So we cannot have one CSV file that stores network relations between the codes?
Finished reading. Incredible work!! I am impressed.
I fixed some obvious typos (like adding many unbreakable spaces) and fix some LaTeX stuff. I set the figure to be much larger (I did not check if space is limited but subsection “Comparing reduction techniques” is a duplicate from a couple of paragraphs above). I also put many comments and suggestions.
Hello @rebelethno. This is to confirm that @bpinaud and @melancon (thanks!) have submitted the latest version of this paper to *Applied Network Science". Here: Overleaf, Online LaTeX Editor
I propose to publish the preprint. What do people think? And: is it possible to publish it on Zenodo instead of the Springer preprint repo?
3 Likes
Why not aim for ArXiV instead? I understand Zenodo hosts datasets (associated with scientific publications it’s true).
Two advantages to Zenodo:
- Indexed by OpenAIRE. EU grant information is a database field, even.
- Supports versioning. Each upload version gets a version DOI, and there is an overall DOI thart resolves to the latest version.
It’s true that maybe the advantage of using the Springer preprint is that the article gets assigned a single DOI with a similar method: when the article is published, the DOI resolves to the journal version. This last is just a conjecture, I have no information.
The paper just passed through the initial technical checks. It is now under peer review. On the journal website, I see that the median speed for a first decision is 61 days (or 70, I am not sure of the number to use).
1 Like
Good point. HAL (the French research publication archive) also has a grant type/number field. But it does not host datasets.
1 Like
And arx.iv does not, right? Between HAL and Zenodo I prefer the latter, it’s more EU- (via OpenAire) and global- (via CERN) centered.
Hi everyone,
Publishing the pre-print before acceptance is not something that is regularly done in my field (people upload pre-prints on their personal websites, Academia.edu pages etc. in order to get paywalled articles out after they are published) so I don’t really have thoughts or an opinion about it because this I am not familiar with the conventions / the practice of doing so.
Based on this, 5 weeks. Elsewhere @bpinaud remarked about 60 days between submission and first evaluation.
1 Like
Hello everyone. I hope you are all great. I have a question. Until the approval of the reports by the European Commission (or who), the reports (final deliverables) are not public, right? I have put them as an output in my university’s evidence of publication and academic outcomes, but as a non-public report. Take care.
As far as I know, reports are public. If the Commission requests changes (unlikely), you can just make another version, itself public. The ethno report you were a part of is already public, with well over 3,000 views.
Good news, @icqe22_authors. Our submission to Applied Network Science received a “revise and resubmit”. The reviewers comments are copied below. The deadline for resubmission is April 17.
We are meeting this morning with Guy and Bruno, since most comments are network science related. We might ask @Nica, @Jan and @Richard for some edits to the paper, especially around the comments of rewiever 2.
Reviewer 1
This paper explores the results of three main methods for network reduction on codes-co-occurrence networks: (1) suppression of edges based on their weights, (2) suppression of edges based on how well their endpoints are part of k-cores, i.e., how well they are connected to other nodes, and (3) extracting the Similian backbone which uses triangle count to measure edge importance.
The reduction methods are simple basic ones. The contribution is more on the interpretation of the results on the specific dataset used by the authors. The paper is coherent, well-structured and nicely written.
Even if I agree with the authors on the fact that visualization is important for human understanding of complex systems, one cannot rely on visualizing reduced networks to draw exact insights from the network. For example, for studying which codes are mentioned together most often it is more interesting to generate embedding for the nodes of the graph, using for example a deep learning approach, and then cluster these embedding. Node embedding may capture more insights on codes than visualization.
Contrarily to the affirmation of the authors about the availability of their data, the CCN network is not available. The links provided by the authors do not provide these networks but just related data that do not seem complete. For example, I have not succeeded to find “contributions” and “informants “ in the PROPREBEL dataset. Could the authors provide the CCN networks they constructed and the code of their algorithms, in a single location, so as to reproduce their results?
