Predicting changes in consumer attention for cultural products, such as books, movies, and songs, is notoriously difficult. Past research suggests intrinsic limits for predicting consumer attention towards individual products. However, little is known about the limits for predicting shifts in collective attention. Here, we analyze five years of nationwide library loan data for almost 3 million individuals, comprising over 136 million loans of more than 750,000 unique titles. We find that culture, as measured by popularity distributions of loaned books, drifts continually from month to month at a near-constant rate, leading to a growing divergence over time, and that drift varies between book genres. By linking book loans to registry data, we investigate the influence of age, sex, educational level, and residential area type on cultural drift, finding heterogeneous effects. Our findings have important implications for market forecasting and algorithmic recommender systems, highlighting the need to account for drift dynamics.

翻译：预测图书、电影、歌曲等文化产品消费者注意力的变化历来极具挑战性。既往研究表明，针对单个产品预测消费者注意力存在内在局限性。然而，对于预测集体注意力转移的边界条件，我们知之甚少。本研究分析了涵盖近300万个体、五年期的全国图书馆借阅数据，涉及超过1.36亿次借阅记录与75万余种独立书目。研究发现，以借阅图书流行度分布为表征的文化现象，每月以近乎恒定的速率持续漂移，导致时间维度上的分歧不断扩大，且漂移速度因图书类型而异。通过将图书借阅数据与登记数据关联，我们考察了年龄、性别、教育水平和居住区域类型对文化漂移的影响，发现异质性效应。本研究成果对市场预测与算法推荐系统具有重要启示，凸显了纳入漂移动态的必要性。