Combining Objective and Subjective Perspectives for Political News Understanding

TL;DR

This work addresses the need to jointly model objective and subjective content in political news. It proposes a multilingual News Processing Framework that combines NLP components, information retrieval, and knowledge bases, and uses target-dependent sentiment classification ($TSC$) to obtain fine-grained sentiments toward entities. The study instantiates the framework on a large French corpus to produce outlet- and topic-level signals, politician-centric analyses, and demographic representations (gender and age). Key findings include a generally negative overall sentiment with outlet- and topic-specific variation, underrepresentation of women despite favorable sentiment toward female politicians, and an age bias toward older figures, reflecting the French political system. The approach provides explainable, scalable insights for multiple stakeholders and is designed to generalize to other languages and countries.

Abstract

Researchers and practitioners interested in computational politics rely on automatic content analysis tools to make sense of the large amount of political texts available on the Web. Such tools should provide objective and subjective aspects at different granularity levels to make the analyses useful in practice. Existing methods produce interesting insights for objective aspects, but are limited for subjective ones, are often limited to national contexts, and have limited explainability. We introduce a text analysis framework which integrates both perspectives and provides a fine-grained processing of subjective aspects. Information retrieval techniques and knowledge bases complement powerful natural language processing components to allow a flexible aggregation of results at different granularity levels. Importantly, the proposed bottom-up approach facilitates the explainability of the obtained results. We illustrate its functioning with insights on news outlets, political orientations, topics, individual entities, and demographic segments. The approach is instantiated on a large corpus of French news, but is designed to work seamlessly for other languages and countries.

Figures (15)

• Figure 1: Average sentiment (blue dots) for the most frequent 40 news outlets in the corpus. The dashed line represents the average sentiment for the entire corpus. Outlet frequency decreases from left to right.
• Figure 2: Fig. A: Deviation of the sentiment associated with major political orientations from the average sentiment of each source. Average sentiment of linked politicians is indicated in parentheses for each source. Fig. B: Distribution of mentions of political orientations in news outlets.
• Figure 3: Distribution of mentions of political orientations for ten impactful political topics.
• Figure 4: Deviation of the sentiment for major political orientations from the average sentiment of each topic. Average topic sentiment is given in parentheses.
• Figure 5: Percentage of gender mentions in the top-10 news outlets and on average.