Dear @mpbraendle and community,
Thank you very much for taking the time to test the IssueSpotlight plugin in CHIMIA and for sharing such detailed feedback. It genuinely helps improve open-source projects 
Regarding your conclusion, I respectfully disagree with the statement that the plugin “degrades the quality of a carefully edited issue.” Its purpose is not to replace human editorial work, but rather to provide an additional and OPTIONAL layer of semantic and visual exploration. I must admit that describing the plugin as something that “degrades quality” feels unnecessarily harsh and somewhat unfair to the spirit of the project itself (open source, collaboration, sharing, etc.).
We have implemented it in several other journals where editorial teams are very happy with the new perspectives it offers their readers, although I completely understand that every journal has its own editorial line and specific needs.
Design / Technical Aspects
-
Hardcoded Spanish terms: You are absolutely right. The plugin relies heavily on the OJS translation system (locale keys), but during development a few terms ended up being written directly into the code. Thank you for pointing that out.
-
Bug on line 233 (IssueSpotlightPlugin.php): Thank you very much!! This is actually a very valuable contribution. The strict comparison (==) can fail in certain OJS ecosystems using custom themes. In my own installations, and in the cases where the plugin has been tested so far, it works correctly because of the template structure we use. However, you are absolutely right that using a regular expression makes the implementation more robust. We will implement it.
-
Author list ordering: At the moment, no specific sorting criteria are applied. The plugin retrieves authors exactly as they appear in the database (following the order of appearance in the issue’s articles). Thank you for the suggestion.
AI Analysis
1. Editorial Summary
The plugin does not aim — and never will aim — to replace a human editor. If your journal already has a dedicated editorial team writing contextualized, thoughtful, and well-crafted editorials, then you certainly do not need this plugin.
However, the reality of the OJS ecosystem is that many journals simply do not have the staff, time, or resources needed to manually create an editorial summary for every issue. For those journals, having a basic AI-generated summary (which does not necessarily mean “bad,” as you suggest) is far better than having nothing at all. Editors currently using our plugin have told us that it is fairly reliable and fills this gap quite effectively.
Regarding your concern about “quite inaccurate” AI-generated information and how to control it:
First of all, this is the first report we have received involving significant inaccuracies. The model generates its responses strictly from the article abstracts and metadata provided to it. As with any LLM, small hallucinations are always possible, but the summaries should not be fundamentally incorrect. This is precisely why the plugin explicitly displays a notice informing readers that the content was generated by AI.
It is also important to consider the underlying technology. IssueSpotlight was designed to be free and accessible. It uses Gemini Flash Lite models, allowing journals to use AI at zero cost. Naturally, the reasoning capabilities of a free model cannot be compared to high-end premium models such as Opus or GPT-5. However, for a cost of €0, the accuracy-to-cost ratio is extremely competitive and more than sufficient for the needs of the vast majority of our users. As I mentioned before, if a human editorial team is already doing this work properly… then this simply is not the right plugin for you.
2. Global Map
You mentioned that, in the case of your journal, the map is usually centered around Switzerland. You are absolutely right: if all or most authors come from the same country, the map adds limited visual value. However, many journals have a highly international scope (for example across Latin America and Europe). For those journals, being able to instantly visualize the global distribution of contributions is extremely useful for readers. If a journal has a purely local scope, the visual impact is naturally less compelling.
Regarding your suggestion of adding links from authors directly to articles: I completely agree. It is an excellent improvement that has already been suggested before, and it is on our roadmap for future versions.
As for drawing collaboration lines between institutions: we actually tested that idea, but the visual results became quite cluttered and did not provide a good user experience. That said, if someone feels inspired to submit a pull request
3. Innovation Radar
You mentioned that you do not see anything in the radar that could not already be inferred by reading the abstracts yourself. Exactly — that is the whole point! If a reader has enough time to read all article abstracts in an issue and manually identify underlying trends, human analysis will always be superior. The plugin exists to provide that overview instantly and at a glance for readers who simply do not have time to read everything.
Regarding the interactivity and the “forces” between the bubbles: the visualization is built using Highcharts’ packed bubble chart library. The interactivity (dragging bubbles around) is simply a built-in Javascript visualization feature intended to improve visual engagement. There are no sophisticated vector forces or confidence metrics behind it; the bubbles are simply grouped according to the categories assigned by the AI (New, Rising, Stable). Perhaps expectations should be kept a bit lower here
As with the map, adding links from these concepts directly to the corresponding articles is an excellent idea for future iterations.
Finally, you pointed out that the keywords identified by the AI differ from those provided by the authors. We actually see that as a positive result. If all we wanted was to display author keywords, then a simple traditional word cloud would be enough. By allowing the AI to analyze the abstracts more freely, it can extract emerging concepts, n-grams, and underlying trends that authors may not have explicitly tagged themselves. Editors currently using the plugin consider these AI-generated trends to be highly accurate and insightful.
4. SDG Impact
Regarding SDG impact: first of all, please note that the AI does not read full PDFs or search for “SDG logos” embedded within articles. It relies strictly on the metadata provided through OJS (titles and abstracts) to infer content. Likewise, when assessing SDG 5 (Gender Equality), the AI does not attempt to guess authors’ genders based on their names — which would be both technically unreliable and ethically problematic. Instead, it detects the research topic itself (for example, Women in Science).
Regarding the percentage distribution: the AI prompt is specifically configured to identify only the 4 to 7 most relevant SDGs within an issue. Therefore, it will never attempt to detect all 16 SDGs, but instead focuses the percentages on the most relevant ones. While this may not perfectly match manual human classification, the most prominent SDGs identified by the AI generally align with the main themes of the issue. Again, this is an approximation intended to provide value where manual tagging is not available.
Final Thoughts
I truly appreciate your contribution, your time, and your suggestions for improvement. However, I still believe it is quite unfair to evaluate the plugin based on the premise that it “degrades quality.”
You may like the plugin or not, and you are completely free to use it or disable it. Given the impressive human expertise and editorial support your journal already has, it is clear that this plugin was not designed for your specific context. If your team can manually align SDGs, write contextualized editorials, and identify trends better than AI, that is fantastic.
IssueSpotlight was created for a different audience entirely: the many journals that do not have that level of support and that can genuinely benefit from a free, automated tool to enrich the reader experience. And of course, if someone feels the results are not good enough, they are perfectly free not to use the plugin.
Thank you again for your feedback.
Best regards,
GitHub Repository: IssueSpotlight AI