseases - orthopedic, mental and cancer. The BERTopic model served as an exploratory method to be used in case labeling and subsequent application in supervised models. The automatic identification of other diseases raises ethical questions about the treatment of health information by machine learning.

Identifying accurate pro-choice and pro-life identity labels in Spanish: Social media insights and implications for comparative survey research.

Although debate remains about the saliency and relevance of pro-choice and pro-life labels (as abortion belief indicators), they have been consistently used for decades to broadly designate abortion identity. However, clear labels are less apparent in other languages (e.g., Spanish). Social media, as an exploratory data science tool, can be leveraged to identify the presence and popularity of online abortion identity labels and how they are contextualized online. This study aims to determine how popularly used Spanish-language pro-choice and pro-life identity labels are contextualized online. We used Latent Dirichlet Allocation (LDA) topic models, an unsupervised natural language processing (NLP) application, to generate themes about Spanish language tweets categorized by Spanish abortion identity labels: (1) proelección (pro-choice); (2) derecho a decidir (right to choose); (3) proaborto (pro-abortion); (4) provida (pro-life); (5) antiaborto (anti-abortion); and (6) derecho a vivir (right to life). We manually reviewed themes for each identity label to assess scope. All six identity labels included in our analysis contained some references to abortion. However, several labels were not exclusive to abortion. Proelección (pro-choice), for example, contained several themes related to ongoing presidential elections. No singular Spanish abortion identity label encapsulates abortion beliefs; however, there are several viable options. Just as the debate remains ongoing about pro-choice and pro-life as accurate indicators of abortion beliefs in English, we must also consider that identity is more complex than binary labels in Spanish.

Comparing human coding to two natural language processing algorithms in aspirations of people affected by Duchenne Muscular Dystrophy

Qualitative methods can enhance our understanding of constructs that have not been well portrayed and enable nuanced depiction of experience from study participants who have not been broadly studied. However, qualitative data require time and effort to train raters to achieve validity and reliability. This study compares recent advances in Natural Language Processing (NLP) models with human coding. This web-based study (N=1,253; 3,046 free-text entries, averaging 64 characters per entry) included people with Duchenne Muscular Dystrophy (DMD), their siblings, and a representative comparison group. Human raters (n=6) were trained over multiple sessions in content analysis as per a comprehensive codebook. Three prompts addressed distinct aspects of participants’ aspirations. Unsupervised NLP was implemented using Latent Dirichlet Allocation (LDA), which extracts latent topics across all the free-text entries. Supervised NLP was done using a Bidirectional Encoder Representations from Transformers (BERT) model, which requires training the algorithm to recognize relevant human-coded themes across free-text entries. We compared the human-, LDA-, and BERT-coded themes. Study sample contained 286 people with DMD, 355 DMD siblings, and 997 comparison participants, age 8-69. Human coders generated 95 codes across the three prompts and had an average inter-rater reliability (Fleiss’s kappa) of 0.77, with minimal rater-effect (pseudo R2=4%). Compared to human coders, LDA does not yield easily interpretable themes. BERT correctly classified only 61-70% of the validation set. LDA and BERT required technical expertise to program and took approximately 1.15 minutes per open-text entry, compared to 1.18 minutes for human raters including training time. LDA and BERT provide potentially viable approaches to analyzing large-scale qualitative data, but both have limitations. When text entries are short, LDA yields latent topics that are hard to interpret. BERT accurately identified only about two thirds of new statements. Humans provided reliable and cost-effective coding in the web-based context. The upfront training enables BERT to process enormous quantities of text data in future work, which should examine NLP’s predictive accuracy given different quantities of training data.

Automatic identification of oncology patients with unsupervised phenotyping and AutoML.

e13547 Background: Recent studies suggest ICD codes often cannot characterise patients with specific diseases like oncology accurately in real-world clinical practice due to a high probability of miscoding, while manual inspection of millions of patients is not feasible as it is highly costly and slow. Therefore, we present an AI driven workflow, which leverages structured discrete values and unstructured textual data from Electronic Health Records (EHRs), to automatically identify cancer patients especially those who might be miscoded or undiagnosed. Methods: The proposed AI driven workflow is trained on 29637 and tested on 29339 patients from MIMIC-III database. First, the workflow builds patient profiles by leveraging the state-of-the-art unsupervised Natural Language Processing (NLP) method to extract phenotypic features from textual discharge summaries. Second, the workflow classifies if a patient has the risk of ovarian cancer (ICD-9: 183.0) or lung cancer (ICD-9: all subcodes of 162) based on their profiles using an Automated Machine Learning (AutoML) framework. Third, we improve the workflow through 100 gold-labelled patients out of the train set who are labelled as positive and negative ovarian (lung) cancer cases by three clinicians with consensus and we subsequently measure its performance against another 100 gold-labelled patients out of the test set and the entire test set with 29339 patients. Results: For ovarian cancer, on the 100 gold-labelled patients (52 ovarian cancer cases) out of the test set, the workflow achieves 0.877 AUCROC with precision 84.7% and sensitivity 96.2% whereas ICD codes have precision 94.6% and sensitivity 74.1%. On the entire test set with 29339 patients, the workflow identifies 111 ovarian cancer patients while ICD codes find only 38 patients which lead to 192% lifting. Similarly, for lung cancer, on the 100 gold-labelled patients (66 lung cancer cases) out of the test set, the workflow achieves 0.814 AUCROC with precision 88.7% and sensitivity 83.3% while ICD codes have precision 96.0% and sensitivity 72.7%. On the entire test set with 29399 patients, the workflow identifies 926 lung cancer patients while ICD codes find 585 patients which lead to 58% lifting. Conclusions: This study shows the AI driven workflow on textual clinical notes can help find more cancer patients with high accuracy especially those who are miscoded and undiagnosed by ICD codes. Since ICD codes are primarily used for billing purposes, this workflow also shows the opportunity to reconcile patient characteristics with ICD codes through the application of such novel AI at scale. Moreover, the results of this study have shown this approach can be scaled across other disease areas within oncology and beyond to help find new intelligence to characterise hard to diagnose conditions and improve patient outcomes through a significant improvement in speed and accuracy of screening and early diagnosis of patients.

