This week Twitter launched Promoted Tweets, an advertising platform that sheds light on its much-discussed business model. The platform takes a page out of Google's advertising playbook by letting advertisers sponsor posts that will appear at the top of Twitter search results.

Twitter hopes to take Promoted Tweets further in the long run by dropping advertising messages into the multithreaded conversation that goes on between its users. But it may find it difficult to ensure that these advertisements are relevant and useful, and it will need to tread carefully so as not to alienate its vocal, opinionated community of users.

Cofounder Biz Stone said in a blog post that Twitter plans to display "relevant Promoted Tweets in your timelines in a way that is useful to you." If the company can do this successfully, it will have solved one of the biggest issues for social networks--turning explosive popularity into explosive revenue.

But even the most popular social network, Facebook, has not yet become a goldmine for its backers. The hope that it would be possible to serve perfectly tailored ads based on users' profiles and activity has faltered because of poor click-through rates. Even so, companies are still searching for the formula that will make social advertising work.

Michael Bernstein, a researcher at the Computer Science and Artificial Intelligence Lab at MIT, has been developing algorithms for automatically identifying the subject of tweets in conjunction with researchers from the Palo Alto Research Center (PARC), including senior research scientist Ed Chi. . The good news, Bernstein says, is that a lot of the interaction on Twitter happens around trending topics (the most popular subjects of conversation at a given moment). Bernstein thinks Twitter could easily insert ads into these conversation streams, much as advertisers already target the audience of a particular show on television.

However, Bernstein points out that this is not how online advertising brings in big amounts of money. A huge percentage of Google's ad revenue comes from ads that only interest a few users a day. So to get a truly successful ad platform going, Twitter will need to identify and target much smaller groups of users who are involved in less popular topics of conversation.

This turns out to be a hard problem. The algorithms designed to extract meaning from a piece of text were intended for longer documents that usually provide plenty of cues to suggest the focus, Bernstein says. For example, a blog post about Apple's iPad will repeat the name of the product several times. In the cramped 140 characters of a tweet, users tend to avoid repetition, making it harder for an algorithm to identify the writer's focus. However, Bernstein says analyzing users' previous messages, as well as those of their networks of contacts, could make the process easier.