Resolving Data Sparsity via Aggregating Graph-Based User–App–Location Association for Location Recommendations

Round 1

Reviewer 1 Report

In this paper the authors propose a graph-based representation model for personalized location recommendations. They improve location recommendation results with app usage records to solve data sparsity problem.

The subject of the article is relevant and worthy of discussion.

The structure of the paper is appropriated.

They describe the motivation and the main contributions of this work.

The proposed approach and the two datasets are described in detail.

They demonstrate the effectiveness of the proposed model by comparing with 8 methods in personalized location recommender systems. The experiments were quite extensive, considered different Sparsity Levels and show that the proposed method presents very positive results compared with other approaches.

The authors should clarify better why different metrics are used to evaluate the experiments in the two datasets (see table 2).

Tables and figures are appropriate.

References are adequate.

Concluding, the topic is interesting and related to the focus of the journal. The approach described presents some differences when compared to other related works.

The article presents some interesting contributions.

Author Response

Dear Reviewer:

      Thank you very much for your advices. Our responses are as follows:

Point 1: The authors should clarify better why different metrics are used to evaluate the experiments in the two datasets (see table 2).

Response 1:  Thank you for your advise. It can be obtained from Table 1 that the number of users and locations of these two datasets are not the same. If we use the same @N metric, the experimental results of different algorithms may not be well compared. We have added our explaination in Section 5.1.1 Metrics (line 310-312), hoping that it can make readers more easily understand. 

Details we added in Section 5.1.1 are as follows:

“We use different @N values to evaluate two diffierent datasets because the number of samples in TalkingData is significantly smaller than that of Telecom Data.”

Reviewer 2 Report

This study considers an important and interesting problem for resolving data sparsity in location recommendations.

I have a few concerns on this manuscript while reading it carefully.

First, the presentation of this manuscript should be improved. It is difficult to figure out the approach of this study. The authors argue that the location recommendations suffer from a data sparsity problem because users typically visit a small portion (e.g., 0.1%) of the entire POIs. Thus, the study proposes app usage records to associate users with locations (POIs). However, it is hard to follow the idea of the proposed solution in Section 4. It is better to provide a running example so that readers can easily find out the key concepts of the proposed solution.

Second, the authors use the mean squared error (MSE) as a loss function in Section 4.5. Can you clarify the reason that MSE is used for this model rather than the cross entropy? I do not think that this model is associated to the regression problem.

Third, several typo errors are found in the manuscript. E.g., Yu et al.[39] -> Yu et al. [39], In [40], Tu et al. -> Tu et al. [40], Mean Square Error -> Mean Squared Error

Author Response

Dear Reviewer:

      Thank you very much for your advises. Our responses are in the attachment. Please see the attachment.

Hoping to hearing from you soon.

Author Response File: Author Response.docx

Reviewer 3 Report

This review’s primary concern aims to give some pointers to the paper’s authors entitled “resolving data sparsity via aggregating graph-based user app location association for location recommendations.” However, it is undeniably true that the article has several strengths. First, the authors address the subject matter (lines 18-33), find ground for realising the research work (lines 33-82; lines 86-70) and set out the general research objective (lines 83-85; lines 70-79) in the introduction. Second, the authors thoroughly review the literature on personalised location recommendation regarding random, deep learning-based and graph-based models (lines 111-159). Third, the authors put forward a specific solution utilising app usage data (lines 160-190).

Nevertheless, the shortcomings outrun the strengths in that the paper structure is confusing and the obtained results are inconclusive.

On the one hand, to address the problems related to the confusing paper structure, a range of measures should be taken. Undoubtedly, a sensible structure moves the writing along logically because there is an order that aids comprehension. Nevertheless, the current contents’ system is disorganised and mistaken. To be specific, let me suggest that the authors take the following measures:

1.       Please, move the main paper contribution content (lines 80-100) to the conclusion section.

2.       Please, put forward a few contrasting hypotheses (lines 226-229). For example, one hypothesis might be that “app usage and user interests are associated.” Another hypothesis might be that “app usage and location characteristics are associated.” Finally, another hypothesis might be that “app usage is associated with location characteristics.”

