Social News Needs a Nuanced ‘Like’, Quickly


If your Twitter timeline has been flooded with messages recently, you are not the only one. The service that converts your timeline into a ‘daily newspaper’ is incredibly popular which makes perfect sense from a UX perspective. The service gives the user a preview of the content behind the link. This makes it easier for us to decide whether we find it relevant for them or not. Other Twitter clients similar ways of adding context to tweets: “Flipboard” creates a social magazine using your timeline,  “Blu” loads your shared photos and gives conversational context for conversations (look into my post about this) and the new web interface (discussed here) uses the side pane for this. While I love the solutions provided, the flood also shows that link sharing is a moving target. A lowered threshold for shared news, leads to more news sharing. So, will we drown in the’s and their counterparts soon? Will we get suffocated by information overload, or is there a way in which we can keep our heads above the water? In this post I would like to speculate about filtering as a solution and its bottlenecks.

Well, the counterforce to sharing is ‘curation’ or its technological equivalent ‘filtering’. Also, in an overload situation you can make money from ‘curation’, ’filtering’ or ‘selection’, so there are many initiatives to try to do this. It is a difficult task though. It is like search but a lot more difficult. There are two extra difficulties. First, like with search, you need to have a way to measure relevance but it needs to be swift. Second, you need to find ‘local maxima’. In search, content that is relevant for everybody is usually also relevant for you personally so a global approach works fine. But news has a much smaller scope in place and time, so there is less data to mine in order to decide relevance.

Googles Page rank algorithm uses hyperlinks to decide on relevance. Google treats each hyperlink as a vote and combing these cleverly gives information about the global relevance of a site. For news hyperlinks are too slow. Bloggers are quick to make hyperlinks, sometimes on the same day, but often in the weeks or months, after content has appeared – but for news this is far too slow. Luckily there are ‘voting systems’ with a lower threshold. For example, social bookmark sites such as delicious, stumble upon and Digg collect bookmarks, and users are quick to make these. Bookmark sharing sites also have a fairly good idea of the interests of their users. Twitter has two popularity mechanisms it has the ‘retweet’ option which is a sign of both popularity and interest to you and your social group. And it has the ‘favorite’ which is a bookmarking option, but slightly different from those on social bookmarking sites. Link shorteners have important data as well. Most link shorteners keep statistics about clicks on the shortened links. This is one of the ways they may make money. Finally there is the ‘Like’ button of Facebook, which gives a measure of popularity. Facebook is trying to scatter like buttons to the whole web, with its ‘open graph’ protocol.

Compared with hyperlinks these faster voting mechanisms have two disadvantages. First, the text around a hyperlink can be used to improve search. But a retweet is less ‘contextualized’. Second, something that has a low threshold to use, is less selective to the quality of the content.  So a second type of context that we need for personalized filtering is your ‘social graph’: who you know and what they are sharing. As news is a social by nature, your social surroundings are valuable in personalizing your news feed. Look at the social graph as a set of pipelines running from you to all your friends and from all your friends to their friends and so on. The algorithm could open or close the valves of these pipelines a bit more, depending on collective voting behavior. If you retweet one of my links, you are more likely to value my news and so are your friends, be it to a lesser extent. All these pipelines can be put a bit more open. If I happen to neglect or ‘view and not act on’ a bit of content, pipelines can be made weaker. This mathematical approach is the equivalent of page rank, and it also helps to solve the problem in a localized rather than global way. No wonder and flipboard print shared content from the people you follow (often close in the social graph).

On top of this basic social information we can build a second layer which helps to pull some of the content someone is sharing apart. To simplify: I follow Bob, Bob shares content about Car and about Politics, I only ‘vote’ for car content, as well as some other people. If the system sees others are acting on a share from ‘Bob’, and these turn out to be the ‘car’ people. The system can decide it is relevant to me. The system can also build general information about the people who are sharing. Klout for instance, measures influence. Klout calls people who reliably share information, that encourages followers to act, ‘syndicators’ and ‘curators’. This is general information about people that can improve the social graph. I guess you can imagine many other ways to classify people, which can help the social graph. Bookmark sharing services can learn about you in a similar way. Using this background information is called Bayesian filtering, and it will improve the social graph even further.

The bottleneck of this system is your ‘voting behavior’, whether it is on Digg or Twitter or Facebook. We share enough, but we are not active in giving information about what we like from others, so the recommendation filters can’t learn about that. Impressed by the numbers of a Pew Internet report about news, Dave Pell called America the “curation nation”. However, he looked at sharing as a form of curation, which gives global background information, like I discussed in the last paragraph. In contrast, however, I would argue that compared with the amount of content which is shared, and the complexity of social graph construction the ‘curation’ levels are inadequate. Also, there is a resonance in the recommendation system. The more the filter does for you, the more news you get which follows the (stochastic) recommendation rules, the more news that follows these rules you will ‘like’. This needs to be compensated for unless it is ‘right’ from the start.

So how can we improve ‘curation’? Any solution needs to be low threshold, close to what people would do with news which they value anyway and preferably it would carry more information than a ‘like’, expressed in a certain social group. To improve ‘like’ we need to probe into the personal, rather than the social domain. I could imagine a twitter client offering ‘boxes’ to put my bookmarked tweets in. This has personal benefits (keeping myself organized), and it could have social benefits (building a folksonomy that adds context to my social graph). Another solution gets even ‘closer’ to my behavior and measures which ‘posts’ I click on and the time I look at a certain post –possibly compared with other uses and uses this as measure of interest. If all your ‘privacy alarm bells’ are going off: good! ‘Curation’ may be improved but the information should not belong to Facebook (see my Diaspora post) or Google (see my Googlization, and Openness according to Google posts). I guess it is time for an ‘open like’ standard. Just let it be a ‘rich’ one.

2 Responses to “Social News Needs a Nuanced ‘Like’, Quickly”

  1. Just a quick addition: Twitter just acquired summify, a clear sign that filtering/curation is an important topic at the company.

    See this gigaOm post.

  1. 1 Reading Lev Manovich’ “The Language of New Media” « @koenvanturnhout MacroBlog

Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Google photo

You are commenting using your Google account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s

%d bloggers like this: