Multi-Intent Natural Language Understanding Framework for Automotive Applications: A Heterogeneous Parallel Approach

Round 1

Reviewer 1 Report

This paper presents an innovative multi-intent parallel interactive framework known as Auto-HPIF, designed to enhance the precision of intent detection in the automotive application domain. By leveraging heterogeneous graphs, the framework facilitates effective intent detection and emphasizes the accuracy of identifying single-intent tasks. A Gaussian prior attention mechanism is introduced to enrich contextual understanding of individual words. Furthermore, the application of a cross-entropy loss function in multi-intent classification enhances the model's adaptability and robustness.

The proposed approach is validated through rigorous benchmarking. Notably, it achieves remarkable performance gains on various evaluation benchmarks, such as MixATIS, MixSNIPS, and CADS. Specifically, there's  3.0% enhancement in overall accuracy on MixATIS, a notable 0.7% improvement in MixSNIPS performance, and a substantial 1.7% advancement on CADS.

While this approach demonstrates effectiveness, it's important to acknowledge potential limitations. The paper acknowledges that the method's evaluation centered on the Chinese Automotive multi-intent Dataset (CADS) may limit its generalizability to other languages and domains. Moreover, the resource-intensive nature of the proposed approach, attributable to its usage of parallel interactive heterogeneous network layers and Gaussian prior attention mechanisms, necessitates significant computational resources.

This paper introduces an advanced framework for multi-intent recognition in the automotive domain, showing improvements in intent detection precision across diverse evaluation scenarios. The comprehensive methodology outlined in the paper, along with its performance benchmarks, positions it as a valuable contribution to the field. While acknowledging its limitations, particularly in terms of generalization and resource requirements, the paper provides an insightful exploration of intent recognition in the automotive context. Given its substantial contributions and potential impact, the paper is suitable for publication with minor revisions (corrections to minor methodological errors and text editing).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Review: Applsci-2531619

The manuscript Applsci-2531619 shows the "Multi-Intent Natural Language Understanding Framework for Automotive Applications: A Heterogeneous Parallel Approach". This work reports an important application about an application of artificial intelligence in automotive Engineering. The problem description needs to be improved by authors, but work is good. Therefore, the manuscript needs of major reviews before publishing. Thus, I am suggesting some main points.

1.    The abstract of work must be re-worked. Authors must provide more details of work results.

2.    I think Authors must present some work highlights in introduction because the introduction is poor in term of important points of article.

3.    The text of work shows moderate English. However, I advise to authors that check the grammatical part of English in paper. After reviewing, the article must be considered for a possible publication.

4.    Work not present its novelty. The novelty of article needs to be detailed in introduction of work. Some novel points can be found in the manuscript body, and then this can be used to describe the novelty.

5.    The problem description needs to be well reported and thus, Authors have to make this. Otherwise, the article is only a research report.

6.    In the item 2 (related work), the text from line 118 up to line 238 must be reviewed. This item reports a revision of the literature regarding topic, but Authors needs to leave the clear text for the reader.

7.    In my point of view, the methodology must also be detailed. On the other hand, the results are not well presented. Usually, a research article reports results and discussion and, therefore, this is not clear in work.

8.    Some symbols at the lines 640, 64, and 642 are shown in other language. However, I suggest to Authors that change to English.

9.    Since, current conclusion does not show well the innovation proposal. Authors need to discuss the proposed theme with its innovation proposal.

 

Moderate English

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper, the authors proposed a multi-intent parallel interactive natural language understanding framework based on heterogeneous graphs for the automotive applications field (Auto-HPIF) to overcome the following two issues: Limited availability of Chinese multi-intent corpus data for research purposes in the automotive domain, and current models, there is commonly a unidirectional information interaction between the slot-filling task and the multiple intent detection tasks, which ultimately leads to inadequate accuracy in intent detection.

The paper’s scope is within the scope of the journal, and it presents an original contribution. The abstract is sufficient to give useful information about the paper’s topic. The proposed algorithm and approach are described and illustrated. The paper is somehow well-structured and written, and the text is clear and easy to read. However, there are some comments we recommend the authors to do:

Make sure to define all abbreviations in the manuscript even if they are well-known. For example, LSTM is not defined, and others.

In the introduction section or where appropriate, you may need to cite and add the following recent references regarding LSTM, deep learning, neural networks, detection and classifiers and their applications:

Wang, Z.; Kim, S.; Joe, I. An Improved LSTM-Based Failure Classification Model for Financial Companies Using Natural Language Processing. Appl. Sci. 2023, 13, 7884. https://doi.org/10.3390/app13137884

Abuqaddom, I.; Mahafzah, B.; Faris, H. Oriented Stochastic Loss Descent Algorithm to Train Very Deep Multi-Layer Neural Networks Without Vanishing Gradients. Knowledge-Based Systems 2021, 230, 107391. https://doi.org/10.1016/j.knosys.2021.107391

Al-Tawil, M.; Mahafzah, B.; Al Tawil, A.; Aljarah I. Bio-Inspired Machine Learning Approach to Type 2 Diabetes Detection. Symmetry 2023, 15, 764. https://doi.org/10.3390/sym15030764

Wang, Z.; Yu, Q.; Wang, J.; Hu, Z.; Wang, A. Grammar Correction for Multiple Errors in Chinese Based on Prompt Templates. Appl. Sci. 2023, 13, 8858. https://doi.org/10.3390/app13158858

Ekolle, Z.E.; Kohno, R. GenCo: A Generative Learning Model for Heterogeneous Text Classification Based on Collaborative Partial Classifications. Appl. Sci. 2023, 13, 8211. https://doi.org/10.3390/app13148211

In Section 5 and before Subsection 5.1, write one small overview paragraph about Section 5 and its subsections.

In Section 5.4, Table 5 must be mentioned in the text. Also, the obtained results in Table 5 need more detailed justification and explanation regarding the algorithmic design point of view.

You need to mention whether your proposed approach suffers from vanishing gradients or not and explain why, where you can cite the above-mentioned reference (Oriented Stochastic Loss Descent Algorithm to Train Very Deep Multi-Layer Neural Networks Without Vanishing Gradients) regarding this issue.

 

At the end of the conclusion section, it is worthwhile to present your best-obtained results as percentages or values in terms of various performance metrics.

The quality of the English language is good. The authors may need to check the whole manuscript for grammar, spelling, and formatting issues in general.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

I can recommend authors to:

- exclude Figure 1. from introduction and to include it in the Related work section;

- review more sources, the current list is not sufficient;

- add aim and objectives of the study, not only contributions in the introduction.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Authors have replied all questions  of this reviewer and, therefore, work can be accepted.

 

Moderate English

Author Response

Dear reviewer,

Thank you very much for your valuable suggestions to this article.We have again refined the English description of this article.