Technology

Cloud Computing

Our system is deployed "in the cloud". This means that rather than filling up our office with servers to accommodate all the data and processing power that we require, our data-serving needs are distributed across the internet. Because of this, our system is scalable and highly resilient.

In addition, cloud-based resources are designed to grow and shrink according to demand. This means that we are able to deliver a highly cost-effective (and green!) solution because it only uses exactly the resources it needs at a given time.

Our systems primarily uses Google App Engine, with some services deployed to Amazon EC2.

Some people may think that Reputation Intelligence has magical powers. But all we're really doing is combining in one easy step what any user with a lot of time and patience could do themselves ...if they didn't mind doing it every minute of every day.

Reputation Intelligence blasts out search requests through API calls to all the major search engines and a slew of minor ones. We also look in all the social places that many search engines don't: Twitter, Flickr, YouTube, Facebook, etc. These searches are refreshed constantly so that you see new results as soon as the search engines do. Then the duplicates are weeded out so that the same results are never displayed more than once.

Essentially this is just like going to Google and typing in a search, then going to Bing and typing in the same search, then Yahoo, and so on - and then scrolling through all the pages of results to pick out the ones you haven't seen before - 24 hours a day.

But the real value of our system is in the data mining we perform on these results to pull out useful market intelligence, like share of voice and automated sentiment analysis.

Automated sentiment analysis

Whenever a new result is found, our Sentimengine™ alogrithmic sentiment analysis engine evaluates whether the mention is positive, negative, or neutral.

Sentimengine uses several methods to analyze the sentiment of a body of text. These methods are under permanent evolution, and they're getting more sophisticated every day. They include:

  • Weighted scoring based on the normative emotional ratings of commonly used English words. To put it crudely, every word in a text has an "emotional score" which, when added up, can give a rough sense of the sentiment of the text.
  • Natural language processing techniques for determining sentiment based on context. This is how a computer can tell the difference between "X kicks Y's ass" and "Y kicks X's ass".
  • Bayesian statistical analysis of the sentiments assigned over time by users. Every time someone marks a result as positive, negative, or neutral,Sentimengine gets a little smarter.