[00:00.000 --> 00:06.120] ...designed in 2006, just after AOL disclosed the search logs of many of its users. [00:06.820 --> 00:14.620] And, well, this reveals that our search logs contain more information than we actually believe. [00:14.840 --> 00:20.400] And also, it shows that even anonymized search logs are not that anonymized. [00:20.520 --> 00:22.380] You can learn a lot about someone. [00:22.540 --> 00:27.000] You can even recover the identity of someone just based on its search logs. [00:29.040 --> 00:31.220] And it's even true today. [00:31.460 --> 00:45.280] I mean, when you read Google privacy policies and how they anonymized search logs after 9 months or 18 months, you don't really know what kind of information they can retrieve just based on your search logs. [00:47.900 --> 00:54.440] And there are even some piece of aggregated logs that they could even share according to their privacy policy. [00:54.440 --> 01:09.680] So, they define aggregated information as information that does not identify you or does not reflect an individual, identifiable, individual person. [01:13.560 --> 01:17.480] So, the idea with TrackMeNot was to provide a simple solution. [01:17.480 --> 01:22.600] I know we can use Tor to just make search anonymously on the web, on Google, on any search engine. [01:22.960 --> 01:27.560] But the thing is, with Tor, no one... well, not everyone can use Tor. [01:28.760 --> 01:29.840] Some of us can. [01:31.320 --> 01:32.900] And it's a great tool. [01:33.060 --> 01:37.640] But not everyone is able to use Tor to search appropriately. [01:38.080 --> 01:43.800] Also, even if Tor is well-configured, you have to use it in an incorrect way. [01:44.760 --> 01:54.200] For instance, recently, there is some search lab in France who made an experiment and they run a Tor exit node. [01:54.780 --> 02:05.280] And 46% of the person running through their exit nodes were logged into Google when they made search on Google. [02:05.640 --> 02:12.140] So, it reveals that even if you configure Tor correctly, it's not sure that you'll really be anonymous. [02:12.140 --> 02:16.780] You have to... well, most of the user may not really configure it appropriately. [02:18.580 --> 02:21.900] And also, the idea was to send a signal to search engines. [02:22.400 --> 02:26.620] Because when you're anonymous, you cannot be counted, actually. [02:27.000 --> 02:35.140] Google has no idea that many users actually do not like their privacy policy and try to go through Tor. [02:35.140 --> 02:38.100] Or when you're using Tor, you cannot be counted. [02:38.420 --> 02:43.560] And the idea was to send a signal that can be seen by search engines. [02:45.280 --> 02:51.100] And so there is no such thing than an opt-out for Google search. [02:52.020 --> 03:00.400] And, by the way, when you opt-out from something like advertising targeting, you just opt-out from advertising. [03:01.460 --> 03:04.220] You're not opt-out from tracking. [03:05.000 --> 03:05.640] So... [03:06.380 --> 03:10.040] So the idea was to use web search obfuscation. [03:10.040 --> 03:17.780] We'll just create nodes in your search history so that the search engine will not be able to really interpret your search. [03:18.060 --> 03:26.340] And we'll have to... well, try to feature some TrackMeNot queries to really find what you're interested in. [03:29.460 --> 03:35.360] So the idea is that no one should be sure that a query in your search history has been generated by you. [03:35.360 --> 03:37.060] So let's take an example. [03:38.080 --> 03:49.140] Either you're a search engine, you're someone with access to Wi-Fi networks of your neighbors, and you want... and you see all the queries that go through the Wi-Fi networks. [03:49.420 --> 03:54.900] And when you see a query, you can assume that the user issued the query, that it revealed something about him. [03:55.920 --> 03:59.880] Well, ideally with TrackMeNot, normally you should not be able to do such thing. [04:00.180 --> 04:04.920] When you see a query, you can no longer assume that it has been made by the user. [04:04.920 --> 04:10.760] You have to also admit the possibility that maybe TrackMeNot generated the query and not the user himself. [04:11.700 --> 04:26.180] So whether you're a search engine or just someone wiretyping on Wi-Fi networks or both, normally you should not be able to be sure that someone made a query and be sure that someone is interested in something. [04:28.380 --> 04:31.140] So that's what I would like to call a reasonable doubt. [04:31.340 --> 04:42.040] We would like to have a property where for any query in your search history, the search engine should not be able to be sure that you made the query. [04:42.460 --> 04:46.860] There should be a negligible probability that TrackMeNot issued the query. [04:48.740 --> 04:58.120] So, normally if I know someone is using TrackMeNot, I should not be able to really be sure that the query is made by him. [04:58.300 --> 05:01.200] But the thing is, what if I don't know who's using TrackMeNot? [05:01.960 --> 05:11.780] How can I then know who's using it and which part of the queries that are receiving my logs are actually made by users? [05:13.280 --> 05:19.740] So that's the point, that normally if TrackMeNot is not detectable, you have to assume that everyone is using it. [05:20.100 --> 05:27.060] Or that an undefined proportion of people are using it. [05:27.240 --> 05:29.820] You cannot know the probability for a user to use it. [05:30.060 --> 05:37.000] And if you receive some query in your search logs, you have to believe that maybe the user is using TrackMeNot. [05:37.160 --> 05:40.640] And if he's using TrackMeNot, maybe this query has been generated by TrackMeNot. [05:43.540 --> 05:44.860] Am I clear at this point? [05:48.340 --> 05:51.820] So, the first version of TrackMeNot has been published in 2006. [05:52.140 --> 05:54.220] And it was just like proof of concepts. [05:55.040 --> 05:56.880] To prove that this approach may work. [05:57.020 --> 06:02.680] That anonymity was not the sole solution to have some privacy when we were doing search on Internet. [06:03.880 --> 06:10.400] So, to provide some technical details, query were just submitted with XML HTTP request. [06:13.320 --> 06:16.920] We used a fixed set of keywords to generate queries. [06:17.060 --> 06:21.700] So, we used a dictionary, I think, about 1500 words. [06:21.940 --> 06:28.440] And so, the search engine could easily filter the queries that were not made by TrackMeNot and detect them. [06:30.260 --> 06:32.540] Also, queries were made at a fixed schedule. [06:34.080 --> 06:39.960] So, it was not too hard for search engine to identify the query made by TrackMeNot. [06:40.020 --> 06:42.520] But at this time, it was just like a proof of concepts. [06:42.760 --> 06:46.520] To prove that this approach might work. [06:49.300 --> 06:51.220] So, this approach has... [06:51.220 --> 06:57.680] Well, TrackMeNot is the first version that's been widely criticized for the reason that I gave early. [06:57.960 --> 07:06.460] And so, from these critics, we tried to improve TrackMeNot to make it more robust and harder to detect. [07:08.800 --> 07:12.960] So, the first thing that we've made is that we use dynamic query list. [07:12.960 --> 07:23.080] Now, when someone is making... is using TrackMeNot, normally, when TrackMeNot make a search on Google or on AOL, being... TrackMeNot will parse the search results and extract some keywords. [07:23.380 --> 07:25.680] It will use these keywords in the next queries. [07:25.920 --> 07:30.020] So, the list of queries will evolve. [07:30.560 --> 07:34.360] And this set of... this process is fully randomized. [07:34.720 --> 07:39.840] So, normally, two users should not have the same set of queries after, like, a week. [07:39.840 --> 07:44.080] And a search engine should not be able to feature, easily, queries made by TrackMeNot. [07:46.660 --> 07:50.900] Another thing that... another thing that we've done is that we use, now, RSS feeds. [07:51.060 --> 07:57.600] So, once a user is running TrackMeNot, you can configure a set of RSS feeds that will be used to extract keywords. [07:57.820 --> 08:06.360] So, when TrackMeNot is running, it will get the feeds, parse them, and extract some keywords. [08:06.360 --> 08:13.500] So, the idea is that, normally, users should be able to select a set of feeds that are related to a topic. [08:14.320 --> 08:16.620] And simulate some interest for that topic. [08:16.740 --> 08:23.440] So, for example, let's assume you're not interested by car, but want to simulate to have Google believing that you're interested by car. [08:23.580 --> 08:27.360] You just find some RSS feeds which speak only about cars. [08:27.360 --> 08:35.680] And that's... and then get these feeds and send query only related to... by... about car to Google. [08:36.640 --> 08:39.160] You can do the same thing for sports, IT. [08:39.880 --> 08:50.420] If you find some website publishing a feed and you can... and you want to simulate an address for this website, you just have to add it to TrackMeNot. [08:52.520 --> 08:56.060] Also, you probably know that search engine gives suggestion to users. [08:56.260 --> 09:03.940] When you're typing something on Google, on Yahoo, you're receiving a set of search... search terms that are suggested. [09:04.260 --> 09:08.820] And, normally, some... well, at some points, you can click on them. [09:08.960 --> 09:12.600] And so, we try to have them also integrate in TrackMeNot. [09:12.700 --> 09:19.460] So, in the same way that you can click on search suggestion, TrackMeNot also will simulate click on the... on these suggestions. [09:22.520 --> 09:25.020] We also changed the way we scheduled queries. [09:25.200 --> 09:27.540] It's not a fixed interval time now. [09:27.700 --> 09:29.620] You can configure... configure it. [09:29.780 --> 09:31.740] And it's randomized also. [09:31.860 --> 09:33.100] You can just set the frequency. [09:33.580 --> 09:40.240] And TrackMeNot will... set... send the query at an average frequency. [09:40.660 --> 09:41.880] And the process is randomized. [09:42.180 --> 09:46.820] So, normally, search engine should not be able to filter the queries just based on timestamps. [09:51.000 --> 09:52.600] Also, there is a query burst. [09:52.780 --> 09:58.920] When TrackMeNot detects that some user is searching on a search engine, it will send a set of query... [09:59.840 --> 10:02.600] to hide the real user search. [10:03.980 --> 10:07.800] And the additional feature is that now we use also dynamic URLs. [10:08.040 --> 10:12.460] So, for example, if I'm searching on Google friends, the URL is not the same. [10:12.600 --> 10:13.660] And TrackMeNot will map that. [10:13.660 --> 10:18.800] And also, we're clicking on some search results. [10:19.020 --> 10:27.860] So, if you... because users also click on some results when they get search results. [10:28.280 --> 10:33.720] And the idea is to try really to mimic all the habit of the user at any granularity. [10:33.720 --> 10:36.380] It's all things that we do not do intentionally. [10:37.000 --> 10:39.400] It's... we do not click on sponsored results. [10:40.080 --> 10:42.820] Um... because we don't... we do not want to click on ads. [10:43.200 --> 10:46.720] It's... it will... it will be assimilated to some sort of click fraud, so... [10:48.000 --> 10:57.960] And, um... actually, also, when TrackMeNot simulate a click on a search link... on search results, uh... it never loads the contents of the search results. [10:57.960 --> 11:00.440] It stops, uh... it stops, uh... the connection before. [11:01.160 --> 11:04.920] So, the... so, the browser will never load untrusted content. [11:08.390 --> 11:12.150] Uh... we also change the way we submit queries to search engines. [11:12.330 --> 11:19.690] Initially, we were using just XML HTTP requests, uh... downloading the HTML content and parsing it. [11:20.190 --> 11:27.450] Uh... but now, search engines contain pictures, feeds, uh... tweets, JavaScripts that are run in the browser. [11:27.450 --> 11:33.730] So, if you... if we just, um... uh... download that without running the JavaScript with... [11:33.730 --> 11:40.250] with, um... with a browser, uh... the search engine might be able to detect, uh... [11:40.250 --> 11:42.230] which queries are submitted by TrackMeNot. [11:44.390 --> 11:49.630] Uh... so, the idea was that we load the search page in... in an iframe. [11:49.810 --> 11:57.650] Uh... in such a way that the search engine normally, uh... will send all the contents to the iframe, as, uh... it will do to the browser normally. [11:59.950 --> 12:04.850] And, uh... also, we simulate keystroke, uh... in the search bar. [12:05.130 --> 12:08.710] So, uh... so, uh... you... you will receive the suggestions. [12:09.150 --> 12:17.310] And the search engine, uh... should not be able, just based on the server logs, on the suggest service, to be able to feature TrackMeNot queries. [12:19.170 --> 12:24.150] We're also downloading the small icons that appears, uh... along the... the URL bar, normally. [12:24.730 --> 12:26.930] Uh... which, for example, for Google is a G. [12:27.570 --> 12:35.890] And, um... so, normally, uh... search engine also, um... is not able to feature TrackMeNot queries based on that. [12:37.870 --> 12:41.890] And, um... the problem with iframe is that it can be easily detected by search engine. [12:41.890 --> 13:03.150] Um... so, what, uh... we changed, uh... this to use a browser element in Firefox, uh... which is exactly like a browser that you use, um... and, uh... it's not... normally, it's not visible to the users, but, uh... it's... still, uh... acts exactly as a visible browser. [13:04.350 --> 13:11.410] And the advantage is that, also, it records, uh... all the search that are made by TrackMeNot in your browser history. [13:11.410 --> 13:19.930] So, uh... normally, it colors the links, uh... that TrackMeNot clicks on, in the same way than, uh... if you really search for them. [13:20.450 --> 13:30.470] So, uh... all the, uh... history-stealing attack based on link coloration should not work, uh... well, or should be mitigated, um... uh... through this approach. [13:31.670 --> 13:37.210] And, uh... yeah, click and, um... and keystrokes are sent through... through DOM events. [13:37.210 --> 13:44.450] So, we are just, um... sending all the events to, uh... the, uh... search engine page. [13:44.850 --> 13:47.110] And, normally, it's, uh... it's more robust. [13:47.330 --> 13:55.970] So, if the search engine changes something in, uh... in the JavaScript or in the interface, uh... normally, it shows, uh... TrackMeNot should still run. [13:56.290 --> 13:59.230] Uh... I just... we'll just try to make a short demo. [14:14.360 --> 14:16.900] So, um... well, normally, this is hidden. [14:17.220 --> 14:17.940] Uh... you don't see it. [14:18.220 --> 14:21.240] And that's, um... well, TrackMeNot process running. [14:21.480 --> 14:25.060] And you can see that it's typing, uh... like, uh... I will normally. [14:25.260 --> 14:26.980] And it receives suggestion at some point. [14:27.040 --> 14:28.480] It will also click on some suggestions. [14:29.200 --> 14:33.440] Um... so, this, uh... just the keystrokes events that are sent to, uh... Google. [14:38.270 --> 14:45.810] And, uh... if, uh... if, uh... it simulates several keyword, uh... several search with, uh... common keywords, it will just... [14:46.450 --> 14:48.270] Well, any kind of query might be generated. [14:50.030 --> 14:57.570] Um... and so, it will try to just remove some, um... some, uh... some terms from the search. [15:11.340 --> 15:13.020] Um... well, actually not. [15:13.380 --> 15:14.060] We're not monitoring. [15:14.320 --> 15:16.300] Well, ideally, uh... I will come to that point later. [15:16.520 --> 15:24.440] But we would like to be able to, um... monitor how fast the users type on the keyboard, or, uh... how you move, uh... move the mouse on the search page. [15:25.880 --> 15:28.340] Um... for... currently, it's not, uh... implemented. [15:28.560 --> 15:32.160] We would like to have something like that to just be able to, uh... really... [15:32.160 --> 15:33.400] to, uh... mimic users' habits. [15:33.640 --> 15:38.340] Also, we don't... we do not know at which granularity, uh... search engines are