The link recommendation model for the add-a-link task continues to being deployed to more and more wikis. For this, @kevinbazira has trained the model for 297 wikis. However, from those only 278 passed the backtesting evaluation and were published T336927. We would like to increase the number of languages for which this model can be deployed.
There are 2 lines of work towards this goal
- fix the language-specific model such that it will pass the backtesting evaluation in additional languages.
- develop a single language-agnostic model. this will facilitate easier deployment and maintenance across languages.
The goal of this task is to scope in more detail the work that is needed for these two lines of work to assess the effort and feasibility.