Reviewer 2
The paper serves as an intersection between network science and sociology. It addresses a meaningful and relevant question: how to properly choose network science (or statistical in the broader sense) tools without posing epistemological challenges. The authors propose four network reduction techniques to analyze how each relates to social science, utilizing three real-world datasets. The authors construct a co-occurrence network of words to understand cultural relationships. The paper clearly addresses the problem and shows how each reduction technique could be semantically related to approaches in anthropology and indicates how qualitative researchers would regard different tools for network reduction in their daily analysis. The paper is easy to follow and well-written. The results are sound and properly presented. However, there are some points of concern that should be addressed to improve the manuscript before recommending publication:
-
In the subsection “Data and pre-processing”, authors are encouraged to give a visual example of contributions, annotations, and codes to decrease ambiguity.
-
In the section “Techniques for network reduction,” authors are encouraged to provide more relevant work on network reduction techniques:
- Zeng, A., & Lü, L. (2011). Coarse graining for synchronization in directed networks. Physical Review E, 83(5), 056123.
- Coscia, M., & Neffke, F. M. (2017, April). Network backboning with noisy data. In 2017 IEEE 33rd international conference on data engineering (ICDE) (pp. 425-436). IEEE.
- Zeng, L., Jia, Z., & Wang, Y. (2019). A new spectral coarse-graining algorithm based on K-means clustering in complex networks. Modern Physics Letters B, 33(01), 1850421.
- Rajeh, S., Savonnet, M., Leclercq, E., & Cherifi, H. (2022, February). Modularity-Based Backbone Extraction in Weighted Complex Networks. In Network Science: 7th International Winter Conference, NetSci-X 2022.
-
The statement “We do this by computing a similarity statistics between the maximal interpretable reduced networks that obtain from applying the different techniques,” do the authors mean that are obtained?
-
When authors suggest that the reduction technique must not foreclose the possibility of updating via abductive reasoning, authors are encouraged to state why it is crucial in their case studies to have
user-defined parameters. In truth, some researchers state that a user-defined parameter in the reduction method is a weakness because it leaves a possibility for biased or inaccurate decisions. -
The statement “they use in a natural way research data” is not clear.
-
In paragraph “With that in mind, we turn introducing our candidate techniques. We claim that … how useful the visualizations they produce are, and how intuitive the method of building them is to ethnographers,” authors claim that they to satisfy conditions 3, 4, and 2. Then they state that their discussions focus on conditions 1 and 2. What is the reason for focusing on these conditions, and not all if they claim a good CNN must respect all conditions based on reference [31]?
-
In subsection “association depth,” authors refer to the result in Figure 1a 1b. However, figure 1 is a demonstration of a co-occurrence edge. Authors probably refer to Figures 2a and 2b.
-
Authors are encouraged to define what the abbreviation SSNA means.
-
Authors are also encouraged to provide an analysis that is not related to religion and investigate how their proposed reduction techniques compare in the application section.
-
More clarity should be given to “Association breadth” when introducing it. It is unclear in terms of its explanation. Authors refer to it as the same two codes occurring over multiple contributions of the same informant. However, it also can be interpreted as given a link between code1 and code2; if five informants use code1 and code2 simultaneously in their contribution, the weight will be five.
-
Using sometimes double open/closed quotations and other times open/closed single quotations. Authors are encouraged to unify their representation.
-
This application is not meant as a full methodological primer. Rather, it means to be a “proof of concept”, and show… → and shows
1 Like
And another proposal for @icqe22_authors: I propose to change the title. Reasons:
- The term “reduction” is misleading for NetSci readers (they think of dimensional reduction). We saw this in the reviewers comments.
- We used this title in two conferences, with published abstracts, proceedings etc.
- Get a funkier title!
So, how about:
Lévi-Strauss in the network: comparing techniques to VERB networks of codes co-occurrence in ethnography.
Yes, it is unfair to the other anthropologists and social scientists cited.
VERB should not be “reduce”. It could refer to becoming lighter, like “trim” or “streamline”. Another possibility is simply “visualize”, but again I fear it might lead that community to thinking about layout algos. Any suggestions? @Richard, you might be the only native speaker of English among us, any idea?
Refine?
2 Likes
How about “distilling” which means “to extract the essential meaning or most important aspects of”
2 Likes
cut out?
How about:
Operationalizing Lévi-Strauss: four techniques of streamlinig networks of coocuring ethnographic codes.
Verb needs to be in a gerung form, I think. @Richard
1 Like
I like Levi-Strauss in the Network because it’s playful, but I think Operationalizing Levi-Strauss will make it seem like the paper is really focused on L-S above all else.