able to, uh... [15:39.020 --> 15:40.180] monitor these kind of events. [15:43.610 --> 15:45.270] Uh... yeah, it's not implemented yet. [15:45.670 --> 15:50.650] Also, um... and we'll... well, it's not, um... [15:50.650 --> 15:55.010] some things we'll like to be able to monitor that and see when the, uh... [15:55.010 --> 15:58.950] how often the user click on did you mean when you receive the search results. [16:06.730 --> 16:07.910] Uh... well, um... [16:08.710 --> 16:09.190] Um... [16:09.190 --> 16:09.510] No. [16:09.810 --> 16:10.550] So far, no. [16:10.710 --> 16:12.890] The solution to really control, uh... [16:12.890 --> 16:17.110] So the question is, are there words that are excluded intentionally, um... [16:17.110 --> 16:18.310] that could give you trouble? [16:18.530 --> 16:19.010] Uh... [16:19.010 --> 16:20.450] And the answer is currently no. [16:20.950 --> 16:21.130] Um... [16:21.130 --> 16:24.130] Normally, if you configure the set of errors as feeds, uh... [16:24.130 --> 16:24.270] Uh... [16:24.270 --> 16:25.990] It should, um... [16:25.990 --> 16:26.330] Uh... [16:26.330 --> 16:30.650] It should allow you to give some direction to the set of keywords that are used. [16:31.530 --> 16:32.010] Um... [16:32.010 --> 16:35.630] But we'll like to have something also allowing users to... [16:35.630 --> 16:36.410] Because, uh... [16:36.410 --> 16:36.430] Uh... [16:36.430 --> 16:38.670] When the set of keywords change, we... [16:38.670 --> 16:43.250] We lose the control about how they change and which are the keywords that are used. [16:43.830 --> 16:44.130] Uh... [16:44.130 --> 16:50.770] We'll like to have some tools to let users filter keywords so they can use, basically, TrackMeNot when they're working, uh... from their office. [17:02.680 --> 17:03.140] Uh... [17:06.760 --> 17:08.460] Normally, you can set the frequency. [17:08.860 --> 17:10.300] So, um... [17:10.300 --> 17:10.440] It [17:24.360 --> 17:26.000] should be created by me, right? [17:26.140 --> 17:28.400] It's probably by TrackMeNot and not by the... [17:28.400 --> 17:29.340] Use of... [17:29.340 --> 17:29.920] Because that should be... [17:29.920 --> 17:30.480] Right? [17:31.140 --> 17:31.600] Uh... [17:31.600 --> 17:31.680] Yeah. [17:31.880 --> 17:33.220] So, um... [17:33.220 --> 17:34.300] You can set the set of... [17:34.300 --> 17:38.880] You can set the frequency 10 per minute, then 5 per minute, and you can change that. [17:38.880 --> 17:39.880] And it's not exactly... [17:40.580 --> 17:41.420] Well, it's not... [17:41.420 --> 17:44.760] If you set 10 per minute, it won't be like 1 every 6 seconds. [17:45.040 --> 17:50.940] It will be more or less like 5 and then 7 seconds later, and it will be like that. [17:51.160 --> 17:51.820] Randomized on that. [17:52.360 --> 17:52.840] Yeah. [17:53.040 --> 17:53.560] So, it's... [17:53.560 --> 17:54.100] It's randomized. [17:54.620 --> 17:55.860] We try to keep the... [17:55.860 --> 17:56.880] The frequency and... [17:58.040 --> 17:58.460] Uh... [17:58.460 --> 17:59.080] Normally... [17:59.080 --> 18:02.220] And also, if you use a burst option, uh... [18:02.220 --> 18:03.880] Normally, it will, um... [18:04.720 --> 18:07.180] Send burst after it detects, uh... [18:07.180 --> 18:07.840] User queries. [18:07.840 --> 18:13.140] So, as a search engine is not able to know when the user made a query, it should not be able to detect burst. [18:13.920 --> 18:14.100] Uh... [18:14.100 --> 18:14.240] Yeah? [18:14.380 --> 18:18.000] All of these automated queries, are they recorded in the browser history? [18:18.360 --> 18:18.660] Yeah. [18:18.980 --> 18:19.220] Uh... [18:19.220 --> 18:20.200] Well, they're recorded. [18:20.440 --> 18:22.580] So far, we just currently remove... [18:22.580 --> 18:25.720] Remove them to not pollute too much browser history. [18:25.980 --> 18:26.460] It's... [18:26.460 --> 18:26.720] We... [18:26.720 --> 18:29.820] We can remove that and let them in the browser history. [18:31.900 --> 18:32.420] Uh... [18:32.420 --> 18:32.660] Uh... [18:32.660 --> 18:32.780] Yes? [18:54.070 --> 18:55.330] Well, it's under development. [18:55.750 --> 18:56.210] Uh... [18:56.210 --> 18:57.210] Actually, we have some prototype. [18:57.210 --> 18:57.810] We do that. [18:58.070 --> 18:58.350] Uh... [18:58.350 --> 18:59.070] Which does that. [18:59.270 --> 18:59.510] Uh... [18:59.510 --> 19:00.310] We'll come to that later. [19:00.810 --> 19:01.270] And... [19:15.060 --> 19:15.520] Uh... [19:15.520 --> 19:15.620] Uh... [19:15.620 --> 19:16.380] So, that... [19:16.380 --> 19:16.960] That was the previous question. [19:17.320 --> 19:17.760] And... [19:17.760 --> 19:19.240] Well, except from... [19:19.240 --> 19:19.560] Uh... [19:19.560 --> 19:21.520] Setting the RSS feeds about a particular topic. [19:21.720 --> 19:22.740] There is no such thing. [19:23.980 --> 19:24.420] Yeah? [19:46.330 --> 19:46.770] Uh... [19:46.770 --> 19:46.790] Yeah? [19:47.150 --> 19:47.590] It's... [19:47.590 --> 19:47.910] Uh... [19:47.910 --> 19:48.650] Well... [19:48.650 --> 19:49.310] Actually, we... [19:50.430 --> 19:51.310] We're using... [19:52.610 --> 19:53.430] We're using... [19:53.430 --> 19:53.870] We're using RSS feeds. [19:54.050 --> 19:59.370] So, normally, what we do is that we just take in the RSS feeds the title of the news, and, uh... [19:59.370 --> 20:00.570] Then remove all the character... [20:01.190 --> 20:01.910] The words that are... [20:01.910 --> 20:02.130] That... [20:02.130 --> 20:03.810] That do not start by capital letters. [20:04.090 --> 20:06.770] We try to assume that, normally, uh... [20:06.770 --> 20:09.550] Words starting with capital letters are no stop words. [20:10.270 --> 20:10.750] Um... [20:10.750 --> 20:14.070] Well, probably, we'll have to have a list of stop words implemented. [20:14.250 --> 20:19.850] But the thing is that if we start to use a list of stop words, we'll have to have different lists for, um... [20:19.850 --> 20:20.390] Every language. [20:20.670 --> 20:21.150] So... [20:34.560 --> 20:35.040] Okay. [20:35.320 --> 20:36.100] So, um... [20:36.100 --> 20:36.200] Uh... [20:36.200 --> 20:37.900] That was the version on Firefox. [20:38.680 --> 20:39.080] Uh... [20:39.080 --> 20:39.900] Well, recently... [20:39.900 --> 20:40.080] This... [20:40.080 --> 20:46.580] This week, actually, we started to have a version of Chrome, which is quite basic now, because we don't have the same API on Chrome than we have on Firefox. [20:47.500 --> 20:47.900] Um... [20:47.900 --> 20:48.580] It's just started. [20:48.680 --> 20:50.680] The development started, like, last week. [20:51.000 --> 20:51.900] And, um... [20:53.240 --> 20:53.500] Uh... [20:53.500 --> 20:54.780] You can denote it from the website. [20:55.060 --> 20:56.840] So, I don't know if you have... [20:56.840 --> 20:57.140] Uh... [20:57.140 --> 21:00.160] Some of you have some experience with Chrome extension development. [21:00.620 --> 21:01.020] Uh... [21:01.020 --> 21:04.640] The thing is, actually, you have a background HTML page, uh... [21:04.640 --> 21:05.160] Which is... [21:05.160 --> 21:05.580] Uh... [21:05.580 --> 21:06.920] Specific to every extension. [21:07.200 --> 21:09.660] And what we do is that we just create an iframe. [21:09.660 --> 21:10.820] So, it's still an iframe. [21:10.920 --> 21:12.580] It's not as good as with Firefox. [21:12.940 --> 21:13.720] And, uh... [21:13.720 --> 21:14.360] We set the... [21:14.360 --> 21:15.640] Your search... [21:15.640 --> 21:15.660] Uh... [21:15.660 --> 21:18.820] The source of the iframe to the search page. [21:19.640 --> 21:20.960] Then, what we do is... [21:20.960 --> 21:22.640] Well, set the current, uh... [21:22.640 --> 21:23.140] Implementation. [21:23.460 --> 21:24.520] And that, uh... [21:24.520 --> 21:27.300] Next step, it will be to inject a content script, uh... [21:27.300 --> 21:27.440] Uh... [21:27.440 --> 21:28.300] In this, uh... [21:28.300 --> 21:28.500] Uh... [21:28.500 --> 21:29.180] Uh... [21:40.360 --> 21:40.840] Uh... [21:40.840 --> 21:42.600] TrackMeNotes that, uh... [21:42.600 --> 21:43.480] Are developed now. [21:43.780 --> 21:44.660] We have, um... [21:44.660 --> 21:46.880] The version of Firefox, which is more complete. [21:48.580 --> 21:49.080] Um... [21:49.080 --> 21:50.060] Uh... [21:50.060 --> 21:50.240] Uh... [21:50.240 --> 21:51.760] So, it runs with... [21:51.760 --> 21:53.460] In a browser, not in an iframe. [21:54.420 --> 21:54.920] Um... [21:54.920 --> 21:55.840] We also, um... [21:57.200 --> 21:58.380] Have a good version for Chrome. [21:58.640 --> 21:59.560] And, uh... [21:59.560 --> 22:02.080] Safari also has an extension system now. [22:02.240 --> 22:03.990] Which is quite similar to the one of... [22:03.990 --> 22:04.150] Uh... [22:04.150 --> 22:06.230] To the extension system of Chrome. [22:07.010 --> 22:08.270] And, um... [22:08.270 --> 22:09.350] So, for, uh... [22:09.350 --> 22:12.390] Internet Explorer, we just have an iframe, uh... [22:12.390 --> 22:13.310] Which loads the content. [22:13.510 --> 22:16.370] We have no possibility to load, um... [22:16.370 --> 22:17.910] Some content scripts, uh... [22:17.910 --> 22:18.590] In the iframe. [22:18.970 --> 22:21.010] Due to some security restriction. [22:21.910 --> 22:22.330] Uh... [22:22.330 --> 22:23.450] The search engines, uh... [22:23.450 --> 22:24.990] You can search on, uh... [22:26.250 --> 22:26.830] Uh... [22:26.830 --> 22:26.910] Uh... [22:26.910 --> 22:26.990] Uh... [22:26.990 --> 22:32.050] You can search on AOL, Beidu, Bing, um... [22:33.390 --> 22:33.970] Uh... [22:33.970 --> 22:34.910] And, uh... [22:34.910 --> 22:36.290] If you search on, uh... [22:36.290 --> 22:38.970] Beidu, the thing is that you can also use, uh... [22:39.490 --> 22:39.830] Uh... [22:39.830 --> 22:40.450] Chinese RSS feeds. [22:40.710 --> 22:41.850] So, it will, uh... [22:41.850 --> 22:44.030] Type Chinese characters, um... [22:44.030 --> 22:44.310] Actually. [22:49.520 --> 22:50.280] Here, uh... [22:50.280 --> 22:52.160] I don't know if it's clear on the... [22:52.160 --> 22:52.360] Uh... [22:52.920 --> 22:53.920] Screenshot on the right. [22:54.740 --> 22:55.240] Um... [22:55.240 --> 22:57.800] It's search on Beidu, and it's, uh... [22:57.800 --> 22:57.940] Uh... [22:57.940 --> 22:58.840] Typing Chinese characters. [22:59.180 --> 23:01.880] So, you just have to, um... [23:01.880 --> 23:02.060] Um... [23:02.060 --> 23:04.440] Configure TrackMina to use Chinese RSS feeds. [23:04.560 --> 23:07.040] And be sure that it's only have, uh... [23:07.040 --> 23:07.120] Uh... [23:07.120 --> 23:07.760] Chinese characters. [23:08.020 --> 23:08.680] And normally, it should... [23:08.680 --> 23:09.380] It should work. [23:15.100 --> 23:16.060] So, um... [23:16.060 --> 23:18.880] What we are also, um... [23:18.880 --> 23:19.920] Trying to bring, uh... [23:19.920 --> 23:20.660] As features now. [23:21.580 --> 23:22.060] Um... [23:22.060 --> 23:22.720] We'll like... [23:22.720 --> 23:23.440] So, uh... [23:23.440 --> 23:24.860] It's related to the questions that you had. [23:24.960 --> 23:25.120] We'll... [23:25.120 --> 23:27.180] We'd like to really simulate, uh... [23:27.180 --> 23:28.000] Uh... [23:28.000 --> 23:28.520] Uh... [23:28.520 --> 23:28.860] User behavior. [23:29.980 --> 23:30.580] So, uh... [23:30.580 --> 23:30.980] Time distribution. [23:31.340 --> 23:33.180] If you're searching more during the weekend. [23:33.440 --> 23:35.780] Or if you're not searching between, uh... [23:35.780 --> 23:36.420] Uh... [23:36.420 --> 23:37.580] 6 and 7 a.m. [23:38.060 --> 23:38.240] Uh... [23:38.240 --> 23:40.380] TrackMina should not search between 6 and 7 a.m. [23:40.480 --> 23:41.710] And if you're, uh... [23:41.710 --> 23:43.280] Searching twice during the weekend. [23:43.400 --> 23:45.220] And that's what you're doing during the week. [23:45.680 --> 23:45.880] Uh... [23:45.880 --> 23:47.160] TrackMina should search in the same way. [23:47.280 --> 23:48.160] Also, if, uh... [23:48.160 --> 23:48.320] Uh... [23:48.320 --> 23:50.180] You're almost always searching on Bing. [23:50.320 --> 23:50.980] And never on Google. [23:51.260 --> 23:54.740] So, proportionality should be the same for TrackMina search. [23:54.900 --> 23:55.710] And for, uh... [23:55.710 --> 23:56.040] Uh... [23:56.040 --> 23:56.720] Your own search. [23:57.910 --> 23:58.460] Uh... [23:58.460 --> 23:59.210] So, uh... [23:59.210 --> 23:59.640] Syntax. [23:59.800 --> 24:01.730] We'd like to get some, um... [24:02.320 --> 24:02.900] Syntax feature. [24:03.300 --> 24:03.710] Uh... [24:03.710 --> 24:05.800] Like how many quotes the user type in. [24:06.100 --> 24:06.600] Uh... [24:06.600 --> 24:08.500] Does he ever type capital letters? [24:08.900 --> 24:09.860] Does he make type... [24:09.860 --> 24:10.160] Uh... [24:10.160 --> 24:10.940] Does he make typos? [24:12.060 --> 24:12.620] Uh... [24:12.620 --> 24:13.460] And there's, uh... [24:13.460 --> 24:15.300] Some search engine-specific, uh... [24:15.300 --> 24:15.620] Syntax. [24:15.710 --> 24:17.720] For example, if you ever, uh... [24:17.720 --> 24:19.180] Search with any URL. [24:19.400 --> 24:20.060] Or file type. [24:20.300 --> 24:21.400] Or site in Google. [24:21.900 --> 24:22.360] Uh... [24:22.360 --> 24:23.730] TrackMina does not... [24:24.640 --> 24:24.800] Uh... [24:24.800 --> 24:25.860] Search like that yet. [24:25.860 --> 24:27.920] And we'd like to be able to, uh... [24:27.920 --> 24:29.100] Also have these kind of, uh... [24:29.100 --> 24:30.320] Elements included in TrackMina. [24:31.100 --> 24:33.320] The number of suggestions accepted by the user. [24:33.520 --> 24:33.820] Uh... [24:33.820 --> 24:34.980] The same thing, uh... [24:34.980 --> 24:35.520] With TrackMina. [24:37.080 --> 24:37.520] Uh... [24:37.520 --> 24:39.870] How many search sessions the user, um... [24:41.040 --> 24:41.480] Uh... [24:41.480 --> 24:41.900] Uh... [24:41.900 --> 24:42.680] Uh... [24:42.680 --> 24:43.740] Well, uh... [24:43.740 --> 24:45.120] By search session, I mean, uh... [24:45.120 --> 24:47.800] When you have two consecutive search with similar terms. [24:48.000 --> 24:49.100] Or, uh... [24:49.100 --> 24:50.300] Terms related to the same topic. [24:50.560 --> 24:51.400] How often does that... [24:51.400 --> 24:52.000] Does that occur? [24:52.240 --> 24:53.440] And, uh... [24:53.440 --> 24:55.800] When you have a search session, How long, uh... [24:56.920 --> 24:57.800] Does it last on average? [24:59.100 --> 25:00.360] And also, um... [25:00.360 --> 25:01.940] Would like to know... [25:01.940 --> 25:02.300] What's the... [25:02.300 --> 25:03.740] What's the profile, uh... [25:03.740 --> 25:05.000] A search engine, uh... [25:05.000 --> 25:05.680] Has about you? [25:05.980 --> 25:08.180] So, I don't know if you heard about Google Dashboard. [25:08.500 --> 25:08.760] Uh... [25:08.760 --> 25:09.360] Basically, they... [25:09.360 --> 25:10.920] They give you raw data about... [25:10.920 --> 25:13.040] Well, all the services that you use from Google. [25:13.210 --> 25:13.720] Like, uh... [25:13.720 --> 25:15.040] Gmail, uh... [25:15.040 --> 25:15.640] Google Search. [25:15.900 --> 25:17.080] And all the service... [25:17.080 --> 25:18.420] All the data that they... [25:18.420 --> 25:18.760] They have... [25:19.460 --> 25:19.900] You... [25:20.580 --> 25:21.020] Um... [25:21.020 --> 25:21.420] Uh... [25:21.420 --> 25:21.880] See... [25:21.880 --> 25:23.240] See them on the services. [25:23.780 --> 25:24.480] But, uh... [25:24.480 --> 25:26.040] The thing is that it's only raw data. [25:26.340 --> 25:26.600] Uh... [25:26.600 --> 25:27.720] It's not interpreted data. [25:27.940 --> 25:28.160] They... [25:28.160 --> 25:28.520] They... [25:28.520 --> 25:28.780] You... [25:28.780 --> 25:31.710] You have no access to the profiles that they build from this data. [25:32.420 --> 25:34.160] And the thing is, um... [25:34.160 --> 25:37.480] For example, Google is able to know if someone has... [25:37.480 --> 25:37.580] Uh... [25:37.580 --> 25:37.600] Uh... [25:37.600 --> 25:39.040] Has the flu before... [25:39.040 --> 25:39.300] Uh... [25:39.300 --> 25:40.320] The user itself. [25:40.760 --> 25:41.120] Uh... [25:41.120 --> 25:43.260] They just can do that based on the... [25:43.260 --> 25:43.760] Uh... [25:43.760 --> 25:44.800] Search time... [25:44.800 --> 25:45.680] Search terms. [25:46.340 --> 25:47.520] If you're looking for... [25:47.520 --> 25:47.710] Uh... [25:47.710 --> 25:49.860] The flu symptoms, it's likely that you have the flu. [25:50.020 --> 25:53.220] And the user alone cannot, uh... [25:53.220 --> 25:53.360] Uh... [25:53.360 --> 25:54.920] Infer that from his own searches. [25:55.480 --> 25:58.000] So, we'd like to have something that will... [25:58.000 --> 25:59.900] Let the user, uh... [25:59.900 --> 26:00.210] Uh... [26:00.210 --> 26:02.000] Have some, um... [26:02.000 --> 26:02.120] Um... [26:02.120 --> 26:03.160] Some insight about... [26:03.160 --> 26:03.440] Uh... [26:03.440 --> 26:04.210] About his profile. [26:06.600 --> 26:08.320] And we're also looking for other... [26:08.320 --> 26:08.900] Uh... [26:08.900 --> 26:09.660] Other source of keywords. [26:10.120 --> 26:10.420] Uh... [26:10.420 --> 26:12.940] To have some things that you can use from your office. [26:13.380 --> 26:13.720] Uh... [26:13.720 --> 26:14.600] And that, uh... [26:14.600 --> 26:16.240] Will not use inappropriate terms. [26:17.600 --> 26:18.120] Uh... [26:18.120 --> 26:18.660] And, uh... [26:18.660 --> 26:18.780] Uh... [26:18.780 --> 26:20.260] We're adding search engines like, uh... [26:20.960 --> 26:21.480] Uh... [26:21.480 --> 26:21.680] Uh... [26:21.680 --> 26:22.840] HTTPS version of Google. [26:23.140 --> 26:23.480] And, uh... [26:23.480 --> 26:23.720] Uh... [26:23.720 --> 26:24.860] We'd like to add also Squalar. [26:27.930 --> 26:28.450] Uh... [26:28.450 --> 26:29.650] So, the big question is... [26:29.650 --> 26:30.490] Does it work? [26:30.930 --> 26:31.450] Um... [26:31.450 --> 26:34.070] Well, we have some reason to believe that it actually works. [26:34.570 --> 26:34.930] Uh... [26:35.510 --> 26:36.750] Look for, uh... [26:36.750 --> 26:39.070] The kind of control measures that might be deployed by search engine. [26:39.550 --> 26:39.970] Uh... [26:39.970 --> 26:41.930] Some paper published by, uh... [26:41.930 --> 26:42.770] People from Microsoft. [26:43.250 --> 26:44.430] Suggests that, uh... [26:44.430 --> 26:47.250] The kind of features they use to identify search boats. [26:47.550 --> 26:48.710] Are, uh... [26:49.510 --> 26:50.010] Uh... [26:50.010 --> 26:50.650] Based on the query. [26:50.810 --> 26:51.450] So, uh... [26:51.450 --> 26:53.170] Most of the times the queries contain spam... [26:53.170 --> 26:53.810] Spam words. [26:54.030 --> 26:55.590] Or refer to adult content. [26:55.890 --> 26:57.650] Or if the queries use, uh... [26:57.650 --> 26:58.690] Very frequently the same... [26:58.690 --> 26:59.810] The same set of keywords. [27:01.250 --> 27:02.430] If the user is... [27:03.090 --> 27:03.590] Uh... [27:03.590 --> 27:03.670] Uh... [27:03.670 --> 27:06.750] The search engine receives too many queries from a browser. [27:07.090 --> 27:07.210] Uh... [27:07.210 --> 27:10.010] We can assume that the browser is actually running, uh... [27:10.010 --> 27:10.350] Uh... [27:10.350 --> 27:10.750] Some search... [27:11.370 --> 27:11.690] Uh... [27:11.690 --> 27:12.010] Some search boats. [27:13.470 --> 27:14.110] Uh... [27:14.110 --> 27:14.970] Some, uh... [27:14.970 --> 27:18.250] Boats just aim at increasing the page rank of... [27:18.250 --> 27:18.510] Uh... [27:18.510 --> 27:19.130] Uh... [27:19.130 --> 27:19.510] Some page. [27:19.730 --> 27:21.490] So they will click always on the same link. [27:21.610 --> 27:24.270] To have the page rank of this page, uh... [27:24.270 --> 27:24.750] Increase. [27:25.530 --> 27:26.430] And if, uh... [27:26.430 --> 27:27.570] This page is too far. [27:27.630 --> 27:29.390] They will have to click frequently on next. [27:29.910 --> 27:30.390] Uh... [27:30.390 --> 27:32.910] So these are the kind of features that, uh... [27:32.910 --> 27:36.310] Are used by search engines to, um... [27:36.310 --> 27:37.950] Detect obvious search boats. [27:38.270 --> 27:38.510] Uh... [27:38.510 --> 27:39.510] They might be other. [27:40.010 --> 27:40.270] Uh... [27:40.270 --> 27:40.830] This... [27:40.830 --> 27:41.990] Well, uh... [27:41.990 --> 27:42.450] I... [27:42.450 --> 27:43.330] The reference here. [27:43.490 --> 27:45.390] The paper has been published in 2008. [27:45.610 --> 27:46.430] So they might... [27:46.430 --> 27:46.930] They might... [27:46.930 --> 27:48.270] They may have, uh... [27:48.270 --> 27:49.530] More elaborated, uh... [27:49.530 --> 27:50.330] Mechanicism now. [27:51.790 --> 27:52.330] Uh... [27:52.330 --> 27:52.610] And, uh... [27:52.610 --> 27:54.090] When we look at, uh... [28:10.270 --> 28:10.810] Uh... [28:10.810 --> 28:11.390] The search profile. [28:11.610 --> 28:13.130] So you can go on this page. [28:13.410 --> 28:13.890] And, uh... [28:13.890 --> 28:15.470] You will see your search profiles. [28:16.430 --> 28:16.950] Um... [28:16.950 --> 28:18.430] Well, I'm not interested in cars. [28:18.690 --> 28:19.470] Or country. [28:19.770 --> 28:20.770] Or real playing games. [28:20.970 --> 28:21.810] So all these, uh... [28:22.450 --> 28:24.510] Interests are actually simulated by TrackMeNot. [28:25.650 --> 28:29.830] And we wanted to be sure that TrackMeNot can actually impact, uh... [28:29.830 --> 28:30.770] User search profile. [28:30.970 --> 28:33.450] So we run an experiment with 10, uh... [28:33.450 --> 28:34.290] Empty search profile. [28:35.730 --> 28:36.770] And, uh... [28:36.770 --> 28:39.870] We generated queries about different topics with these profiles. [28:39.870 --> 28:40.910] And, uh... [28:40.910 --> 28:42.370] Tried to see, um... [28:42.370 --> 28:43.090] If, uh... [28:43.390 --> 28:43.990] The query... [28:43.990 --> 28:45.190] The search profile were... [28:45.190 --> 28:46.870] Was impacted by, um... [28:47.590 --> 28:48.110] These queries. [28:48.410 --> 28:49.510] And so we... [28:49.510 --> 28:51.150] We had some results where, um... [28:52.630 --> 28:54.170] The profile, uh... [28:54.170 --> 28:55.170] Are actually impacted. [28:55.850 --> 28:56.150] Um... [28:56.450 --> 28:58.710] Normally, if you try to, uh... [28:59.310 --> 28:59.910] Um... [28:59.910 --> 29:01.890] Search about a certain topic with TrackMeNot. [29:02.470 --> 29:03.070] Um... [29:03.070 --> 29:04.830] The search engine will, uh... [29:04.830 --> 29:05.710] Modify your profile. [29:05.910 --> 29:07.050] But will not, um... [29:07.470 --> 29:09.710] Consider all the categories that are, uh... [29:09.710 --> 29:10.930] Related, uh... [29:10.930 --> 29:12.430] That are connected to this topic. [29:13.070 --> 29:15.830] And also, it will add some, uh... [29:15.830 --> 29:18.410] Categories that are not connected to the Track... [29:18.410 --> 29:20.870] The topic that TrackMeNot is, um... [29:20.870 --> 29:21.650] Searching about. [29:39.090 --> 29:39.690] Uh... [29:39.690 --> 29:41.350] We'll have different set of keywords. [29:41.910 --> 29:42.690] But, uh... [29:43.590 --> 29:43.630] Uh... [29:43.630 --> 29:44.990] Well, the default matters. [29:45.390 --> 29:46.430] So, uh... [29:46.430 --> 29:46.530] Uh... [29:46.530 --> 29:46.710] We... [29:46.710 --> 29:47.550] We have to... [29:47.550 --> 29:48.910] Find some way to improve that. [29:50.590 --> 29:51.070] Um... [29:51.070 --> 29:54.330] Also, search engine can detect colorations about thousands of users. [29:54.610 --> 29:54.930] Uh... [29:54.930 --> 29:55.870] We're not able to do that. [29:55.990 --> 29:56.550] We don't have... [29:56.550 --> 29:58.770] We do not have access to all these profiles. [29:59.310 --> 29:59.790] Um... [29:59.790 --> 30:00.450] I think... [30:00.450 --> 30:01.790] Last week, um... [30:02.630 --> 30:02.930] Uh... [30:02.930 --> 30:04.530] Some research paper from, uh... [30:04.530 --> 30:05.570] Yahoo has been slash dotted. [30:05.890 --> 30:06.990] And they... [30:06.990 --> 30:07.290] They... [30:07.290 --> 30:09.790] What they did is that they correlate, um... [30:10.610 --> 30:12.150] Search queries with, uh... [30:12.150 --> 30:12.870] Demographic data. [30:13.110 --> 30:15.430] And they find that there are some, uh... [30:15.430 --> 30:18.070] Some queries that are really characteristic of, uh... [30:18.070 --> 30:18.890] Certain kind of population. [30:19.650 --> 30:21.010] For