@Makyen That pattern looks like it's already caught by Potentially bad keyword in answer and Potentially bad keyword in body; append -force if you really want to do that.
@Makyen That pattern looks like it's already caught by Potentially bad keyword in answer and Potentially bad keyword in body; append -force if you really want to do that.
@Makyen That pattern looks like it's already caught by Potentially bad keyword in answer and Potentially bad keyword in body; append -force if you really want to do that.
@Makyen That pattern looks like it's already caught by Potentially bad keyword in answer and Potentially bad keyword in body; append -force if you really want to do that.
@Makyen That pattern looks like it's already caught by Potentially bad keyword in answer and Potentially bad keyword in body; append -force if you really want to do that.
### TypeError on teward/Osiris at 2020-07-11T03:57:54Z: 'NoneType' object is not subscriptable File "/home/smokey/SmokeDetector/chatcommunicate.py", line 461, in __call__ result = self.__func__(*processed_args) File "/home/smokey/SmokeDetector/chatcommands.py", line 1969, in feedback post_url = url_to_shortlink(post_url)[6:]
### WebSocketConnectionClosedException on teward/Osiris at 2020-07-10T11:38:20Z: Connection is already closed. File "/home/smokey/SmokeDetector/excepthook.py", line 44, in run_with_except_hook
[ SmokeDetector | MS ] Bad ip for hostname in body, bad keyword in body, bad keyword in title, body starts with title and ends in url, link at beginning of body, +4 more (727): Keto Beach France by joseve on askubuntu.com
@πάνταῥεῖ Potentially bad keyword in body - Position 13-145: buy a laptop from Italy and use it for programming, but I will use the keyboard in English input and I have a question about it.<a
I asked a legitimate question on the Christianity stack exchange and within less than a minute it was downvoted and deleted as "spam". I am utterly baffled. The question was completely on topic for the site, simply asking about the practice of clerical celibacy among orthodox and catholic christi...
[ SmokeDetector | MS ] Bad keyword in answer, bad keyword in link text in answer, pattern-matching website in answer, potentially bad ns for domain in answer, potentially bad keyword in answer, +1 more (248): How can i blog component in Angular? by siyaram yadav on stackoverflow.com
Thanks, random spammer! Doesn't seem to have worked after all, though. Potentially because I don't have a puzzling.SE account.
Nope, created one and still erroring...
"FIRE requires your metasmoke account to be write-authenticated with Stack Exchange in order to submit spam flags. Your metasmoke account doesn't appear to be write-authenticated. Please open this page to authenticate with Stack Exchange."
@Makyen I have, although I didn't see that checkbox earlier and I removed write access, planning to re-add it to troubleshoot, but now I can't figure out how to re-add write access.
np. I have a minor tweak to the Flagging preferences page, which should make the actual function of that checkbox more clear, and add a link to the status page from which you can re-auth. Hopefully I'll get it finished in the near future. :;
@πάνταῥεῖ That one is a bit tricky. Because the answerer disclosed their affiliation and the question requests resources similar to what the answerer provided, we can't consider that spam, or even NAA.
While we're on the topic of tricky calls, what's the call on something like metasmoke.erwaysoftware.com/review/posts/70343? It's a completely off-topic SEO question, with a link to their site, and absolutely no effort put into asking it. TP because we don't want this posted on SO and it's mildly spammy, or FP because it's in the form of a technical question?
folks, there seem to be a pattern at The Workplace that some trolls pick user names containing "Hitler", "idiot", "sucks". Could SmokeDetector be extended to trigger warning on similar names at this site?
I'm not sure who to alert on this, but metasmoke.erwaysoftware.com/review/posts/137008 picked up the trailing part of the URL (the filename) 100.run to analyze for "Potentially bad ns for domain in answer" when the domain was actually nvidia.com. Seems like a small bug in the domain-finding code?
@RyanM (cc @double-beep) I would say that one is FP. While the link in the question does, effectively, promote their site, and they are, clearly, wanting to promote their business in their profile, the question appears to be an actual question, which is at least in the ballpark of what the site covers. So, it doesn't look like the question was intended as spam to me. The OP's responses in comments also argue against spam being their intent.
