Semantics sits at the Core of Data
Innoradiant “technology stack” is based on a stratification of multiple technologies. The basic idea is to have a background layer integrating big data oriented technologies (basically distributed storage and analysis) with language aware technologies able to add intelligence to the data to be processed (social networks texts). On top of that, the user experience layer facilitate the exploitation of extracted information. We call all this "Big Data and Semantics for New Product Development".
Mixing Artificial and Human Intelligence
In order to be able to extract popular product features and to associate user attitudes to them, the machine must be able to understand the language. This is a typically hard task: years of experience working on analogous tasks prove that it can be tackled only by mixing machine learning algorithms with a human driven, rule based encoding of knowledge.
In recent years there have been spectacular advances in the field of Machine Learning, with the advent of Deep Learning Technologies. Thanks to adoption of a neural network architecture VoU is able to detect and classify human attitudes towards product features. Also we adopt single layer neural networks (aka “Word embedding”) to retrieve terms that should be considered synonyms in specific domains.
As clever as unsupervised AI techniques might be, they have limits in terms of performance and user-driven behavioral change. That’s why our linguists have delivered and keep refining every day grammars for handling peculiarity of language. This is an important phase of our hybridization technology and a crucial one to achieve high quality. This is also the reason, why serious user voice analysis cannot be language independent. We currently deal with English, French and Italian.
Big Data Processing Link
Yes we are big data compliant.
For Innoradiant the ability to move into a big data processing framework it is not a fashionable tendency but a need. Even by focusing on forums only, the amount of information to be processed in enormous. Just to have an idea, web places where people discuss about cosmetics are in the order of 14 Million, just for the English, indexed part of the web (often access by robots, such as indexing engines, is forbidden in forums).
We achieve the ability to produce time effective analysis on estimated product success by adopting a distributed secured architecture where we can add nodes in couple of minutes depending on the need of our customers.
We also exploit big data aware technology such as Apache Pig for data mining and Elastic Search for document oriented storage.
We know that your main goal is to understand if the conception of your future product will meet the approval of the market. For this reason our presentation layer let you discover answers to your questions rather than just jiggle with data. Of course you can always go one step further and play around with the complexity of the information we extract, but the point is that you don't have to.
In order to achieve this result we have adopted a user friendly, responsive user interface based on Kibana, which is in turn based on NodeJS and allows real time communication with the data storage (Elastic Search). As for the configuration of the platform, we have rather opted for a more corporate oriented framework such as Java Server Faces.