This is multi-layered CNN for text classification with hierarchical classes.
Classes structure:
[
Consumer Electronics | Tablets & E-Books | Tablets,
Consumer Electronics | Games, Consoles & Software | Game Consoles,
Hobbies and Leisure | Bicycles | Parts and Accessories,
Hobbies and leisure | Hunting and fishing,
...
]
The task was not only consider hierarchical structure of classes but also calculate accuracy on each level.
Level of hierarchy varies from 2-4.
Data was represented as title, description, price and category_id, where category_id is multi-labeled with "|" as separator.
>>> cnn_model = CNNClassification(df,df1,df2)
>>> cnn_model.predict() # see main() to understand how to interpret results
- Extend from 4-level to n-level
- Try to use stopwords