KaKoo

Nov 05 2011

How Algorithms and Editors Can Work Together to Burst the “Filter Bubble”

The algorithms that surface content for us on Facebook and Google are miracles of modern programming. But Eli Pariser, author and chairman of the board at MoveOn.org, has concerns.

In March, Pariser gave a popular TED talk about “filter bubbles” — the idea that when search and social networks only serve us content that we “like,” we’re not seeing content we need. He cited examples where liberal-leaning Facebook friends only see fellow liberals in their “Top Stories,” or a frequent traveler only got tourism results when Googling “Egypt” in the midst of the Arab Spring.

As users increasingly get their news from curated social channels, this trend has the potential to isolate us and damage our world view.

At Friday’s Mashable Media Summit, Pariser offered some solutions, and focused on how human editors and algorithms can work together to get users clicking on content that matters.

7 Things That Personalization Algorithms Do Poorly

Pariser pointed out the critical things that social personalization gets wrong when it comes to content.

Anticipation: If there’s a small story about a meeting of the Greek parliament today, a human editor could anticipate that stocks might tumble tomorrow. Algorithms are rarely good at making this kind of abstract correlation.
Risk Taking: For an algorithm to be successful, it needs to be right most of the time. Suggestion engines almost always offer up “safe” content within a very narrow spectrum. Human editors have the will to take risks on content that might be wildly successful (or fail miserably).
Big Picture: Algorithms seldom connect the dots between specs of content to form a big picture of current events. An editor can create a front page (today, a homepage) that shows the news of the day in context, and arranged by importance.
Pairing: Human editors can draw you in with something “clicky” and get you to stick around by pairing that item with something of substance. This can be an art more than a science, which is why algorithms come up short.
Social Importance: Algorithms are good at surfacing what’s popular but not necessarily what’s important. The war in Afghanistan may not be “likeable” or “clickable,” but a human editor can ensure that stories about it get seen.
Mind-Blowingness: Pariser spoke about the Napoleon Dynamite problem on Netflix. Users either loved the movie (rated it five stars) or hated it (one star). Because the Netflix algorithm doesn’t like making risky recommendations, it often eschewed Napoleon from suggestion lists — even though people who like the movie really like the movie.
Trust: People learn to trust good editors. If something seems boring or irrelevant but a trusted editor says it’s important, you’ll heed. Algorithms may never be so trustworthy.

How Do We Fix It?

In his talk, Pariser noted that nearly every major online media company and platform is moving toward some level of personalization. And why not? It drives clicks and engagement, which drives revenue.

But how can we create balance? For his book The Filter Bubble, Pariser asked the big platforms (Facebook, Google and Netflix, among others) about the difference between implicit and behavioral intent.

“In this era where we have data about everyone, do you trust behavioral data, or what people actually say they want?” he posed from the stage of the Media Summit. “If you don’t trust what users say they want, then users lose agency. You’re just sending them things they will click on.”

It’s clear the current platforms don’t get us there on their own. But by striking a balance between editors and code, Pariser thinks we can get the best of both worlds.

“The great thing about the Internet is that it’s a very malleable thing,” he said. “It’s not a medium, it’s a meta-medium.”

By hooking people with content users like and pairing it with content users need, editors can drive traffic and value simultaneously.

“How do we make hard news as irresistible as LOLcats? That is what news is competing with. We need to find new ways of packaging it,” Pariser asserted.

“The Internet can go either way. It can encapsulate us in a little bubble of our narrow interests, or it can connect us to new people and ways of thinking.” The latter is what we all hoped for, Pariser said. And his hybrid media strategy might be one way to save us from creating “a bubble of one.”

Media Summit 2011

The Mashable Media Summit on Nov. 4 at the Times Center in New York City attracted professionals in digital, tech, advertising, sales, marketing, mobile and publishing from all over the world.

Click here to view this gallery.

Presenting Sponsor: AT&T

More About: eli pariser, journalism, mashable media summit, Media, News, personalization, Tech

For more Social Media coverage:

Follow Mashable Social Media on Twitter
Become a Fan on Facebook
Subscribe to the Social Media channel
Download our free apps for Android, Mac, iPhone and iPad

Categories: News, eli pariser, journalism, mashable media summit, media, personalization, tech

Jun 03 2011

Why Web Personalization May Be Damaging Our World View

By Josh Catone

Think about all the websites you visit each day. Think about all the blog posts and articles you read, all the videos you watch and the pictures you view. Now think about how you find all that stuff. Some of it is probably recommended by friends, some of it you find directly by visiting sites you know you like. But a growing portion is being curated by robots — computer algorithms that are filtering content and deciding what we get to see.

Almost all popular websites, from search engines to social networks to media outlets, are now utilizing filters in some way to personalize content for visitors.

On the surface, there are clear benefits to filtering for both businesses and consumers. Personalization equates to greater relevance, and for web publishers, relevant content and ads means more clicks and ultimately more money. For users, relevance means less time spent finding content they’ll enjoy.

Who really has time to sift through the 161 million results Google sends back for a search about “Miley Cyrus,” anyway? If the search engine knows you’re probably looking for her latest music video or her tour schedule, what’s the harm in showing you those results on the first page?

According to former MoveOn.org executive director and current board president Eli Pariser, however, there are dangerous, unintended consequences to filtering. In this new book, The Filter Bubble, Pariser argues that all this filtering is starting to isolate us. When websites show us only what we like, we get cut off from the diverse points of view that can enrich our understanding of the world. That might be relatively harmless when you’re searching for Ms. Cyrus’ latest single, but what about when you’re trying to find information about pending legislation in Congress or news about revolution in another country?

We had a chance speak with Pariser about how filters are changing the Internet as we know it.

Q&A with Eli Pariser, Author of The Filter Bubble

Eli Pariser The promise of the Internet is that it can connect people from different backgrounds, with different beliefs and across disparate locations. How is the trend toward personalization impeding the fulfillment of that promise?

We’re more connected than ever to stuff “people like us” like — not just ads and products, but increasingly content as well. Yahoo News, for example, personalizes which articles it shows to which visitors.

There’s a simple psychological logic to this: We like to be surrounded by the familiar, and by information that confirms what we already believe. It drives up pageviews and gets visitors coming back. But it’s a problem because it means you’re less likely than ever to be confronted with information that challenges your views, or gets you out of your comfort zone. Your own point of view follows you wherever you go.

In an increasingly complex and vast media landscape, filters can provide relevance and combat information overload. Can some level of personalization be useful? What are we missing that we need to see?

Some amount of algorithmic personalization is necessary — there’s just too much stuff to sort through for humans to do it all. And tools like the Netflix movie recommendation engine can help people find content they wouldn’t otherwise be aware of.

“Like” isn’t a neutral word — it’s easy to Like “I just finished a marathon,” and hard to Like “cell phones may cause cancer.”

But if these systems aren’t designed carefully, you can miss a lot. Take the Facebook “Like” button — the main way that information gets spread on Facebook. “Like” isn’t a neutral word — it’s easy to Like “I just finished a marathon,” and hard to Like “cell phones may cause cancer.” So some kinds of information get through, and others don’t, and when that’s happening in the Facebook News Feed, where an increasing number of folks get their news, it’s a real problem.

Isn’t seeking out a diversity of information a personal responsibility? And haven’t citizens always lived in bubbles of their own making by watching a single news network or subscribing to a single newspaper?

There are a few important ways that the new filtering regime differs from the old one. First, it’s invisible — most people aren’t aware that their Google search results, Yahoo News links, or Facebook feed is being tailored in this way.

When you turn on Fox News, you know what the editing rule is — what kind of information is likely to get through and what kind is likely to be left out. But you don’t know who Google thinks you are or on what basis it’s editing your results, and therefore you don’t know what you’re missing.

Yes, we should all seek out diverse information flows. But great media makes that easy, even fun — combining information vegetables and information dessert and making sure you get enough of both. A lot of the personalization that exists today just serves up information junk food. It may be delicious, but it doesn’t feed the soul.

