Latent Multi-View Semi-Nonnegative Matrix Factorization with Block Diagonal Constraint

Round 1

Reviewer 1 Report

This paper deals with Latent Multi-view Semi-Nonnegative Matrix Factorization with Block Diagonal Constraint.

A new algorithm named latent multi-view semi-nonnegative matrix factorization with 5 block diagonal constraint is proposed.

Experiments on several multi-view benchmark datasets demonstrate the effectiveness of this approach.

 

 

The article is well written and interesting. However, it could be further enhanced, following the comments below.

First of all, the material and methods section should be explained in greater detail.

Moreover, I think the authors should provide further information regarding the method used.

Furthermore, the conclusions appear a bit poor and should be more thoroughly investigated.

 

Literature should be broadened. In particular, it is suggested to consider the following reference:

 

https://www.researchgate.net/publication/328266643_Cluster_Analysis_An_Application_to_a_Real_Mixed-Type_Data_Set

 

I encourage the authors to refine their paper to make it available for publication in the journal.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

SUMMARY

The authors propose a new multi-view clustering algorithm, the main features of which are a combination of latent representation learning and Semi-NMF, and a block diagonal constraint.

 

The manuscript provides a detailed description of the algorithm and justification of its performance. The used notation is clearly described, which enhances the positive impression of the paper. The results of testing the algorithm on a number of widely used multi-view clustering problems are presented. The results are compared with the state-of-the-art algorithms. It is shown that the proposed algorithm shows a higher accuracy of clustering.

 

COMMENTS

1. Perhaps, the authors should slightly expand the description of examples of multi-view clustering problems (lines 16-17).

2. Please, explain whether it is possible to modify the algorithm if the number of clusters is not known in advance?

3. The results shown in Table 3 are also recommended to be shown using visualization.

4. It is not entirely clear how to interpret the visualization (Fig. 4) before binarization.

5. Will there be a deterioration in the efficiency of the algorithm if the number of views is higher than 10?

6. The authors should add an analysis of the computational complexity of the algorithm: a comparison of its performance (speed and memory) with other algorithms, as well as the dependence of performance on the dimension of the problem (numbers of clusters, samples, features, views).

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 3 Report

In this paper,  a new algorithm named latent multi-view semi-nonnegative matrix factorization with block diagonal constraint (LMSNB) is proposed. The concept of this work is good. However, it needs improvement in terms of organization and also some technical details.

1- Abstract must be very professional. The novelty of the work must be clearly needed in the abstract. 

2- Simulation parameters are not defined correctly in the evaluation section, so looking forward to seeing these in the evaluation results part. 

3- I also do not see the finding and limitations of this work, so it is advised to add these details in the revision. 

4- Simulation results must have a detailed explanation for more understanding 

5- Related work must discuss the existing methods with their advantages and disadvantages. You can modulate the one para about existing limitations and proposed ideology. The author is advised to add a number of examples related to this manuscript as follows:

A) Rostami, et al, 2022. A Novel Time-aware Food recommender-system based on Deep Learning and Graph Clustering. IEEE Access.

B) Azadi et al,. Graph-based relevancy-redundancy gene selection method for cancer diagnosis. Computers in Biology and Medicine. 2022 Aug 1;147:105766.

c) Saberi et al,. 2022. Dual Regularized Unsupervised Feature Selection Based on Matrix Factorization and Minimum Redundancy with application in gene selection. Knowledge-Based Systems, 256, p.109884.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 4 Report

This paper proposes an algorithm named latent multi-view semi-nonnegative matrix factorization with block diagonal constraint (LMSNB). It first uses latent representation learning and Semi-NMF to get a lower-dimensional representation with consistent information from different views.  It also uses the block diagonal constraint and graph regularization to produce a new representation that captures the original data's global and local structure. The experiment was conducted on six multiview datasets to compare the performance of the proposed algorithm with the other four algorithms using some external clustering validation metrics.

In general, the paper is well-written and organized. It seems that the paper has received other reviewers' comments, and the authors have revised the paper. In this version, the authors should consider the following points to further improve the quality of the paper.

- The Introduction is quite long. It is better to move some paragraphs that review some related works. It is better to summarize the limitations of previous approaches and highlight how the current method can address these limitations.

- In addition, give some real-life applications of the proposed algorithm.

 

Author Response

Please see the attachment.

Author Response File: Author Response.pdf