A COMPARATIVE STUDY OF WORD REPRESENTATION METHODS WITH CONDITIONAL RANDOM FIELDS AND MAXIMUM ENTROPY MARKOV FOR BIO-NAMED ENTITY RECOGNITION
Bio-Named Entity Recognition (Bio-NER) is the process of identifying and semantically classifying biomedical technical terms and named entities in Biomedicine literature. Therefore, it is a major task in biomedical knowledge acquisition. Meanwhile, Natural Language Processing (NLP) plays an important role in Bio-NER in the biomedical domain. The first and most essential biomedical literature mining task incorporates biomedical entity recognition such as protein, gene, and chemicals. The most recent Bio-NER methods rely on predefined traditional features, which attempt to capture the specific surface properties of entity types. However, these empirically predefined feature sets differ between entity types and are manually constructed and complicated, which means developing them is costly. In this paper, we systematically present a comparative evaluation study of three methods, which are: the traditional feature representation method, the continuous bag-of-words (CBOW) model, and a new prototypical representation method with two popular sequence-labeling approaches (Conditional Random Fields (CRFs) and Maximum Entropy Markov Models (MEMM)). We evaluated these models with two major Bio-NER tasks, which involve the JNLPBA and GENETAG corpora. This paper examined the prototypical word representation method and found that Word2Vec can be successfully used for Bio-NER. Our results show that the new prototypical representation method improved the performance of the two machine learning models with different datasets. Also, the new prototypical representation method performed better than the traditional feature representation method and CBOW model for both datasets. Finally, our experiment proved that the CRF classifier with the new prototypical representation method achieved the best results when 90% data was used as training data, yielding overall F-measure values of 0.79% and 0.85% for the JNLPBA corpus and GENETAG corpus, respectively. In comparison, the results achieved using the ME classifier yielded overall F-measure values of 0.76% and 0.78% for the JNLPBA corpus and GENETAG corpus, respectively. Keywords—biomedical named entity, prototypical representation, data representation methods, Word2Vec. 1.0 INTRODUCTION Biomedical Named Entity Recognition (Bio-NER) is the process of identifying and semantically classifying biomedical technical terms and named entities in Biomedicine literature; it is a major task in biomedical knowledge acquisition, and is challenging because of the complexity of biomedical terminology. These challenges have motivated many researchers to design efficient techniques for biomedical information extraction and text mining. Unlike general named entities (e.g. person, location, date and time), biomedical named entities have inherently complex structures. This makes it very challenging to identify and to classify these entities during biomedical information extraction. The field of Bio-NER is vast, but there is still a wide gap in performance between the general NER and existing Bio-NER [1][2][3][4][5][6][7]. Therefore, there is room for improvement in this field given that the recognition accuracy of named entities basically hovers around a 10-point F-measure. In addition, the ability of biomedical researchers to manage, integrate, and analyze biomedical data is often limited due to a lack of tools, accessibility, and training [8]. The difficulty and potential importance of this task has attracted many researchers [9]. To date, most NER tools such as (Stanford NER, Illinois NET, and OpenCalais NER WS) have been able to capture the characteristics of different entity classes using feature engineering, which finds the set of features that best help distinguish entities of a specific type from other entity classes. Currently, this procedure is manually constructed and often optimized for a specific gold standard corpus. A Comparative Study Of Word Representation Methods With Conditional Random Fields And Maximum Entropy Markov For Bio-Named Entity Recognition. (Special Issue 2018) pp. 15-30 16 Malaysian Journal of Computer Science. Information Retrieval And Knowledge Management Special Issue, 2018 These traditional features do not have the powerful ability to capture words with comparable synonyms that appear in the same context [10]. The task of text recognition becomes difficult as most biomedical names contain different characteristics [11] such as: Descriptive nature of entity names; One head noun shared by two or more entity names; Several spelling forms for one entity name; Frequent use of ambiguous abbreviations; and Authors of biomedical texts often do not keep track of standards for naming entities and they choose abbreviated forms according to their personal preference. All these characteristics make for trickier identification of biomedical entities compared to the identification of traditional entities. Therefore, it is necessary to explore effective word representation and learning methods to deal with the special characteristics of text in the biomedical domain. Recent works have explored and implemented new representation planners—as part of the traditional feature engineering method—with Bio-NER to address the complex structures of biomedical entities. In this paper, we developed several Biomedical Named Entity Recognition models. The paper presents a comparative evaluation between the traditional feature representation method, continuous bag-of-words (CBOW) model, and a new prototypical representation method. The classification performance of these representation methods was scrutinized using two popular Natural Language Processing (NLP) approaches (Conditional Random Fields (CRFs) and Maximum Entropy Markov Models (MEMM)). All models were evaluated via two standard Biomedical Named Entity Recognition datasets, namely the JNLPBA corpus and GENETAG corpus. The goal of this study is to determine the representation method with the most influence on classification performance, and the classifier that most suits Bio-NER and whether or not the training data size will have an effect on overall performance. We review the related works in this field in Section 2. Then, we discuss several techniques and approaches on this subject matter in Section 3. In Section 4, we present the implementation of the setup for the models examined in this study. We report our findings in Section 5. Finally, we discuss our conclusions in Section 6.