What is your biggest fear of runaway personalization? What are the consequences?

Being a politically-minded person, my biggest fear is probably that important but un-sexy problems — from homelessness to the war in Afghanistan — fall out of view entirely.

My biggest fear is probably that important but un-sexy problems — from homelessness to the war in Afghanistan — fall out of view entirely.

Clay Shirky points out that while most newspaper readers read the internal sections (Sports, Home and Garden, whatever), at least they had to flip by the front page which let them know if something important was going on that they should know about. Now it’s possible to live in a bubble where that stuff doesn’t ever show up — you’d never know it’s happening.

And the challenge there is: We can lose sight of our common problems, but they don’t lose sight of us.

While most personalization on the web is algorithmically driven, aren’t we implicitly informing the algorithms based on the choices we’ve previously made interacting with content? Couldn’t you then, in theory, manipulate the filter so you see what you want to see or are there too many factors beyond our control?

Yeah, a lot of people want to “trick” the algorithm — there’s some allure to psyching out Google. But there are probably too many factors at work to do it effectively.

Consider: Even if you’re completely logged out of Google, on a new computer, the company can track 57 signals about you — from what kind of laptop you’re using to what your IP address is to what the font size in your browser is. Already, that gives a lot of important clues about age, income and demographics.

It’s ironic — the promise of personalization is that it gives us our own personal view of the world. But the challenge is that a lot of the time, it’s actually pushing us toward a stereotyped, simplified version of ourselves: “This person is male, so we’ll show him more gadget and car news.”

Many of the major social, discovery and media sites on the Internet now implement some type of personalization. Do you feel these sites have a responsibility to educate consumers about how their information is being filtered? Do you think users should be able to opt out of personalization?

Yes, on both counts.

Marshall McLuhan, the media theorist, famously described media as an extension of the nervous system — television is literally eyesight from afar. The systems by which media bring information to us shape how we understand the world. And so understanding those systems and having some control over them is critical.

The companies that are doing this filtering have a huge responsibility — the same responsibility, in many ways, as the old-school media institutions they’re replacing. And they need to wield it well — to educate their users about how the filtering works, to give them some control, and to build algorithms that have this sense of civic purpose embedded in them.

You’ve said the solution to the dangers of too much personalization is to build ethics into the algorithm. How does one teach a robot gatekeeper to act with a sense of civic responsibility?

A lot of the danger here comes from relying too heavily on a few specific indicators of what people want — clicks, Likes, video views, whatever.

“Want” is complicated word, because we all have lots of conflicting desires — we want to eat cake and be thin, we want to be educated about the world and watch Jersey Shore. Looking at a variety of signals that encompass that range would be a start — for example, adding an “Important” button alongside the Facebook “Like” button. Otherwise, you can end up with your own personal Jersey Shore marathon.

How is the “filter bubble” related to privacy?

It’s sort of the inverse of privacy. Privacy is about controlling what the world is allowed to know about you. This is about controlling what you’re able to see of the world — what your filters let through and what they don’t.

The common thread is that they both have a lot to do with personal data. Personalization couldn’t exist without the massive dossiers of personal data being collected by big companies online these days. And it’s a problem because consumers don’t have much control over that. The current laws around personal data just don’t contemplate a world in which a click on one website changes what you see on an entirely different one. (Or indeed, websites at all.)

Another emerging trend in online media is the rise of human-based content curation using social media tools. Could human curation work in tandem with algorithmic filtering to help us avoid getting trapped in “filter bubbles?”

Yes, I think that’s part of the solution. There are lots of things that humans are great at that algorithms are still very bad at: anticipating what’s going to be a story, pairing different stories together, giving a representative sense of what’s happening in the world.

That’s the happy ending that’s possible here if these tools become more transparent and controllable: humans and robot algorithms, living in peaceful harmony, giving us both entertainment and the information we really need.

