Knowledge Graph

Knowledge Graph

Statistics of the co-occurrence of normalized forms of particular words and phrases are the basis of all NLP. But to capture the meaning in a way that matches the language-independent thinking of an expert we need to have something more than just these descriptions of surface form distribution. To achieve this we regard a set of keywords or a piece of content as a guide for navigation in a semantic model for a particular domain. A query, a web page, or any other content, will be mapped to a specific neighbourhood in a semantic space representing expert knowledge for a particular domain. This mapping is based on entity and relationship recognition, using Deep Learning and other Machine Learning techniques. This higher level approach also allows us to free ourselves from being bound to text only. Images and video are mapped to exactly the same structure as natural language.

Segmentation and sentiment

There is also an independent but important contextual semantics element required for proper segmentation of the audience: sentiment. We enable free combination of sentiment and entities for fully empowered contextual segmentation.

Explainable AI

So when we need to compare for intent matching purposes a keyword segment, a keyword list, or a set of URLs against each other, we do this via mapping the items to the expert knowledge model. This enables us to also discover new relationships between concepts and to discover novel contextual structures in unstructured content. Our model is also fully transparent and readable to a human expert — we feel Explainable AI is a critical requirement for validation of the predictions of our model, and it opens up exciting opportunities for experts to discover new knowledge both automatically and understandably.

The architecture

Our approach is based on an architecture of scalable microservices, using modern frameworks, and hosted in the cloud.

The team

NinaData’s technology team combines extensive industry experience with technical know-how and experience in building a scalable high-value and high-performance architecture for AI analytics of media content. Our data scientists and management combine methodological skills with practical application experience. They have worked in the past as AI engineers, Machine Learning researchers, AI business managers and co-founders of successful startups.