Amazon-Book and MovieLens-1M

MovieLens-1M: This dataset contains user ratings for movies from the MovieLens website. It consists of 6,040 users and 3,952 movies. Users rate movies on a 5-star scale, and for our analysis, we convert the ratings into binary signals (positive and negative feedback) using a threshold of 3.5. The training set contains 456,138 positive feedback interactions and 337,990 negative feedback interactions, while the testing set consists of 111,412 positive feedback interactions.Amazon-Book: We select the Amazon-Book dataset from a large crawl of product reviews on Amazon. This dataset comprises 35,736 users, 38,121 items, and 1,960,674 5-star ratings. We used a threshold of 3.5 to convert ratings into binary signals. The training set contains 1,252,292 positive feedback interactions and 302,056 negative feedback interactions, while the testing set consists of 327,682 positive feedback interactions. The dataset reflects real user behavior in an online retail environment and poses challenges related to exposure bias.

MovieLens-1M：本数据集收录了MovieLens网站的用户影视评分数据，涵盖6040位用户与3952部影视作品。用户采用5星评分制进行评级，为开展本次分析，我们以3.5为阈值，将原始评分转换为二元反馈信号（正向反馈与负向反馈）。其中训练集包含456138条正向反馈交互样本与337990条负向反馈交互样本，测试集则包含111412条正向反馈交互样本。

Amazon-Book：我们从亚马逊（Amazon）大规模爬取的商品评论数据中筛选出Amazon-Book数据集。该数据集包含35736位用户、38121件商品及1960674条5星评分。我们同样以3.5为阈值，将评分转换为二元反馈信号。其中训练集包含1252292条正向反馈交互样本与302056条负向反馈交互样本，测试集则包含327682条正向反馈交互样本。该数据集反映了在线零售场景下的真实用户行为，同时存在曝光偏差（exposure bias）相关的建模挑战。