Inflating Topic Relevance with Ideology: A Case Study of Political Ideology Bias in Social Topic Detection Models
This work highlights a problem of biased topic relevance for users of social topic detection models, demonstrating how human-selected input biases can negatively impact model accuracy.
This paper investigates how political ideology biases in training data propagate through widely-used NLP models, leading to a deterioration of retrieval accuracy. The authors propose learning a text representation invariant to political ideology to mitigate this bias.
We investigate the impact of political ideology biases in training data. Through a set of comparison studies, we examine the propagation of biases in several widely-used NLP models and its effect on the overall retrieval accuracy. Our work highlights the susceptibility of large, complex models to propagating the biases from human-selected input, which may lead to a deterioration of retrieval accuracy, and the importance of controlling for these biases. Finally, as a way to mitigate the bias, we propose to learn a text representation that is invariant to political ideology while still judging topic relevance.