Blau Spaces

We leverage a distinctive machine learning pipeline known as socio-network analysis, wherein user characteristics and follower graphs are intricately woven into embeddings within the multi-dimensional Blau Space. Harnessing more than 50 socio-economic and behavioral factors, meticulously labeled through clustering or LLM analysis—spanning continuous variables like age and location, as well as discrete static properties such as gender and birthplace—we embark on a comprehensive exploration.

(The organizing force in Blau space is the homophily principle, which argues that the flow of information from person to person is a declining function of distance in Blau space which is really just a set of socio-economic characteristics)

Our approach entails modeling affiliations as latent variables, intricately gauging the similarity between alters based on shared profile information. Employing innovative unsupervised methods, we discern the dimensions of profile similarity that foster densely linked circles.

What sets our methodology apart is the simultaneous consideration of social network structure, coupled with insights derived from cutting-edge clustering and LLM analysis. This synergy culminates in a highly performing and distinctive technique, enabling us to pinpoint and define what we term "Socio-network circles."

Each "Socio-network grouping" is endowed with a unique definition of profile similarity. Whether formed around friends from the same school, location, or subtle concepts introduced to measure trust, these circles capture nuanced relationships. Our machine learning models are trained by concurrently optimizing node circle memberships and profile similarity functions, ensuring the most effective explanation of observed data. This fusion of innovation and precision propels our method to the forefront of identifying and understanding intricate social structures.