BERT based Transformers lead the way in Extraction of Health Information from Social Media

Ramesh, S.; Tiwari, Abhiraj; Choubey, Parthivi; Kashyap, Saisha; Khose, Sahil; Lakara, Kumud; Singh, Nishesh; Verma, Ujjwal

doi:10.18653/v1/2021.smm4h-1.5

Cited by 6 publications

(5 citation statements)

References 8 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The parser pipeline achieved an accuracy of 95.1% on the OntoNotes 5.0 corpus. The en_core_web_trf dependency parsing pipeline has found application in various research studies for the initial reduction in irrelevant information [47][48][49]. This utilisation helped to reduce the model's processing time when dealing with extensive datasets and is concurrently leading to improvements in overall accuracies.…”

Section: Dependency Parsingmentioning

confidence: 99%

Knowledge-Based Intelligent Text Simplification for Biological Relation Extraction

Gill,

Chetty,

Lim

et al. 2023

Informatics

View full text Add to dashboard Cite

Relation extraction from biological publications plays a pivotal role in accelerating scientific discovery and advancing medical research. While vast amounts of this knowledge is stored within the published literature, extracting it manually from this continually growing volume of documents is becoming increasingly arduous. Recently, attention has been focused towards automatically extracting such knowledge using pre-trained Large Language Models (LLM) and deep-learning algorithms for automated relation extraction. However, the complex syntactic structure of biological sentences, with nested entities and domain-specific terminology, and insufficient annotated training corpora, poses major challenges in accurately capturing entity relationships from the unstructured data. To address these issues, in this paper, we propose a Knowledge-based Intelligent Text Simplification (KITS) approach focused on the accurate extraction of biological relations. KITS is able to precisely and accurately capture the relational context among various binary relations within the sentence, alongside preventing any potential changes in meaning for those sentences being simplified by KITS. The experiments show that the proposed technique, using well-known performance metrics, resulted in a 21% increase in precision, with only 25% of sentences simplified in the Learning Language in Logic (LLL) dataset. Combining the proposed method with BioBERT, the popular pre-trained LLM was able to outperform other state-of-the-art methods.

show abstract

Section: Dependency Parsingmentioning

confidence: 99%

Knowledge-Based Intelligent Text Simplification for Biological Relation Extraction

Gill,

Chetty,

Lim

et al. 2023

Informatics

View full text Add to dashboard Cite

show abstract

“…Table 5 shows the performance of uni-modal and bi-modal models on English tweets. Since there are strict limits with 3 submissions per day for the official 6, our official SMM4H 2021 submission combined the results of ten models with the same settings using a simple voting scheme achieved the results on par with the best-performing team [43] that utilized the RoBERTa model with undersampling and oversampling.…”

Section: Ade Text Classificationmentioning

confidence: 99%

Multimodal model with text and drug embeddings for adverse drug reaction classification

Sakhovskiy

Tutubalina

2022

Journal of Biomedical Informatics

View full text Add to dashboard Cite

“…Text formatting was initially done to extract the therapeutics from physician notes using Named Entity Recognition(NER), details of which are provided in (Supplementary Figure 1). [21] The encrypted information ([**word**]) along with single-letter words (example F, M etc), present in the MIMIC data were removed from the text. Text was converted to lowercase followed by standard pre-processing such as removal of whitespaces, stopwords, punctuations, digits, and words with less than or equal to two letters.…”

Section: Shockmodes Pipelinementioning

confidence: 99%

ShockModes:A Multimodal Model for Prognosticating Intensive Care Outcomes from Physician Notes and Vitals

Pal

Patel

Bhatnagar

et al. 2022

Preprint

View full text Add to dashboard Cite

Objective. Shock Index (SI) is widely used for prognosticating outcomes in ICU and emergency settings. We aimed to create a multi-modal early warning system (EWS) for development of abnormal shock index using routinely available vitals and clinical notes. Material and Methods. 17,294 ICU-stays in MIMIC-III data were scored for SI. A new episode of abnormal SI was defined as SI > 0.7 for >30 minutes AND preceded by >=24 hours of normal SI. ICU stays with <24 hours admission, or SI >0.7 within the first 24 hours of admission, or missing SI in >50% in the 24 hour early warning window were excluded, leaving a final cohort of 337 normal and 84 abnormal SI instances. 3117 features from vitals time-series combined with BERT-based features from clinical notes were used to train a battery of machine learning models. The best multimodal pipeline (ShockModes) was assessed for interpretability using SHAP features. Results. Vitals-based, notes-based and multi-modal classifiers achieved the best sensitivity of 0.81, 0.81, and 0.83 with corresponding specificity of 0.92, 0.99, and 0.94 respectively, thus demonstrating the potential of ShockModes for early detection, while preventing false alarms. Global SHAP values revealed Fourier-features of heart rate and heparin sodium prophylaxis as top features. Sensitivity of early detection was highest in acute respiratory failure and chronic kidney disease patients. Conclusion. The multimodal, interpretable early warning system ShockModes can be used for prognosticating SI based outcomes in ICU and emergency settings.Objective. Shock Index (SI) is widely used for prognosticating outcomes in ICU and emergency settings. We aimed to create a multi-modal early warning system (EWS) for development of abnormal shock index using routinely available vitals and clinical notes. Material and Methods. 17,294 ICU-stays in MIMIC-III data were scored for SI. A new episode of abnormal SI was defined as SI > 0.7 for >30 minutes AND preceded by >=24 hours of normal SI. ICU stays with <24 hours admission, or SI >0.7 within the first 24 hours of admission, or missing SI in >50% in the 24 hour early warning window were excluded, leaving a final cohort of 337 normal and 84 abnormal SI instances. 3117 features from vitals time-series combined with BERT-based features from clinical notes were used to train a battery of machine learning models. The best multimodal pipeline (ShockModes) was assessed for interpretability using SHAP features. Results. Vitals-based, notes-based and multi-modal classifiers achieved the best sensitivity of 0.81, 0.81, and 0.83 with corresponding specificity of 0.92, 0.99, and 0.94 respectively, thus demonstrating the potential of ShockModes for early detection, while preventing false alarms. Global SHAP values revealed Fourier-features of heart rate and heparin sodium prophylaxis as top features. Sensitivity of early detection was highest in acute respiratory failure and chronic kidney disease patients. Conclusion. The multimodal, interpretable early warning system ShockModes can be used for prognosticating SI based outcomes in ICU and emergency settings.

show abstract

BERT based Transformers lead the way in Extraction of Health Information from Social Media

Cited by 6 publications

References 8 publications

Knowledge-Based Intelligent Text Simplification for Biological Relation Extraction

Knowledge-Based Intelligent Text Simplification for Biological Relation Extraction

Multimodal model with text and drug embeddings for adverse drug reaction classification

ShockModes:A Multimodal Model for Prognosticating Intensive Care Outcomes from Physician Notes and Vitals

Contact Info

Product

Resources

About