Hidden gender bias in big data as revealed by neural networks: Man is to woman as work is to mother?

Abstract

Social events become big data. The analysis of big data becomes knowledge about society. If the big data is biased, this bias is transmitted to the analysis and to our knowledge. We propose a tool to discover gender biases and, potentially eliminate them from big data prior to analysis. We use neural network analysis via word embedding. This is the first time that this technique has been tested on a Spanish data body. As proof of concept, the neural network was fed with half of the Wikipedia in Spanish. More than 28 million words. We describe the techniques and specialized knowledge necessary to discern gender bias and examined whether it is possible to divide the analysis work into externalizable microtasks.