Multi-Meta Information Embedding Enhanced BERT for Chinese Mechanics Entity Recognition
Round 1
Reviewer 1 Report
This paper proposed a Chinese Named Entity Recognition model based on a modified pre-trained BERT, which combines lexical and radical information. The results of contrast experiment, generalization experiment and ablation experiment verified the performance of the proposed model.
This paper shows that the pre-training large model has a good application prospect in the vertical field of mechanics entities recognition.
The paper should also compare and analyze the proposed model with the baseline model in terms of model complexity or running time.
There are many English and writing mistakes in the paper, such as:
line 75, the means of "I-MOT" and "B-MOT" should be explained
line 93 "it can can"
line 156 "maed del"
line 160 "FALT" or "FLAT"
line 198 "lexicon embedding, lexical embedding and radical embedding" or "Character embedding, lexical embedding and radical embedding"
line 420 "in the data"
line 480 "Third" should be "Fourth", " except for LE-BERT and MMIEEE-BERT have better performance on balanced data than other models." the mean is not clearly.
line 640 "feature" should be "future"
Author Response
We would like to thank the reviewer for the constructive feedback. We appreciate the thoughtful and positive comments, which have certainly helped to improve the presentation and quality of our paper. We have updated our paper according to the suggestions and performed more experiments as requested by the reviewers.
Comments of reviewers in black
Answers to the reviewers in orange
Modifications of the manuscript in blue italics
Reviewer 2 Report
I think the subject of study might be of interest.
Indeed, the subject of study is of interest given the recent impact that Artificial Intelligence, machine learning and the use of different algorithms are currently having in different areas of knowledge, which gives a special value, an added value to this work.
The procedure carried out, through its comparison with existing algorithms, to arrive at the proposed model is well described and explained graphically.
The context where the study is carried out, the region, the city must appear in the abstract.
The work is fine, the procedure is fine, the topic as I say is of interest, so I think the work should be published.
Author Response
We would like to thank the reviewer for the constructive feedback. We appreciate the thoughtful and positive comments, which have certainly helped to improve the presentation and quality of our paper. We have updated our paper according to the suggestions.
Comments of reviewers in black
Answers to the reviewers in orange
Modifications of the manuscript in blue italics
Author Response File: 
Author Response.docx
Reviewer 3 Report
In the article, the authors introduce a novel approach designed to improve Chinese Named entity recognition in mechanic domain.
The "Introduction" and "Related work" Sections introduce the topic and the related problems quite clearly, and provide an overview of approaches commonly used in the literature.
Moreover, the proposed approach is scientifically sound and the authors' results show an improvement compared to the other models used as baseline. However, there are some points that need to be addressed to improve the overall quality of the work. First of all, an extensive editing of the English language is required as the many typos (to name a few: extractin, fist, resent studies, etc) and numerous phrases with ambiguous meaning make reading the paper difficult.
Furthermore, although the authors present the architecture with the aim of improving Chinese NER in the mechanical domain, the proposed architecture appears to have no domain-specific components. In fact, the improvement in performance compared to the baseline also occurs in different domains. The authors are asked to explain this point better.
The figure describing the entire proposed architecture is difficult to read. Even the comparative tables are difficult to read, as each time you have to review in the text which architecture base 1, base 2, etc. corresponds to. Furthermore, it would be useful to bold or underline the text which approach obtains the best performance for each test.
an extensive editing of the English language is required as the many typos (to name a few: extractin, fist, resent studies, etc) and numerous phrases with ambiguous meaning make reading the paper difficult.
Author Response
We would like to thank the reviewers for the constructive feedback. We appreciate the thoughtful and positive comments, which have certainly helped to improve the presentation and quality of our paper. We have updated our paper according to the suggestions.
Comments of reviewers in black
Answers to the reviewers in orange
Modifications of the manuscript in blue italics
Author Response File: Author Response.docx
