Abstract: 【Purpose/significance】The data scale of government agencies has been growing unprecedentedly in the digital era, which al?
so brings challenges to the automatic processing of multi category government data. In this context, this paper introduces a multi-level text classification method to solve the problem of automatic classification of import and export commodities.【Method/process】Based on the hierarchical structure of HS coding, a three-level classification model is constructed, which classifies text by layer-by-layer dis? crimination and accumulation. At the same time, the classification effect of SVM and TextRNN algorithm is compared.【Result/conclu? sion】The overall effect of multi-level classification models is fine for solving the problem of commodity classification. In the case of sufficient data, the effect of TextRNN is better than SVM (93.00% > 92.90% in the first layer and 96.46% > 96.38% in the second lay? er). However, in the case of insufficient learning, SVM has greater advantages (92.49% < 95.92% in the third layer). SVM achieves the best superposition accuracy of 85.88%.【Innovation/limitation】This paper attempts to solve the problem of automatic commodity classi? fication based on multi-level classification, but the data scale and application scenario still need to be expanded.