Reviewer #1
Questions
1. Summary Please briefly summarize the main claims/contributions of the paper in your own words. (Please do not include your evaluation of the paper here).
This paper aims to address a challenging problem in the field of AI-driven fashion pairing recommendations, which is the understanding and accurate interpretation of compatibility relationships between fashion items. The paper introduces a dataset called Pair Fashion Explanation (PFE) and an innovative two-stage pipeline model to utilize this dataset for generating explanations that convey compatibility relationships between items.
2. Strengths and Weaknesses Please provide a thorough assessment of the strengths and weaknesses of the paper, touching on each of the following dimensions: - Novelty (how novel are the concepts, problems addressed, or methods introduced in the paper) - Quality (is the paper technically sound?) - Clarity (is the paper well-organized and clearly written?) - Significance (comment on the likely impact of the paper on the AI research community, i.e., explain how the paper might impact its own sub-field or the general AI community.).
Strong Points:
1. Dataset Contribution: The introduction of the PFE dataset is a significant contribution to this research, as it provides researchers with a unique resource to delve deeper into compatibility relationships. This will help improve the understanding of relationships between fashion items.
2. Model Innovation: The proposed two-stage pipeline model is an innovative work that holds the potential to enhance the quality of explanations, making them better at conveying compatibility relationships between items.
3. Evaluation Methods: The paper mentions the use of both automatic metrics and human evaluation to assess the explanations generated by the model. This multi-faceted evaluation approach helps validate the quality of the explanations.

Weak Points:
4. Insufficient Model Training Details: For the training processes of the Cross-Attn model, most of the details are placed in the appendix, and it is recommended to condense and include them in the main body of the paper. The training process of the Rationale Extraction model also lacks details. It is advisable for the authors to illustrate these model training processes using diagrams to enhance reader comprehension and reproducibility of the research.
5. Lack of Statistical Information for PFE Dataset: The article does not provide certain statistical information about the PFE dataset, such as the statistics regarding the number of data entries for different types of clothing. This statistical information is crucial for subsequent researchers to better understand the dataset and its utility. It is recommended that the authors provide more statistical information about the dataset to enhance its comprehensibility.
6. Unclear Writing Details: In some sentences in the section on constructing the PFE dataset, such as "To filter out irrelevant sentences, we use the named entity recognition capabilities of the spacy package," it seems unclear. It is suggested that the authors rewrite these sentences to make them more understandable and eliminate potential ambiguities.
3. Questions for the Authors Please carefully list the questions that you would like the authors to answer during the author feedback period. Think of the things where a response from the author may change your opinion, clarify a confusion or address a limitation. Please number your questions.
Please see the above comments on the weaknesses.
4. Reproducibility Does the paper provide enough information to be reproducible? If not, please explain (It may help to consult the paper’s reproducibility checklist.)
Yes
5. Resources Are there novel resources (e.g., datasets) the paper contributes? (It might help to consult the paper’s reproducibility checklist)
Yes
6. Ethical considerations Does the paper adequately address the applicable ethical considerations, e.g., responsible data collection and use (e.g., informed consent, privacy), possible societal harm (e.g., exacerbating injustice or discrimination due to algorithmic bias, etc.)? If not, explain why not. Does it need further specialized ethics review?
Yes
7. OVERALL EVALUATION Please provide your overall evaluation of the paper, carefully weighing the reasons to accept and the reasons to reject the paper.
Weak accept: Technically solid paper where reasons to accept, e.g., good novelty, outweigh reasons to reject, e.g., fair quality.
8. CONFIDENCE How confident are you in your evaluation?
Quite confident. I tried to check the important points carefully. It is unlikely, though conceivable, that I missed some aspects that could otherwise have impacted my evaluation.
Reviewer #3
Questions
1. Summary Please briefly summarize the main claims/contributions of the paper in your own words. (Please do not include your evaluation of the paper here).
To explain compatibility relationships between fashion items, this paper designs a Pair Fashion Explanation (PFE) dataset and proposes an innovative two-stage pipeline model, which allows the model to generate explanations that convey the compatibility relationships between items. The experiments are conducted to verify the effectiveness.
2. Strengths and Weaknesses Please provide a thorough assessment of the strengths and weaknesses of the paper, touching on each of the following dimensions: - Novelty (how novel are the concepts, problems addressed, or methods introduced in the paper) - Quality (is the paper technically sound?) - Clarity (is the paper well-organized and clearly written?) - Significance (comment on the likely impact of the paper on the AI research community, i.e., explain how the paper might impact its own sub-field or the general AI community.).
Strength:
1. This paper releases a dataset for pairwise explanations, which can promote the development of this field.
2. This paper proposes a reasonable method to generate informative and diversified explanations and offline experiments are performed to show its effectiveness.

