My superb colleagues have built ROLF, an exemplary GUI that demonstrates among other things the usefulness of explanations (https://rolf.scalac.io). Without installing the package, you have quick access to the default pipeline that the absa.load function returns. The backend solely wraps up a pipeline into the clear flask service which provides data to the frontend. Write your text, add aspects, and hit the button “analyze”. In a few seconds, you have a response, a prediction together with an estimated explanation.
On the right-hand side, for any given aspects, there is the overall sentiment, and below, the sentiment within independent spans (in this case, sentences that come from the sentenzier). Click on a span, and the inspect mode window pops up. This is a visualization of patterns that come directly from the basic pattern recognizer. Review an explanation by clicking on different dots (importance values of patterns).
Please be aware that the service runs on a CPU “minimal-resource” machine, therefore, the inference time is extremely high compared to a service working on well-adjusted modern computational units. The adjustment is straightforward having once defined the service requirements.