Research Articles
ZHANG Meng, WANG Jingkai, YANG Jiaming
Chinese Journal of Journalism & Communication.
2026, 48(2):
115-137.
Focusing on “algorithm aversion” within the context of news consumption, this study employs a controlled experiment to systematically examine the differences in news consumers’ subjective evaluations of algorithm-recommended versus human-recommended texts. It further explores the moderating effects of demographic characteristics and content preferences on recommendation preferences. A total of 800 participants were involved in the experiment, with three dimensions—willingness to accept recommendations, perceived personalization, and overall evaluation—measured using Likert scales. Data were analyzed using independent samples t-tests, multiple linear regressions, and path analyses. The statistical results reveal that human recommendations significantly outperform algorithmic recommendations overall, eliciting higher subjective satisfaction in the dimension of users’ overall evaluation, thereby verifying the existence of “algorithm aversion” in the Chinese news consumption sector. Although the algorithm group generally scored lower in the dimensions of recommendation willingness and perceived personalization, these differences did not reach statistical significance. Multiple linear regressions and interaction effect analyses indicate that variables such as occupation, gender, daily news reading time, and active news-seeking behavior significantly impact subjective evaluations across different recommendation modes. Notably, significant or marginally significant interaction effects were observed in certain dimensions for “occupation × recommendation mode” and “content preference × recommendation mode”, indicating that occupation and news content preferences moderate users' subjective evaluations of recommendation approaches. For instance, among users with a high preference for international news, the satisfaction gap between the algorithmic and human recommendation groups is narrowed. Furthermore, the three-dimensional path analysis demonstrates that overall evaluation partially mediates the effect of perceived personalization on the willingness to accept recommendations. The findings suggest that when algorithmic recommendations demonstrate a precise understanding of user interests, users are more likely to tolerate algorithmic opacity, which enhances their overall evaluation, whereas relying solely on technical optimization yields limited outcomes.