| Summary: | 碩士 === 國立虎尾科技大學 === 資訊工程系碩士班 === 107 === Food safety has always been an issue greatly concerned by people, and a lot of information about food and medicine can be found on the Internet. Nevertheless, the information on the Internet is quite complicated, some of which might even be incorrect. Therefore, the topics to be explored in this paper are how to find out the information of food safety from this massive information and avoid being misadvised by the incorrect information of food safety on the Internet. In this study, the Integrating Machine Learning on Social Platform and Chatbot based on Cloud Computing Framework (IMSCCCF) was proposed to solve the aforesaid problems. The Food and Medicine Safety Information Platform (FMSIP) was further created to verify the feasibility of the IMSCCCF. In this study, an attempt was made to find out the incorrect information of food safety by the following measures. Firstly, the government's Open Data and the Internet news related to food safety as sources of training data were used to train a binary classification model. Secondly, the posts from the recent Facebook users of the fans page were also input to the binary classification model to identify the potential food safety information. Lastly, the said potential food safety information was compare against the Open Data stored in the database through the Latent Semantic Indexing (LSI). In this study create a Chatbot which users can get the correct knowledge of food safety with simple operations. This paper preprocess the content entered by the users, and then compare against the back-end database through LSI. Finally, comparison of results were returned to the users. The computational time increases as the amount of data increases. In order to improve the computational speed of the system, a Spark cluster architecture and Tensorflow were adopted in this study. The performance and accuracy of different binary classification algorithms running in different operational environments were also compared. In order to test whether the binary classification model can be accurately classified under real cases, this paper makes a systematic evaluation. According to the systematic evaluation, the accuracy of machine learning and deep learning algorithms are 0.769 and 0.726, respectively, which indicates that the binary classification model can be accurately classified in this case, and the machine learning method is better than deep learning.
|
|---|
