学术文献库

Traditional approaches to classify the political bias of news articles have failed to generate accurate, generalizable results. Existing networks premised on CNNs and DNNs lack a model to identify and extrapolate subtle indicators of bias like word choice, context, and presentation. In this paper, we propose a network architecture that achieves human-level accuracy in assigning bias classifications to articles. The underlying model is based on a novel Mesh Neural Network (MNN),this structure enables feedback and feedforward synaptic connections between any two neurons in the mesh. The MNN ontains six network configurations that utilize Bernoulli based random sampling, pre-trained DNNs, and a network modelled after the C. Elegans nematode. The model is trained on over ten-thousand articles scraped from AllSides.com which are labelled to indicate political bias. The parameters of the network are then evolved using a genetic algorithm suited to the feedback neural structure. Finally, the best performing model is applied to five popular news sources in the United States over a fifty-day trial to quantify political biases in the articles they display. We hope our project can spur research into biological solutions for NLP tasks and provide accurate tools for citizens to understand subtle biases in the articles they consume.