Monday, April 5, 2010

Personalized news: The short history of staying informed online

I've been passionate about information dissemination on the Internet ever since I first came in touch with it. In a country like India, where information was often hoarded and brokered, the Internet seemed to be the holy grail that would cause disintermediation - keep people informed in a relevant and timely manner.

When there was a flurry of activity around news sources going online, with blogging opening up millions of content creators, certain products stood out. Those that could aggregate opinions and topics across disparate sources, and send them down personalized channels for consumption by commoners.

These products could be classified based on how they worked:

1) News and blog aggregators: Those that used natural language processing, or crowd sourcing to get the pulse of the world, and create an acceptable frontpage based on consumer or editorial votes. Google news, Slashdot, Digg, Techmeme all fall into this. The challenge has mostly been around getting the right algorithms to figure out the most relevant story, and push stories that were getting stale out, to make way for the young ones. The news and blog aggregators have been a huge hit. They allow the community to democratically decide whats relevant, and timely, and allow a good mix of sources to facilitate a diversity of opinions. I believe this is a great service to humanity, where people have mostly been misguided by a few individuals. The innumerable awards Google News got are testimony to this.

2) Personalized News: People have different tastes. The low cost of publication, and the notion of syndication allowed people to demand news that suited their tastes. Until now, editors of news papers controlled and decided whats relevant for the populace. Products started arriving that learnt user preferences over time, and either personalized or customized the news. Findory is a famous example. Some may have even known about PurpleYogi, which started as a personalized news service and later changed direction to become an eDiscovery company, Stratify. In personalized news, the system tries to learn the user's preferences automatically - probably by their click or rating behavior. In a customized news site, the user is allowed to specify his tastes - largely through topics and queries of his choice. Both result in a tailor-made site, through implicit or explicit choice. Many sites have come and gone here - often failing to get profitable. DayLife, NewsTin - both changed their focus to provide news to media and businesses; Findory and PurpleYogi died; MeeHive - an effort by Kosmix that is still around.

ReadWriteWeb has a nice survey on personalized news.

Why has personalized news been such a hard market to crack?

Laziness. People know their tastes. But they either dont want to put it out, or cannot enumerate it. Every personalized site demands users to specify or train the site first up. This is boring. I can easily think of a 100 topics I'd like to read news on, and this is not exhaustive. Everyday I spend time reading about more thats not even been in this list!

3) Curation through social networks: Humans lean towards trusting others (like editors, elders, seniors, gurus, peers) because they are lazy to figure out things for themselves. Its just too much work to do the due diligence on our own. We've evolved to be reactive - listening to interesting stuff others tell (and passing it along), or blindly following what the editors decided we should read. Humans place implicit trust on personalities and social contacts - to decide whats relevant. I found an interesting book on this subject. News and the Human Interest Story. Now, is this a coincidence that the newest form of information dissemination through Twitter and Facebook have been runaway successes?

Studies show that most messages passed around are links to other reading material - more so on Twitter and Facebook.

"The survey also shows that people of their social networks and social networking products and services relevant technology filter used to assess and react to the news. They rely on traditional e-mail and comment on stories and other tools on stocks. Among those who go online for news, 75 percent receive news via e-mail or post on social networking sites and 52 percent share links with other caused them by these means."

These networks perfectly emulate the way we have received news for generations - from someone we know and look up to. Infact, what we like and want is often influenced by the people we look up to. Is it surprising that a simple way to spread information through online societies would almost solve the personalized news problem for good?

I've been involved in the information dissemination area for quite some time. During the early years of my Phd, we had reckoned that social networks are the perfect mechanism to spread any useful information, and went on to prototype a system MetaVines, that emulated "social information filtering". In social networks, we have three phenomena that help the transfer of relevant information. Benevolence, Context and Reputation. When I get to know something, I dont mind passing it on to some one in my network, even if it costs some effort to do so. I am benevolent to the information needs of another in my network. When I get some information, I know exactly who will benefit from that, and send it only to them. When I receive information from many sources, I know how much to trust that - based on who its coming from, and my past experience with him. We built a system that used relevance feedback to assist the transfer of information across instant messaging networks. While no existing popular network has this capability right now, you can only hide or unsubscribe from a contact, the news I receive from my network is extremely topical and relevant - exactly for these reasons. Benevolence, Context and Reputation.

After years of work behind products like Google News, and on news filtering at HiveFire, I feel I've not seen a news edition thats more interesting than Twitter Times is a fantastic service that builds a personalized news paper based on the links that my sources and their sources tweet. In well-built networks with good clustering co-efficient, this ensures a high degree of relevance. Twitter Times has taken this one step further, and created news pages for Twitter Lists - carefully curated lists of people who write on topics. Twitter has enabled millions of people to be curators of online information. Twitter lists allow curation of related curators. Twitter Times allows you to build your own personalized news paper based on your influencers.

Our company Dhiti too has a labs product, called Intweetion that converts your incoming twitter stream into a navigable library of news and information. Try my own stream at

Has the holy grail of personalized news finally been solved?

No comments:

Post a Comment