Video & Audio Deep Content Search not like Text
Posted by p-air on October 2, 2006
Yesterday, Michael Arrington over at TechCrunch put up a post titled “All the Cool Kids are Deep Tagging“. What was interesting about this post is how it really shows what aspects of keyword search are broken, by describing how several of the companies playing in video and audio search are overcoming those issues. Specifically, it’s the idea that in podcasts and videos, there may be things we’re specifically looking for, but if we left it up to keywords to be the locator of those, it could be a long afternoon of viewing videos from our results set. Instead, what we’re really looking for is meaning, for example “show me which podcasts (and where on these podcasts) Mike Arrington talks about video search technologies”. Given the number of ways that this subject might be discussed in a podcast, typing a search where Mike says the words “video search technologies” could be somewhat fruitless. But finding places where Mike discusses the various companies that tag videos to help users find what they’re looking for, or finding podcasts where Mike talks about companies that were funded and whose products help media companies better manage what content is contained in their videos, is actually very important but far more complex than keyword search.
What’s currently lacking in today’s funded search and categorization technology companies are those that have an inherent ability to effectively machine tag the transcribed content, and map that back to the location on the video or podcast where it is located. The advantage that automating the tagging would provide, is the consistency of the tagging that will be done. This becomes more and more important when needing to have machines find things in more automated ways, as is being proposed the W3C’s Semantic Web framework. It’s also important for humans needing to know how to find what they’re looking for consistently, not ad hoc based on how someone may or may not have tagged a piece of content.
Folksonomy has it’s role in search, but not at the exclusion of proper taxonomies and the application of ontologies, given that these provide structure for solving problems, especially in business. This revolves around the issue of context, and without it it’s tough to solve real problems. Context lies in the semantic relationship between the nodes of a taxonomy. While it’s very nice that people are taking the time to tag their content, as I frequently do here, there are plenty of times that they do not do so consistently or with completeness. Machine tagging remains important despite what some critics of taxonomies will some times have you believe (I found some hysteria around this from some of the bigger proponents of folksonomies).
Video and audio search, can benefit from folksonomy (as there’s a lot that is not expressed with words, but can only be described by seeing or hearing it, and machines aren’t so good at that yet, where people are), but as it relates to addressing business problems, there’s a need for taxonomies to play an important role here. What’s needed is technology that can index the meta-information along with the content, and that technology is not far from seeing the light of day in a big way soon.
Tags: video search, podcast search, semantic search, tagging
