Topic modeling
The topic export plugin automatically analyzes the text of the article along with citizens' comments and recognizes the relevant thematic categories from a predefined classification. The sections appear on the public page of the article, so that visitors can see at a glance the main issues of the consultation.
How it works Topic modeling;
1. The plugin is automatically activated every time a consultation article is published or updated.
2. The title and text of the article are collected. In a later development of the tool, it will be possible to collect feedback submitted by citizens.
3. The language of the content (Greek, English, German supported) is automatically located so that the analysis and the tags of the topics correspond to the language of the consultation.
4. Using a multilingual embeddings model (mpnet) a numerical representation of the overall content of the consultation is created.
5. The classification of the system includes around 100 pre-defined topics, organised into general categories (parents) and sub-categories (leaves), curated to cover topics of public consultation; each has its own numerical representation calculated in advance.
6. Using the cosine similarity and z-score normalization index, it is calculated how much the content of the consultation is linked to each topic of the classification.
7. Only subjects whose relevance exceeds a predetermined threshold are retained, and grouped under the general category to which they belong.
8. The result is stored permanently in the article and appears on the public page, just below the title and above the text of the article, in a compact format "General category: subcategory 1, subcategory 2", indicating that they have been generated automatically (Created with AI).
English