Abstract: In order to investigate the formation mechanism of odors and compounds in foods（such as wines and alcoholic beverages）, especially in foods containing complex aromas, a new method of applying LDA model to mine the relationship between the odor in red wine and chemical molecules is proposed. The method combines the aroma and molecules in the flavor data set, regards the red wine as the document, the odors and molecule as the word. The hidden wine features are excavated through the LDA theme model. According to the distribution of red wine and chemical molecules in wine, the method is combined with Apriori algorithm for correlation analysis, ultimately to find the relationship between odor and chemical molecules, which lays the foundation for designing an electronic nose that can recognize food odor by testing the chemical molecules. The experimental data are provided by the Nantes Oniris Laboratory. The experimental results partially confirm the feasibility of applying LDA model to mine the relationship between wine odors and chemical molecules.

Key words: Latent Dirichlet Allocation（LDA）, odor of red wine, molecule of red wine

摘要： 为了调查食品尤其是包含复合香气的食品（如葡萄酒和酒精饮料等）中的气味活性化合物的构成机理，提出了一种将LDA模型应用于红酒气味与化学分子关系挖掘的方法。该方法在红酒风味数据集上，将红酒看作文档，气味和化学分子看作词语，通过LDA主题模型挖掘隐含的红酒特征；根据红酒与化学分子在红酒中的分布进行聚类，并结合Apriori算法进行关联分析，最终找出气味与化学分子之间的关系，为设计一个能够通过测试化学分子识别食品气味的电子鼻打下基础。实验数据由法国南特大学Oniris气味实验室提供，实验结果部分地证实了将LDA模型应用于红酒气味与化学分子关系挖掘的可行性。

关键词: 潜在Dirichlet分配（LDA）, 红酒气味, 红酒分子式