大宗商品供应链风险识别

赵昕，戚丹阳，刘华琼^*

山东交通学院交通与物流工程学院，山东 济南  250357

摘要：针对传统大宗商品供应链风险识别方法的识别角度不全面，识别结果准确度较低等问题，采用文本挖掘方法，建立包括数据采集及语料库构建、数据预处理、风险识别的大宗商品供应链风险识别模型框架，从中国知网、万方数据知识服务平台中获取大宗商品供应链管理相关研究文献，构建包含不同文献数的3个语料库，对语料库数据进行词频分析、N-gram分析、相关性分析、累计词频-信息熵（term frequency-information entropy，TF-H）降维及潜在狄利克雷分配（latent Dirichlet allocation, LDA）主题建模，并将风险识别结果与传统供应链风险识别方法的识别结果进行对比，验证方法的有效性。结果表明：通过LDA主题模型生成20个大宗商品供应链风险主题，每个主题从不同角度展示当前大宗商品供应链面临的风险，将识别出的大宗供应链风险分为市场风险、物流风险、金融风险、环境风险、管理风险、合作风险6种类型；文本挖掘方法与其他传统风险识别方法的风险识别结果具有较强的耦合度，且识别维度更全面、识别结果更准确。文本挖掘技术可全面、准确地识别供应链风险因素，可为大宗商品供应链风险识别提供理论支撑。

关键词：大宗商品；供应链管理；风险识别；文本挖掘；LDA

Risk identification of bulk commodity supply chains

ZHAO Xin, QI Danyang, LIU Huaqiong^*

School of Transportation and Logistics Engineering, Shandong Jiaotong University, Jinan 250357, China

Abstract: To address issues in traditional risk identification methods for commodity supply chains, such as incomplete perspectives and low accuracy of identification results, a text mining approach is employed to establish a bulk commodity supply chain risk identification model framework, which includes data collection, corpus construction, data preprocessing, and risk identification. Research papers related to bulk commodity supply chain management are collected from China National Knowledge Infrastructure(CNKI) and Wanfang Data Knowledge Service Platform. Three corpora with different numbers of texts are constructed. These corpora undergo word frequency analysis, N-gram analysis, correlation analysis, term frequency-information entropy(TF-H) dimensionality reduction, and latent Dirichlet allocation (LDA) topic modeling. The results of the risk identification are compared with those from traditional supply chain risk identification methods to validate the effectiveness of the proposed approach. The results show that the LDA topic model generates 20 bulk commodity supply chain risk topics, each reflecting the risks faced by the current bulk commodity supply chain from different perspectives. The identified risks are categorized into six types: market risk, logistics risk, financial risk, environmental risk, management risk, and cooperation risk. The text mining approach demonstrates a strong correlation with traditional risk identification methods, while offering a more comprehensive identification dimension and more accurate results. Text mining technology can comprehensively and accurately identify supply chain risk factors and provide theoretical support for bulk commodity supply chain risk identification.

Keywords: bulk commodity; supply chain management; risk identification; text mining; LDA