Beyond Relevance: Factor-level Causal Explanation for User Travel Decisions with Counterfactual Data Augmentation

• DOI: 10.1145/3653673
• Publication Date: 2024-03-22T12:01:27Z
• Authors: Hanzhe Li; Jingjing Gu; Xinjiang Lu; Dazhong Shen; Yuting Liu; YaNan Deng; Guoliang Shi; Hui Xiong
• Keywords: Relevance; Factor (programming language); Counterfactual conditional; Causal model

Point-of-Interest (POI) recommendation, an important research hotspot in the field of urban computing, plays a crucial role construction. While understanding process users’ travel decisions and exploring causality POI choosing is not easy due to complex diverse influencing factors scenarios. Moreover, spurious explanations caused by severe data sparsity, i.e., misrepresenting universal relevance as causality, may also hinder us from decisions. To this end, article, we propose factor-level causal explanation generation framework based on counterfactual augmentation for user decisions, named Factor-level Causal Explanation User Travel Decisions (FCE-UTD), which can distinguish between true false generate explanations. Specifically, first assume that decision composed set several different factors. Then, preserving structure with joint contrastive learning paradigm, learn representation detect relevant Next, further identify constructing generator, particular, it only augment dataset mitigate sparsity but contribute clarifying other cause Besides, dependency learner proposed each scores. Extensive experiments conducted three real-world datasets demonstrate superiority our approach terms check-in rate, fidelity, downstream tasks under behavior The extra case studies ability FCE-UTD choosing.