A Densely Connected GRU Neural Network Based on Coattention Mechanism for Chinese Rice-Related Question Similarity Matching

Round 1

Reviewer 1 Report

This paper presents a method called Coattention-DenseGRU for the detection of semantic similarities between questions in rice-related Chinese question and answer communities.

The paper is confuse. Firstly, English should be revised in order to ease interpretation, as same phrases do not make sense. Secondly, even the images contain more than one typo, and should have been better revised before submission, like Figure 2. Finally, sections and subsections should also be better organized, for instance related work and state of the art is mixed with the experimental setup. Parameter settings, which is a decision took prior to obtain results, appears inside results. 

However, the reason which made me the most to refrain my recommendation is that it seems to exist a confusion with evaluation metrics. Although in tables appear a column called "P", which seems to be related with precision, a well-know metric in text related problems, in the body it is described as accuracy, which is a completely different metric. I do not know which one is used, and if it is accuracy, as stated, I cannot perceive the relevance to use accuracy, recall and F1, and not to use precision. And evaluation metrics directly compromise results and thus the contribution.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

The paper is interesting and promising. However, paper requires a lot of improvements. Especially, the English style is not on an expected level which does not help in understanding the paper. Since I find the topic potential, here are my suggestions to improve your paper:

1. In Lines 28-29 "The accuracy and F1 values are 96.3% and 96.9%", those two values of F1 accuracy values of which methods? Please be more precise and accurate.  
2. Try to avoid personal forms  Line 138, Line 141 or Line 164
3. Lines 175-176: "According to the statistics, 99.9% of the questions contains less than 100 words in length, so we can set the length of the questions to 100." require citation to understand which statistics you are mentioning.
4. What is in Line - 179 "(2)". Is it indicated numeration of the chapter or subchapter? 
5.   The calculation formula of GRU must be explained better. It is not clear to the reader what is h tilde and W supposed to be weight and U is weight matrix etc...
6. The formula (6) and (7) are the same, you could use one of those formulas to present a hidden layer of yours model.
7.  In the formula (9), what is BiGRU(d_i)? Please provide more details about this function of the vector. Maybe here should be your model function with a vector?
8.  Formula 10 and 11 are the same and not indicated in the text? 
9. The conclusions are poorly described, and the authors should improve them.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The previous submitted comments were corrected.