以文字探勘為基礎之食品安全事件分類模型研究

近年來食安問題層出不窮，每年皆有許多食安事件搶占新聞版面，網路新聞媒體也是民眾獲取食安資訊的一個重要管道，如果能有效地將網路新聞媒體中的食安事件進行分類，將有助民眾分辨食安事件的嚴重性，減少不必要的恐慌，因此本研究將透過文件探勘技術結合機器學習方法，針對網路新聞媒體上非結構化文字資料進行分析，使用類神經網路、支援向量機以及簡單貝氏分類器等監督式機器學習方法建構食安事件分類模型，比較各種機器學習方法對於食安事件分類的有效性，並且驗證非監督式機器學習方法中的隱含狄氏分布應用於食安事件分類的可行性。經本研究實驗證實，類神經網路模型、支援向量機模型、簡單貝氏分類器模型等三種監督式分類模型以及非監督式機器學習方法的隱含狄氏分布模型對於網路食安新聞分類都具有極佳的效能，F1-Measure皆達0.8以上，其中以支援向量機模型的分類效能最為優異達到0.980。

關鍵字

食品安全；機器學習；文字探勘

並列摘要

In recent years, food safety issues have emerged in an endless stream. There are lots of food safety issues have been reported in newspaper and internet every year. The online news media is an important channel for people to obtain food safety information. If people can effectively classify food safety incidents in online news media, it will help the public to distinguish the seriousness of the food safety incident and reduce unnecessary panic. Therefore, this research will use text mining and machine learning to analyze the unstructured text data of news on the internet for food safety classification. Using supervised machine learning technologies, such as neural network (NN), Support Vector Machine (SVM) and Naive Bayes (NB) classifiers, to construct the classification model about the event of food safety. And using unsupervised machine Learning of Latent Dirichlet Allocation (LDA) to verify the feasibility in the classification model about the event of food safety. Confirmed by this research experiment, three kind of supervised machine learning and one unsupervised machine learning are all perform excellent result in the classification model about the event of food safety. Their F-Measure are all higher than 0.8, and the Support Vector Machine model has the best result. Its F-Measure can reach 0.980.