External Slot Relationship Memory for Multi-Domain Dialogue State Tracking

Round 1

Reviewer 1 Report

In this study, the multi-domain problem is examined using a data-driven approach that uses time interval self-attention in dialogs. A Dialogue State Monitoring model based on External Slot Relationship Memory is proposed by the authors to overcome the difficulties of learning relationships between slices from a single sample, which may cause noise. and can reduce efficiency. We're also eliminating time complexity by using a small filter to discard slot information unrelated to the current dialog state. The findings of our study are remarkable and point to promising results.

 

To improve the manuscript:

1-The abstract should be restructured to provide a concise summary of the study's objectives and outcomes.

2-Although the mathematical background is satisfactory, the equations in Section 3.4, specifically those pertaining to External Slot Relationship Memory, should be elaborated upon with detailed explanations.

3-The results of the study, although innovative, should be presented in a clear and organized manner.

4-The result analysis section should be written in a descriptive and comprehensible manner, providing a more detailed interpretation of the results.

Overall, I believe that this study will make a valuable contribution to the existing literature.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

I have finished my review of this manuscript and below are my comments:

1)"Experimental results on the MultiWOZ2.0 and MultiWOZ2.1 benchmarks demonstrate significant improvements while reducing the time complexity to O(n)"

Include empirical results to support the claim above.

2) At the end of Section 1, the contributions of this manuscript must be articulated in point form. This should be followed by the description of how this paper is organized.

3)The authors should offer relevant critique of all the models/solutions presented in Section 2. This will help the readers appreciate the contributions of the current work.

4) "Although the above methods consider slot relationships to some extent, learning slot relationships from a single sample narrows the model’s learning scope and makes it difficult to capture the fundamental connections between slots. There is often a problem with introducing interference information."

Describe how the proposed approach solves the challenges above.

5) In Section 3, a pseudo-code and data flow diagram need to be included to depict the operation of the proposed approach.

6) In sub-section 5.3, include a table that shows the percentage improvements attained by the proposed method, using the baselines models in Table 2.

7) In the list of references, recent works published in the year 2023 are missing. Therefore, you need to incorporate the most recent and relevant works carried out over the recent past.

The quality of English language used in this manuscript is fine.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf