katana
katana copied to clipboard
Use Heuristic Inference to improve form-fill capability
Please describe your feature request:
Automatic form filling without context is a hard task. After implementing a series of robust standard rules, it would be interesting to investigate further strategies to infer the form category from the page:
- What is the topic of the page - semantical analysis (form to book an airplane, form to subscribe to a newsletter, login form, etc)
- Can we use prior knowledge to classify similar forms? (forms from web frameworks, from taken from snippets on the web)
- Form filling can be a multistep operation: we need an autonomous approach (eg. with a fitness function that rewards the most promising filling paths)