ArCAR: A Novel Deep Learning Computer-Aided Recognition for Character-Level Arabic Text Representation and Recognition

Round 1

Reviewer 1 Report

The manuscript is centered on a very interesting and timely topic, which is also quite relevant to the themes of this journal. Organization of the paper is good and the proposed method is quite novel. The length of the manuscript is about right but the abstract is a bit too long and it should be trimmed a bit. 

The manuscript, moreover, does not link well with recent literature on sentiment analysis appeared in relevant top-tier journals, e.g., the IEEE Intelligent Systems department on "Affective Computing and Sentiment Analysis". Also, latest trends in multilingual sentiment analysis are missing, e.g., see Lo et al.’s recent survey on multilingual sentiment analysis (from formal to informal and scarce resource languages) and Oueslati et al.’s review of sentiment analysis research in Arabic language. Finally, check recent resources for multilingual sentiment analysis, e.g., BabelSenticNet.

Authors seem to handle sentiment analysis simply as a binary classification problem (positive versus negative). What about the issue of neutrality or ambivalence? Check relevant literature on detecting and filtering neutrality in sentiment analysis and recent works on sentiment sensing with ambivalence handling.

The manuscript presents some bad English constructions, grammar mistakes, and misuse of articles: a professional language editing service is strongly recommended (e.g., the ones offered by IEEE, Elsevier, and Springer) to sufficiently improve the paper's presentation quality for meeting the high standards of this journal.

Finally, double-check both definition and usage of acronyms: every acronym should be defined only once (at the first occurrence) and always used afterwards (except for abstract and section titles). Also, it is not recommendable to generate acronyms for multiword expressions that are shorter than 3 words (unless they are universally recognized, e.g., AI).

Author Response

First of all, we would like to thank reviewer#1 for his insightful comments and suggestions to improve the manuscript. Really, the reviewer comments improve the manuscript and gave it more highlight availability. Thanks again.

Here, we have addressed the all points raised by the reviewer and tried our best to answer them.

Kindly, find the comment's answers as in the attached PDF file. 

Author Response File: Author Response.pdf

Reviewer 2 Report

 

The introduction lacks of focus. I would suggest to list the limits of the approaches published so far and how the authors designed their new method to overcome them.

section2.

• "Arabic text in nature is unstructured data and it must transform and represent to be more understandable for machine learning algorithms." What do the authors mean with "arabic text is unstructured data"? Is ENglish text structured data?
• "bic words, and they fail to capture the semantic dependencies of the words at the same time Different Arabic sentences may have exactly the same representation regardless of the words’ meaning"
  may be this is obvious to people who know the Arabic language, could the authors elaborate a bit more on this concept of ambiguity?
• page 3. I could not understand "They proposed new term weighting scheme called Term Class Weight-Inverse Class Frequency (TCW-ICF) to extract the most discriminating features from the input Arabic texts. Their approach was used to reduce the input feature dimensionality since their input dataset has more than 5000 Arabic farmer complaints. At the 155 end, they outperformed the TFIDF achieving overall accuracy of 85% [36]."
• Section 3.1 moves too fast across the applications of deep learning to arabic text. May be the authors should introduce concepts and algorithms a bit more slowly.
• page 4 "In 2004, Kordi ... to classify Arabic text documents classification". I did not understand.
• Section 3.1 Please provide a public link for each dataset or comment how data can be obtained by people interested in replicating the results.
• Section 3.2 I could not understand the "quantization" process. It looks like 1-hot encoding of the characters. What is the real difference?
• Secton 3.3. THere are several works on Western languages based on character occurrences rather than classical linguistics pipeline. The authors seem to suggest that English and other western languages have not been investigated with character-level processings.
• Section 3.4 It is not very clear what the two new datasets are. What are the hypertrainable parameters?
• Section 3.5 NB layers may be are the BN layers?
• 3.5 summarizes years of research in CNN in few lines. May be the authors could add references to the most important works.
• 3.6 how are the values like mini-batch size determined? What are the references after .. deviation of 0.01? Why shuffling guarantees that examples are only seen once per epoch? 
• Eqs 1--7: please center.
•  
•  

 

 

 

 

• page 1, line 37. What do the authors mean with 28 basic alphabets?
• page 2, lines 57-59: repetition of an entire sentence
• page 4, line 104: supper performance. Supper?
•  

Author Response

First of all, we would like to thank reviewer#2 for his insightful comments and suggestions to improve the manuscript. Really, the reviewer comments improve the manuscript and gave it more highlight availability. Thanks again.

Here, we have addressed the all points raised by the reviewer and tried our best to answer them.

Kindly, find the comment's answers as in the attached PDF file. 

Author Response File: Author Response.pdf