Images courtesy of Jen Campbell and iStockphoto, dibrova

More About: eli pariser, filters, interview, personalization

For more Social Media coverage:

Follow Mashable Social Media on Twitter
Become a Fan on Facebook
Subscribe to the Social Media channel
Download our free apps for Android, Mac, iPhone and iPad

Categories: eli pariser, features, filters, interview, personalization, social media, web

Apr 21 2011

Why Social Media Falls Short of True Web Personalization [OPINION]

By Hank Nothhaft

This post reflects the opinions of the author and not necessarily those of Mashable as a publication.

Hank Nothhaft is the co-founder and chief product officer of Trap.it, a personalized content discovery platform launching early this summer. Trapit was incubated at SRI and the CALO project.

To the credit of Mark Zuckerberg and Co., the Like button may very well be the most ingenious creation of the Web 2.0 world.

It is the simplest way to express a sentiment of approval and to broadcast that information to your social graph. The Like button has quickly become the dominant protocol for disseminating posts, pages, apps and more. And it’s not surprising that Google has countered with its own +1 concept.

But we still see the demand from users for more granular options, even in something in the form of a “dislike” button or a “love” button. The act of liking, predictably, means different things to different people.

For Facebook, “liking” really just means allowing the liked item to be a part of your network and world. And they’re sticking to it — so much so that Facebook has used the Like button to replace other gestures, like becoming a fan.

Indeed, one button to rule them all.

Except, the Like button and +1 button aren’t ultimately that useful to anyone who’s not Facebook or Google. It’s an early-inning opportunity in the face of something much bigger — true personalization.

Right now we’re all liking things around the web. As a result, my Facebook newsfeed has, ironically enough, come to resemble a Google Reader. It’s filled with updates from media outlets, bands and brands. Oh, and some updates from my actual friends are mixed up in there somewhere, too.

In the best case, the Like button has turned Facebook into the world’s largest, glorified RSS feed. When you click the Like button, you are simply subscribing to an all-or-nothing feed of content from a source. Facebook doesn’t select only those posts with a high likelihood of personal relevance. Facebook has become a broadcast platform. It’s no longer about me.

What we really need is a Me button.

High-Fidelity Interests

Now, Facebook certainly does have some insight into what I care about, articulated through my stated interests and my likes around the web. But these measures are roughly-hewn at best. They’re not adaptive, and they’re not granular.

In fact, they are more aligned with outward expressions, contrived statements of an identity I wish to project, and attributes used in targeting advertising at me.

We are constantly shifting and evolving individuals. Our focus and interests change in real time. Any system that proposes to really know me needs to have an ambient and authentic understanding of how I traipse about these more abstract concepts called “interests.”

The Like button on the other hand is a blunt instrument. Although Facebook can be said to be building a massive interest graph behind all of those likes, we’re talking about a relatively low-fidelity understanding of “me.”

Social is Only Part of the Answer

Conceptions of personalization today have become convoluted, distorted and diluted. The term has been co-opted to describe many things.

But let’s be clear about one thing: The “socialization” of the web is not personalization.

Telling me how many of my friends like an article or page is not personalization, it’s a new form of passive peer pressure or groupthink. Facebook’s idea of personalization has become the Web 2.0 equivalent of the Web 1.0 portal. Saying Facebook knows me because I like NPR and photography is like saying my “My Yahoo” page was personalized because it showed me San Francisco weather forecasts and local sports scores.

Personalization implies a much greater level of control and a significantly more tailored experience than social can ever offer.

Many companies recognize this and are attempting to address the issue. Yet even these services that most purport to establish relevance based on the interest graph fail to provide compelling personalized content recommendations.

These services all rely too heavily on collaborative techniques and misguided and simplistic interpretations of the social graph, or they simply lack the sophistication to scale beyond the same 100 (or 500) media outlets and professional blogs already saturating the echo chamber.

All of these approaches tend to converge within localized topic areas, and that, to put it bluntly, is just boring. It’s also why none of these efforts in so-called personalized content and news have succeeded or caught on with mainstream audiences.

The Me button on the other hand, starts from a fundamentally different understanding of relevance. The Me button is as easy as the Like button. One click to personalize, and to pivot.

It’s All About Me

The Me button understands I’m interested in complex, multi-dimensional topic areas or concepts, not a just sources or keywords.

The Me button moves with me as the content it discovers evolves my interest and knowledge in a topic simply by the act of being consumed. The Me button adjusts volume based on engagement.

The Me button not only makes sure I don’t miss the biggest stories of the day but is also constantly canvasing the nooks and crannies of the web to uncover for me the gems my social network never could.

The Me button combines promised precision and reach of search with the perceived serendipity of social discovery.

It will delight me day after day by proactively and accurately discovering for me the stuff I know I want and the stuff I really want but didn’t know. It will adapt to my evolving interests, continually improving and proactively exploring these boundaries to build a better understanding of … well, me.

In fact, it’s not about understanding the content as much as it’s about understanding me first and foremost. It sounds like a simple shift, but it is radically difficult to achieve, technologically speaking. What I describe is one of the hardest engineering problems of our time.

Therein lies the rub. True personalization is primarily about people, not about what people read, watch or listen to. People grow, evolve and change. We never sit still. We’re always in flux, by nature.

But this doesn’t mean that such a fine-tuned level of personalization isn’t achievable. The Me button is basically a natural evolution in content discovery and has even been positioned by some as the next major phase in the life of the consumer web. And though it’s early in the game, some of the best minds in technology are dedicating themselves to cracking this nut.

From my perspective, the most successful approach to the Me button requires tapping into a balance between its two core components — deep focus on the individual and great sourcing. Though each one on its own is a worthy challenge, the balance between the two is the ultimate and necessary goal.

Deep personalization truly places you at the center of the equation by building a unique and dynamic interest model for each and every user. This model is based on your actual tastes and preferences, not what’s trending, not what’s popular with your friends or people that look like you on paper and not solely on your outward representation in the social graph.

The other piece, great sourcing, has two facets: quality and scope — that is, the required ability to tackle a significant chunk of the real-time web, uncovering those golden nuggets outside the echo chamber, and doing so with a strong focus on the best and most relevant content.

That’s why the Me button doesn’t stop at news. Sure, it delivers to me the best and most relevant news and blog posts on my favorite topics and interests, but it also recommends deals and product information, things to do and even media like videos and podcasts, all picked just for me and based on my current context — from time of day, to location, to the platform I’m using.

The Me button is really a virtual personal assistant in the end — a prosthetic that helps me own the web, once and for all.

Shaken, Not Stirred

Don’t take me to be hating entirely on social, though. A blended approach that uses social for what social is good for is ideal. It’s just that knowing what my friends think only really matters once you know what I think.

Putting my friends before me is putting the cart before the horse. MoveOn.org board president Eli Pariser’s recent TED talk outlines this point pretty well.

He notes that what we need is an information world that “gives us a bit of Justin Bieber and a bit of Afghanistan,” marked by controls that let us filter content by any number of factors — relevance, importance, comfort (topics that can be difficult to discuss or read), challenge level and points of view (with an option to see “alternative”).

The key to successful personalization is remembering that I am both the origin and the target of the system’s relevance. But that said, it needs to expand and routinely test my bounds if I am to grow.

Ultimately, the Me button should operate in a similar fashion, giving me a hearty main course of content selected just for me with sides of the best of what’s currently important in my social network and the biggest or most important stories of the day at a macro level.

The next generation of the web will be naturally and necessarily driven by the potential of personalization, in all aspects from mobile to media. The social web has brought us so far, but in doing so has also emphasized its own limitations in this area. When we finally achieve this goal, the result will be as important and ubiquitous as search or social networking has been in other stages of the web’s evolution. But let’s not water it down with lightweight parameters and misrepresentations of personalization’s true potential.

Interested in more Social Media resources? Check out Mashable Explore, a new way to discover information on your favorite Mashable topics.

More About: facebook, like button, Opinion, personalization, social graph, social media

For more Social Media coverage:

Follow Mashable Social Media on Twitter
Become a Fan on Facebook
Subscribe to the Social Media channel
Download our free apps for Android, Mac, iPhone and iPad

Categories: Opinion, contributor, facebook, features, like button, personalization, social graph, social media