Hierarchical Classification Algorithm Based on FastText

At present, methods for automatic assigning labels for literature using the Chinese Library Classification system are mostly based on machine learning methods, and many use the build knowledge base to improve classification effect. These methods are applicable to small-scale dataset, and as the categorical numbers increase, the classification effect will decrease and the time cost will increase dramatically. The paper proposes a hierarchical classification algorithm based on fastText text classification tool to solve classification indexing problem on large-scale literature dataset. The algorithm trains the two-layer classifier from the top-down method and designs an optimization algorithm to reduce the error transmission between the two layers. Experimental results showed that the optimized hierarchical classification technique improved the classification effect compared to the traditional classification technique.