今天收到icassp的三个审稿人意见了，评分转化过来是19、15、19，只能说很虚，rebuttal要好好做，其中15的审稿人没有提供意见（没啥好说的，我也觉得这篇论文有点答辩），19的两篇其中一篇说要不要拿这个idea试试IQA，但其实IQA的目的和AQA还是有很大区别，AQA这个方法牵过去不大可能有好效果，所以着重关注这个审稿人的意见（超多，而且审稿人明显是懂哥，意见一针见血的，正确的，中肯的）

根据ChatGPT4给出的建议，在rebuttal做下面几个方面的修改：

Based on the reviewer's comments, here are some suggestions for revising your ICASSP paper on zero-shot image aesthetic assessment (IAA):

Clarify Experimental Setting (Comment 1)

• Clarify the Use of Datasets: Ensure the description of how you use the AVA database and your proposed MMA database is clear and consistent. Explain why only the test dataset of AVA was used and clarify the relationship between the MMA database and the AVA training set.

Expand and Justify the MMA Database (Comment 2)

• Justify Image Selection: Explain the rationale behind choosing only 5,996 images for the MMA database. Discuss any constraints or specific reasons for this selection.

• Consider Expanding the Dataset: Address whether including more images could potentially improve the zero-shot performance of your model. If feasible, consider adding more images to strengthen your findings.

Comparative Analysis with VILA (Comment 3)

• Direct Comparison with VILA: Provide a detailed comparison with VILA’s approach, especially since VILA has reported better performance. Highlight the differences and potential advantages of your KZIAA method over VILA.

• Advantages of Your Method: Clearly articulate the unique strengths or features of KZIAA that might not be present in VILA.

Rationale for MMA Database Construction (Comment 4)

• Compare with PCCD Database: Discuss why you chose to reconstruct the MMA database instead of using an existing one like PCCD. Highlight the unique aspects or benefits of your MMA database.

Clarify Figure 3 and Knowledge Graph (Comment 5)

• Explain Figure 3: Clearly explain the meaning of solid and dashed lines in Figure 3.

• If your multi-modal knowledge graph G is static, clarify this point and discuss the implications. If it's capable of updates, explain how this is achieved.

Comprehensive Review of Existing IAA Methods (Comment 6)

• Update your literature review to include other significant works like the one mentioned by the reviewer (Personality-assisted Multi-task Learning for Generic and Personalized Image Aesthetics Assessment, TIP 2020).

• Ensure that your review comprehensively covers key developments in the field to contextualize your work effectively.