Researchers propose a 'Densing Law' to describe the relationship between data scale and tokenization capacity in user representation learning. They analyzed the Alipay dataset and found that tokenization can mitigate performance degradation at large scales. The law suggests an approximately linear relationship between tokenization capacity and input data size, with varying scaling slopes depending on tokenization method and data source. An adaptive tokenization method called ALGN was developed based on this law, outperforming existing baselines in experiments. --- Why it matters: This matters to AI researchers because it provides a quantitative framework for understanding the trade-offs between data scale and tokenization capacity, which is crucial for large-scale user representation learning applications. Source: https://arxiv.org/abs/2608.23392