Moved to a new topic for improved readability. See reviewers comments here.
Reading the first paper now. It is clear that semantic network analysis as per Jiang et. el. is closely related to semantic networks as studied in Guy @melancon’s own work on, for example, analysis of news coverage in France. When @brenoust developed his notion of entanglement, he was working on such networks.
However, the framing and the language here look quite different. For example, if you consider the “Research procedure” section (p. 3717-3718) it slips from “most occurring words” to “co-occurring concepts”. I have to re-read several times to make sure I know what I’m looking at. Additionally, there is the puzzling move to throw away everything except the 150 highest occurring words, which appears to destroy plenty of information without any consideration of how this would influence results. This might be because these authors, unlike Guy and his colleagues, do not like to deploy math to deal with large networks. The semantic networks that Guy showed me are very large. This seems corroborated by sentences like:
This study used eigenvector centrality as the criterion measure because it indicates a word’s overall network centrality.
Which looks imprecise to me. There is no such thing as “overall” network centrality!
Another important difference is whether analysis considers the keywords associated to news pieces or their text. The reference to “stop words” in the paper seems to point to the latter method, with some simple NLP being deployed at step 1.
Am I wrong? Are there two separate traditions of building networks from word co-occurrence? If so, we should highlight it, writing a “related work” section of the paper that would be structured like this: