It's a little solution to compare some museums with each other. It is implemented by aggregating museum reviews, determining them sentiment and extracting keyphrases.
Pitching presentation
Example of output of museum review:
For sentiment determining we use rubert-base-cased-sentiment model
For keyphrases extraction we use keyt5-base model
The easiest way to reproduce the solution is to open colab, switch on GPU mode, load museum_feedback.xlsx dataset and start all notebook.
You can run notebook local. To do this you should install all dependencies, like pandas and torch. And probably you'll need GPU to run this code.
We collected the dataset manually from the Yandex maps website
Dataset stored in data folder and look like this:
| museum_id | feedback_text | visitor_mark | museum_name |
|---|---|---|---|
| 1 | Отличный музей, ... | 5 | Музей истории Томска |