A Context Awareness Hierarchical Attention Network for Next POI Recommendation in IoT Environment

Round 1

Reviewer 1 Report

This is an intriguing paper, and the authors attempt to add significantly to the existing literature. However, it is poorly written and needs to be improved in order to be more readable.

Introduction: It does not give a good context for your work. Your ideas must be organized while maintaining the flow of your arguments.

Work related: This section was very well written. However, it would be beneficial if the authors could highlight the gaps in previous work in detail.

Section three: There are a plethora of equations. Please carefully define all terms and double-check the accuracy of all equations. Some of the equations and terms were difficult for me to understand.

The section on Results: You must interpret all of your findings in a meaningful way. The authors have only reported the results, which is insufficient.

 

It would be beneficial if the authors could highlight the implications/importance of your findings in the conclusion.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

Although the  paper  is  well  written. But, I found the manuscript need some revisions as

 

1-       The presentation of this paper needs to be improved, especially the Abstract part such that, the authors said "The experimental results … show  that the AUC, precision … are better than the comparison models…” . What are the differences between this work and the previous works?  Deep explanation is needed about the difference.

2-       Comparison with previous work should be included more.

3-       Writing all equations and mathematical symbols in the text of paper by italics,

4-       Some remark words on computation complexity of the results should be given.

5-       The resolution of the Figure 1 is to bad.

6-       There are problems with referring the References. For example, page 2, line 72, " Zhang et al.[2]" But, the Ref. [2] belong to Sarwat et al. .  Please check all of them, not only my example.

7-       In page 3, line 11, “Zhao et 110 al.[12]”,  the citation is not in the journal format.

8-       In subsection 3.1 “Problem Definition” Could not trace  for any Refs in the text?

9-       In the whole paper, reduce to use of the words "we", avoid such type language.

10-   There are some typos, Comma, full stop and spacing between words should be checked seriously from beginning to last up to references. For example, in abstract, line 10, “…LBSN).In view of …”

11-   Put dot (.) at the end of the sentences in the citation part of the all figures and  tables.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In the article, the authors describe the design of a system that, based on available (geographical and temporal) data and user behavior, will recommend another point-of-interest (POI). The solved problem is not new, there are several methods to solve it, but the achieved results are not the best, so it is important to look for new, more accurate solutions. An overview of experimental results, with which it is possible to compare own achieved results, can be found in [1].

The authors called their system Context Geographical-Temporal-Social Awareness Hierarchical Attention Network (CGTS-HAN), in which they connected geographic data with time data, creating a sequence of triplets (longitude, latitude, time), which is used in further calculations. The proposed system contains innovative ideas and consists of 4 parts: Context iteration layer, Geo-temporal social attention network, Co-attention network, Sorted POIs. The individual parts are described in more detail, but the descriptions need to be supplemented with explanations of the symbols used and more precise expressions.

The experimental part was done on 2 published datasets: Foursquare dataset and Yelp dataset. It would be appropriate to indicate what data these datasets contain. The results of the experimental part are processed in the style as it was done in the article [2], which is good, but why are these results not compared to each other (the Foursquare dataset is used in the article). In the experimental part, it would be appropriate to give specific information about the model on which the data were calculated, for example, the size of the neural network in the Context iteration layer, and so on.

In order for the article to be publishable, it is necessary to revise it and compare the achieved results with the existing ones.

Recommendations:

1. Page 4, Fig. 1: There are symbols in the picture that are not explained in the text, for example ci, Su.

2. Page 4, l. 169: what does the expression "the context in the context" mean

3. Page 5, l. 184: Not explained what "g" is.

4. Page 6, l. 220, l. 231: Transition[12] needs to be explained. What is Transition and can be found in the article [12]. Analogous to Transformer[26].

5. Page 6, l. 237: What does "matrix as follows, denoted MxM" mean?

6. Page 6, formula (8): Reformulate the definition. In what relation is the independent variable "dis" to the dependent variables.

Page7, l. 264: Explain what hi in H represents.

Page 8, formula (20): Notation ||*|| is used to calculate the norm. How is the norm calculated if the argument is a set?

Page 8, formula (21): Explain what "X" is, because "x" is not in the expression. The procedure for obtaining formula (21) from formula (18) is not clear.

Page 12: It is surprising that the achieved results in accuracy on the test set are better than on the training set. Could you explain it?

 

[1] PABLO SÁNCHEZ, ALEJANDRO BELLOGÍN: Point-of-Interest Recommender Systems: A Survey from an Experimental Perspective. ACM Comput. Surv., Vol. 1, No. 1, Article. Publication date: June 2021

[2] Wang X, Liu Y, Zhou X, et al. A Point-of-Interest Recommendation Method Exploiting Sequential, Category and Geographical Influence. ISPRS International Journal of Geo-Information 2022, Volume.11, 80

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 3 Report

The paper looks much better. I can't understand why the text of algorithm 1 is centered (it looks crazy).