Paper
Full text
A COMPARATIVE STUDY OF WORD REPRESENTATION METHODS WITH CONDITIONAL RANDOM FIELDS AND MAXIMUM ENTROPY MARKOV FOR BIO-NAMED ENTITY RECOGNITION
Semantic Scholar · Computer Science · 2018
Abstract
Bio-Named Entity Recognition (Bio-NER) is the process of identifying and semantically classifying biomedical technical terms and named entities in Biomedicine literature. Therefore, it is a major task in biomedical knowledge acquisition. Meanwhile, Natural Language Processing (NLP) plays an important role in Bio-NER in the biomedical domain. The first and most essential biomedical literature mining task incorporates biomedical entity recognition such as protein, gene, and chemicals. The most recent Bio-NER methods rely on predefined traditional features, which attempt to capture the specific surface properties of entity types. However, these empirically predefined feature sets differ between entity types and are manually constructed and complicated, which means developing them is costly. In this paper, we systematically present a comparative evaluation study of three methods, which are: the traditional feature representation method, the continuous bag-of-words (CBOW) model, and a new prototypical representation method with two popular sequence-labeling approaches (Conditional Random Fields (CRFs) and Maximum Entropy Markov Models (MEMM)). We evaluated these models with two major Bio-NER tasks, which involve the JNLPBA and GENETAG corpora. This paper examined the prototypical word representation method and found that Word2Vec can be successfully used for Bio-NER. Our results show that the new prototypical representation method improved the performance of the two machine learning models with different datasets. Also, the new prototypical representation method performed better than the traditional feature representation method and CBOW model for both datasets. Finally, our experiment proved that the CRF classifier with the new prototypical representation method achieved the best results when 90% data was used as training data, yielding overall F-measure values of 0.79% and 0.85% for the JNLPBA corpus and GENETAG corpus, respectively. In comparison, the results achieved using the ME classifier yielded overall F-measure values of 0.76% and 0.78% for the JNLPBA corpus and GENETAG corpus, respectively. Keywords—biomedical named entity, prototypical representation, data representation methods, Word2Vec. 1.0 INTRODUCTION Biomedical Named Entity Recognition (Bio-NER) is the process of identifying and semantically classifying biomedical technical terms and named entities in Biomedicine literature; it is a major task in biomedical knowledge acquisition, and is challenging because of the complexity of biomedical terminology. These challenges have motivated many researchers to design efficient techniques for biomedical information extraction and text mining. Unlike general named entities (e.g. person, location, date and time), biomedical named entities have inherently complex structures. This makes it very challenging to identify and to classify these entities during biomedical information extraction. The field of Bio-NER is vast, but there is still a wide gap in performance between the general NER and existing Bio-NER [1][2][3][4][5][6][7]. Therefore, there is room for improvement in this field given that the recognition accuracy of named entities basically hovers around a 10-point F-measure. In addition, the ability of biomedical researchers to manage, integrate, and analyze biomedical data is often limited due to a lack of tools, accessibility, and training [8]. The difficulty and potential importance of this task has attracted many researchers [9]. To date, most NER tools such as (Stanford NER, Illinois NET, and OpenCalais NER WS) have been able to capture the characteristics of different entity classes using feature engineering, which finds the set of features that best help distinguish entities of a specific type from other entity classes. Currently, this procedure is manually constructed and often optimized for a specific gold standard corpus. A Comparative Study Of Word Representation Methods With Conditional Random Fields And Maximum Entropy Markov For Bio-Named Entity Recognition. (Special Issue 2018) pp. 15-30 16 Malaysian Journal of Computer Science. Information Retrieval And Knowledge Management Special Issue, 2018 These traditional features do not have the powerful ability to capture words with comparable synonyms that appear in the same context [10]. The task of text recognition becomes difficult as most biomedical names contain different characteristics [11] such as: Descriptive nature of entity names; One head noun shared by two or more entity names; Several spelling forms for one entity name; Frequent use of ambiguous abbreviations; and Authors of biomedical texts often do not keep track of standards for naming entities and they choose abbreviated forms according to their personal preference. All these characteristics make for trickier identification of biomedical entities compared to the identification of traditional entities. Therefore, it is necessary to explore effective word representation and learning methods to deal with the special characteristics of text in the biomedical domain. Recent works have explored and implemented new representation planners—as part of the traditional feature engineering method—with Bio-NER to address the complex structures of biomedical entities. In this paper, we developed several Biomedical Named Entity Recognition models. The paper presents a comparative evaluation between the traditional feature representation method, continuous bag-of-words (CBOW) model, and a new prototypical representation method. The classification performance of these representation methods was scrutinized using two popular Natural Language Processing (NLP) approaches (Conditional Random Fields (CRFs) and Maximum Entropy Markov Models (MEMM)). All models were evaluated via two standard Biomedical Named Entity Recognition datasets, namely the JNLPBA corpus and GENETAG corpus. The goal of this study is to determine the representation method with the most influence on classification performance, and the classifier that most suits Bio-NER and whether or not the training data size will have an effect on overall performance. We review the related works in this field in Section 2. Then, we discuss several techniques and approaches on this subject matter in Section 3. In Section 4, we present the implementation of the setup for the models examined in this study. We report our findings in Section 5. Finally, we discuss our conclusions in Section 6.