A linguistic/game-theoretic approach to detection/explanation of propaganda
作者：
Highlights：
• Constructs and tests propaganda detection models on 205,000 news articles.
• LightGBM with C-index = 0.9 and F1-score = 0.84 outperforms alternative models.
• Computes Shapley values to explain every propaganda score to enhance user trust.
• Aggregated Shapley values suggest a linguistic profile for propaganda.
• Examples of linguistic markers include first-person plural, certainty, and religion.
摘要
•Constructs and tests propaganda detection models on 205,000 news articles.•LightGBM with C-index = 0.9 and F1-score = 0.84 outperforms alternative models.•Computes Shapley values to explain every propaganda score to enhance user trust.•Aggregated Shapley values suggest a linguistic profile for propaganda.•Examples of linguistic markers include first-person plural, certainty, and religion.
论文关键词：Propaganda,Linguistic analysis,Shapley values,Prediction explanation,Gradient boosting
论文评审过程：Received 31 March 2020, Revised 1 February 2021, Accepted 10 October 2021, Available online 15 October 2021, Version of Record 30 October 2021.
论文官网地址：https://doi.org/10.1016/j.eswa.2021.116069
