Recommendation Is Still the Holy Grail For News


If you count all the different media-related apps for the iPhone and iPad (s aapl), there are almost an infinite number of ways to consume the news — some are specific to a particular newspaper or magazine, while others try to aggregate different sources of news for you. What the media industry really needs is some way to filter all of that information in useful ways, and recommend things you might not have read yet. But so far, despite some new entrants on the scene such as News360, which launched an iPad news-reading app at DEMO today, social networks like Twitter and Facebook have proven to be the best way of getting recommended content.

News360, which is based in Russia, has had an iPhone app for some time, but launched a much more ambitious iPad version of its service at the DEMO conference in California. The company says that its news app uses “sophisticated AI technology” to pull in articles from a thousand worldwide news sources, at which point these pieces of content are “algorithmically evaluated for news freshness and source credibility.” The primary user interface for the app can be switched based on which way you hold the iPad: in portrait mode it is a list of headlines, and in landscape mode it is a series of photos representing the different news stories, which sort of float past in a stream.

The app provides an excerpt of the story, with a full version that can be read in a built-in browser, and it also groups related stories from other publications that are about the same topic (although it suffers to some extent from the “Google News problem,” in which multiple versions of the same newswire piece show up under the names of dozens of different media outlets). And in an interesting twist, News360 adds links from common keywords or names to information tabs that pull in content from Wikipedia and Freebase. Users can also link to a TripIt account and have the service pull in news from locations where they will be travelling.

The problem with News360 is essentially the same problem I have with other so-called smart aggregation services and apps, including a soon-to-launch project from The Washington Post (s wpo) called Trove (others launching similar attempts include AOL (s aol) and Yahoo (s yhoo)). Although aggregation services are nice, they are of limited use unless they can provide filtering and recommendation — in other words, cut through the noise and learn what I like. And that is very difficult to do. Even the New York Times (s nyt), which has information about me based on my participation in its Times People social network, does only a so-so job of this, in the sense that what it recommends often seems fairly random.

Arguably the biggest source of recommendation-type data is Facebook, which can see when your friends like something, when they share it, etc. Huffington Post has driven a lot of traffic to the site by making smart use of Facebook integration to recommend stories that people you follow have liked or read. The reality is that social networks like Twitter and Facebook are inherently farther ahead when it comes to recommendations because of all the social signals that are embedded in my social graph, the relationships with the people I follow and my friends and social network.

There are some services and apps that make use of the links that get passed around via Twitter and Facebook — there’s Tweeted Times and for showing you links from Twitter, and PostPost does something similar for Facebook links, producing a kind of personalized newspaper., the social news platform that Betaworks and the New York Times are launching soon, takes a different twist. And Flipboard pulls in links from your Twitter stream, your Facebook graph and RSS feeds and shows it to you.

But no one is really doing much when it comes to recommendations. I’ve tried playing with Trove, and it is not much better than a random sampling of news that is being shared on the web — and News360 seems equally haphazard. It’s possible that they could get better over time, of course, although there doesn’t seem to be any way to tell either service that it is wrong when it suggests a particular story to you. And Flipboard in particular seems well placed to do something smart with the content it is pulling in, since it acquired a semantic engine called Ellerdale, but so far hasn’t done much with it.

Others are trying to solve the recommendation conundrum, including a startup that is still in stealth mode called Woven, which former Flightcaster staffer Bradford Cross is developing as a kind of social twist on the idea of a “smart agent” that adapts as it learns about you. And Google (s goog) has tried to add social signals to Google News, but without much success. So far, if you want recommendations about what to read, your Twitter stream and Facebook graph are probably the best solution — and anyone who wants to do better is going to have to leverage both of them to do it.

Related GigaOM Pro content (sub req’d):

Post and thumbnail courtesy of Flickr user Zert Sonstige


Todd Martin

Hi Mathew. You might want to give the site I work on ( a try. FeedsAnywhere was featured on Web Worker Daily last summer and its a web based RSS/Atom feed
reader that learns what you like so you can read the most interesting stuff first. If something is incorrectly thought to be interesting you can correct it. As with any machine learning I’ve seen it
takes a little time before you get accurate recommendations, but in my opinion the results are worth it. Syncing with Google Reader can give it a jump start.

FeedsAnywhere also tries to tackle what you call the “Google News problem” by hiding near exact duplicates and grouping similar stories so you can mark them all read at once. Though there is still a lot of room to improve in this area. If you decide to check it out your feedback is appreciated.

Guillaume Decugis

I think the big difference between Facebook/Twitter and personalized News is the status it puts users on: active vs passive.

