Long- and Short-Term Preference Modeling Based on Multi-Level Attention for Next POI Recommendation
Round 1
Reviewer 1 Report
For long, I have been hoping to see more attention focused on the statistically challenging methods (a "family") of multi level analysis. Thus, I am pleased for your paper as an excellent example of such an approach with hopes of many references and followers.
Author Response
Thanks for your comment
Author Response File: Author Response.docx
Reviewer 2 Report
My major concern is the novelty of this paper. 1) Two observations are quite trivial in the view of the research of human mobility. For the first observation, I am not quite clear about how this relates to the main contribution of this paper. Besides, that is the motivation of applying Bayesian-based approach or any low-dimensional representation approach, but I cannot see it clearly motivates any part of the design of the model. For the second observation, it has been well discussed and well-studied in the field of human mobility prediction (some references and the existing studies I listed below)
2) applying attention mechanism in modeling the long and short term dependencies are well-studied by many existing studies such as (but not limited to):
Yang, Dingqi, et al. "Location prediction over sparse user mobility traces using RNNs: Flashback in hidden states!." Proceedings of the Twenty-Ninth International Joint Conference on Artificial Intelligence. 2020.
- Xia, J. Feng, Y. Qi, F. Xu, Yong Li, D. Guo, F. Sun. AttnMove: History Enhanced Trajectory Recovery via Attentional Network, in AAAI 2021.
Feng, Jie, et al. "Deepmove: Predicting human mobility with attentional recurrent networks." Proceedings of the 2018 world wide web conference. 2018.
These works are missing in both related work to discuss the novelty or compared with as baselines.
The symbols are quite confusing. For example, on the left of Equation 11.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Reviewer 3 Report
The research is designed in an appropriate way and the paper provides a well-designed description of the proposed method. The role of each section is clear, the figures support the understanding. Validation od the proposed method is acceptable and the results are promising. The proposed LSMA framework has novelty and provide significant contribution to this field.
Author Response
Please see the attachment
Author Response File: Author Response.docx
Round 2
Reviewer 2 Report
I think the revised version addressed my concerns.