Trapit and Prismatic. There’s also the crop of social newsreaders Flipboard, Zite, and Pulse.
After launching in April, Wavii found significantly more people were using the mobile version than the desktop/Web. The company spent the last six months improving the mobile app, with changes to the look and feel, facial recognition to cleanly present the relevant parts of photos, and more news events and topics to follow. (Many of the features promised for the new iPhone app did not appear to be up and running during a recent visit to the Web version.)
Register with Facebook or Twitter and Wavii scours your profile for details to help identify news and events you’re interested in. (You can also register with e-mail—a new feature along with Twitter registration—creating more of a blank slate Wavii profile.) Naturally, you can search; comment on and share what you’re following both on Wavii and other social networks; see what other Wavii users are following and saying; and quickly follow new people, topics, and broad categories, which Wavii will suggest based on your recent activity.
“We’re smart enough to help introduce you so you don’t have this myopia problem where it’s like the things I know and that’s all I get,” Aoun says.
Wavii also lets you follow types of stories, such as interviews, startup acquisitions, movie trailer releases, product launches, or celebrity engagements.
The company’s system searches for patterns in language and information relevant to specific types of news events. In the case of an engagement, for example, it has been programmed—through a subset of machine learning called active learning—to filter out instances when one party is, say, “engaged” in a heated debate with another, and only surface wedding engagements. “It will learn these subtle differences of language,” Aoun says.
Having learned the language, Wavii’s system begins to perform the tasks of a human news editor, selecting relevant details and making judgments about credibility. (As such, it stands in contrast to the efforts of Circa, for example, which uses teams of human editors to put easily digestible news chunks on mobile devices.)
Unlike other information extraction systems, which may recognize the verb in “A married B,” Aoun says Wavii’s system knows what married means and will look for dozens of relevant details to fill out a template for news of a wedding: Who was wed? Who was in the wedding party? How big was the ring? Who designed the dress?
It’s (relatively) easy to aggregate details about an event that’s been covered thousands of places on the Web, but to be valuable as a news source, Wavii has to gather those details early in the news cycle, when information is scarce, and with a high level of accuracy. To accomplish this, Wavii models the reliability of particular news sources for particular topics to help establish a confidence score. CNN would rate higher on politics than Dan Lewis’ blog, for example. It also filters out wire services stories that appear in hundreds of different places with only a few words changed to avoid skewing the confidence score through repetition.
“We use these things, what we think of as like multiple evidence points to lead us to conclusions, and when things pass certain thresholds, we say, now our system is going to believe it,” Aoun says.
If an event fails to meet that threshold, or sources appear to be speculating, Wavii adds a purple “rumor” tag, until the feed is updated with more data.
