Between Social and Semantic Networks: A Case Study on Classroom Complexity

Round 1

Reviewer 1 Report

The paper is very well structured and written.

Both the introduction and the theoretical background include relevant and actual literature.

The article presents an innovative methodology to gather information from interpersonal and representational domains. This is the main positive aspect of this research.

 

 

 

Author Response

Dear “Reviewer 1”,

 

Thank you very much for your time to reading it and for the comments.

Reviewer 2 Report

The authors investigated the interplay between social contacts and knowledge structuring within a classroom setting, interpreted as a complex system. To this aim, the authors constructed conceptual networks of word co-occurrences on five topics and self-reported networks of social ties, then using PCA for investigating relationships between the two considered network structures. This represents a very interesting approach, addressing at the same time the cognitive and social aspects of classroom settings by using single-layer networks. The work is theoretically grounded and the authors took great care in explaining the relevant literature from the social sciences and education research. The authors have to be praised for the transdisciplinarity of their approach, spanning Education Research, network science, computational social science and psycholinguistics.

I highly appreciated the manuscript and would have a few suggestions that could be useful for further improving it, hence my recommendation for minor revisions. Let me underline that I consider the manuscript already of high quality.

In the cognitive sciences, word evoking is a technique known also as free associations. De Deyne et al., Behav. Res. Meth. (2013) outlined a continuous free association game where an individual could report up to the 3 first words coming to their mind when reading a given cue word. Although the authors used 5 rather than 3 as a limit in their analysis, they could refer to De Deyne's paper for the cognitive science audience, underlining also that the networks built here are not "standard" free association networks, i.e. there is no link between the cue and the evoked words, but rather co-occurrence networks, built according to Russell's scheme. It is important to underline this difference because free associations do not display network cores but rather tend to glue together different semantic clusters, whereas word co-occurrences can give rise to highly clustered network regions (see also the different semantic network layers in Stella, BDCC, 2019 and the work by Kenett et al., PNAS, 2018). In the analysis of the "Physics" conceptual network, the authors identify strong relationships between physics and equations or other mathematical jargon. This is in agreement with the approach by Stella et al., Plos ONE (2019), who reconstructed the students' mental perception of physics through free associations. The authors might briefly reference this work in their Discussion for additional comparisons with previous results. In the network analsysis, what do the authors mean by "closest links"? Do they mean that in terms of closeness centrality or edge betweenness? For future research, would a multi-layer network approach be viable? Multi-layer networks mix together different types of networks made of potentially different nodes, see also Kivela et al., JCN (2014). One layer could be a social one, another layer could be a semantic one and inter-layer links would connect together words elicited by specific actors. This point could relate to the difficulty outlined in the conclusions. The level of detail of the authors' writing is impressive, providing several useful references for the journal's audience. Unfortunately there are several typos and language issues, which complicate the interpretation of a few passages. I would recommend a careful revision of the text. Here are a few typos I spotted:
shade some light -> shed some light - Page 5
have had follow -> followed
can not be cleared detached -> cannot be clearly detached
words networks -> co-occurrence networks
would must to be produced -> must be produced

Author Response

Dear “Reviewer 2”,

thank you for the time and for the careful reading of our work. All comments given were very enlightening. That helped us to see new possibilities in our research. We toke in consideration all given suggestions and we tried to embrace them in the best way.

Below a brief detail of the modifications we did, upon your suggestions.

--> Reference De Deyne et al. (BRM, 2013) was included, pointing the difference between theirs and our networks (Lns 212-216).

--> Declaration that there is no directional link between the cue and the evoked words was included (Lns 236-237).

--> References Kivela et al., (JCN, 2014) and Stella and Kenett (BDCC, 2019) and declaration of possibility to use multilayer network perspective as further research steps were included. (Lns 210 and 447).

--> Reference Stella et al., (PloS1, 2019) comparing our findings to their results was included (Lns 371-372).

--> Term “closest relation” was modified to “closest words” and “closest pairs” in order to avoid double interpretation. In our networks, proximity between words accounts for Russel’s co-occurrence similarity index.

--> Pointed typos and others were revised.

Reviewer 3 Report

The manuscript addresses an important topic of how in learning science the content of learning and social relations both affect the learning situation. The authors introduce an approach based on combination of analysis of sematic and social networks. The theoretical underpinning how to conceptualize the learning situation is refreshingly original combination of Chavellard’s views about didactic transpositions and Moscovici’s views. The way the views are blended, I find very interesting and well suited to conceptualize the problem at hand. This is good example, as I think how theory is fitted to phenomenon to be investigated instead of fitting the phenomenon to pre-exiting rigid theoretical framework (which, as I think, too often happens in science education research). The chosen method for analysis, which combines sematic and social networks, is also well designed to cohere with the theoretical background. These are strengths of the present manuscript.

I have some minor notions and suggestions authors may consider.

The use of term ”unities” may need some explanation. Why “unities” instead of “unites”. There is difference in meaning, so preference for somewhat lesser used expression “unities” raises expectations of some implied reasons for this choice.  

In line 20, it might be good idea to connect the first paragraph to paragraphs defining complexity and network. In First paragraph, the need for suitable methods is raised. Why not continue like “Complex networks provide the methods....” and then define complexity and networks, as authors have done (not in separate short paragraphs). At least for me, that would increase the flow of text.

In general, in Introduction, the flow and fluidity of the text may be improved by considering joining some of the very short paragraphs, which in my opinion, are nevertheless so much about same theme that joining is possible.

In lines, 48-50 authors tap on very important issue; should we considered certain phenomena as separate or entangled. The allusion they make that sometimes certain phenomena are consider as “ontologically” separate, may too often reveal itself rather as a methodological constraint than something inherent in the phenomenon itself.  Because this is important point and argument for using complex systems perspective and complex networks, and because authors seem to be well aware of this, I urge them to expand this point.

In section 2, authors introduce the Chavellard’s view on didactical transpositions. I find this very welcome, since this viewpoint has received in Anglo-American science education research much less attention that it deserves, also, in German didactic tradition it is still less appreciated than e.g. Duit’s didactical reconstruction. However, as the authors show, Chavellard’s views have many strengths that should be better know.   

In paragraph starting on line 79, where Moscovici’s views are outlined (which, again, I find very welcome and good choice, enriching the viewpoints utilized in science education research) authors could again consolidate the text by joining short paragraphs to larger ones; after, the theme remains broadly the same. This is of course a matter of taste, but at least, authors could consider such changes throughout the manuscript.

The way the authors’ couple the Chavellard’s and Moscovici’s views to construct theoretical underpinnings for their work, I find really well motivated and constructed. This also fits nearly seamlessly to the way the sociosemantic network approach can model the situation. To me, this seems to be the most important innovation and idea in the manuscript, and thus, I think it could be already mentioned  more explicitly in the introduction.

In Figure 1 authors show schematically how they have conceptualized the phenomenon they approach by using sociosemantic networks. In that, question arises about the role of teachers and how it is taken into account. It appears that while other aspects are quite thoughtfully taken into account, the role of teacher is rather disconnected. I think authors should say more about this and explain in more detail how they conceive the role of teacher. I think that it is justified choice to leave teacher’s role with less attention and there is no need to modify the analysis, but something must be said about the choices.

In section 3.2.1. the authors explain how they have constructed the semantic network. It uses the wor-association type methods. Here, I think, a bit more could be said how word associations can be assumed to be related to semantics. For example, two recent works by Stella et al might be useful in this:

Stella, M.; De Nigris, S.; Aloric, A.; Siew, C.S. Forma mentis networks quantify crucial differences in STEM 422 perception between students and experts. PloS one 2019, 14.

