Research Of A Multi-Label Classification Model For Industries

Nowadays the economy is developing rapidly in a mixed manner.Besides,the scope of business is also seeing multi-sectoral mixed trends.In the field of information management,it is of great importance to automate the enterprise classification problem.A reference solution has been proposed for the comprehensive data characteristics and model error examples.The main research in this thesis is as follows.1.A deep learning model for multi-label classification of nearly 1,000 types of enterprise industries is proposed.In this thesis,an Albert-BiGRU-Attention model is constructed based on deep learning to solve the multi-label classification problem of nearly 1,000 categories of enterprise industries.Comparative experiments are conducted to verify the effectiveness of this deep learning model and the results of the test set are analyzed for error samples.Seven types of error samples are analyzed,and two improvement solutions are proposed to address the problem.2.An improved solution for the industry and industry chain multi-task learning approach is proposed.The industry multi-label classification task is complemented by the effective use of additional industry chain information for learning.Due to the strong correlation between industry classification and industry chain classification and the use of the same features for decision-making,the multi-task learning method is improved and optimized using the multi-task learning method.The industry and industry chain multi-task learning approach shares the underlying structure,makes top-level tasks independent,and outputs industry category labels.This is experimented with and validated for effectiveness,significantly improving over singletask learning.3.A multi-task learning method that includes hierarchical information is proposed.A multi-task learning method incorporating hierarchical information is designed,taking into account the relevance of the labels to each other,and the effect of complementing the model with knowledge and modeling.A soft sharing mechanism of multi-task learning is used for learning,and the hierarchical classification task is multi-tasked with the industry and industry chain task,keeping independent models and sharing some parameters.Experiments prove that this improvement of the method is effective.The experiments on the above model methods are performed based on the evaluation of accuracy,precision,recall and other indicators.The outcome of the experiments shows that all the above three models are effectively solving the industry multi-label classification problem.It can predict the correct and reasonable industry categories,and each improvement of the model method has a significant improvement,which proves the effectiveness of the scheme.