We see a reasonable number of questions like this one. We really should give the OP the benefit of doubt. OTOH, if the user continued to post multiple questions linking to their site, where the link wasn't an inherent part of the question, or gave other indications that the intent was spam, then I could agree with calling it TP, but I don't see those here.
As to their profile being spammy, yes, it is, but it doesn't seem gratuitously so. Profiles are permitted to have a reasonable amount of promotion. It's really only where they go considerably over the top, have R/A/NSFW content, or are here only to promote (and post Q&A content) when moderators will step in. After all, the "about me" portion of the profile, and the main profile page in general, are intended as a way to softly promote, or at least introduce, the user.
@SmokeDetector While this post reads like a "me too" NAA, the linked blog post is actually a blog describing how to solve the problem. I can't say that the blog was actually written by the OP (there's a partial name match, but the matched portion is very common), but their profile indicates they are definitely affiliated. This one looks more like spam, at least due to the undisclosed affiliation, to me. However, my impression is that it really was posted with the intent to promote the blog post.
@AndrewT. If you don't mind a bit of additional functionality, the alpha version of SOCVR's Request Generator provides a button on posts to open a popup with the ability to send a few different commands to SmokeDetector, including !!/report (with provided reason). If you use the userscript, you'll probably want to select the option to have all commands for SD sent to Charcoal HQ.
It also has the ability for you to set a point to revisit a post after an arbitrary amount of time. The next time you visit an SE question or review page after the time you set, the page for the post you set for a revisit will be opened in a new tab.
Note that the alpha version of SOCVR's Request Generator does not work with Greasemonkey 4 on Firefox, but does work with other userscript managers on all tested browsers. [Disclosure: I'm the primary author of the alpha version of this script.]
@JeffSchaller I haven't looked at the code in detail with respect to this occurrence, but I expect that it pulled the 100.run from the filename used in the two lines after the actual URL. The code does try to find valid domains in text outside of properly formatted URLs. Unfortunately, detecting domains outside of well formatted URLs is quite error prone, with a significant rate of detecting things that are not even domains, let alone spam domains.
You're right that the code could use work to do a better job at rejecting some common situations where it detects something as a domain, but where it's not a domain, or even something intended to be a domain.
@Makyen ahh, I see -- I think that's right. I just now noticed that MS did find nvidia in the "Contains domains" section. Agreed that it's hard to distinguish domains outside of a URL reference, so I retract my "bug" report :)
@JeffSchaller np. :) I do consider it a bug/feature request that the code there be improved, as there are at least a few situations that are common enough that it would be worth not getting reports for them.
I'm happy to help, if I can. I know I file a good number of "fp" on reports on the Unix-y sites I frequent, because filename extensions often look like domains.
Not sure if there's tangible benefit in reducing the 'fp' rate on that metric, though?
I suppose it'd mean fewer reports, if that was the only metric that flagged a post.
I'd really have to take a look at what the FP rate is for each type of mis-detection. We wouldn't want to reduce the sensitivity to the point of not detecting TP, but overall there's a few thousand FP for that report reason. I haven't looked to see how many of those are a result of detecting a domain in a situation which is programmatically easy to exclude as somewhere a domain would be located. Obvious filenames is one case in point. There are a few others which I've noticed.
It's not a high enough FP rate that it's a high priority, but it's something that would be nice to look at, and could be done at the same time as consolidating the code for detecting domains.
@gnat Could you provide some examples? There appear to be several valid users on TWP with usernames which match those keywords. It's not enough users such that we couldn't exclude them, but it does complicate the situation and indicate that those are not going to be perfect indicators.
If there are troll posts which SD isn't catching, we would appreciate the posts being manually reported. From searching on metasmoke (1, 2), it appears we've never had a manual report which didn't trigger other reasons about a post with a username matching the words you've provided.
@Makyen so, this is technically doable to have warnings like that, thanks. As for details and examples, being a regular user I don't have much of these so I passed this to TWP moderators in No Smoking chat. I think it now up to them to decide whether the site needs that or not
