GraphBrain is a research toolbox created for the purpose of studying collective human knowledge. It consists of a number of tools for both social and semantic networks analysis, knowledge retrieval from varying sources, knowledge integration, natural language processing, exploration and visualization. All of these tools are meant to be building blocks, organized around a central idea of the recursive hypergraph as a knowledge base.
This project is a fundamentally interdisciplinary effort, drawing from fields such as the Social Sciences, Artificial Intelligence and Computer Science, Epistemology, Computational Linguistics and Cognitive Science.
In its current state, Graphbrain provides the following:
- An hypergraph knowledge base, capable of representing and efficiently querying complex relationships between entities;
- Knowledge extraction algorithms for popular data sets such as WordNet and DBPedia;
- Entity disambiguation algorithms;
- An entity extraction algorithm for free-form text, such as news items and blog posts;
- A domain specific language to define knowledge extraction rules for free-form text;
- A web-based interface to navigate the knowledge graph.
The main ambition of this software project is to provide a robust base upon which new methods of socio-semantic analysis of digital media can be easily defined. At its core, GraphBrain contains an expressive knowledge representation system that easily lends itself to the organization of entities in terms of relationships, sources and conflicting beliefs.
The GraphBrain knowledge graph is designed to avoid notions of ground truth, instead allowing for the representation of all beliefs as relative to a source. Then, it is possible to define algorithms to determine consensus according to certain criteria or methods. This matches the current situation of digital spaces, where a very large amount of information is available, but knowledge discovery by human actors is constrained by centralized algorithms with poorly understood dynamics (e.g. search engines and recommendation systems).
Beyond the scientific study of collective knowledge phenomena in digital media, GraphBrain could enable the creation of systems of knowledge exploration that are less centralized, less dependent of private actors and more transparent to the users.
Here’s a demo of the current interface: