CETD: Counterfactual Explanations by Considering Temporal Dependencies in Sequential Recommendation
Round 1
Reviewer 1 Report
This paper proposes a counterfactual explanation model based on a VAE for the sequential recommendation. The matter is significant and the experimental results look good concerning the effectiveness of the proposed model. The article is technically innovative and clearly structured. Before the article is accepted, authors are requested to carefully revise the following issues:
1. In the Abstract and Introduction, the authors would do well to strengthen their explanation of the motivation for the article, which is now presented as two challenges but without a clear statement of the shortcomings of the existing work.
2. In Sec. 3, the article only lists the pseudo-code and lacks an explanation of the code; it is suggested that more work be done to describe the pseudo-code.
3. In Sec.4, the comparison methods chosen for the article are slightly dated, and it is suggested that more relevant work from the last 3 years be added.
4. In Sec.4, regarding the results of the experiment, the results of all models in [31] are far from CR-VAE and CETD, so what is the point of comparing the algorithms in [31]? In addition, [9] is only a conference paper with questionable quality of algorithms.
The writing quality of the article is acceptable.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 2 Report
Comments for author File: Comments.pdf
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Reviewer 3 Report
An interesting topic is selected by the authors. The work proposes a study on Counterfactual Explanations by Considering Temporal Dependencies (CETD), a counterfactual explanation model that utilizes a Variational Autoencoder (VAE) for sequential recommendation and takes into account temporal dependencies. For improvement of explain ability, CETD employs a Recurrent Neural Network (RNN) when generating counterfactual histories. Although the overall presentation of the work is good the work requires a revision for a chance of acceptance.
Authors stated that, “Thorough experiments conducted on two real-world datasets demonstrate that
The literature overview of the work must be improved. There are total of 31 cited works, however there is only 1 cited work from 2023 and 5 from 2022. The ratio of the recently cited works (after 2022) must be at least 25% of the total citation more than 40 cited work.
The proposed CETD consistently surpasses current state-of-the-art methods.” The state of the art methods is not clear. Authors must present a table of comparison with counterpart works from the literature. What are the advantages and disadvantages of the proposed approach to the counterpart work from the literature?
The results given in Table 1 should be explained with more deetials. Why given results from [31] are so low while the results from [9] and CETD are so higher is the metrics different or a difference of presentation? It would be better to extent the given example and use results for high rank journal than the results from conference in this table.
Also another important matter when studying Artificial intelligence algorithms the hyper-parameter is a crucial parameter that can significantly affect the overall performance of algorithms. Authors should also add a section about this matter and novel approaches in this topic for the possible readers and clearly explain that if the hyper-parameters are not selected accurately of properly even the state of the art methods would fail. Authors are welcome to use following or similar works for the mentioned comment.
[A1] Calik, Nurullah, Filiz GüneÅŸ, Slawomir Koziel, Anna Pietrenko-Dabrowska, Mehmet A. Belen, and Peyman Mahouti. "Deep-learning-based precise characterization of microwave transistors using fully-automated regression surrogates." Scientific Reports 13, no. 1 (2023): 1445.
[A2] Karaman, Ahmet, Dervis Karaboga, Ishak Pacal, Bahriye Akay, Alper Basturk, Ufuk Nalbantoglu, Seymanur Coskun, and Omur Sahin. "Hyper-parameter optimization of deep learning architectures using artificial bee colony (ABC) algorithm for high performance real-time automatic colorectal cancer (CRC) polyp detection." Applied Intelligence 53, no. 12 (2023): 15603-15620.
Author Response
Please see the attachment.
Author Response File: Author Response.pdf
Round 2
Reviewer 3 Report
no further comments the work can be accepted as it is.