Online News Media Bias Analysis using an LDA-NLP Approach

It is widely recognized that every media outlet has its own ”spin” on news, and this bias has been described in many ways and at many levels. In political news for example, the bias can be liberal, conservative, moderate, corporate, etc. In addition, recent research has focused on the ’sentiment dimension’ to further identify and categorize news bias. This is achieved through analysis of the adjective and adverb terms found in the news texts. The accuracy and generality of these models depend on the evaluation methods used to appraise the intensity and emotional weights of the adjectives and adverbs, thus rendering the results open to controversy. In this paper we propose a unifying system to extract information from political news texts and analyze it within a cognitive network. We view the different news media sources as agents with unique personalities, which we assume are latent within their texts. We use a combination Latent Dirichlet Allocation (LDA) and natural language processing (NLP) methods to identify the different agents’ personality traits with respect to various topics or concepts. An agent’s personality traits affect its inclination to word a certain event in a specific way. Using the common concepts stored in the cognitive network, our system can compare the different agents on a unified and normalized platform.