Today we're doing the putting the final touches on LikeLines: Video highlights via web-scale aggregation of moments that viewers like, our entry to the mozilla Drumbeat Unlocking Video challenge, which is closing tomorrow.
The challenge addresses the question "How can new web video tools transform news storytelling?" Our answer to this question is the paradigm of distributed directing that allows news reports to be generated automatically, but without a central reporter. The raw material is footage captured by individuals with cameras and mobile phones who witness an event. One challenge faced is how to filter this footage: in particular, how to find the most interesting points? LikeLines gives the answer to this question.
The LikeLines concept is basically a heatmap that shows how many people found certain portions of a video interesting. You use it if you don't want to watch a video all the way through. Instead, you click the heatmap to jump in to just to the places that are worth watching start watching from there. What's worth watching is decided on the basis of what other viewers found worth watching -- either they tag those segments explicitly by clicking a "like" button or else they let the player record their stop, starting and cuing behavior. We also want the player to be able to make use of multimedia content analysis (visual analysis or speech recognition) in order to be able to "seed" interesting moments. This sort of seeding user contributions with multimedia content analysis has been used by our colleagues:
Ewine Smits and Alan Hanjalic. 2010. A System Concept for Socially Enriched Access to Soccer Video Collections. IEEE MultiMedia 17, 4 (October 2010), 26-35.
Our entry is in the form of video:
The video has been finished for a while now, and now I am just adding some text to make it clear that the idea is elegant, but also quite clever in that you combine user input and multimedia content analysis, which allows you to bootstrap from raw video.
It's a little crazy trying to write, because I need to switch out of research paper mode in to the mode of "hey, this will really work" and "hey look everyone, this is totally needed, totally non-trivial and totally does not exist anywhere yet". I am working now (when I stopped to write a blog post) on a sentence communicating that we can address the cold start issue with content analysis based seeding. And that verification using content analysis will help to control spam. And that if all goes well the whole thing should be able to learn by itself: It will require some R&D effort, but all the pieces of technology needed already exist.
Also, I had a little bit of trouble getting the right tone for the biography. So we're big shots at a cool technical university in the Netherlands? I guess that's important to communicate. But how to say that we are also passionate about supporting distributed and democratic news? Do I divulge that the first draft of LikeLines was churned out on a bus from Boston to Portland, the video was recorded in a long after-hours effort, and the whole thing has been discussed in every detail in chat sessions?
And how to communicate that we are doing this because it's what we love to do? We had some light-hearted lines in the bio to convey this tone (about our cat-video habit on YouTube and about me largely eschewing social media for the traditional postcard), but those got dumped in favor of some harder hitting facts about our experience in this area: right people, right skills, right place, right time...
In the end, I'm also in this to experience the crowdsourcing aspect of working on innovation in a open collaboration environment. What a breath of fresh air in the daily grind of publish and perish. And the giddy joy of communicating a concept that is ripe, feasible and useful.
Distribution of paper citations over time
2 months ago