Show HN: I made a neural net that analyzes privacy policies • Hacker News Board

Re: Show HN: I made a neural net that analyzes privacy policies

#91

Post by murukesh_s » Mon, Sep 23, 2019, 5:12 AM UTC

Cool tech and good use of NLP. but isn't the privacy policy system entirely broken? It's like driving on unmarked, unpaved roads, why don't we have a global template that's list in the beginning a checklist that is human understandable/comprehensible quickly like - Do we share your data : (Y/n)

Etc

Without something like that the problem is the companies could change the wordings and the neural net could not detect until trained again which is potentially more dangerous!

It's time we get some standard for user web privacy policy docs like gdpr

Re: Show HN: I made a neural net that analyzes privacy policies

#92

Post by jph00 » Mon, Sep 23, 2019, 5:18 AM UTC

rameerez wrote: post #72

goldemerald wrote: post #71

Earlier quoted context omitted.

Does your work integrate pretrained LMs like BERT or GPT2?

Yes, but not Transformer-based like these two, rather LSTM-based like ULMFiT

Interesting! Did you have to make any substantive changes to the ULMFiT approach to make it work for this problem? Did you use the fastai implementation, or write something from scratch?

(Disclaimer: I'm a co-author on the ULMFiT paper.)

Re: Show HN: I made a neural net that analyzes privacy policies

#93

Post by shlant » Mon, Sep 23, 2019, 5:18 AM UTC

epoch_100 wrote: post #28

Hi. A related project that takes a more human-powered approach is PrivacySpy ( https://privacyspy.org ). Would be neat to see how these tools intersect. PrivacySpy is open source, community run, and more about grading policies on a standardized rubric (as opposed to entrusting that to ML), so these tools might complement one another. (Full disclosure: I'm a contributor to PrivacySpy.)

We have something similar as well over at https://privacy.commonsense.org/.

We have taken different ideas from many different implementations and applied them more specifically to ed tech products.

Re: Show HN: I made a neural net that analyzes privacy policies

#94

Post by raynr » Mon, Sep 23, 2019, 5:19 AM UTC

foxes wrote: post #87

The "game" (training) that asks you to analyse "privacy threats" is a bit strange. It feels like it takes two random excerpts from a privacy policy and asks you to compare them, but with this, it feels like it is missing some global information, you are just looking at local details. Eg one policy might be disclosing what they do (but its actually relevant to collect data, eg password manager) while the other just sa…

Yeah, it's not comparing like for like. Feels like the system is trying to collect training data from users.

To elaborate on your last sentence, context is critical in assessing whether a clause is pro- or anti- privacy. Is the collection of information critical to the provision of the service? What is collected, and how much? And so on.

Re: Show HN: I made a neural net that analyzes privacy policies

#95

Post by a2x » Mon, Sep 23, 2019, 6:08 AM UTC

Letting people doing the training without any knowledge and context seems to makes no sense.

How do you ensure, that it isn’t a bot, or an army of bots?

What is the neuronal network doing?

How do you unsure your personal integrity, those of your team members and the overall integrity of your ‘system’?

Re: Show HN: I made a neural net that analyzes privacy policies

#96

Post by digitalengineer » Mon, Sep 23, 2019, 6:08 AM UTC

rameerez wrote: post #51

epoch_100 wrote: post #34

Earlier quoted context omitted.

For a site trying to fight for privacy, don't you think it would be better to not use Google Analytics to track the people who visit your site?

Yes, definitely. I mention it in Guard's own privacy policy, I don't like using it either, but reasons are: (a) it's the simplest and as far as I know one of the few free analytics tools available, (b) not having a measure of the website activity will make me effectively blind and unable to make decisions, (c) I don't send any personally identifiable event (and, for this matter, I don't send any events apart from pag…

I’m on mobile, so I can’t check if you’re already using this technique, but you could always use the anonimize IP function. This way the last 3 numbers of the IP data will not be send through the Google Analytics script. More info: https://www.jeffalytics.com/gdpr-ip-addresses-google-analyti...

Re: Show HN: I made a neural net that analyzes privacy policies

#97

Post by lab00002 » Mon, Sep 23, 2019, 6:13 AM UTC

Holy Crap , Tinder shares your profile with potential employers ( or rather companies contracted by those employers).

So glad I gave up on online dating a long time ago.

But really this is amazing.

The next step would be be to have a lawyer write a small opinion piece on the most popular sites.

Re: Show HN: I made a neural net that analyzes privacy policies

#98

Post by VvR-Ox » Mon, Sep 23, 2019, 6:32 AM UTC

Though it's nice you built this I'd wish we wouldn't need it because of governments who do their job and protect the people they were originally intended to serve.

How long can we keep up arming us in the battle against powerful and rich entities who steal our data and buy politicians to have direct access to the process of law-making?

Re: Show HN: I made a neural net that analyzes privacy policies

#99

Post by gempir » Mon, Sep 23, 2019, 6:38 AM UTC

simongr3dal wrote: post #42

the_pwner224 wrote: post #41

I went to the homepage and ctrl-clicked a company card (Mozilla) to open the details in a new tab. Instead, the site jacked my ctrl-click and instead tried to navigate to the link in the same tab. Middle-clicking does not work at all. It then went to an error page instead of loading the details for Mozilla, but while it's an interesting idea I'm not sure how useful it is. I don't usually create an account on a websit…

More and more websites are so advanced that they can’t even use an tag anymore. Instead they do some convoluted onclick-scripting that breaks all standard behavior and accessibility functionality.

I think this is mostly a problem of the tools used. An anchor tag by definition is an inline element, so it shouldn't really be a giant box that's clickable, so you default back to an onClick.

onClick in popular frameworks just means left click and nothing more, which makes sense, except for that use case of opening in a new tab with middle mouse button or right clicking etc. So you have to add a lot of logic to support all that.

If you break the html spec and make an anchor tag a block element then you have to deal with catching and stopping the event from it otherwise it would work as a normal link but you actually just want to change state in your JS app.

So I think tools like Angular, React, Vue etc. should get a better way to create links on website that just change state .

Re: Show HN: I made a neural net that analyzes privacy policies

#100

Post by burnaway » Mon, Sep 23, 2019, 6:54 AM UTC

djsumdog wrote: post #62

rameerez wrote: post #51

Earlier quoted context omitted.

Yes, definitely. I mention it in Guard's own privacy policy, I don't like using it either, but reasons are: (a) it's the simplest and as far as I know one of the few free analytics tools available, (b) not having a measure of the website activity will make me effectively blind and unable to make decisions, (c) I don't send any personally identifiable event (and, for this matter, I don't send any events apart from pag…

Have you considered hosting your own Matomo server? You could just scrape logs or use a Javascript tracker. But the data will stay local to you.

+1 for Matomo, it's a no brainer for any 'privacy service' to stay far away from Google Analytics.