[Paper Review] Improving Multi-Word Entity Recognition for Biomedical Texts

Hamada Nayel, H. L. Shashirekha+2 more|arXiv (Cornell University)|Aug 15, 2019

Biomedical Text Mining and Ontologies31 references10 citations

TL;DR

This paper proposes FROBES, an enhanced IOBES tagging scheme for biomedical named entity recognition (BioNER), improving multi-word entity detection in biomedical texts. Using a BiLSTM-based model on i2b2/VA 2010 and JNLPBA 2004 datasets, FROBES outperforms standard IOBES and baseline models, especially for entities longer than two words, with ensemble voting further boosting performance.

ABSTRACT

Biomedical Named Entity Recognition (BioNER) is a crucial step for analyzing Biomedical texts, which aims at extracting biomedical named entities from a given text. Different supervised machine learning algorithms have been applied for BioNER by various researchers. The main requirement of these approaches is an annotated dataset used for learning the parameters of machine learning algorithms. Segment Representation (SR) models comprise of different tag sets used for representing the annotated data, such as IOB2, IOE2 and IOBES. In this paper, we propose an extension of IOBES model to improve the performance of BioNER. The proposed SR model, FROBES, improves the representation of multi-word entities. We used Bidirectional Long Short-Term Memory (BiLSTM) network; an instance of Recurrent Neural Networks (RNN), to design a baseline system for BioNER and evaluated the new SR model on two datasets, i2b2/VA 2010 challenge dataset and JNLPBA 2004 shared task dataset. The proposed SR model outperforms other models for multi-word entities with length greater than two. Further, the outputs of different SR models have been combined using majority voting ensemble method which outperforms the baseline models performance.

Motivation & Objective

• To address the challenge of recognizing multi-word biomedical entities, especially longer ones, in clinical and research texts.
• To improve the representation of multi-word entities in sequence labeling by extending the IOBES tagging scheme.
• To evaluate the effectiveness of the proposed FROBES model on standard biomedical NER benchmark datasets.
• To enhance overall BioNER performance through model ensemble techniques using majority voting.

Proposed method

• Propose FROBES, a modified IOBES tagging scheme that better captures the boundaries and internal structure of multi-word biomedical entities.
• Design a BiLSTM-based neural network model as the baseline system for sequence labeling in BioNER.
• Train and evaluate the model on two standard biomedical NER datasets: i2b2/VA 2010 and JNLPBA 2004.
• Apply majority voting ensemble by combining predictions from multiple SR models (including IOB2, IOE2, IOBES, and FROBES) to improve robustness and performance.
• Use standard evaluation metrics such as precision, recall, and F1-score to compare model performance across different tagging schemes.

Experimental results

Research questions

• RQ1 Can an enhanced IOBES tagging scheme improve the recognition of multi-word biomedical entities compared to standard IOBES?
• RQ2 How does the FROBES tagging model perform on long multi-word entities (length > 2) in biomedical text?
• RQ3 Does combining outputs from multiple sequence representation models via majority voting improve overall BioNER performance?
• RQ4 Is the proposed FROBES model more effective than baseline models on standard biomedical NER benchmarks?

Key findings

• FROBES significantly outperforms standard IOBES and other SR models in recognizing multi-word entities with length greater than two.
• The BiLSTM-based model with FROBES tagging achieved higher F1-scores on both i2b2/VA 2010 and JNLPBA 2004 datasets compared to baseline models.
• Majority voting ensemble of different SR models further improved performance, surpassing individual models including the FROBES baseline.
• The improvement was most pronounced for longer entities, confirming the effectiveness of FROBES in capturing complex entity boundaries.