The rise of social media has led to an exponential proliferation of content online and widespread demand for tools to filter that information. Popularity and relevance are the most common metrics through which to filter that content – but are they the best?
We asked three people building cutting-edge social software what they think the relationship between relevance, popularity and filtering is going to be in the future. They offered three very different responses. What do you think the future of information filtering will look like?
This article is brought to you by Thomson Reuters and the Knowledge Effect. To learn more, please click here.
Nick Halstead is the founder of Twitter-tracking service Tweetmeme and data mining startup DataSift. He thinks that both relevance and popularity will be important for filtering information in the future, but that what these concepts mean is changing dramatically.
“Relevance is changing,” he says.
“We used to think of relevance as matching a query with a result, in most part via search engines performing keyword searches. And popularity was based on single metrics (such as Pagerank) that were based on global behavior.
“What is evolving is the next evolution of both, and both are being changed by the social graph. Relevance is defined now by an understanding of a person’s likes and dislikes – Facebook and Twitter both have immense data on behavioral patterns around what you like. And content is now understood and categorized semantically. So relevance in the future will be determined by how your likes match content.
“Popularity is changing as well – services like Klout, PeerIndex etc. are proof that social authority is becoming important. The first company to find that magic ‘rank’ that suddenly means ‘ad-targeting’ (and therefore cash) will become huge.
“So the same combination of relevance plus popularity will still drive the future. See what Quora wrote about how they plan to choose content, for example. But it is going to be based on new social metrics – and fundamentally you won’t ‘search’ – you will ‘follow’ content that is matched to you. With or without needing to tell the computer.
Relevance vs Popularity in Filtering, Throughout History
Pre-press days: Religious leaders kept the content on-topic by selecting it themselves. But at the same time, they had to keep their audiences in mind. As Sy Safransky said in the most recent Sun magazine, if Jesus had been a sourpuss, he wouldn’t have been able to draw a crowd.
Days of the Printing Press: The battle between editorial responsibility and the need to sell papers pitted relevance against popularity. The most popular content subsidized the most relevant, though.
The Early Web: Easy publishing enabled so much niche content that observers worried people would become insulated in self-justifying info-ghettos. Hyper-personalized, with no counter-balance of concern for popularity?
Early Social Web: The promise of mega-distribution through Facebook and Twitter has prompted publishers to emphasize link-bait and other content that’s all about popularity.
The Future?: Nearly infinite data, through easily published content, exhaust data from online behavior and meta-content built by people and machines, based on the patterns manifested online. That’s a whole lot of content to choose from in serving Web users. How should relevance and popularity be used in deciding which of it to deliver?
Google knows it needs to adapt, quick.”
Edwin Khodabakchian is the creator of Feedly, a magazine interface for content subscribed-to from around the web. Feedly released its iPhone app this week and we called it possibly the best iPhone feed reader on the market.
Khodabakchian says that there are different kinds of web users and they need different kinds of filtering. Feedly uses relevance for discovery of sources and popularity of items for filtering and prioritization.
“We break users into 3 categories,” Edwin told us.
- Facebookers – they spend hours in Facebook. They love to connect with friends and content is just an excuse to interact, be cool, feel part of the tribe.
- Passionate Users – They care about topics, they have a specific connection with the author and brand they read. They love predictability.
- Twitter Users and Bloggers – who live in information and crave for real time.
“I think filtering is different for these different classes of users of information. For Facebookers, relevance and popularity is about: can I find something really funny or different which I can share with my friends and be cool. For passionate people, they have already sources or sites they trust – this is what make it predictable. Filtering for them is a mix of their favorite sources, what is popular, and suggestions.
“Twitter is more interesting because it has an implicit interest graph under the hood and it is time based. It is great to know what is happening right now…and it is great to get content from sources you might already know and feel passionately about.
“We focus on passionate people who want to feed their minds. Who look for a predictable experience…with a tiny bit of discovery.
“When you are on a source page in Feedly desktop, there is a ‘you might also like’ list of sites. This is collaborative filtering…similar to the Amazon you might also like.
“In doing that, we are trying to create a relationship between a user and a source or author, not a one-off article display. About 70% of the new Feedly users do not know about RSS or Google Reader. The hard part of the personalization nut is that you have 30 seconds to help them build something relevant and you can not ask them because people go blank. And it has to not feel like work.”
“People care more about presentation and packaging than the actual set of parameters behind the info feed.” -Ouriel Ohayon
It surprises me that even Feedly users who are passionate about a topic need to have so little asked of them as this before they throw up their hands. I’m not sure what this says about the value proposition of feed-based content, or the emotional weight of information overload.
Ouriel Ohayon is a serial entrepreneur, most recently the founder of mobile app recommendation service Appsfire. He argues that this emphasis on user experience is so important that it overrides both relevance and popularity as a concern when it comes to filtering information.
“The problem is not algorithmic or metrics,” Ohayon says.
“The problem is experience. You can build the best filters [but] it does not matter if the tools for experimenting them and UI for displaying them are awful. For example: Google Reader vs Flipboard.
“People care more about presentation and packaging than the actual set of parameters behind the info feed. It requires of course elements of contextual serendipity and personalization: social graph, geography (very important), freshness, interestingness (rated if you will). But eventually users don’t want to hear about that – they want a nice experience. Especially on mobile.”
As a technologist, I find that a little bit depressing. I don’t think I agree with it. I like parts of all three of these responses, but feel most closely myself like reality lies somewhere in between what Halstead said and what Khodabakchian said. I’m drawn to processes that determine quality sources of information through relevance detection, personalized as appropriate, but then filtered on the item-level by popularity or variable weight of sources.
I also want to be able to lift the hood and fiddle with the calculations as appropriate. Perhaps that points to the subjectivity of this question – because most people certainly don’t want to do that.
I expect that information overload will be dealt with in a variety of ways in the future, and for most people the end result will be to kick back and have the best content sent to them automatically, however what’s best is determined.
Do you think there’s a relationship between relevance, popularity and filtering content? What do you think it will look like in the future?