Learning of the user behavior structure based on the time granularity analysis model

The construction of a consumption pattern can realize the analysis of consumer characteristics and behaviors, identify the relationship between commodities, and provide technical support for commodity recommendation and market analysis. However the current studies on consumer behavior and consumption patterns are very limited, and most of them are based on market research data. This method of data collection has high cost, low data coverage, and lagging survey results. The algorithm proposed in this article analyzes purchasing data from e-commerce platforms and extracts short- and long-term consumption matrices of consumers. By further processing these two matrices and removing the difference in granularity in time and marginal substitution rate, these matrices are finally integrated to form one consumption pattern matrix that can describe the characteristics of consumer consumption behavior in a period of time. Extensive experiments on various domains demonstrate that our proposed method outperforms state-of-the-art baselines on synthetic and real-world datasets.