What really drives delivery, particularly at the consumer mailbox providers, is engagement.
The big drivers of engagement are
- having permission to send email and
- sending mail users want to receive and interact with.
Authentication is there so that the filtering engines know what mail is actually from you. It allows them to be really harsh on spam forging your domain or sent without your authority and still delivering your legitimate mail to the inbox. If your mail is fully authenticated and still going to the bulk folder, then the problem is related to your email. Something you’re doing, whether it’s a permission problem or an engagement problem or whatever, is making the filters think your mail isn’t wanted.
Fixing authentication isn’t going to fix delivery problems caused by authenticated email.
“They’re doing what they’re supposed to do,” said Wedbush analyst Michael Pachter. “They’re growing revenue and controlling operating expenses.”
Central to Twitter’s changes is an effort to promote healthy discourse, after the company has struggled to rein in toxic behavior. It has also been working to make the platform more conversational.
Twitter Chief Executive Jack Dorsey told analysts Tuesday that the company is now taking a more proactive approach to addressing abuse and its effects on the platform.
It has made changes that are “taking a bunch of the burden away from the victims of abuse and harassment on our service and we’re making the agents that we have working on this process much more effective and much more efficient,” Mr. Dorsey said.
Earlier this month, Twitter announced that it was using machine learning algorithms to proactively surface abusive tweets to its teams for review. Previously, Twitter users had to flag offending tweets before the company would remove them. Now, about 38% of abusive content that Twitter takes down is surfaced by the algorithms.
This isn’t the first time Twitter has used machine-learning technology, but previously it was applied to tracking spam.
Twitter is also now using a new appeals process to better prioritize tweets it should remove that contain private information about its users. Twitter now removes 2.5 times more tweets that share personal information since the launch of this feature.
We’ve gotten to the point, particularly with Google but also with the other webmail providers, where the bulk of egregious spam is blocked. What’s left is not some spammer sending 10MM messages, but a much more difficult problem. Spam that reaches the inbox is sent in much smaller quantities. It’s also heavily targeted. Spammers are trying to look like legitimate marketers but still sending mail without permission.
This targeted spam is something I’ve been thinking about a lot lately. Mostly because anti-spammers did a pretty good job making not-spamming look like it was beneficial to senders. Many deliverability recommendations boil down to stop spamming but phrased in a way that makes the advice more palatable. Much of the type of spam that’s getting caught in the new filters follows deliverability recommendations. The piece it misses is that it’s not being sent with the permission of the recipient.
.. Believe it or not, spam filters started out as protecting users from mail they didn’t ask for. As the internet as grown and email has become a channel for crime the focus of filters have changed. But, fundamentally, deep down, the original purpose of keeping mail boxes useful by stopping unsolicited mail is still there. The ML filters are giving Google, and others, tools to actually address that mail better.
The bad industries ones will be transformed before the good ones. What I mean by that is that computer vision applied to medical imaging would be huge. But the detection/classification isn’t accurate enough for that field, just yet. Yes, results are amazing on standard datasets such as ImageNet but they fail to become equally good when there are orders of magnitudes less amount of data. And in the field, accuracy is very important, a net classifying cancer correctly 90 % of the time is likely useless.
One exception is automated language translation which is getting very good. I’m noticing that some of the articles papers I’m reading are machine translated. They appear to apply machine translation to English articles and then have some editor doing manual touch-ups which seldom is enough.
The “bad” industries such as spam and SEO can definitely benefit from ML as it exists today. There are ML algorithms (LSTM) that can generate faked web sites with images that, from Googlebot’s point of view, are completely indistinguishable from real sites. Another use would be to generate realistic looking accounts in social media to steer the conversation, perhaps for political purposes. Porn obviously, could also use ML due to the huge amount of data (the porn itself and user interactions) available.
.. I think it’s pretty safe to say finance will be a big one. Finance has a large amount of individuals and firms researching the applications of ML methodologies to financial indicators. With the semi-recent rise of quant firms, I think this research is only going to get more aggressive, and HFT will become more lucrative and more automated as long as regulation does not get in the way.
.. HFT – yes. But longer-term investment (i.e. Buffett – or even with a horizon of a couple of years) is unlikely to be transformed soon – ML needs vast historical data, which is very slow to generate. Waiting 10 years only gives you 10 years of history, which is 5 non-overlapping 2-year forward returns, and maybe 1 or 2 economic/financial regimes.
This is also a problem with new datasets being generated – there is not nearly enough history available to test them or feed them to a ML system.
Furthermore, arguably, longer-term investment requires forward-looking modelling of scenarios, based on the kinds of inputs that were not seen in history. ML is not very applicable when you get big covariate shifts.
So I would say human financial analysts are not going anywhere, and any improvements would be relatively small and incremental... HFT is not profitable. Its completely commoditized.