The Arc de Triomphe, Wrapped: Measuring Public Installation Art Engagement and Popularity through Social Media Data Analysis

Round 1

Reviewer 1 Report

The work aims to improve understanding of how a popular art installation elicits emotions and online interactions. The article analyses textual user-generated content from Twitter and Instagram to achieve this aim. The authors set forth to answer four research questions by leveraging popular classification algorithms. Overall the article is well organized, defines the research questions, and addresses a topical problem regarding public art experiences and engagements via social media. Furthermore, the article mainly uses CLI tools accessible to non-technical experts (e.g. Digital Humanities students). 

Nevertheless, there are numerous critical limitations that I list below. 

General Limitations. 

- Authors do not discuss related and prior works that use similar research methods (i.e. ML and social media analytics) to address similar questions. Therefore, there is no baseline to help readers understand how this work advances the field. 

- Authors provide minimal and poorly documented details about the data collection steps.

- Authors have not shared a data repository, so others cannot reproduce their analysis and findings.  

- The use of machine learning methods is not justified, and prior works using similar approaches are not discussed. Currently, the wording of the research questions strongly indicates that a statistical approach (e.g. t test, Pearson's r, chi square) would have been a more appropriate solution. Please update the RQs and improve the scope of the article to make a strong case for the ML experiments. 

- No background information is provided about the APIs to help readers and non-expert understand the full range of data one can extract from Instagram and Twitter posts.  

- No background information is provided about the user base of these two social media platforms. Prior research indicates that different demographic groups use the two platforms. 

Methodology 

Line 235 – please remove the statement about using cutting-edge methods since you mainly use CLI tools (e.g. Instaloader and Twarc) and lexicon/rule-based sentiment analysis tools. 

Explain what data these CLI tools extract and how (i.e. the API methods used). 

Line 248 – Be more specific about what “data” you are processing with OpenRefine. 

Line 253 – Unclear what JSON keys or csv/dataframe columns were stored as categorical data types. Also, it is unclear what data was transformed to “numbers”. Do you mean that you one-hot encoded posts and comments? 

Lines 264:269 – The fact that VADER is detecting only positive sentiment indicates an error. Did you try to test your data with another tool? Please cite the version or deployment of VADER that you are using. Does VADER even detect sentiments like triumph, attraction, and surprise? Unclear what OpenRefine does. Please provide details. 

Lines 271:286 – Move this paragraph to the methodology section. Please explain how you searched and collected the hashtags and how you used these hashtags in your analysis. Also, please explain how exactly the CLI tool you are using retrieves these 7,078 Instagram posts and 3,776 Twitter posts. Are you getting historical or streaming data? Are you searching the API using specific keywords, and how could you define the custom time range for the search?

Author Response

Please see the attachment

Author Response File: Author Response.docx

Reviewer 2 Report

1. This paper analyzed the social media data on the Arc de Triomphe and compared the data before and after wrapping. The paper provides relevant and useful information for further studies and for practical applications in this field.
2. The paper is generally written well but does not meet the publication standard. There are many grammatical errors and typos. Some examples (not all) are given here for authors’ reference: in Line 57, “cities improves…” should be “improve…”; Lines 248, 255 and other places, “Error” information should be fixed; etc.
3. Some references are not mentioned in the text; such as references 2, 6, 21, 22, 31, etc. In addition, please put the reference numbers in a bracket.
4. It is not clear what data are used for training in the supervised learning algorithms and what criteria are used to evaluate the performance of the algorithms.
5. It is recommended for the authors to carefully proofread the paper before the next submission.

Author Response

Please see the attachment

Author Response File: Author Response.docx