Knowledge-Guided Prompt Learning for Few-Shot Text Classification

Round 1

Reviewer 1 Report

The paper has a strong poin in comparing some methods with the proposed one.

However, there are points that should be revised:

1.  The authors on;y shows basic computation/formula. It would bebeter if they support the formula transformation step by step.

2. the term " prompt learning guided by knowledge." should be more explsained

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The topic of the paper is very timely. However, leaving aside that the structure of the paper is difficult to follow, and some English terms should be avoided (e.g., repeating "so on", several times in consecutive paragraphs), there's also room for improvement in other sections.

The authors should indicate more clearly what has been done and why this manuscript is necessary in the Introduction. As it is now, it is unclear what is novel here.

Some of the general aspects of what is now "Methods" should be in the Introduction.

It was unclear whether the comparison across procedures is actually fair? All all the tested models proposed to do this task? Would it work similarly in alphabetic languages like English?

The conclusion section needs some work. Some section of limitations is necessary.

Overall, there's some potential, but the authors need to indicate clearly the novelty of the present approach over previous studies, and why their proposed procedure appears to work better than other well-known procedures.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

Overall, the manuscript looks okay but needs to be incorporated the following suggestions; Please try to accommodate the following suggestions:

1.      Provide a few sentences to mention the rationale for "why this problem in hand needs zero-shot learning to be resolved', - mention it in the introduction section to showcase the importance of the Zero-shot (ZS). This should address- why ZS, the importance of ZS in text classification, and how ZS is the best suitable than the other Text classification techniques.

2.   If there are any future scope and limitations (maybe I missed them) then please address that in the conclusion section.

3. From your relevant work section, please get two to three closely related papers and compare their results with your results, as you did with BERT-based models.

4. Please mention, what tools/programming libraries, you have used to implement your model and on what configured machine. 

5. Please clear, if you have used manual labeling then how did you establish the interrater agreement?

Thank you for the submission and looking forward to the revised version.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

I am happy with the changes. The authors may want to check the ms for some expressions and typos:

Line 39. “have received a lot of attention”….I believe “Increasing attention” would work better

Lines 343-344. “ However,There is a wide gap in the performance of all the prompt-based 344 models. The result demonstrates ”. A spaces is necessary after “however,” and in lowercase. Also, “The result” should be “This result”.

In general, some proofreading is necessary, but the authors have addressed my concerns on the content and structure.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

I appreciate the author's attempt to reiterate the paper by accommodating he suggestion. However, a few comments were not accommodated efficiently and alternate work has been supplied.

All the best.

Author Response

Please see the attachment

Author Response File: Author Response.pdf