example, um... [30:21.010 --> 30:22.010] Well, uh... [30:22.010 --> 30:23.750] If you are searching for, uh... [30:23.750 --> 30:25.050] Radeon, uh... [30:25.050 --> 30:26.310] X, uh... [30:26.310 --> 30:28.150] 800, you're probably, um... [30:28.150 --> 30:28.570] A male. [30:28.870 --> 30:29.530] And, uh... [30:29.530 --> 30:30.010] So, uh... [30:30.710 --> 30:30.830] Uh... [30:30.830 --> 30:32.410] If you're looking for, uh... [30:32.410 --> 30:33.250] I think it's, uh... [30:33.850 --> 30:33.990] Uh... [30:33.990 --> 30:35.010] JSU's song lyrics. [30:35.410 --> 30:35.530] Uh... [30:35.530 --> 30:37.630] You're probably originating from Asia. [30:37.970 --> 30:41.030] And all these kind of things that, um... [30:41.030 --> 30:43.630] Search engine can infer from, uh... [30:43.630 --> 30:45.170] This huge amount of data they have. [30:45.370 --> 30:45.670] We... [30:45.670 --> 30:47.010] We cannot, uh... [30:47.810 --> 30:48.170] Uh... [30:48.170 --> 30:48.430] Uh... [30:50.210 --> 30:50.730] Uh... [30:50.730 --> 30:51.070] Uh... [30:51.070 --> 30:51.190] What? [30:52.750 --> 30:53.290] Um... [30:53.290 --> 30:54.450] I think it's search demo... [30:54.450 --> 30:55.030] Uh... [30:55.030 --> 30:55.350] Search demographics. [30:56.030 --> 30:56.570] Um... [30:56.570 --> 30:57.190] Um... [30:57.190 --> 30:58.570] I don't really have, uh... [30:58.570 --> 30:58.850] The name. [30:58.970 --> 30:59.950] I can find it later. [31:03.370 --> 31:06.170] And also, TrackMeNot cannot hide your identity. [31:06.390 --> 31:09.610] So, if you search for your own name on Google, we will... [31:09.610 --> 31:12.710] Well, probably it will appear as, um... [31:12.710 --> 31:14.090] As different of, uh... [31:14.090 --> 31:15.150] From the TrackMeNot searches. [31:15.490 --> 31:16.970] The thing is, um... [31:16.970 --> 31:20.270] At some point, we had a feature which generated, uh... [31:20.270 --> 31:21.330] Some sort of, uh... [31:21.890 --> 31:22.050] Uh... [31:22.050 --> 31:22.950] Social security number. [31:23.190 --> 31:25.430] So, that it will submit, uh... [31:25.430 --> 31:26.350] This security... [31:26.350 --> 31:28.670] Social security numbers to, uh... [31:28.670 --> 31:28.910] Uh... [31:28.910 --> 31:29.250] Search engines. [31:30.330 --> 31:31.590] And, uh... [31:31.590 --> 31:32.170] But we... [31:32.170 --> 31:33.270] We did not, uh... [31:34.250 --> 31:34.570] Uh... [31:34.890 --> 31:35.350] Develop... [31:35.350 --> 31:36.430] Develop that, uh... [31:36.430 --> 31:36.790] Anymore. [31:37.010 --> 31:37.930] And maybe we... [31:37.930 --> 31:38.110] We will. [31:39.570 --> 31:40.930] Also, uh... [31:40.930 --> 31:43.970] TrackMeNot does not let you delete search in your... [31:43.970 --> 31:45.510] In a search engine search logs. [31:46.130 --> 31:46.630] Um... [31:46.630 --> 31:48.130] The thing is, uh... [31:48.130 --> 31:48.270] Uh... [31:48.270 --> 31:50.930] When I run TrackMeNot for a long time, um... [31:50.930 --> 31:51.050] Um... [31:51.050 --> 31:53.890] My very old queries are deleted from my, uh... [31:53.890 --> 31:54.750] Google search profile. [31:54.990 --> 31:55.850] And, uh... [31:55.850 --> 31:56.030] Uh... [31:56.030 --> 31:57.270] These are not only the, uh... [31:57.270 --> 31:59.630] Query generated by TrackMeNot, but all, um... [32:00.370 --> 32:00.670] Um... [32:00.670 --> 32:00.690] Um... [32:00.690 --> 32:01.830] My, uh... [32:01.830 --> 32:02.890] Search, uh... [32:02.890 --> 32:04.230] Are deleted by Google. [32:04.970 --> 32:05.530] Um... [32:05.530 --> 32:06.110] But, uh... [32:06.110 --> 32:08.390] It probably stays in the search logs. [32:11.170 --> 32:12.190] And, uh... [32:12.190 --> 32:13.550] So, actually, um... [32:13.550 --> 32:13.650] Um... [32:13.650 --> 32:15.050] Well, the answer is, uh... [32:15.050 --> 32:15.630] It depends. [32:15.890 --> 32:18.210] So, it depends on what exactly, uh... [32:18.810 --> 32:19.090] Uh... [32:19.090 --> 32:19.330] Work means. [32:20.030 --> 32:22.210] Do we want TrackMeNot to be indetectable? [32:22.410 --> 32:22.650] Uh... [32:22.650 --> 32:24.710] As I said in the beginning, uh... [32:24.710 --> 32:24.890] Uh... [32:25.530 --> 32:25.970] We'll... [32:25.970 --> 32:27.130] Like to have, uh... [32:27.130 --> 32:29.150] TrackMeNot being able to, uh... [32:29.150 --> 32:33.910] Well, we'd like to have the search engine consider that maybe every user is using TrackMeNot. [32:34.470 --> 32:34.910] Uh... [32:34.910 --> 32:40.310] If that's the case, as soon as TrackMeNot is detected by a search engine, as soon as it's used, uh... [32:41.110 --> 32:43.190] Well, we can consider that we failed. [32:45.250 --> 32:45.690] Um... [32:45.690 --> 32:47.290] So, is that a case of failure? [32:47.830 --> 32:48.170] Uh... [32:48.170 --> 32:52.050] What if 99% of TrackMeNot queries can be flagged by a search engine? [32:53.450 --> 32:53.890] Uh... [32:53.890 --> 32:55.370] Should we consider that it's still working? [32:55.530 --> 32:57.070] I mean, there is still 1%? [32:57.870 --> 32:58.310] Uh... [32:58.310 --> 32:59.110] Well, probably not. [32:59.290 --> 33:04.730] But what if 50% of my queries, my own search queries, also flagged and deleted by, uh... [33:04.730 --> 33:05.530] In the search logs? [33:05.910 --> 33:07.950] Actually, there are, um... [33:07.950 --> 33:09.030] There is a paper, uh... [33:09.610 --> 33:09.750] Uh... [33:09.750 --> 33:10.750] Which used... [33:10.750 --> 33:11.090] Uh... [33:11.090 --> 33:13.810] Machine learning algorithm to try to feature, uh... [33:13.810 --> 33:14.550] TrackMeNot queries. [33:14.850 --> 33:17.290] And on average, um... [33:17.290 --> 33:20.090] They deleted, like, 50% of, um... [33:20.830 --> 33:21.270] Uh... [33:21.270 --> 33:21.650] User queries. [33:21.850 --> 33:22.670] But, uh... [33:22.670 --> 33:27.130] Being able to remove, like, 99.9% of, uh... [33:27.130 --> 33:27.730] TrackMeNot queries. [33:27.830 --> 33:31.810] Well, it was on default configuration and on a previous version of TrackMeNot, but still. [33:31.810 --> 33:32.850] And, um... [33:33.330 --> 33:33.430] Um... [33:33.430 --> 33:34.950] Well, normally for, uh... [33:34.950 --> 33:37.150] Spam filter, if it detects, uh... [33:37.150 --> 33:39.130] 99% of spam, but, uh... [33:39.130 --> 33:41.670] Deletes 50% of your, uh... [33:41.670 --> 33:43.030] Real search, uh... [33:43.030 --> 33:43.510] Real message. [33:44.390 --> 33:45.130] Real email. [33:45.310 --> 33:45.430] Sorry. [33:46.310 --> 33:46.790] Uh... [33:46.790 --> 33:48.390] Well, we consider that it's not working. [33:48.570 --> 33:50.010] But the thing is, uh... [33:50.010 --> 33:55.410] Well, what if a search engine dropped 90% of my queries, but is still able to profile me quite accurately? [33:57.310 --> 33:57.790] Um... [33:57.790 --> 33:58.310] Maybe I just... [33:58.310 --> 33:59.690] Maybe I want to add my profile. [33:59.810 --> 34:00.970] I'm not interested in the search engine. [34:01.170 --> 34:02.350] I'm just deleting, uh... [34:02.350 --> 34:03.630] Some specific search queries. [34:06.510 --> 34:06.910] Um... [34:06.910 --> 34:07.930] And, uh... [34:07.930 --> 34:08.450] So... [34:08.450 --> 34:09.890] One of our objective, what... [34:09.890 --> 34:12.450] Was to make it harder for a search engine to, uh... [34:12.450 --> 34:13.370] Read user logs. [34:14.010 --> 34:14.410] And... [34:14.410 --> 34:15.570] The question is, uh... [34:15.570 --> 34:16.970] What should be the cost for a search engine? [34:17.170 --> 34:17.990] How many queries... [34:17.990 --> 34:18.290] Uh... [34:18.290 --> 34:20.250] What ratio of queries should it drop? [34:20.550 --> 34:21.490] To, uh... [34:21.490 --> 34:23.930] Be able to, uh... [34:23.930 --> 34:24.750] Get some information. [34:25.110 --> 34:25.510] If... [34:25.510 --> 34:26.930] Well, as I say, if... [34:26.930 --> 34:28.270] It drops 90%. [34:28.270 --> 34:28.990] Uh... [34:28.990 --> 34:31.310] Should we consider, well, that we succeed at some point. [34:32.890 --> 34:35.130] And do we want to make the profile unreadable? [34:35.310 --> 34:37.510] Generate noise and queries, uh... [34:37.510 --> 34:37.930] Without, uh... [34:39.210 --> 34:39.370] Uh... [34:39.370 --> 34:40.150] Specific targets. [34:40.550 --> 34:41.610] And about any topic. [34:41.850 --> 34:43.630] So, at this point, uh... [34:43.630 --> 34:46.570] Well, the search engine won't be able to infer anything from, uh... [34:46.570 --> 34:47.330] Our search logs. [34:47.530 --> 34:50.870] Or do we want to be able to modify the search engine... [34:50.870 --> 34:51.290] Uh... [34:51.290 --> 34:53.390] The search profile based, uh... [34:53.390 --> 34:53.930] On track... [34:53.930 --> 34:54.490] On track... [34:54.490 --> 34:54.850] Not search. [34:57.410 --> 34:58.290] And, uh... [34:58.290 --> 35:01.350] Well, do we really want to have any kind of query, uh... [35:01.350 --> 35:02.730] Sent by track me nuts? [35:03.210 --> 35:03.630] Uh... [35:03.630 --> 35:05.890] If we want to have something like, uh... [35:05.890 --> 35:06.930] Allowing, uh... [35:06.930 --> 35:07.270] Any... [35:07.270 --> 35:08.590] Every user to, uh... [35:09.170 --> 35:11.790] Say that its own query has been generated by track me nuts. [35:12.030 --> 35:13.390] Even if it's not using track me nuts. [35:13.610 --> 35:13.850] We... [35:13.850 --> 35:15.890] We would like to have any, uh... [35:15.890 --> 35:16.670] Potential, uh... [35:17.090 --> 35:18.710] Search, uh... [35:18.710 --> 35:20.190] Being generated by track me nuts. [35:21.310 --> 35:22.130] And, uh... [35:22.130 --> 35:23.590] If not, uh... [35:23.590 --> 35:23.710] Uh... [35:23.710 --> 35:25.010] Well, uh... [35:25.010 --> 35:25.470] Does this... [35:25.470 --> 35:27.910] Should the query made by track me nuts seem... [35:27.910 --> 35:28.050] Uh... [35:28.050 --> 35:30.050] Be similar only to your queries? [35:30.370 --> 35:31.430] That's more realistic. [35:31.790 --> 35:32.710] But, um... [35:35.150 --> 35:35.590] Um... [35:35.590 --> 35:36.770] So, um... [35:36.770 --> 35:37.970] Well, we have all these questions. [35:38.250 --> 35:39.690] And we would like to have your feedback. [35:39.870 --> 35:40.530] So, uh... [35:40.530 --> 35:40.990] We have... [35:40.990 --> 35:43.050] We're building community on get satisfaction. [35:43.550 --> 35:43.750] Uh... [35:43.750 --> 35:44.650] The URL is here. [35:44.870 --> 35:46.910] Get satisfaction dot com slash track me nuts. [35:46.910 --> 35:48.250] And, um... [35:48.250 --> 35:49.990] So, we're open to feedback. [35:50.310 --> 35:51.050] If you have any feedback. [35:51.510 --> 35:52.010] Like, uh... [35:52.490 --> 35:55.010] Even things that this kind of query are generated by track me nuts. [35:55.130 --> 35:56.970] And they're not really realistic. [35:57.370 --> 35:57.610] Uh... [35:57.610 --> 35:59.350] So, we could add some, uh... [35:59.350 --> 36:00.790] Regular expression to feature them. [36:02.230 --> 36:02.710] Um... [36:02.710 --> 36:03.750] If, uh... [36:03.750 --> 36:05.110] You would like, uh... [36:06.430 --> 36:06.910] To... [36:06.910 --> 36:09.010] If you would like to have track me nuts being, um... [36:10.390 --> 36:12.250] Able to have some additional features. [36:12.530 --> 36:12.750] Uh... [36:12.750 --> 36:13.710] You can suggest it. [36:14.950 --> 36:15.430] Um... [36:15.430 --> 36:15.710] Uh... [36:15.710 --> 36:16.650] Also, if you... [36:16.650 --> 36:19.030] If there are some critics about track me nuts that, uh... [36:19.030 --> 36:19.670] You want to formulate. [36:19.950 --> 36:21.150] This is a kind of suggestion. [36:21.490 --> 36:21.760] And, um... [36:22.730 --> 36:22.930] Um... [36:22.930 --> 36:23.670] Well, this... [36:23.670 --> 36:23.970] Uh... [36:23.970 --> 36:26.070] Track me nuts also evolve based on, uh... [36:26.070 --> 36:26.430] Critics. [36:28.110 --> 36:28.830] And, uh... [36:28.830 --> 36:30.090] If you have any, uh... [36:30.090 --> 36:31.950] Attach schemes that you want to try on track me nuts. [36:32.470 --> 36:32.890] Uh... [36:32.890 --> 36:33.990] Please feel free to do so. [36:34.430 --> 36:34.770] Uh... [36:34.770 --> 36:35.890] We don't have, uh... [36:36.470 --> 36:36.890] Any... [36:36.890 --> 36:38.050] Set of data. [36:38.170 --> 36:39.790] So, we cannot test, uh... [36:39.790 --> 36:40.270] Track me nuts. [36:40.510 --> 36:41.750] On, uh... [36:41.750 --> 36:42.010] User data. [36:42.010 --> 36:43.470] See if, uh... [36:43.470 --> 36:45.450] Track me nuts queries are different of user, uh... [36:45.450 --> 36:45.810] Queries. [36:46.310 --> 36:46.790] Um... [36:46.790 --> 36:49.470] But if you want to do that on your own queries, or, uh... [36:49.470 --> 36:49.630] Um... [36:49.630 --> 36:51.570] I could even give a set of, uh... [36:51.570 --> 36:54.010] My own search logs, um... [36:55.390 --> 36:55.870] Uh... [36:55.870 --> 36:56.010] Uh... [36:56.010 --> 36:58.770] If you want to try to train some, uh... [36:58.770 --> 37:02.310] Mechanism to detect track me nuts queries, or, uh... [37:11.510 --> 37:11.990] Uh... [37:11.990 --> 37:13.350] To the websites, uh... [37:13.350 --> 37:14.310] URL of the websites. [37:14.610 --> 37:15.470] And if you have any questions. [37:27.570 --> 37:28.010] Um... [37:28.010 --> 37:29.810] I'm a little bit curious about the burst mode. [37:30.230 --> 37:33.290] And about specifically how it detects... [37:33.290 --> 37:34.810] When you're actually sending a query. [37:35.750 --> 37:36.190] Um... [37:36.190 --> 37:38.970] Does it detect that you open the, like, browser page... [37:38.970 --> 37:41.270] Like, the browser search page itself, and then start... [37:41.270 --> 37:43.090] Send the algorithm, uh... [37:43.090 --> 37:45.360] Searches down query? [37:46.460 --> 37:46.900] Um... [37:46.900 --> 37:48.040] It detects, uh... [37:48.040 --> 37:49.580] Well, actually, it's based on the URL. [37:50.400 --> 37:52.000] So, there is a, uh... [37:52.000 --> 37:53.800] Regular expression which detects the URL. [37:54.640 --> 37:55.080] Um... [37:55.080 --> 37:56.880] Normally, it's the URL of the search. [37:57.460 --> 37:58.560] So, uh... [37:58.560 --> 38:01.240] As soon as you make a search, normally, it will, uh... [38:01.240 --> 38:01.900] Trigger a burst. [38:02.260 --> 38:02.700] Okay. [38:02.940 --> 38:03.480] So, it's... [38:03.480 --> 38:07.660] The burst is triggered after you get the URL back that you've searched something. [38:07.920 --> 38:08.100] Right? [38:08.420 --> 38:08.580] Yeah. [38:08.880 --> 38:13.700] So, essentially, the first search before a burst will always be the user's search. [38:13.800 --> 38:14.060] Yeah. [38:14.540 --> 38:15.780] So, that's kind of the technical one, isn't it? [38:16.100 --> 38:16.460] Uh... [38:16.460 --> 38:16.600] Yeah. [38:16.680 --> 38:19.060] That's why we have also the frequency which... [38:19.060 --> 38:21.860] Well, we have both the frequency and the... [38:23.340 --> 38:23.780] Um... [38:23.780 --> 38:25.080] It will generate, uh... [38:25.080 --> 38:26.020] Search, uh... [38:26.020 --> 38:26.660] At, uh... [38:26.660 --> 38:27.320] Given frequency. [38:27.680 --> 38:28.780] And, um... [38:29.560 --> 38:32.900] We're not sure the search engine will be able to know if, uh... [38:32.900 --> 38:35.400] Which exactly is the search that triggered the burst. [38:36.200 --> 38:36.720] So... [38:36.720 --> 38:39.600] How fast your search... [38:39.600 --> 38:39.820] Yeah. [38:40.000 --> 38:43.200] If you set the frequency, like, one every hour, uh... [38:43.200 --> 38:45.240] And then, you have, uh... [38:45.240 --> 38:47.160] Use a search in the middle of the hour. [38:47.460 --> 38:47.800] It's... [38:47.800 --> 38:48.540] It won't work. [38:53.390 --> 38:54.630] Work with, uh... [38:55.130 --> 38:56.910] Moxie Marlon spikes Google sharing. [38:57.870 --> 38:58.310] Uh... [38:58.310 --> 39:01.710] Would, would query generated from tracking on to Google? [39:03.850 --> 39:04.290] Uh... [39:04.290 --> 39:04.330] Uh... [39:04.330 --> 39:04.430] Uh... [39:05.030 --> 39:05.470] Uh... [39:05.470 --> 39:06.110] Yeah, normally, yeah. [39:06.230 --> 39:08.710] Yeah, yeah, um... [39:08.710 --> 39:09.190] I think... [39:09.190 --> 39:10.890] I, I know exactly how Google sharing works. [39:11.050 --> 39:11.870] I think it, uh... [39:11.870 --> 39:13.030] It's based on the reality. [39:13.250 --> 39:15.530] But