Abstract: To make better use of the unstructured chart data in collaborative manufacturing enterprises,a chart analysis approach was proposed based on natural language generation.The Optical Character Recognition (OCR) and key point detection network were used to identify and extract the text and data information of the chart.Then,the corresponding text descriptions were generated by trained natural language generation model according to the different intentions of users for helping to analyze the charts more intelligently and accurately.The proposed method was applied to the interactive platform of the manufacturing enterprise,and the result showed an accuracy of 88.6% for chart extraction and an evaluation score of 86.4% for text descriptions.The application cases in enterprises and related research also showed that the proposed method could accurately analyze different types of charts.

Key words: data analysis, deep learning, chart data, natural language generation

摘要： 为充分利用协同制造企业在生产过程中的非结构化图表数据,提出一种基于自然语言生成的图表数据分析方法。首先使用光学字符识别技术和关键点检测网络对图表中的文本信息和数据信息进行识别和提取;随后将用户需求作为输入,通过自然语言生成模型输出相应的图表文本描述,使其可以根据用户不同的意图,生成智能和准确的图表分析结果。该方法图表提取的精度为88.6%,文本描述的评估得分为86.4%。通过在企业的应用案例和相关调研也表明该方法能够根据用户需求对不同类型的图表进行准确的分析。

关键词: 数据分析, 深度学习, 图表数据, 自然语言生成