About PriBot

PriBot is based on the paper:

PriBot:
Answering Free-form Questions about Privacy Policies with Deep Learning
,

by Hamza Harkous (EPFL), Kassem Fawaz (UMich), Rémi Lebret (EPFL),
Florian Schaub (UMich), Kang G. Shin (UMich), Karl Aberer (EPFL)




How it works!

If people read all the privacy policies they face online, they will need 255 hours per year. Thus, it is rarely the case. With the emergence of the small screen and voice-activated devices, this problem is further amplified: privacy policies cannot be easily communicated in such channels. We propose to fix these issues with a conversational interface.

PriBot is the first question-answering (QA) chatbot for privacy policies. It takes a previously unseen privacy policy and uses it to answer, in real time with high accuracy and relevance, user questions that are posed in free form. It further simplifies the policy with high-level summaries generated from the legalese text. Its applications range from customer service settings to giving the customers a way to compare various providers based on a specific question. Its main advantage is saving humans work that that can take from minutes to hours.

Go chat with it!

Under the Hood

For the detail-savvy, here is what is ongoing under the hood.

Policy Segmentation

PriBot parses the HTML of the privacy policy in order to produce a set of candidate answers. Each of these answers is constructed to be coherent, i.e., discussing a single topic.

QA Ranking

PriBot ranks the candidate answers according to their relevance to your question.

Deep Learning Algorithm

For ranking, PriBot uses a hierarchy of 21 deep learning classifiers that we have developed. This hierarchy allows PriBot to differentiate between a case of first party collection and third party sharing, even when they use similar wording.

QA Interface

Finally, PriBot gives you the answers in a chatbot interface.

Team

Got a question or feedback?

“Shoot'em at hamza.harkous@gmail.com

Hamza Harkous