Weakness:
1. The method looks straightforward and less innovative.
2. Some important design principles and detailed descriptions are missing. For example, the architecture of extraction model and generation model.
3. The paper has many grammatical problems, which are not conducive to reading.
3. Questions for the Authors Please carefully list the questions that you would like the authors to answer during the author feedback period. Think of the things where a response from the author may change your opinion, clarify a confusion or address a limitation. Please number your questions.
1. Compared with the existing methods, what is the innovativeness of the paper?
2. Please introduce the structure and design motivation of the extraction model and generation model.
4. Reproducibility Does the paper provide enough information to be reproducible? If not, please explain (It may help to consult the paper’s reproducibility checklist.)
No
5. Resources Are there novel resources (e.g., datasets) the paper contributes? (It might help to consult the paper’s reproducibility checklist)
Yes
6. Ethical considerations Does the paper adequately address the applicable ethical considerations, e.g., responsible data collection and use (e.g., informed consent, privacy), possible societal harm (e.g., exacerbating injustice or discrimination due to algorithmic bias, etc.)? If not, explain why not. Does it need further specialized ethics review?
Yes
7. OVERALL EVALUATION Please provide your overall evaluation of the paper, carefully weighing the reasons to accept and the reasons to reject the paper.
Weak Reject: Technically solid paper where reasons to reject, e.g., poor novelty, outweigh reasons to accept, e.g. good quality.
8. CONFIDENCE How confident are you in your evaluation?
Not very confident. I am able to defend my evaluation of some aspects of the paper, but it is quite likely that I missed or did not understand some key details, or can't be sure about the novelty of the work.
Reviewer #4
Questions
1. Summary Please briefly summarize the main claims/contributions of the paper in your own words. (Please do not include your evaluation of the paper here).
The paper introduces the Pair Fashion Explanation (PFE) dataset, which aims to address the challenge of understanding compatibility relationships between fashion items in AI-driven outfit recommendations. The authors propose a two-stage pipeline model that leverages this dataset to generate explanations that convey the compatibility relationships between items. The experiments demonstrate that the model can produce knowledgeable and informative descriptions, as evaluated by automatic metrics and human evaluation. The paper also mentions that the code and data will be released upon publishing.

2. Strengths and Weaknesses Please provide a thorough assessment of the strengths and weaknesses of the paper, touching on each of the following dimensions: - Novelty (how novel are the concepts, problems addressed, or methods introduced in the paper) - Quality (is the paper technically sound?) - Clarity (is the paper well-organized and clearly written?) - Significance (comment on the likely impact of the paper on the AI research community, i.e., explain how the paper might impact its own sub-field or the general AI community.).
Strengths：
1. The paper introduces a valuable dataset that is specifically designed for pair-matching explanations.
2. The experimental design in this paper is thorough and persuasive, providing strong evidence to support the claims made.
Weaknesses：
The technical approach utilized in this paper appears to be relatively simple, lacking significant innovation.
3. Questions for the Authors Please carefully list the questions that you would like the authors to answer during the author feedback period. Think of the things where a response from the author may change your opinion, clarify a confusion or address a limitation. Please number your questions.
Why not use joint training instead of a two-stage approach? How can you ensure that the performance of the first-stage model is good?
4. Reproducibility Does the paper provide enough information to be reproducible? If not, please explain (It may help to consult the paper’s reproducibility checklist.)
Yes
5. Resources Are there novel resources (e.g., datasets) the paper contributes? (It might help to consult the paper’s reproducibility checklist)
Yes
6. Ethical considerations Does the paper adequately address the applicable ethical considerations, e.g., responsible data collection and use (e.g., informed consent, privacy), possible societal harm (e.g., exacerbating injustice or discrimination due to algorithmic bias, etc.)? If not, explain why not. Does it need further specialized ethics review?
No further specialized ethics review is needed.
7. OVERALL EVALUATION Please provide your overall evaluation of the paper, carefully weighing the reasons to accept and the reasons to reject the paper.
Weak accept: Technically solid paper where reasons to accept, e.g., good novelty, outweigh reasons to reject, e.g., fair quality.
8. CONFIDENCE How confident are you in your evaluation?
Somewhat confident, but there's a chance I missed some aspects. I did not carefully check some of the details, e.g., novelty, proof of a theorem, experimental design, or statistical validity of conclusions.
Go Back