So I'm trying to make sense of the hashtag usage data and I think today's trends are #pixelfed, #rubykaigi, #ff and #vendredilecture and it seems like those actually kind of make sense
Current trends according to the new algorithm I'm trying to implement. Deciding values are unique accounts instead of total usage per time period (1 day), threshold of 5 unique uses in current time period, sort by (O-E)^2/E where O > E (O for unique uses during current time period, E for unique uses during past equivalent time period)
Well here's the code, feedback appreciated https://github.com/tootsuite/mastodon/pull/7697/files
@Gargron I wonder if it's worth comparing the percentage of unique accounts tooting a hashtag out of the baseline of total unique accounts tooting that day? otherwise I think it might be easier to get a hashtag "trending" on days with lower traffic (since hashtags maintaining or even dramatically increasing their percentage of toots from a high-traffic day to a lower-traffic day will have lower raw counts compared to the previous day than a new hashtag that hadn't been trending the day before?)
@enkiv2 @aparrish This is all a bit beyond my expertise. I think total toots per day should have no effect on trends, since every hashtag has a base frequency it should be equally easy to trend on a low traffic and high traffic day as long as there's a genuine increase in frequency (threshold=5 accounts tooting about it)
I don't know what TFIDF stands for, sorry
here's what I mean (numbers indicate unique users)
day | #foo | #bar | total
--- | ---- | ---- | -----
1 | 100 | 5 | 1000
2 | 150 | 25 | 500
according to your formula, on day 2, #foo has a score of 25 ((150-100)^2/100). #bar has a score of 80 ((25 - 5)^2/5). however, as a total *percentage* of unique users using the hashtag, #foo increased from 10% on day 1 to 30% on day 2 (+20%), while #bar only increased from 0.5% on day 1 to 5% on day 2 (+4.5%).
@dataKnightmare @aparrish @Gargron
Trending hashtags are only visible to instance admins, right?
If so, I don't see how they would benefit from this distinction (though they might benefit from local vs federated rates)
@dataKnightmare @aparrish @Gargron
I might be misunderstanding.
@enkiv2 what do you mean only admins? afaik they are (will be since momentarily hidden) a standard features for users @dataKnightmare @aparrish @Gargron
@charlyblack
Alright, I guess I misunderstood.
@dataKnightmare @aparrish @Gargron
@enkiv2 yeah it's already implemented on m.s, now hidden, but yesterday we saw it tested for some time. I don't know how many other instances are already running the latest release though.
@aparrish @enkiv2 I'm not sure bar *isn't* deserving of being considered a trend under those circumstances