Searching the web requires that either you know the name of what you’re looking for, or someone looked for it the same way you created your narrative, it using the same words. This is problematic for a number of different use cases. Say you saw a movie, but you have no idea what it was called. Or who was in it. This often happens with old movies, at least for me. So I want to find out more about it, but, uh oh. I’d have to go to Reddit or Yahoo Answers or something similar and give my narrative of the film, hoping someone else has seen it.

Crowdsourcing answers is kind of a fad these days, but it seems to me like with the vast analysis Google has done on the web, and with the massive databases it has amassed with its keyword searches over the years, I have to imagine that they should be able to cross-reference this information to allow for narrative search.

I expect it would work something like this: the user types their narrative. Using keywords in the narrative, a script categorizes the narrative based on hundreds of thousands of tags generated by users. With some noise reduction and some false positive refinement, this could create an extremely accurate categorization with relatively little resources. The categorization tags could then be hashed. That hash could then search the current “keyword” web, and as the results come back, they could be indexed like the original narrative was. The results are then ordered by how close the result hashes are to the hash of the original narrative.

People use natural language to find information. It’s easier, and it makes more sense. If we want to open up the web more than ever, we need to let people find information the same way they think it. And that’s in narrative format.

I’m not a programmer. I’m not a database expert. And I’m certainly not a web search expert. But given how willing people are to freely tag and give information, the database could be easily built, and constantly optimized as people report on the accuracy of their searches and add more tags, resulting in more accurate hashes.

Please post any questions, or if you have any suggestions on how to make this a reality, please hit up the comments. I’d love to see it happen.

I think I’m going to have to try this one out.