Recommendation System Algorithms on Location-Based Social Networks: Comparative Study

Round 1

Reviewer 1 Report

This manuscript presents a comparative study of three algorithms, SVD, SVD ++, and nonnegative matrix factorization. While SVD confirms its efficiency with a lower error rate, SVD++ has a lower error rate in MAE. However, the current version needs significant improvement before it can be considered for publication. 

The formulation of the research problem and motivation is not clear. Currently, readers are not able to understand the importance and reasons for investigating these three algorithms. As we all know, there are many different types of recommendation system algorithms. Matrix factorization is one of the popular algorithms. However, without a clear description, the readers are hard to realize why the authors focused on MF algorithms and why particularly focused on location-based social networks.

The current literature doesn't provide sufficient background and motivation for this study. As the matrix factorization is the main idea to develop the study, the discussion of the algorithm is relatively short compared to content-based and collaborative filtering recommendations. There is also a lack of reasons for selecting SVD and SVD ++ for further investigation. The authors also need to check the redundant content in the section 2.4 and 3. For example,"In this study, we will compare........latent and neighbor models."

In terms of experiment, dataset's citation is required. The detail information for selecting the sub-dataset is also required. For example, why the description of the dataset contains 10 metropolitan areas but there is only 9 areas in table 3? 

In addition, the current findings appear to be expected and obvious.  The findings are more with respect to people's intuitive understanding without further elaboration. Therefore, the significance of the research appears to be rather weak. 

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The article by the authors deals with a very interesting topic related to social networks and their use. The authors attempted to compare the three algorithms. The paper presents the current state of art, but it seems to me that it is not thorough and sufficient.
The very description of the base, as it seems to me, i.e. the tested algorithms, leaves much to be desired. These descriptions are too laconic ala wikipedia. Therefore, these descriptions should definitely be corrected. (chapters 3.1, 3.2, 3.3)
The structure of the article itself is also not convenient for the reader. No discussion of the results obtained. The summary, i.e. the conclusions, are also insufficient. In fact, the authors did a pretty good scientific job, but they couldn't sell it at all. They did not show what the results obtained in their simulations give. It is bad practice to do the test for the sake of comparison without naming the application of these results.
I believe that the article after major corrections is worth publishing because the research done by the authors is promising but must be better described.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 3 Report

In this study, the authors give a comparison of three matrix factorization techniques, SVD, SVD++, and nonnegative matrix factorization. The deployed recommendation system's main goal is to estimate each user's restaurant rating and give recommendations based on that forecast. Two performance indicators, RMSE and MAE, are used to evaluate this experiment. SVD++ has a lower error rate than SVD in RMSE. SVD++ has a lower error rate than SVD in MAE.

 

The main focus of this study is on the individuals who share ideas, activities, events, and hobbies over the Internet in social networks. Those individuals also share their whereabouts and location-related content, such as photographs and reviews, through location-based social networks (LBSNs). Thus, individuals who investigate how people use geosocial tools and strategies have been paying close attention to LBSN-based studies.

 

This research contributes significantly to the present state of the art. Users tend to share their experiences of places with others, just as they do on location-based social networks (LBSNs) like Foursquare, Yelp, and TripAdvisor, and they also visit places based on ratings and reviews. The complexity of picking users' points of interest (POI), such as restaurants, hotels, stores, and tourist attractions and places, is increasing as these networks rapidly grow with large amounts of data of various forms. Therefore, I believe the main question (a comparative analysis of three matrix factorization methods, namely, SVD, SVD++, and nonnegative matrix factorization) has been comprehensively addressed.

 

Overall, the article is well-written, with clear, easy-to-read wording. In addition, the findings are in line with the evidence and reasoning offered. The topic is unique and relevant to current research in this sector. Thus, I believe it is a timely study and an excellent work of scholarship.

Study justification is needed. There are also some typos that need to be corrected.

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

The authors have addressed previous review feedback properly. However, the current version still misses some important content. The current literature is not sufficient for readers to realize the concept of matrix factorization. It still lacks reasons for adopting the most three common approaches, NMF, SVD, and SVD++. Although the authors provided some more details for each algorithm, they are not the only common approaches in the domain of matrix factorization. As a common approach, it is not a sufficient reason. In terms of the current findings, the authors responded that they include more information in the discussion section but there is no discussion section in the paper. 

In addition, there are several typos or weird sentences in the current manuscript. For example, "Two common challenges affecting the implementation of a recommendation system 48 included The first consideration is the size of the processed datasets." in the introduction. "Each type of the dataset is available In in the form of separate JSON objects." in the experiments.

 

Author Response

Please see the attachment

Author Response File: Author Response.pdf

Reviewer 2 Report

The authors made corrections. They may not be outstanding, but in my opinion they give a chance to say that the article could be published.
While Conclusion should be more developed to better show why their research is great, it's the same problem as before. They did a good job but they can't sell it.

Author Response

Please see the attachment

Round 3

Reviewer 1 Report

In the current revision, the authors had made satisfactory improvements to address them. Acceptance is recommended.