2. Answers, not analytics. This was what I felt could be achieved with a grammar based, natural approach to
text parsing. I felt that we could treat every statement out there to be an answer that corresponds to a
question. An ask-web or a find-web are perhaps more attractive alternatives to a search-web.
3. Search: Categorized and Socialized.
With categorized search, information becomes a library, not an overload. Here, you can see the conferences
that are happening on machine learning. And it’s easy to create and delete categories.
I resolved a credit card issue with my bank by being able to escalate the problem using socialized search. It
needs a Linked-In login for this feature though.
Categorized Search:
What in machine
learning?
Socialized Search:
Who in machine learning?
4. Analysis of social traffic and extraction of important entities from Twitter stream for “machine learning”.
Blogs
Gigaom
Quora
Wired
Even though we only query
Twitter, social media doesn’t
live in silos. The content
moves from one platform to
another, and can be captured
pretty much anywhere.
All these entities were picked
up from the actual content
and not through hashtags or
any other metadata.
5. Crowd wisdom can be quite easily harvested, and the algorithm works on all kinds of text data, structured or unstructured.
6. Building knowledge by looking around the neighbourhood of the term.
Simple rules, like the prefix-suffix rule, or the rule that tells the program to look for more “meaningful”
sentences and publish them first etc., give us more control over the relevance of the retrieved information.