Abstract: With the rapid growth of Chin’s green consumer groups, green consumption has shown great environmental value while driving consumption upgrades. The impact of energy-saving information exposure on green consumption behavior has also attracted scholars’ attention. Data-driven and model-driven methods have been combined. Natural language processing technology has been adopted to construct energy-saving information exposure indicators, and propensity score matching and quantile regression havebeen used to conduct empirical research on 5,889 household appliances consumption and comment data on an e-commerce platform in China. It has been indicated that the overall energy-saving information exposure of e-commerce platforms in China is not high. The marginal effect of energy-saving information exposure on the sales of energy-saving home appliances shows a monotonous upward trend, and the inflection point of information redundancy has not yet appeared. At the same time, inverted U-shaped change in the impact of home appliances with different sales scalesto energy-saving information exposure has been found, which has a greater impact on home appliances with medium sales scale. Finally, the adjustment effect of the “platform effect” on energy-saving information exposure has been further verified, which provides a reference for the research on green consumption behavior based on large-scale multi-source heterogeneous data.

Key words: energy-saving information exposure;natural language processing; propensity score matching;green consumption behavior