From hardware to software: in-vehicle interface complaints in China, 2019-2026 (dataset and code)

This repository contains the aggregated data, validation labels, analysis code, and figure code for the study "From hardware to software: a seven-year computational analysis of in-vehicle interface complaints in China." The study examines how owner complaints about in-vehicle interfaces changed between 2019 and 2026. We assembled 23,699 owner reviews of 120 vehicle models from 41 brands in the Chinese market and coded each review for the presence and category of interface complaint using a large language model, validated against 150 independently human-coded reviews (Cohen's kappa = 0.757). Contents:- data/model_level_aggregates.csv: model-by-period aggregates (review counts, interface complaint counts by category, median screen size, mean purchase price)- data/validation_150.csv: human and model labels for the validation sample (no review text)- data/sample_120_models.csv: the sampled models and their metadata- code/: retrieval, coding, and analysis scripts, including the full coding prompt- figures/: source data, plotting code, and 300-dpi figures All files are deposited at the top level of this record: the paths data/, code/, and figures/ in README.md correspond to the file names listed above. Raw review text is not redistributed because it is user-generated content owned by its authors. The retrieval scripts allow the corpus to be reconstructed from the public source, subject to the platform's terms of service. Code is released under the MIT License; data and figures under CC BY 4.0.