Could you add the screenshot please? Thanks!

Oops, sorry.

Deleted: Screenshot_20160815-093617.png 109 KB

	Screenshot_20160815-093617.png 109 KB View Download

@mastiz: Seems like a ML issue. Can you investigate pls?

@mdw: Can you pls tell us the output of https://www.google.com/chromesuggestions?

)]}'
{"provider":"CONFIG_BASED_MIXER","experiment_id":0,"suggestions":[{"thumbnail":"https://www.google.com/webpagethumbnail?c\u003d63\u0026d\u003dhttps://news.ycombinator.com/\u0026r\u003d3\u0026s\u003d296:188\u0026a\u003dXFwSvYG2qkZ-I8VaM7Ply6LsfEI","feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":151},"title":"Hacker News","url":"https://news.ycombinator.com/","providers":["BROWSING_HISTORY"]},{"thumbnail":"https://www.google.com/webpagethumbnail?c\u003d63\u0026d\u003dhttps://www.facebook.com/\u0026r\u003d3\u0026s\u003d296:188\u0026a\u003defcizMTExbO14vSZwCYusAnxs1I","feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":133},"title":"Facebook","url":"https://www.facebook.com/","providers":["BROWSING_HISTORY"]},{"thumbnail":"https://www.google.com/webpagethumbnail?c\u003d63\u0026d\u003dhttps://plus.sandbox.google.com/\u0026r\u003d3\u0026s\u003d296:188\u0026a\u003dVjYm9AXXrUMI_tacsTOdxOr6RdU","feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":88},"title":"Google+","url":"https://plus.sandbox.google.com/","providers":["BROWSING_HISTORY"]},{"thumbnail":"https://www.google.com/webpagethumbnail?c\u003d63\u0026d\u003dhttp://www.huffingtonpost.com/\u0026r\u003d3\u0026s\u003d296:188\u0026a\u003dz9fYHhYrcPPE5DyBSdaqLQmhQQQ","feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":83},"title":"Huffington Post","url":"http://www.huffingtonpost.com/","providers":["BROWSING_HISTORY"]},{"feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":81},"title":"Teams","url":"https://teams.googleplex.com/","providers":["BROWSING_HISTORY"]},{"thumbnail":"https://www.google.com/webpagethumbnail?c\u003d63\u0026d\u003dhttp://www.cnn.com/\u0026r\u003d3\u0026s\u003d296:188\u0026a\u003dRwXf8qO2Zq_q8AhvKo_qFL6JDVw","feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":80},"title":"CNN - Breaking News, Latest News and Videos","url":"http://www.cnn.com/","providers":["BROWSING_HISTORY"]},{"feedback_stats":{"times_shown":2643,"times_clicked":0,"times_another_clicked":0,"pre_score":43},"title":"Popular \u2013 Memegen","url":"https://memegen.googleplex.com/popular","providers":["BROWSING_HISTORY"]},{"thumbnail":"https://www.google.com/webpagethumbnail?c\u003d63\u0026d\u003dhttp://www.nytimes.com/\u0026r\u003d3\u0026s\u003d296:188\u0026a\u003dSVL5mT818zXlWyfR-RFYeLVJHyo","feedback_stats":{"times_shown":1775,"times_clicked":0,"times_another_clicked":0,"pre_score":39},"title":"The New York Times - Breaking News, World News \u0026 Multimedia","url":"http://www.nytimes.com/","providers":["BROWSING_HISTORY"]}],"ranker_config":{"feedback_enabled":false,"num_days_for_staleness":17,"experiment_id":0,"max_sorting_error":0.3,"min_history_entries_mobile":40,"profile_expiration_days":12,"min_history_entries":40,"feedback_rctr_exponent":1,"transition_typed_weight":1,"feedback_prior_times_other_clicked":24,"feedback_prior_times_clicked":3,"time_of_day_window_secs":7200,"min_clusters_mobile":8,"time_of_day_scoring_boost":1,"min_clusters":4,"reject_tab_reloads":true,"feedback_ctr_exponent":1,"max_suggestions":8,"queryref_penalty_threshold":8,"suggested_link_scoring_weight":2,"max_history_entries":11200,"feedback_prior_times_shown":48},"cluster_stats":{"entry_count":32096,"examined_entry_count":17500,"cluster_count":227,"blacklisted_count":0,"matching_device_entry_count":15614,"filtered_entry_count":1556},"timestamp":1471362833899835}

Hi mdw@, thanks for reporting this issue!

Our pipeline produces two separate sets for desktop and mobile, and your dump above reflects desktop tiles which are not useful. 

You can only fetch proto-wire-format-version of the tiles for mobile, in:

https://www.google.com/chromesuggestions?t=2

If you are multi-logged-in, please make sure you choose &authuser= appropriately.

Please attach the binary dump so we can debug this further.

Hi Matt,

I suspect you are seeing one suggestion for www.huffingtonpost.com and one for m.huffingtonpost.com. I had the same issue with www.spiegel.de and m.spiegel.de. I thought that this is a bug in ML (mixing desktop and mobile suggestions), but it turned out I was actually using both URLs inadvertently.

Using the URL in Comment 5 you should see both show up as suggestions.

HTH

Patrick

For the record, we also filed the internal b/30910181, which could be a dup.

It is likely that solving your case requires a rather drastic change in the suggestions pipeline, which we'll need to launch behind an experiment/variation.

I am not sure that I am using both URLs knowingly. On mobile, going to huffingtonpost.com redirects you to m.huffpost.com. I do see m.huffpost.com as an omnibox completion (3rd or 4th rank) for "huff", though, so perhaps this did happen at some point - I am not sure.

As far as the user is concerned, though, this is a bug. I am not sure of the right way of fixing it; perhaps you can look for identical favicons and similar hostname patterns.

Requested file attached. Let me know if you need any other information for help debugging this.

Deleted: chromesuggestions 1.6 KB

chromesuggestions 1.6 KB View Download

Thanks again for sharing details!

Agreed this is a bug from the user's perspective, although it's not particularly easy to reproduce.

I extended our future experiment plan (https://docs.google.com/document/d/1xcvmrIIPyjnZv2uj9lHo2LSPAuDNINUMYft2kUWRKPg) with a dedicated experiment. I imagine two possible solutions:
1. (less intrusive) Use redirect information to merge tiles.
2. Never process redirected pages and instead replace them with their redirects.

I will let you know once the Finch experiment is running so you can verify whether it solves your particular case (we can't verify that without having access to your whole history, which is not a reasonable think to ask for).

Unfortunately, the ETA for the experiment is second half of September, since we're blocked on some infrastructure issues meanwhile, as well as M54 reaching beta (where our UMA metrics have been polished).

Sorry for the delay. mdw@: can you verify if the bug still reproduces for you? (not that we fixed anything, but we would rule out a one-off weirdness if still reproduces)

Well, since I have reported this bug back in August I have switched devices, so on my current device I don't see this -- but that's not to say the bug is fixed.

Switching devices shouldn't be a problem if you didn't clear your synced Chrome history.

We now have a bunch of strategies to prevent this kind of dups, but they're not bullet-proof and I'm not very confident that it fixed your particular issue. My guess is that you at least once requested the desktop version of the page from mobile, which could explain the scenario.

Depending on how much time you're willing to spend debugging, we could select a date in the past and try to reproduce your issue in Most Likely's debug console (mostlikely.corp.google.com, select "mobile" in the algorithm tab and a date in the past, like 2016-09-01; beware of multilogin). If you can see the two duplicate tiles, we can double check if the fixes in place are effective. For this, you'd need to enable a experiment manually (http://mostlikely.corp.google.com?experiment_id=3310888).

Great

Update: we've rolled out some improvements to all users that fix *some* duplicates. Unfortunately, I believe it shouldn't fix your issue because the domain differs (huffingtonpost.com vs huffpost.com).

As I mentioned earlier, experiment 3310888 would solve this problem. We've been running this for a while and unfortunately the metrics we see don't back up a launch to all users.

I'm trying to collect other kind of metrics that would support the launch of 3310888, but I'm afraid I don't have an ETA.

Sorry for the long delay, a quick update: we've rolled out a few improvements to dedup tiles but this particular class of problems with multiple domains involved is something we've consciously avoided due to potential risk of overmerging (across unrelated sites).

It also seems to be a one-off issue as a result of requesting the desktop version, although we never confirmed this suspicion.

Hence, I'll mark this as won't fix, please reopen if you disagree, and thanks for reporting.