Human-AI Interaction: Human Behavior Routineness Shapes AI Performance

These are codes for the routineness model and its corresponding data, plus a subset of the anonymized and processed data for the prediction models containing 10.000 users.

Data

dataset.tar.gz — Data for the routineness model, including 151 .parquet files, each of them containing 1,000 anonymized users' hourly behavioural observations.

subset_train.csv — A subset of the mobility data for training on prediction models, containing 10,000 anonymized users.

subset_test.csv — A subset of the mobility data for testing, containing the same 10,000 anonymized users.

Code

routineness.stan — A Stan program, that allowed us to measure the weight of the routine/random behaviours of each individual and to calculate their routineness.

run.py — A Python file to run the Stan program. Noted that a package, CmdStanPy, must be installed. Under the hood, CmdStanPy uses the CmdStan command line interface to compile and run a Stan program.

本数据集包含常规性模型（routineness model）的配套代码与原始数据，以及面向预测模型的、包含10000名匿名用户的匿名化处理后数据子集。

数据
dataset.tar.gz — 常规性模型所需数据，包含151个.parquet文件，每个文件均包含1000名匿名用户的每小时行为观测记录。
subset_train.csv — 用于预测模型训练的移动性数据子集，包含10000名匿名用户。
subset_test.csv — 用于预测模型测试的移动性数据子集，与上述10000名匿名用户完全一致。

代码
routineness.stan — 一款Stan程序，可用于测算每位个体的常规行为与随机行为权重，并计算其常规性得分。
run.py — 用于运行上述Stan程序的Python脚本。需注意，需提前安装CmdStanPy依赖包；该工具底层通过CmdStan命令行界面编译并运行Stan程序。