This article takes French news from 2023 to 2025 as the object, constructs a cross media news corpus and audience interaction database, uses French Transformer representation, topic clustering, supervised classification, and econometric modeling methods to identify the main discourse patterns in news texts, and examines their differential impact on audience participation at different levels. Research has found that French news mainly forms five stable modes: institutional/governance type, conflict/opposition type, testimony/character story type, data/expert interpretation type, and crisis/risk warning type. Among them, institutional/governance type accounts for the highest proportion, conflict/opposition type is more likely to bring clicks, browsing, and sharing, testimony/character story type has the strongest promoting effect on comments and long discussion chains, and data/expert explanation type, although not achieving the highest superficial popularity, can steadily enhance deep participation. Further analysis shows that the performance of different modes is not consistent between mainstream media and local media, serious news and soft news, official websites and social platforms, and crisis and normal periods. Starting from the context of French news, this article combines discourse analysis, computational text research, and audience quantification evaluation to provide empirical evidence for understanding news expression adjustment and audience relationship reconstruction in a platform based communication environment.
