MOLI: Smart Conversation Agent for Mobile Customer Service

Round 1

Reviewer 1 Report

This submission presents a study on a Smart Conversation Agent for Mobile
Customer Service. The authors state that "In our approach we follow the architecture proposed by He et al. [6]", but they also should highlight how they diverge from such ana approach or what they have done more than what possible within the work of He et al. Such a comparison should not only done in a qualitative manner within the second section, but also in a quantitative manner in section 4, and the analysis of the related literature should be strengthen. I believe the proposed work is innovative, but the reader should have all the information to come to such a consideration.

Some presentation issues are present, such as a poor use of puctuation (with long sentences without a comma to define subordinates), some grammar errors (we find -> we found, to tremendous -> to the tremendous), the acronyms (such as CNII) never introduced before being used, a lost reference at line 180, or text in Figure 5 with a too small size.

Author Response

Comments and Suggestions for Authors

This submission presents a study on a Smart Conversation Agent for Mobile
Customer Service. The authors state that "In our approach we follow the architecture proposed by He et al. [6]", but they also should highlight how they diverge from such ana approach or what they have done more than what possible within the work of He et al. Such a comparison should not only done in a qualitative manner within the second section, but also in a quantitative manner in section 4, and the analysis of the related literature should be strengthen. I believe the proposed work is innovative, but the reader should have all the information to come to such a consideration.

Answer:

Thank for you very good advice.

According to your advice, we have added more about our work different from He[6]’s. In addition, we have done some comparison in a qualitative manner and a quantitative manner. We have shown the results of comparison with He[6]’s work named MPCNN in Table 6.

Some presentation issues are present, such as a poor use of puctuation (with long sentences without a comma to define subordinates), some grammar errors (we find -> we found, to tremendous -> to the tremendous), the acronyms (such as CNII) never introduced before being used, a lost reference at line 180, or text in Figure 5 with a too small size.

Answer:

Thank you very much for your advice.

We have modified all errors and add introduce of CNII. Also, we add lost reference at line 180 and adjust the size of Figure 5.

Reviewer 2 Report

This article proposes an interesting natural-language application for supporting dialogues with customer services, reducing errors in comparison to other similar models according to their experiments. Their system queries the user about the missing information, for properly replying customer questions.

Sections should be started numbering in one instead of zero. See section "0. Introduction".

Second paragraph of introduction should be rewritten in a more formal way. It starts with an exclamation ended with three exclamation marks ("!!!"), which is not very formal and outside common scientific writing style.

The related work section is very brief (only half page). Notice that related work section should provide a general view of the most relevant works, indicating which gaps of the literature are still not covered and will be addressed in your contribution.

In order to properly address replies to customer questions, this approach proposes to ask the missing information. Similar works obtains health-related indicators for providing the right recommendation, like the following work about obtaining the missing Heart Rate Variability information for providing customers with a right neighborhood recommendation:
Lacuesta, R., Garcia, L., García-Magariño, I., & Lloret, J. (2017). System to recommend the best place to live based on wellness state of the user employing the heart rate variability. IEEE Access, 5, 10594-10604.
Which are the contributions of your work over works like the aforementioned one?


Section 2 properly provides the problem formalization of conversational question answering with an illustrative example, although the authors could improve how the formalization matches with the natural language example in Figure 1.

Section 3 formally introduces their approach with enough visual content, although the distribution in subsections is not very appropriate in my opinion. I suggest reorganizing this section for avoiding to have so short sections as sections 3.3.1 and 3.3.2. Notice that each of these sections only has three lines.

Line 180: There is a missing reference when mentioning GRU.

Table 2 should include more information, such as some related percentages (e.g. relative frequencies) or more metrics. In its current form, its information is very poor.

However, the remaining part of the experimentation section (i.e. section 4) is well introduced and the obtained results are worth of being published.

In section 5, could you further describe your future research lines instead of just enumerating them?

Author Response

Comments and Suggestions for Authors

This article proposes an interesting natural-language application for supporting dialogues with customer services, reducing errors in comparison to other similar models according to their experiments. Their system queries the user about the missing information, for properly replying customer questions.

Sections should be started numbering in one instead of zero. See section "0. Introduction".

Answer:

Thank you very much for such reminder.

We started from zero instead of one according to request of MDPI.
Second paragraph of introduction should be rewritten in a more formal way. It starts with an exclamation ended with three exclamation marks ("!!!"), which is not very formal and outside common scientific writing style.

Answer:

Thank you very much for this suggestion. We have deleted “!!”.

The related work section is very brief (only half page). Notice that related work section should provide a general view of the most relevant works, indicating which gaps of the literature are still not covered and will be addressed in your contribution.
Answer:

Thank you very much for this suggestion.

We have added related work to provide more general view of current relevant works.

In order to properly address replies to customer questions, this approach proposes to ask the missing information. Similar works obtains health-related indicators for providing the right recommendation, like the following work about obtaining the missing Heart Rate Variability information for providing customers with a right neighborhood recommendation:
Lacuesta, R., Garcia, L., García-Magariño, I., & Lloret, J. (2017). System to recommend the best place to live based on wellness state of the user employing the heart rate variability. IEEE Access, 5, 10594-10604.
Which are the contributions of your work over works like the aforementioned one?
Answer:

In our work, we presented a first approach for conversational question answering in the complex and little-explored domain of technical customer support. Our approach matches a user’s question with the most relevant answer from knowledge base.

Section 2 properly provides the problem formalization of conversational question answering with an illustrative example, although the authors could improve how the formalization matches with the natural language example in Figure 1.

Section 3 formally introduces their approach with enough visual content, although the distribution in subsections is not very appropriate in my opinion. I suggest reorganizing this section for avoiding to have so short sections as sections 3.3.1 and 3.3.2. Notice that each of these sections only has three lines.

Answer:

Very good suggestion!

We merge them together under 3.3.

Line 180: There is a missing reference when mentioning GRU.

Answer:

Yes, we have added the reference.

Table 2 should include more information, such as some related percentages (e.g. relative frequencies) or more metrics. In its current form, its information is very poor.
Answer:

Very good advice.

We have added percentages.

However, the remaining part of the experimentation section (i.e. section 4) is well introduced and the obtained results are worth of being published.

In section 5, could you further describe your future research lines instead of just enumerating them?

Answer:

Very good suggestion.

In the future, we will increase some common sense knowledge graph and intent knowledge graph to help understand our user. Also, we will increase conversation context to record users’ behavior to help understand users’ associated incomplete representations.

Round 2

Reviewer 1 Report

This submission has been considerably improved as the authors proofread it and resolved all the highlighted language and presentation issues. The addition in the manuscript made it more clear and interesting. I think it can be accepted as it is.