- Rating
- 4(+1)
- Confidence
- 4
- Informativeness
- 3
- Presentation
- 3
- Soundness
- 2
Read review
Paper summary
This paper focuses on two coupled issues, namely Feature Entanglement and Optimization Inconsistency in MLNC, and proposes , which tackles the two issues by modeling homogeneous and heterogeneous information as low- and high-frequency signals, respectively. A JROC module, including LFR and GCC, is introduced to align representation and optimization spaces, where LFR is introduced to disentangle features and clear decision boundaries and GCC quantifies optimization consistency and calibrate the fusion process. Extensive experiments on five datasets show achieves superior performance over state-of-the-art MLNC baselines.
Critical strengths
-
Unlike existing MLNC methods suppressing heterogeneous signals to avoid semantic confusion directly, this paper claims the benefits from the heterogeneous signals and appropriately uses them to mine complementary information with a technically sound framework. And the idea is interesting.
-
The model effectiveness is evaluated on five datasets with existing MLNC methods, and the results show the advantages against the baseline models. Additionally, the authors also provide the statistics analysis to verify the superiority.
Critical weaknesses
None.
Questions for authors
-
Though the paper has demonstrated the promising model performance, the model efficiency has not been evaluated. Could the authors provide a comparison of the training and inference time (e.g., seconds per epoch) and GPU memory usage between and the baselines.
-
The authors should provide the experimental results that the model uses either homogeneous information or heterogeneous information to further demonstrate the rationality of combining both types of information.
-
The fusion weight is generated from both the frequency signals and the label representation. Whether the performance improvement is attributable to the learnable label representations but not the frequency modeling? The authors could provide experiments to explain this question.
-
The authors don’t provide the setting of data split, which makes it hard to reproduce the experiments.
3 comments
Responses to Questions 3 & 4
Response to Q3:
To clarify the source of the performance improvements, we conduct experiments that decouple the frequency modeling and the label representations. We compare the full model ("W/ F&L") against variants lacking frequency modeling ("W/o F") and lacking label representations ("W/o L"), where "F" means the frequency signals and "L" means the label representations.
For constructing the variant without frequency modeling ("W/o F" variant), we employ a GCN to extract node representations and treat the output as the structural state representation , which then interacts with the label representation matrix . For DF$^2$Net without label embedding ("W/o L" variant), we remove the label representation matrix , and map the structural state representation to the class dimension.
The experimental results below indicate that removing either component degrades performance. The best performance is achieved when both components are included, confirming that the performance gain relies on both frequency modeling and label representations.
| Datasets | Components | HL | RL | Ma-AP | Mi-AP | Ma-AUC | Mi-AUC | LRAP |
|---|---|---|---|---|---|---|---|---|
| HumLoc | W/o F | 7.33±0.21 | 14.20±0.62 | 27.93±1.16 | 52.78±1.63 | 71.45±1.49 | 86.85±0.69 | 66.56±1.32 |
| W/o L | 7.29±0.06 | 13.06±0.57 | 28.14±1.21 | 54.13±0.61 | 73.99±1.25 | 88.36±0.38 | 67.29±0.89 | |
| W/ F&L | 7.29±0.07 | 12.54±0.19 | 28.39±0.77 | 54.69±0.55 | 74.24±0.72 | 88.76±0.10 | 68.46±0.45 | |
| PCG | W/o F | 12.79±0.05 | 25.94±0.30 | 22.80±0.50 | 33.33±0.56 | 62.85±0.69 | 73.16±0.16 | 50.86±0.25 |
| W/o L | 12.76±0.03 | 25.54±0.27 | 23.39±0.44 | 33.82±0.33 | 63.60±0.25 | 73.56±0.17 | 51.30±0.47 | |
| W/ F&L | 12.68±0.05 | 25.37±0.30 | 25.31±0.34 | 34.62±0.08 | 65.29±0.24 | 74.32±0.19 | 51.38±0.38 |
Response to Q4:
In our experiments, all datasets are randomly partitioned into training, validation, and testing sets with a 6:2:2 data split. This data split has been implemented in previous works [1] and [2]. We have added the data split in the refined version.
[1]Correlation-aware graph convolutional networks for multi-label node classification. In SIGKDD 2025.
[2] Multilabel node classification with label influence propagation. In ICLR 2025.
Dear Reviewer Ly3Y,
Thank you very much for your thoughtful feedback and for taking the time to consider our responses. We sincerely appreciate your recognition of our work and your encouraging evaluation.
Thank you again for your time and consideration. We wish you all the best!
Authors
Responses to Questions 1 & 2
Response to Q1:
We evaluate the training time, inference time, and GPU memory usage for one epoch, and the experimental results are provided below. The results demonstrate that DF$^2$Net achieves promising computational efficiency. We have included these results in the refined version.
Response to Q2:
To demonstrate the necessity of integrating both types of information, we conducted experiments that isolate the high-frequency (α=0) and low-frequency (α=1) components. The experimental results show that the combination approach ("High-frequency+Low-frequency", namely the adoption of the learnable α) achieves the best performance. This confirms that combining homogeneous and heterogeneous information captures complementary information.