利用类别引导词的投诉文本分类

Complaint Text Classification Based on Guiding Words

【目的】针对投诉事件文本具有信息量大、非结构化、规律性不强等特点,当前城市投诉信息管理亟需寻找一种高效的分类方法,提高管理人员的工作效率。【方法】分析投诉事件特点进而对其进行文本预处理;借助句法分析器、同义词林,并通过文档贡献度过滤引导词;采用TF-IDF计算引导词权重系数,并以VSM表示,最后通过SVM对处理后的投诉事件文本进行分类。【结果】在多个类别投诉事件测试文本中,该方法查准率和查全率平均值达到82.1%和82.3%。【局限】投诉事件文本的稀疏性在一定程度上影响分类结果。【结论】实验证明该方法在投诉事件文本分类中是有效的、可行的,能够提高投诉文本分类效果。

[Objective] For complaint text has the characteristics of informative, unstructured, weak regularity etc., the current information management of city complaint needs an efficient classification method to improve the efficiency of the management staff. [Methods] Analyze the characteristics of complaints and go for text preprocessing; Then use the parser, synonyms forest, and through the contribution of the document to filter guide word; At last, calculate the guide word weighting coefficients with TF-IDF, use VSM model to represent guide words and use SVM model to classify the complaint text. [Results] In multiple categories of complaint text, the average precision of the method is up to 82.1% and the average recall is up to 82.3%. [Limitations] The sparsity of complaint text affects the classification results to a certain extent. [Conclusions] The experiment results show that the method is effective and feasible in the text classification of complaints, and it can improve categorization effect of the complaint text.