Assessing Together the Trends in Newspaper Topics and User Opinions: A Co-Evolutionary Approach

Background

Prior to technically construct the model that will be discussed in this chapter, some theoretical foundations have had to be laid. Assuming a large enough but fixed number of topics, for a newspaper like the New York Times, it can be hypothesized that all topics are discussed at all time in a minimal form. Hence, all regular subtopics aren’t changing the words distribution within the way every topic is covered (even more so if data are preprocessed to remove under- and over- occurring words). Henceforth no punctual trend can emerge without a variation in either the way the topic is addressed or the way it is perceived (although the latter would end up influencing the former). The model proposed here is able to capture these variations. Whilst they, on their own, can enhance the comprehension of all events by identifying the link between semantic and attitudes’ shifts, a whole new dimension is brought to the analysis by the introduction of the trend index. Being able to quantify both in duration and in valence one of those shifts adds to society’s understanding of the way discussions emerge. To examine trends as defined hereabove (that is to say peaks of interest in online discussions/publications), it is necessary to observe shifts in relatively short time slices. Scaling down in time, while it might soften punctual trends, will show general dynamics and lead to deduce relative importance of a topic over the others. Hence, the temporal component of such an analysis is both key to the understanding of evolving power balances and a tool to displace the examination’s perspective.