Maximal Instance Algorithm for Fast Mining of Spatial Co-Location Patterns

Round 1

Reviewer 1 Report

The revision is of a much higher quality overall. There are a few concerns/issues for the authors to address: 

1. the effect of min_d. It seems that co-locations are easier to be detected while min_d keeps increasing. But the goal of applying such methods is not to detect as many co-location patterns as possible. So what would you suggest to pick out the optimal min_d parameter value?
2. The comparison between different methods heavily focuses on the efficiency rather than effectiveness. When you design synthetic data experiments, you probably designed what the desired results would be, e.g., how many co-location patterns should be detected and where they are. So I'd suggest adding the result quality comparison before the efficiency comparison. Do all methods generate the same results? Any false-positive or false-negative errors?
3. The new real dataset is better than the previous transportation one, because the features seem to be more independent. But the interpretation part is still weak. What do the detected co-location patterns, e.g. {U,H}, {U,L} mean in the real-world context? Do the patterns make sense or counterintuitive? The paper should set an example to showcase the practical usefulness of the method.
4. please double check syntax or grammatical errors. I spotted typos like "jion" in several figures. 

Author Response

Thanks for your review. Please see our response in the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This is fine for me. I accept the revision.

Author Response

Thanks for your review.

Round 2

Reviewer 1 Report

I still don't understand why you didn't design your experiment dataset with targeted results. Then how do you evaluate the correctness of your results? How to confirm that all (or most of) your detected patterns are meaningful co-location patterns? How can someone be confident of using your method if the resulting quality is not guaranteed?

Author Response

Thanks for your comments, please see my response in the attachment.

Author Response File: Author Response.pdf

Round 3

Reviewer 1 Report

satisfactory response

This manuscript is a resubmission of an earlier submission. The following is a list of the peer review reports and author responses from that submission.

Round 1

Reviewer 1 Report

I would appreciate to have more scientific discussion in Conclusions section. You should compare your results with other authors. The last section showing the application on real data is valuable and should be more emphasised.

Please check English spelling and style. Some suggestions from my side in the text. Formatting also should be more careful.

Comments for author File: Comments.pdf

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 2 Report

The paper presents a new clustering algorithm for fast mining of spatial co-location patterns. The algorithm is detailed step by step. However, the paper didn’t present convincing evidence to support the claimed advantages over the existing methods. Some details of the algorithms need to be clarified. Please find my comments/questions below.

1. The presented algorithm tries to identify low-size row instances from high-size row instances, which contradicts some of the existing methods. It borrows the concepts of hierarchical clustering, yet little literature from this field has been referenced. Despite the argument on page 3 that claims join-based methods are inefficient, no evidence is provided to make this point sound enough. Therefore, I wasn’t convinced that this top-down approach is clearly better.
2. A fundamental uniqueness of spatial dataset is spatial relationships. In the case of identifying co-location patterns, how to define and use the spatial distance is one of the most important factors. This paper briefly discussed the usage of Euclidean distance on page 7. And the approach is rudimental as it relies on the user-defined minimal distance threshold. In addition to your discussion of the effect of min-dist parameter, I believe it is necessary to add a discussion of how to define the distance in different scenarios. Since Euclidean distance is less optimal in many cases, let alone it is difficult to calculate it for spatial features that are not symbolized as points.
3. Where is the experiment with synthetic data mentioned in the abstract?
4. The choice of the test dataset is poor. All the facilities are transportation-related, which are endogenous. It is hard to comprehend the usefulness of the identified co-location patterns. Plus, the “facilities” are not clarified in the paper, e.g. what does a rail facility mean? Station? Railway? I’d suggest conducting the experiment with more meaningful datasets, especially since the types of features are not obviously dependent. For instance, a previous co-location study that came across identified co-location patterns between lung cancer cases and pollutant sources.
5. The USA map has several flaws: inappropriate projection; clearly wrong scales; useless colored basemap; missing Hawaii & Alaska.
6. My biggest problem with the experiment is the lack of comparison with existing methods. Without comparisons, how can the readers believe that this method is better to some extent?

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 3 Report

The authors present a co-location pattern discovery algorithm in an efficient using maximal instances. The idea of the paper is interesting, however several issues should be addressed: 

1- The introduction of the paper should be re-written, please elaborate more in the introduction by explaining well the motivation and the contribution of the paper. 

2- More comparison with the baseline algorithms should be performed. 

3- Presentation should be improved for instance algorithm 1 should be re-written. 

4- More references are needed: 

Highly efficient pattern mining based on transaction decomposition

Maintenance of discovered high average-utility itemsets in dynamic databases

 

 

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Reviewer 4 Report

Maximal Instance Algorithm for Fast Mining of Spatial Co-location Patterns                                                

The bottom up approach of co-location feature-based pattern recognition in spatial databases is replaced with top down. The presentation is clear and algorithms are described and testing is done using synthetic and real data obtained from authentic data bank US National Transportation Atlas Database with Intermodal Freight Facilities (NTAD_IFF)

The 24 references are sufficient  and quite connected.

There are  many typo and grammatical errors. It needs to be grammar checked.

The symbol pi  is never defined and pi is used and pr is not defined ( I think it is probability)

 

I have one concern, it is not this paper authors, but the “whole co-location community”. Apparently, they are redoing or recreating the work already done and published extensively.  This is part of spatial databases: One area is qualitative spatial reasoning, another area is in Graph theory that deals with frequent subgraphs, maximal subgraphs - here you call them row instance patterns, cliques, maximal row instances, association rules, interX, intraX co-location, RIT, candidate co-location, apriori, pruning etc. They have used the same terminology with the exception:   Row instance vs  frequent subgraph.

 

So I find nothing new in this paper or everything is new.

Author Response

Please see the attachment.

Author Response File: Author Response.docx

Round 2

Reviewer 1 Report

I accept your response.

Reviewer 3 Report

Thanks for the revision.