Scholarlink

<p>Recommender systems underpin many online services, yet improving user acceptance and engagement remains a critical challenge beyond optimizing accuracy. This study introduces a novel method for generating personalized persuasive explanations in recommender systems by integrating users’ demographics, Twitter-derived Big Five personality traits, and social network data. Addressing the limitations of one-size-fits-all explanations and static personality mappings, the proposed approach leverages multi-label machine learning models to infer users’ dominant persuasive strategies based on the Cialdini principles of influence. A rich dataset was collected from Persian-speaking Twitter users, including a unique corpus of tweets annotated with personality traits and a second dataset incorporating persuasion effectiveness ratings. Personality traits were identified through multi-label classification, using text representations generated by ParsBERT and FastText. Based on inferred personality traits, demographic information, and social network data, multi-label classifiers were used to predict users’ preferred persuasive strategies, which were then used to generate personalized persuasive explanations. In experimental evaluations on datasets of 1936 participants for personality detection and 415 participants for persuasive strategy prediction, our method achieved up to 0.88 and 0.95 weighted F1 score for personality trait detection and persuasive strategy detection, respectively, and outperformed the examined rule-based baselines in this evaluation. In a controlled user study with 132 participants, participants assigned to our personalized persuasive explanation condition reported substantially higher willingness to follow the recommendations than those exposed to the two rule-based personalization approaches, with increases of 68.6% and 52.6%. The personalized explanations also received higher user evaluations of perceived usefulness compared to the rule-based baselines, producing improvements of 56.8% and 32.3%. Taken together, these findings suggest that personalized persuasive explanations can improve users’ reported willingness to follow recommendations and perceived usefulness in similar contexts.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Personalization of persuasive explanations in recommender systems: leveraging users’ demographics, social network data, and twitter-derived personality traits

  • 期刊论文

  • 发表日期：

    2026年6月23日

查看全文
  • Mahshid Keivandarian,
  • Fakhroddin Noorbehbahani,
  • Azadeh Mohammadi

摘要

Recommender systems underpin many online services, yet improving user acceptance and engagement remains a critical challenge beyond optimizing accuracy. This study introduces a novel method for generating personalized persuasive explanations in recommender systems by integrating users’ demographics, Twitter-derived Big Five personality traits, and social network data. Addressing the limitations of one-size-fits-all explanations and static personality mappings, the proposed approach leverages multi-label machine learning models to infer users’ dominant persuasive strategies based on the Cialdini principles of influence. A rich dataset was collected from Persian-speaking Twitter users, including a unique corpus of tweets annotated with personality traits and a second dataset incorporating persuasion effectiveness ratings. Personality traits were identified through multi-label classification, using text representations generated by ParsBERT and FastText. Based on inferred personality traits, demographic information, and social network data, multi-label classifiers were used to predict users’ preferred persuasive strategies, which were then used to generate personalized persuasive explanations. In experimental evaluations on datasets of 1936 participants for personality detection and 415 participants for persuasive strategy prediction, our method achieved up to 0.88 and 0.95 weighted F1 score for personality trait detection and persuasive strategy detection, respectively, and outperformed the examined rule-based baselines in this evaluation. In a controlled user study with 132 participants, participants assigned to our personalized persuasive explanation condition reported substantially higher willingness to follow the recommendations than those exposed to the two rule-based personalization approaches, with increases of 68.6% and 52.6%. The personalized explanations also received higher user evaluations of perceived usefulness compared to the rule-based baselines, producing improvements of 56.8% and 32.3%. Taken together, these findings suggest that personalized persuasive explanations can improve users’ reported willingness to follow recommendations and perceived usefulness in similar contexts.