When looking for content that you would like to see, you might expect to get the best recommendations from someone with similar tastes to yours. The problem though, is how to find such a person.
Collaborative filtering uses techniques for finding people with similar interests to yours and makes content recommendations for you on this basis.
For this to work:
(paraphrased from AT&T Labs 2001 Research Paper)
Steem is a growing network, and with more than 50,000 weekly active users, probably qualifies is this respect. More users would help improve performance though.
Yes it does! Votes are the most obvious way to provide this information, but resteems and follows might be other signals that could also be used.
These algorithms do tend to require a bit of work, and significant computational resources, but I believe there is plenty that competent programmers can do to improve user experiences. With the above in place.
In order to have a really good set of data for collaborative filtering, we not only need to have the recommendations (votes data), but also some way of indicating if users have seen the content (views data). The main reason is so we can distinguish between people who just haven't seen the content, and those who did see the content, but didn't vote for it. The latter group could be regarded as giving the content a mild negative rating, and this is valuable information.
The votes data for all content is publicly available from the blockchain, but the views data for Steemit.com is held in an 'off-chain' database owned/managed by Steemit Inc. I'm not sure whether the other front-end projects also collect views data in their own databases, but since their traffic is much lower than Steemit, I suspect it will have little value for this purpose.
I know Steemit Inc has a lot to be working on, and if they don't have time to be implementing these kinds of additions to steemit.com, perhaps they might think about making the content views data available via an API, so that 3rd party developers (like myself) can innovate in this direction to try and create great new user experiences.
I understand there may be concerns about the privacy aspects of releasing such data, but releasing it under some form of non-disclosure agreement, or perhaps even in some anonymised form would be far better than nothing.
Any chance 
Releasing views data may also enable analytics interfaces, and allow authors to evaluate which content is attracting most external views, where these are coming from and perhaps boost their SEO skills. This would surely help drive more people towards Steemit too.
