Research on Chinese Medical Entity Relation Extraction Based on Syntactic Dependency Structure Information

Round 1

Reviewer 1 Report

The paper’s subject is relevant and interesting. The research in text mining in national languages is developed in the present time and it is very important for the medical application according to the intensive informatization of the healthcare domain. The proposed approach for this problem decision based on dependency trees is acceptable.  

Comments and recommendations:

-        The analysis of the problem state should be improved:

o   The three directions are indicated as the background of the proposed method (see section 2). Why are these approaches considered only? Why SVM or decision tree based methods are ignored? Do you consider the application of fuzzy classifiers? I suppose that analysis of these methods should be included in the analysis of the problem state, for example, based on these publications:

§  Tian, J.-Z., Zhang, X.-L., et al.       Application of data mining technology in the picture archiving and communication system of structured report, Chinese Journal of Clinical Rehabilitation, 10(45), 2006,pp. 108-110+117

§  V. Levashenko, E. Zaitseva and S. Puuronen, "Fuzzy Classifier Based on Fuzzy Decision Tree," EUROCON 2007 - The International Conference on "Computer as a Tool", 2007, pp. 823-827

§  N. M. Abu-halaweh and R. W. Harrison, "Rule set reduction in fuzzy decision trees," NAFIPS 2009 - 2009 Annual Meeting of the North American Fuzzy Information Processing Society, 2009, pp. 1-4

§  Elmurngi, E., Gherbi, A. An empirical study on detecting fake reviews using machine learning techniques,          7th International Conference on Innovative Computing Technology, INTECH 2017, 8102442, 2017,  pp. 107-114

§  Huang HL, Hong SH, Tsai YC. Approaches to text mining for analyzing treatment plan of quit smoking with free-text medical records: A PRISMA-compliant meta-analysis. Medicine (Baltimore). 2020 Jul 17;99(29)

§  Nguyen, A.N., Lawley, M.J., et al., Symbolic rule-based classification of lung cancer stages from free-text pathology reports, Journal of the American Medical Informatics Association, 17(4), 2010, pp. 440-445

o   Is BiLSTM developed by authors? In the analysis of the problem states (lines 161-178) I didn’t find an answer to this question

-        How the dependency tree is constructed (section 3.1.1)? Could you describe in more detail your interpretation of the dependency tree in the context of this investigation?

-        Could you explain the proposed pruning strategy in more detail (section 3.1.2)? How is the shortest path tree developed?

-        All experimental studies are implemented for NNs and the proposed model with some modifications. Could you compare your result with the investigation based on other approaches?

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper is well written an well organised. It proposes a Chinese medical relation extraction method based on syntactic dependency structure information. The proposed architecture is state of the art and introduces some original discourse structures in the IE architecture. The experimental results for Chinese medical entity relation extraction dataset are promising, showing that syntactic dependency information is important in the relation extraction task. Some suggestions: line 19 - 'Medical entity relation extraction aims to extract structured entity relations from medical texts.' Explain better the task Medical entity relation extraction line 20 - These relations exist in the form of triples (<subject, predict, object>), predict -> predicate line 21 - Relation extraction is the basis of many downstream tasks, as well as the difficulty and focus of information extraction. refrase sentence line 40 - At present, Chinese word-cutting methods include classical mechanical word-cutting methods, statistical word-cutting methods, and neural network methods. cite some work for each method line 91 rule-based approaches, machine learning approaches, and deep learning approaches. cite some work for each approach

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

I thank authors for the consideration of my recommendation and comments and the paper modification.