Everyday, the top executives at tech companies are quoted by thousands of websites. Oversaid has built a way to automatically extract those quotes, and attribute them to the right person.
First, we gather online sources, and extract articles from them frequently. Then we clean up the articles and extract quotes from them using state-of-the-art Natural Language Processing (NLP). Finally, we go through each quote and attribute them to executives we're tracking, or to someone we're not tracking yet that should be on our radar.
Oversaid is tracking a long list of online tech news sources. We periodically pull all new articles from each sources, and clean each web page so that we are left with the text of the article, ready to be inspected.
Each article is parsed using NLP tools. We use patterns to detect if quotes are present, go through each one to further decide whether it's a proper quote as opposed to an expression in quotes or a movie title, etc.
Once we have all the proper quotes, it's time to figure out who they're from.
In what is the trickiest part of this whole process, we try to attribute each quote to an executive or a company. First we look at the quotes explicitly attributed to a person or a company (Ted Sarandos in the example).
Then, we go through the quotes attributed to the third person ("he" or "she"). We try to refer it to an entity previously announced in the article, using an NLP technique called coreference. In the example below, it allows us to detect that "he" is Ted Sarandos.
Of course, Oversaid doesn't know every single executive out there, so as we extract more quotes we start tracking new executives and companies.
The Oversaid newsletter features the most interesting quotes of the week and commentary to help you make sense of it all. We are currently working on the Oversaid platform, which will be a Software-as-a-Service platform for professionals.
Oversaid will let you search through quotes and conduct your own analysis. We're hoping to answer these questions:
If this data and those questions are of interest to you, please email us at [email protected] and help us build a better product.