The upgrade we performed to TechCrunch’s Messenger bot turns 1 month today. It’s been 4 weeks since we integrated our NLP middleware into the existing bot architecture to make it benefit from our query rewriting technology, so it’s time to look at the effects it is taking.
We have received a lot of feedback during the last weeks so we carefully analyzed all of it and extracted some awesome insights that, cautiously enough, show that the upgrade is a change for the better.
Firstly, people are actually using NLQ (natural language queries) which weren’t properly addressed before, but are successfully answered now thanks to Bitext technology:
Secondly, some unexpectedly well answered queries. You see, in Machine Learning they like to talk about generalization and how it makes the system so powerful. It is essentially the base of learning, because it allows algorithms to solve tasks even when they deal with data unseen before in the training phase.
But our approach is equally powerful. There are infinite ways to purport the same meaning and our parser is robust enough to analyze them all, so anything built on top of that inherits that ability. That’s why we see a difference between the previous queries and the next ones: the following are unplanned structures that, nevertheless, the bot handles perfectly thanks to the flexibility of the query rewriting solution.
Of course, we want to thank everyone that has sent their feedback and will keep on working to solve queries as exquisite as “Feed me about what's happening in the technology industry” or “Hi, I want to ask you about Artificial Intelligence” that we’ve also received. Users make the product!