Unsupervised Predominant Sense Detection and Its Application to Text Classification

Round 1

Reviewer 1 Report

Authors proposed a new method for identifying domain-specific word senses for unsupervised predominant sense detection and successfully evaluated their DSS model in text classification task.

There are a number of comments on the article:

1) It is not clear why the section "Related work" has number 5. It should be placed after section "1. Introduction”, because it is the basis for substantiation the novelty and practicability of developing the method proposed in the subsequent sections of the article.

2) In addition, in the "Related work" section, there are no references to scientific publications of 2019-2020, while the topic of text classification is a rapidly developing area in Natural language processing. For example, resource https://paperswithcode.com/task/text-classification demonstrates big amount of recent publications on the topics of text classification. Authors should expand their «Related work» section.

3) In the table 4 the number of CNN model output categories is not specified, as well as the shape of input tensor.

4) In the figure 3 “Percent” label is recommended to replace with “Ratio of topmost of senses, %”.

5) It is interesting to see information about the inference time of key parts of the DSS framework developed by the authors.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Reviewer 2 Report

This paper describes a method of sense attribution based upon Markov Walks, that is overall able to detect the predominant sense of words in specific domains in an unsupervised manner.

The paper is pleasant to read and qualitatively good. The experimental results are presented well, and easy to access. I recommend some further references to works regarding automated classification of documents, in particular some recent investigations, such as, for instance, 

@article{Cristani201867, Author = {Cristani, M. and Bertolaso, A. and Scannapieco, S. and Tomazzoli, C.}, Document_Type = {Review}, Doi = {10.1016/j.ijinfomgt.2018.01.010}, Journal = {International Journal of Information Management}, Pages = {67-75}, Source = {Scopus}, Title = {Future paradigms of automated processing of business documents}, Volume = {40}, Year = {2018}}   I recommend proof reading for some typos (not significant) and adding the above reference.

Author Response

Please see the attachment.

Author Response File: Author Response.pdf

Round 2

Reviewer 1 Report

All comments have been corrected by the authors. Therefore, I think that the article can be accepted for publication.