Dual-Channel Heterogeneous Graph Network for Author Name Disambiguation

Round 1

Reviewer 1 Report

The paper proposes a methodology about author name disambiguation. The subject of the paper is very interesting and related to the journal goals.  On the other hand, authors should correct the following in order their paper is technically sound.  

 It would be very useful the authors to insert a link in footnote when they mention DBLP, IEEE and ACM in order the reader can see those databases.  In line 15 it would be nice if the authors clarify the number 2.5 if they refer to all the publications in DBLP because in https://dblp.org/statistics/recordsindblp.html the number is close to 6 million publications and also mention the source of the 2.5 million they refer. Moreover, in line 16 I suggest the authors mention the exact rapid rate at least for the 3 years so the reader understands the growth.  In line 18-19 the authors should correct the phrase use the same name to share the same name. Moreover, the authors add an example with the shame name in dblp digital library and explain. In line 29 please insert reference to related works. In line 30 please correct the word higher to high and also insert reference for the statement. In line 34 please analyze the methods that use heterogeneous graphs and insert bibliographic reference.  In related work it would be very useful to provide a literature review about the related bibliography on the problem of name ambiguity in general and conclude the reasons the authors choose that method they propose. In the Related work section the authors should remove the word reference or references and replace with the surname of the first author for example for the reference with number 1 in text should appear Shi et al. and et al. in italics (et al. is the abbreviation of the phrase and others).After the replacement of the words reference or references the authors should rewrite paragraphs 2.1 and 2.2 in order to be suitable for academic paper. The authors should improve generally their references and replace phrases like academic papers or other with the last name of the first author and et al. in italics or rephrase the sentence and insert the number in the end.  I suggest the authors to read more academic paper to understand how to write the review. In section 3 please replace the title with a title that will describe better the section such as materials and methods or problem formulation.   Moreover, the authors use the term Graph Neural Network, term is related to Neural Network, but in the paper is not clear if the authors used Neural Networks so if the researchers used Neural Networks they have to mention more details in paper or replace Graph Neural Network with Graph Network.  

The authors propose a methodology about a name disambiguation, but it is not clear if the method is about name disambiguation between authors in different organizations or disambiguation in the same organization. Since in definition 1 includes author organization, so in definition 2 if a paper does not belong to one author means that the same belong to the same organization since the papers are grouped based on organization. So the authors should clarify that point.  In addition, it would be very helpful in the definitions the authors mention an example and analyze in order the reader fully understand the propose methodology.  It is vital the authors to give more technical details about the proposed methodology, describe for example if they used python or other program to create the graphs, how they collect the data and give the related code for the readers. Please give a link for fastText.  The authors should explain why the used Metapath2vec instead of Hin2Vec.  

Author Response

Please see the attachment.

Author Response File: Author Response.zip

Reviewer 2 Report

The article presents a novel approach for the disambiguation of author names based on the use of  Dual-channel Heterogeneous Graph Network graphs.  Graph networks have been used in the past to similar problems however the authors deviate from previous constructions by elegantly combining previously proposed solutions. The main idea is to use fastText to generate the semantic vector that represents each paper, then employ the heterogeneous graph to generate the paper relationship vector and finally, the semantic similarity matrix and the relationship similarity matrix are merged, and clustered by the DBSCAN.

The presented idea is interesting though it is mainly a combination of previous schemes and such that the novelty seems to be limited. However, the combination seems to be non-trivial and the fact that the experimental results depict superior performance designates that the article should be accepted.  On the other hand, it could be interesting to see how other competitive algorithms would perform in similar datasets, either be extra experiments or by a theoretical discussion (that will in essence explain why the specific alternatives were chosen as baselines and not others). Henceforth my proposal is to accept the article subject to revision, where the revision refers to an extra comparison of the work with similar works in the literature. This comparison can be only theoretical justifying the choice of the specific baselines that the authors use for the experimental comparison. 

Author Response

Please see the attachment.

Author Response File: Author Response.zip

Round 2

Reviewer 1 Report

The authors fulfilled all the necessary modifications.