3.       Please, create a methodology section to include research designs (lines 192-213), formulations (lines 223-231) and proposition (lines 232-297). Needless to say that this methodology section might be divided into different subsections whose headings should be named accordingly to their contents’ subject matter.

4.       Please, create an analysis of the results section with appropriate contents and subsections. For example, table 2 and lines 336-373 seem suitable for this section. Similarly, let me suggest that the authors create preliminary, core, and contrasting subsections.

5.       Please, move the summary (lines 373-381) to the conclusion section. 

On the other hand, to address the problems related to inconclusive results, several measures might be taken as follows:

6.       Although the authors find ground to perform the research work from a professional perspective, I wonder if they might pin down the research gap from a scientific point of view. So, please, address the research gap in the introduction by referring to other research works published in high-impact factor journals.

7.       The conclusion section is tiny and is nothing more than a summary. Please, highlight your contribution by bringing the contents mentioned above from the introduction. Similarly, let me suggest that the authors develop practical implications loosely based on the obtained results. Likewise, put forward future lines of research. Finally, acknowledge the paper’s limitations.

I hope these comments help improve the paper and encourage the authors to move forward.

Author Response

Dear reviewer:

Thank you very much for your advises. Our responses are in the attachment, please see it to check our revision. Hoping to hearing from you soon. 

Author Response File: Author Response.docx

Round 2

Reviewer 2 Report

The authors have cleared my concerns. I recommend that the manuscript would be accepted for publication in the journal.  

Author Response

Dear Reviewer:

    Thank you very much for your constructive advises！

Reviewer 3 Report

This second review’s primary concern is to give some pointers to the paper’s authors entitled “Resolving Data Sparsity via Aggregating Graph-based User-App-Location Association for Location Recommendations.” Fortunately, the authors have followed the reviewer’s recommendations and pinned down the research gap in the introduction. Similarly, they have formulated a few hypotheses in the introduction. Besides, they have reordered the contents to strengthen the methodological section. Equally, they have restructured the analysis of the results section. Finally, they have enriched the conclusion section.

However, there is room to improve the paper. To be specific, the shortcomings are as follows:

(1)    I would instead say the authors did not count the chickens before they were hatched when they developed the paper contribution in the introduction (lines 79-100). This point is more important in the context of a weak conclusion. Therefore, of the benefits to this paper of moving these paragraphs to the conclusion section, the most significant is to strengthen the conclusion section. So, please, I must insist on moving these paragraphs to the conclusion section.

(2)    It is correct without any doubt that a good paper needs to address a research gap. The authors address the research gap but they do not refer to another research work to back up their statements. Please, try to find ground for your statement by considering other research works.

(3)    Although the authors have created three hypotheses in the introduction (lines 231-233), the way they have formulated them is wrong. Therefore, please, preface the hypotheses wording with the suitable symbolised term, for example, “H1:…,” so that each hypothesis can be identified numerically and precisely.

(4)    I wonder why the authors put forward three hypotheses in the introduction section and neglected their empirical contrast in the analysis of the result section. Please, go back over the analysis of the result section to empirically contrast these hypotheses. Are your hypotheses verified or rejected? Please, state and demonstrate it in the analysis of the results section.

(5)    Unfortunately, I am not qualified to judge the English language because I am not a native speaker, nor is this foreign language my expertise. Nevertheless, some drawbacks should be revised. For example, the expression “researches” sounds poor and should be replaced with “research works.”

I hope these comments help improve the paper and encourage the authors to move forward.

Author Response

Dear Reviewer:

    Thank you for your advises! We have revised our paper according to your advises. The detailed responses are in the attachment.

    We are hoping to hearing from you soon.

Author Response File: Author Response.docx