Session-Based Recommendations for e-Commerce with Graph-Based Data Modeling

Round 1

Reviewer 1 Report

1. Some writing should be carefully edited and corrected, for exam, in abstract, ""We compared our approaches against state-of-the-art Graph Neural Network (GNN) models 10 using four session-based datasets one of which contains data collected by us from a leather ". Please check the whole paper carefully to avoid writing issues.

2. The motivation and the targeted problem are not clear for me. The gap of existing work is not clear either. I suggest the authors to well illustrate these important aspects in the introduction so that this paper can have a clear and strong motivation. 

3. The contributions and novelty look not sufficient enough, I suggest the authors to highlight them.  

4. The ref[4] was formally published at IJCAI 2021, so the information should be updated. Some related work on session-based or sequential recommendations should be reviewed and discussed. For instance, 

[1]. Sequential Recommender Systems: Challenges, Progress and Prospects

[2]. Modeling multi-purpose sessions for next-item recommendations via mixture-channel purpose routing networks

[3]. Hierarchical attentive transaction embedding with intra-and inter-transaction dependencies for next-item recommendation

[4]. Modelling local and global dependencies for next-item recommendations

 

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

The topic is very interesting; the recommender system is still hot in Big Data, e-commerce, and social media. Recommender systems are also an essential factor to increase revenue for e-commerce business companies, so this is very important research to support academics and experts in handling several problems in business, industry, and research.

 

1.     Abstract: In part of the abstract section, there is no explanation of the exact problem in this research field. This part is essential to what the actual problem will be solved, so authors need to develop novel algorithms or models that aim to handle the problem.

 

Moreover, it would be better if the author explained the achievement of this research, including a summary of evaluation metrics and comparison results.

 

2.     To my best knowledge, the Recommender system faces several essential problems, including sparsity data, cold start, serendipity level, long tail, diversity product, etc. Unfortunately, the authors fail to explain what exact problem in their research field specifically.

 

3.     In the proposed model, the second model adopts a recurrent approach, is this model subclass of a recurrent neural network?

 

4.   See list number 244; why to use the symbol    ? it has a specific mathematic symbol. It is very inappropriate to use it as a representation of a set of data.

 

5.     For every symbol, equation and formula are better to add a specific explanation or description (for example, formula 5,6,7).

 

6.     Session based recommendations have already been proposed by another researcher in previous work. How can you claim your work?  (see https://arxiv.org/pdf/2103.16104v2.pdf) and (https://arxiv.org/pdf/1706.04148v5.pdf).

 

7.     In the result and discussion section, there is an essential shortcoming in this manuscript where the authors fail to explain and analysis of the result clearly. Why your proposed model achieved better? What factors influence them? This is important to define and claim the finding. Essentially, aim to answer "why" the model performs. The author only shows the graphic and table without critique and analysis of the graphic and table clearly.

 

8.     In the result and discussion, the author also claimed the model achieves better effectiveness in computation cost. On the other hand, the proposed model performs better in effectiveness. Unfortunately, there needs to be an analysis and explanation of what the important factor that influences the effectiveness of computation cost.

 

9.     In my opinion, the literature review section will be better to show the position of your novel algorithm using table presentation, as you mention in the manuscript reference in part no. 4 (see reference). You can adapt Table 2 on literature no. 4.

 

 

10.  In the conclusion section,  it needs improvement with the future work even with 1 paragraph.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The Manuscript is well written , the proposed method provides results in considerably short time than compared model. However, some justificatios are needed.

1) Introduction, please provide what is the difference of this work from other graph based approaches 

2) Any reference for HSP The Hierarchical Sequence Probability 

2) In experiment part, please mention whether the items do not appear in train dataset or test dataset are removed from test or train dataset, or not?

3) Line 319 'Typically, an MRR > 0.2 indicates that' Is there any reference for this?

4) Please justify why a second metric to evaluate the proposed model is not used?

5) What are the most closest work to this one in literature? I think the following one is the most closest one. If so what is the improvement over this paper?

Delianidi, M.; Salampasis, M.; Diamantaras, K.; Siomos, T.; Katsalis, A.; Karaveli, I. A Graph-Based Method for Session-Based 441

Recommendations. In Proceedings of the 24th Pan-Hellenic Conference on Informatics; Association for Computing Machinery: 442

New York, NY, USA, 2020; PCI 2020, p. 264–267. doi:10.1145/3437120.3437321.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

The authors have well addressed my concerns raised in last round and thus I think the work could be accepted.