 

Open Access Stella, M (2020) Forma Mentis Networks Reconstruct How Italian High Schoolers and International STEM Experts Perceive Teachers, Students, Scientists, and School. Educ. Sci. 2020, 10(1), 17; https://doi.org/10.3390/educsci10010017

 

In general, I find the methodological part very carefully constructed and completed. The way it allows to combine the semantic and social aspects in learning settings is very well motivated and to my understanding, also novel idea in many respects, especially that it opens way to explore the interaction between social and semantic development. Here, discussion could be strengthened by discussing at appropriate place paralleling attempts (although not in context of learning) reported in:

 J Saint-Charles & P Mongeau. Social influence and discourse similarity networks in workgroups. Social Networks 52 (2018) 228–237.

 

In line 270, authors introduce “word’s similarity network”.  I would say rather “proximity” or “relationship” based on Russell’s similarity index. Similarity, I think, should be reserved for comparisons of neighborhoods of words or networks. Of course, if authors like to use the condensed expression, it is OK but way to use the expression “similarity” here needs to be clarified before figures are introduced.

Lines 271- 284 provide essentially summaries what is contained in networks. The readability might be improved by putting this information on Table instead of listing it in the running text. Then, running text could give a condensed overview of importance of the highlighted parts in Table.

In Figure 5, authors may consider explaining in more detail what is shown in adjacency heat map. It might be that not all interested readers are familiar with heat maps. In addition, the plots that show distributions could use larger fonts for better visibility. Here, as in case of Figure 4, the opening line of the paragraph could explain a bit more what is shown in Figure (although this is explained in methods section, it is not bad idea to repeat some points to remind the readers).

 

The Discussion and Conclusion parts are well aligned with theoretical underpinnings and the importance of results is discussed clearly. I would, however, suggest that authors take a look on the Stella’s work I mentioned above. I can see here many connection points to that work. Considering it would enhance the usefulness (which already is very high) of the work.

In summary, I find the research very insightful and novel, and I am convinced it is very important contribution to science education research and has much potential to open up new type of methodological approach. In principle, the manuscript is publishable as it is but I hope that authors take time to consider my suggestions and possibly expand the discussion in some places.  

 

Author Response

Dear “Reviewer 3”,

thank you very much for the reading and for all inputs and all constructive comments. All comments were important not only for refining the text but also for rethink several points in our research. We appreciate the presented references and we worked our best to embrace all suggestions.

Below a brief detail of the modifications we did, upon your suggestions.

--> the term “unities” was used as plural of unity, refereed to elementary entities of a complex system. The second phrase of the opening paragraph of the paper was modified to make it more explicit.

--> A phrase was added (Lns 22-24)

--> Paragraphs were joined (Lns 20-29)

--> Paragraphs were joined (Lns 30-38)

--> Although the last paragraph of Introduction is tiny, we opted for leaving it this way in order to highlight the aim of the work.

--> An argument for using the complex network approach was included (Lns 49-53)

--> Paragraphs were joined (Lns 86-94)

--> Teacher’s role and research choice were included (Lns 149-158)

--> Paragraphs were joined (Lns 161-169)

--> Word associations and semantics: suggested references included and the paragraph was rewritten (Lns 207-216)

--> Saint-Charles and Mongeau reference was included in the same paragraph where we present other works combining social and though domains (Lns 60-64)

--> Word’s similarity network replaced by Word’s co-occurence network in figure 2, in figure 2 caption, in line 291 in figure 4 caption and in line 305 “similarity” was deleted.

--> Term similarity was clarified before figures (Lns 211-212).

--> Table suggested was inserted between lines 283 and 284.

--> Opening phrases for figures 4 and 5 were expanded (Lns 292 and 309).

--> Fonts in figure 5 were enlarged

--> Heat map was replaced by heatmap

--> Heatmap was explained (Lns 313-314)

--> Stella’s works were included in Discussion (Ln 372) and in Conclusion (Ln 447) sections.