it, it, it, it, it must redirect, uh... [39:16.110 --> 39:16.550] Uh... [39:16.550 --> 39:16.750] Uh... [39:16.750 --> 39:18.850] The traffic through the Google sharing process. [39:19.110 --> 39:22.290] So, normally, I think it, it, it, it also works with, uh... [39:24.170 --> 39:24.570] Google sharing. [39:24.590 --> 39:25.070] All right. [39:25.170 --> 39:25.290] Yeah? [39:31.870 --> 39:32.270] Um... [39:32.270 --> 39:33.830] So, I think it's 20%. [39:33.830 --> 39:34.730] So, the, uh... [39:34.730 --> 39:36.910] The number of queries that are click through are, uh... [39:36.910 --> 39:37.430] Twenty percent. [39:38.170 --> 39:40.790] I will also, as I say, will, so, uh... [39:40.790 --> 39:44.530] Like to add some things that will monitor the frequency of, uh... [39:44.530 --> 39:46.450] Click through, and eventually, uh... [39:46.450 --> 39:50.470] Which is the rank of the page which is clicked on the search results. [39:50.470 --> 39:53.470] So, if, for example, the user is only clicking on the, uh... [39:54.030 --> 39:54.390] Uh... [39:54.390 --> 39:55.110] First, uh... [39:55.110 --> 39:57.570] Search results, have something like that, also in TrackMeNot. [39:58.730 --> 39:59.130] Yes? [40:04.400 --> 40:04.800] Yeah? [40:14.070 --> 40:14.470] Uh... [40:14.470 --> 40:14.950] Can you repeat? [40:15.110 --> 40:15.710] I, I didn't... [40:29.010 --> 40:29.810] So, um... [40:29.810 --> 40:30.030] So, [40:33.170 --> 40:33.930] um... [40:33.930 --> 40:35.490] Yeah, normally, it receives the creation. [40:35.670 --> 40:37.050] We, we, we're, we're simulating, uh... [40:37.050 --> 40:37.210] Uh... [40:37.210 --> 40:38.970] Well, so, that's why, um... [40:38.970 --> 40:40.930] We, we, we are simulating keystrokes in the search bar. [40:40.930 --> 40:48.090] So that it will automatically get some suggestions from Google and from other search engine as well. [40:49.970 --> 40:50.490] Yes? [41:00.160 --> 41:00.680] No. [41:02.200 --> 41:09.220] Well, actually, currently, it may not work for an enterprise because we're not able to really... [41:09.220 --> 41:14.460] Well, if someone wants to use TrackMeNot from its enterprise, it's risky because there are these kind of keywords. [41:16.600 --> 41:18.840] But the thing is... [41:18.840 --> 41:21.300] Well, as I say, with RSS feeds, it will work. [41:23.780 --> 41:31.140] I don't know if we set TrackMeNot on a proxy from an enterprise, how many queries it should generate to be realistic. [41:31.380 --> 41:34.820] Probably too much, and maybe it will be flagged. [41:35.120 --> 41:35.880] Maybe not. [41:38.600 --> 41:39.160] But... [41:39.160 --> 41:42.040] Well, it might be an interesting idea. [41:42.040 --> 41:42.800] So... [41:44.480 --> 41:44.940] Yeah? [41:45.260 --> 41:49.320] Also, whenever you were talking about how you're currently implementing... [41:51.140 --> 41:58.160] Matching user creators, so, you know, if you do a lot of queries on the weekend, that's when it will really fill in to cover you up. [41:58.260 --> 42:01.540] What about functionality for the opposite of that? [42:01.620 --> 42:05.090] What if you like to mask your mood? [42:08.660 --> 42:11.180] Well, that's the current mode, actually. [42:11.700 --> 42:12.360] So that's... [42:12.360 --> 42:13.940] It will... [42:14.960 --> 42:15.420] Currently... [42:15.420 --> 42:16.040] That's what it's doing now. [42:16.060 --> 42:16.300] Yeah. [42:16.940 --> 42:17.160] It's... [42:17.160 --> 42:20.620] It's a fixed frequency and... [42:21.660 --> 42:22.120] Yes? [42:28.250 --> 42:33.130] One, two, three, four. [42:33.130 --> 42:34.290] No, two, three. [42:34.870 --> 42:35.090] Yes. [42:35.090 --> 42:35.370] Yes. [42:44.140 --> 42:47.140] What kind of URLs could... [43:02.750 --> 43:03.350] Okay. [43:03.590 --> 43:04.690] On the... [43:04.690 --> 43:07.650] So you would like to be able to add a custom search engines. [43:08.770 --> 43:17.150] Well, actually, to add a search engine, we just have to add some regular expression which corresponds to the search engine search URL. [43:19.030 --> 43:23.490] Normally, it's not such a big effort to add a single search engine. [43:23.690 --> 43:28.990] We just... I think we have to just add some regular expression. [43:29.110 --> 43:42.930] If you know the URL and know where the search terms are, it's... Well, currently, it's not possible from the option window of TrackMeNotes, and maybe we'll make that possible. [43:43.590 --> 43:43.790] It's... [43:43.790 --> 43:47.910] It's... I think it should be... Yeah, shouldn't be too hard. [43:55.850 --> 44:01.650] Well, the thing is, normally, you're making search on Google only when Firefox is open, so... [44:01.650 --> 44:08.170] Like that, we're able to map the user... the user habits, so... [44:09.230 --> 44:14.510] If a user is... only have his browser open during the... the weekend, normally, it should... [44:14.510 --> 44:18.070] It should work, and we will not have to be... to monitor the frequency. [44:19.810 --> 44:20.250] Um... [44:20.250 --> 44:24.730] You can have the browser running in... in background, and... [44:24.730 --> 44:25.490] It's not running as a service. [44:26.650 --> 44:27.090] What? [44:27.210 --> 44:28.470] It's not running as a service. [44:28.810 --> 44:29.230] No, no. [44:29.530 --> 44:31.170] It's... It's... Well, the thing is by... [44:31.170 --> 44:31.810] By... [44:31.810 --> 44:38.490] As an extension run on the browser, we're able to have all the parameters from the browser, the cookies, all the user agents, and... [44:38.490 --> 44:39.490] So, it's really... [44:39.490 --> 44:44.010] We're really able to simulate the same sorts of... of... [44:44.010 --> 44:45.870] Parameters and user would. [44:45.870 --> 44:46.850] And, [44:52.530 --> 44:57.270] as I said, if you made four searches, that basically they can identify your computer. [44:57.990 --> 45:00.870] So, I'm just curious how... [45:00.870 --> 45:03.550] By implementing TrackMeNot, that you can sort of... [45:07.270 --> 45:07.790] I... [45:07.790 --> 45:08.570] Could you repeat? [45:09.150 --> 45:09.670] I... [45:09.670 --> 45:18.350] In other words, a Google executive said that if you made four general searches on a blank computer, that they can basically identify your computer by the searches you made. [45:18.350 --> 45:35.690] In other words, if you're of a particular interest, that they're searching algorithms, smart enough to sort of allocate and sort of uniquely identify your particular searches with, you know, you're obviously tracking IP addresses and everything else. [45:36.150 --> 45:40.050] My question is, how does TrackMeNot stop this? [45:43.570 --> 45:44.130] Um... [45:44.130 --> 45:44.370] Well... [45:44.370 --> 45:45.150] Can I explain that correctly? [45:45.250 --> 45:47.370] Yeah, yeah, no, it's clear. [45:48.650 --> 45:49.170] Um... [45:49.170 --> 45:57.110] Well, if, by any chance, TrackMeNot is able to generate these four queries that will identify normally you, um... [45:57.110 --> 46:03.210] You can assume that Google will reconnect you with the wrong, uh, computer. [46:04.910 --> 46:05.430] Um... [46:05.430 --> 46:15.090] But, uh, as I said, we don't have access to all the data, so we're not able to know exactly what are the, uh, queries that are characteristic for some users. [46:15.090 --> 46:18.990] What, what, what, what is the signature, uh, of, uh, users? [46:20.350 --> 46:20.750] Um... [46:20.750 --> 46:31.650] Uh, I'm not really sure, actually, that, um, we do generate such, uh, specific, uh, we, we, we do some, uh, such specific search, uh, that often. [46:31.990 --> 46:33.690] I, I don't know if... [46:33.690 --> 46:43.690] Uh, when you take a look at, at your own, uh, search history, most of the time, um, this, this is, this is, well, I, I, I can only speak for my own search history. [46:43.970 --> 46:45.910] It's quite generic searches, so. [46:46.450 --> 46:55.310] But, um, we, we, without having access to, uh, all the, um, search engine data, we're not able to really answer to that. [46:59.000 --> 46:59.440] Okay. [47:01.180 --> 47:01.780] Thank you. [47:01.780 --> 47:01.820] Thank you. [47:01.920 --> 47:01.960] Thank you.