Chapter 4: Metadata Magic 123 4.10 The Dream: When Metadata Really Works In the preceding sections, we described different opportunities and challenges of metadata. As one desirable state, consider the following scenario starring Bob the activist: THE MAGICAL CHAMONIX TOUR Bob is kicking himself. What seemed like a simple job is turning into a nightmare. He had volunteered to compile the material from their Chamonix trip, edit it a little and publish it on their cell’s site. He had not anticipated the amount of manual effort that goes into linking the relevant pieces to each other. Of course he could just dump the stuff to the site, but since they had taken all that trouble to collect the material, it would be a bit disappointing to let it just sit there, an unordered jumble of fi les. For instance, it was simple enough to index the images and video by location with the help of this public domain geotagging site, but it was quite a task to enter the locations by hand, linking to those items that lacked the coordinates. Also, the creator metadata in the material was text-only. People’s names were not consis- tent. It was Mike or Michael in the videos, O’Brannen in the images, MOBR in the blogs. Of course it was possible to relink the stuff but Bob’s life started to seem too short for that. When he vented off some steam on the task on the cell’s blog, Trisha suggested that Bob talk to this guy from the University of Cambridge. They had some media analysis software that might help in the task. Bob followed the lead, found this very co-operative research group, and in a week Bob was running their Experience Engine software. That proved to be one of the most useful applications ever to land on his hard disk. The software was designed to take in a large collection of ill-structured media fi les, cluster the fi les for potential relations and invent mean- ingful metadata labels, and suggest how to link the fi les into what they called an Experience Base. Just what Bob needed. After a night or two of crunching the material (and crashing twice), the applica- tion suggested some clusters based on time, location, and communication patterns. Bob was pleased to see that the application had divided the people into two groups: those who were there in Chamonix to experience the trip and those who stayed home and just received messages from the group. Bob had defi ned the various aliases of the group members, and it seemed that the software had fi lled in most of the creator metadata quite nicely. On the fi rst skim the clusters also seemed to make a lot of sense. On two occa- sions Bob had to shift video between episodes (why did the software associate Mont Blanc with dinner time? Ice cream cones with whipped cream on top? Not quite, Bob thought), and had to repair some weird automatic links between the images and video. Most of the time the software had beautifully found the correspondence between those images and the video clips, giving the video the proper geolocation tags as a by-product. Those blurry night time shots after some wine, raclette, and some more wine were quite hopeless, of course. Bob could not blame the software, as often he had no clue either to what those clips were about. In any case, the metadata made sense, and with the help of the