Physical Activity, Sedentary Behavior, and Sleep on Twitter: Multicountry and Fully Labeled Public Data Set for Digital Public Health Surveillance Research.

BackgroundAdvances in automated data processing and machine learning (ML) models, together with the unprecedented growth in the number of social media users who publicly share and discuss health-related information, have made public health surveillance (PHS) one of the long-lasting social media applications. However, the existing PHS systems feeding on social media data have not been widely deployed in national surveillance systems, which appears to stem from the lack of practitioners and the public’s trust in social media data. More robust and reliable data sets over which supervised ML models can be trained and tested reliably is a significant step toward overcoming this hurdle. The health implications of daily behaviors (physical activity, sedentary behavior, and sleep [PASS]), as an evergreen topic in PHS, are widely studied through traditional data sources such as surveillance surveys and administrative databases, which are often several months out-of-date by the time they are used, costly to collect, and thus limited in quantity and coverage.ObjectiveThe main objective of this study is to present a large-scale, multicountry, longitudinal, and fully labeled data set to enable and support digital PASS surveillance research in PHS. To support high-quality surveillance research using our data set, we have conducted further analysis on the data set to supplement it with additional PHS-related metadata.MethodsWe collected the data of this study from Twitter using the Twitter livestream application programming interface between November 28, 2018, and June 19, 2020. To obtain PASS-related tweets for manual annotation, we iteratively used regular expressions, unsupervised natural language processing, domain-specific ontologies, and linguistic analysis. We used Amazon Mechanical Turk to label the collected data to self-reported PASS categories and implemented a quality control pipeline to monitor and manage the validity of crowd-generated labels. Moreover, we used ML, latent semantic analysis, linguistic analysis, and label inference analysis to validate the different components of the data set.ResultsLPHEADA (Labelled Digital Public Health Dataset) contains 366,405 crowd-generated labels (3 labels per tweet) for 122,135 PASS-related tweets that originated in Australia, Canada, the United Kingdom, or the United States, labeled by 708 unique annotators on Amazon Mechanical Turk. In addition to crowd-generated labels, LPHEADA provides details about the three critical components of any PHS system: place, time, and demographics (ie, gender and age range) associated with each tweet.ConclusionsPublicly available data sets for digital PASS surveillance are usually isolated and only provide labels for small subsets of the data. We believe that the novelty and comprehensiveness of the data set provided in this study will help develop, evaluate, and deploy digital PASS surveillance systems. LPHEADA will be an invaluable resource for both public health researchers and practitioners.

Empowering digital pathology applications through explainable knowledge extraction tools

Exa-scale volumes of medical data have been produced for decades. In most cases, the diagnosis is reported in free text, encoding medical knowledge that is still largely unexploited. In order to allow decoding medical knowledge included in reports, we propose an unsupervised knowledge extraction system combining a rule-based expert system with pre-trained Machine Learning (ML) models, namely the Semantic Knowledge Extractor Tool (SKET). Combining rule-based techniques and pre-trained ML models provides high accuracy results for knowledge extraction. This work demonstrates the viability of unsupervised Natural Language Processing (NLP) techniques to extract critical information from cancer reports, opening opportunities such as data mining for knowledge extraction purposes, precision medicine applications, structured report creation, and multimodal learning.SKET is a practical and unsupervised approach to extracting knowledge from pathology reports, which opens up unprecedented opportunities to exploit textual and multimodal medical information in clinical practice. We also propose SKET eXplained (SKET X), a web-based system providing visual explanations about the algorithmic decisions taken by SKET.SKET X is designed/developed to support pathologists and domain experts in understanding SKET predictions, possibly driving further improvements to the system.

Web-Based Text Analysis of the Patient Safety Concerns of Various Healthcare Stakeholders.

Patient safety is a fundamental aspect of the quality of healthcare and there is a growing interest in improving safety among healthcare stakeholders in many countries. The Korean government recognized that patient safety is a threat to society following several serious adverse events, and so the Ministry of Health and Welfare of the Korean government set up the Patient Safety Act in January 2015. This study analyzed text data on patient safety collected from web-based, user-generated documents related to the legislation to see if they accurately represent the specific concerns of various healthcare stakeholders. We adopted the unsupervised natural language processing method of probabilistic topic modeling and also Latent Dirichlet Allocation. The results showed that text data are useful for inferring the latent concerns of healthcare consumers, providers, government bodies, and researchers as well as changes therein over time.

Natural language word embeddings as a glimpse into healthcare language and associated mortality surrounding end of life

ObjectivesTo clarify real-world linguistic nuances around dying in hospital as well as inaccuracy in individual-level prognostication to support advance care planning and personalised discussions on limitation of life sustaining treatment...