Bias Detection and Classification

A custom artificial intelligence solution is used to classify the bias of our articles in order to provide more neutral and impartial with results by eliminating the human bias factor of these ratings.

This classification is based solely based on the content itself and no other factors, such as author viewpoints.

The calculation of bias scores is based on a wide variety of factors. We utilize the paper “Computationally Detecting and Quantifying the Degree of Bias in Sentence-Level Text of News Stories by C.J. Hutto” ‘s statement based model, as well as a custom-built recurrent neural network to analyze overall article bias.

Our model runs on python and utilizes TensorFlow and Keras, as well as various text sentiment and natural language processing libraries. It works by utilizing sentence-based classification and combining the results to give an overall bias, as well as an LSTM TensorFlow neural network which tokenizes and breaks down the article in order to analyze its bias.

The model is trained on data from AdFontMedia’s dataset.

We are constantly working on training and improving the accuracy of our model. However, due imperfect nature of AI, we do intervene if we feel the bias scores aren’t representative of the bias in the article.