Social Networks put you in an active role: you build your network and your incentivized to publish. News readers naturally put you in a passive mode: you just read.

From there on, it’s much harder keeping you engaged and you don’t have the same amount of affinity with the content on news readers than social networks. And without engagement, affinity and usage, any recommandation strategy is likely to fail.

Now there’s however also some relevancy issues with social networks: because they’re people-centric, social networks are not good at making recommandations on niche topics or topics I don’t share with my community.

We’ve hit that wall developing a social news reader on smartphones and that’s why we decided to take a different approach for our new product (disc: I’m the CEO of The approach consists of building a topic-centric social media where people can publish and share content on topics (rather than with all their friends or followers). Eg: I can follow CleanTech but decide not to follow Mobile Industry News. Making it possible for niche topics or topics I don’t share with my friends/followers to emerge and increasing relevancy to me.

Josh Davis

Matthew, thanks so much for this article. It is the best roundup along with detailed commentary I have seen on personalized news.

This has been a dream concept for me since I got on the Internet. For the highly inquisitive types, the idea of getting personalized news makes so much sense, and I feel like we are getting closer.

In various forms I have tried RSS feeds (via Google reader and other), AllTops, various custom news sites and spent years fine tuning my categories on Reddit. I was certainly able to achieve pages that provided content that was better than what I would have achieved at a general interest news site, but I still felt like there were gaps.

The best I have found in a daily digest form is which you mentioned. If you treat your follows as a highly curated list, it does a fairly good job.

I wrote up my experience here:

I think one big advantage for me is they provide signal aggregation with news stories, but at the same time, they capture all the images and videos that are shared. Being able to see all the pictures of accounts I follow on Twitter has been reason enough for me to keeping coming back.

I have loaded up all the services you shared and am going to do some reading this morning to see if any of these services might offer some improvements to my news reading.


excuse my self plug :), but we at (news In-a-Gist) are trying something in this regard.

one of the potential problems with relying on our twitter or facebook friends is that we could miss out stumbling upon new and interesting things. and when it comes to filtering tweets from a news source, user approval (by way of retweeting) still seems the best way to identity good ones from the run-of-the-mill news.

we have used both the above signals (discovery and crowd-sourcing) and some more. lets see how it goes.


Hi Mathew,

thank you for your ideas! I don’t exactly understand your point why the Twitter- and Facebook-timelines itself are still the best place for recommendations and the TweetedTimes or PostPost fail with recommendations: they only aggregate the link-recommendations from your social graphs, so it’s basically the same in my eyes . ?
Another trend next to social aggregation of news (including algorithms) could be the social curation, what scoopIt, storify and other’s do and what our plattform does in a collaborative way with twitter-links. The german newspaper Berliner Morgenpost actually uses storify to create social-reports, an interesting step in my eyes.
In my eyes the advantage of social curation against social aggregation is, that it is more topic-related and less noisy, and it benefits from the reputation of the curators. I am interested in your thoughts about that.

Mathew Ingram

Thanks, Neil — but I don’t want recommendations from just one site, I want recommendations from every conceivable source of information all at once :-)

Joe Pemberton

Until there’s a perfect algorithm, the next best thing already exists in Twitter and Facebook.

But where the Twitter and Facebook interfaces suffer — by giving readers a firehose of links threaded with empty status updates — Flipboard innovates. Flipboard is like a filter, serving up the articles and images associated with URLs.

Roman Karachinsky

Matthew, thanks for the insight into this

As part of the team that developed News360, I can definitely appreciate your sentiment on news recommendation, and this is something we’re working on. Because of the unstructured and time-sensitive nature of the news it is very difficult to create a good system to do this – you can’t just take the algorithms that work for netflix, or pandora, or amazon, and try to adapt them to news.

Our approach is to first create a platform that is really good at identifying what the news are really about, and then try to leverage this information together with your social graph and reading patterns to give useful recommendations. We already have taken the first steps – if you connect to Facebook in News360, we’ll try to analyze your Facebook profile and create a custom feed for you based on the pages you’ve liked, your work history, where you live and so on. We are working to improve this, and start using other datasources from your social graph, and we hope to make this accurate enough to be useful.

Right now, though, I think there is still value in our implementation – try creating a new category (or use My Stories) and putting some of the stuff you like in there. I have mine set up with Google, Apple, Microsoft and a bunch of other tech companies, and it does a remarkably good job about giving me a daily stream which covers all of the tech news I’m interested in, without duplicates and overlapping coverage that I get in my Google Reader.

Comments are closed.