以Instagram資料實做一套餐飲評分機制

Abstract

隨著網際網路的發達，各種社群平台如雨後春筍般的興起，如：facebook、instagram、ppt、twitter、dcard...等，大眾僅需透過手機連上網即可隨時隨地發表自己的評論意見，討論著每一件事、每一個事物，因此對於美食愛好者而言，能夠透過大眾的意見找到真正的美食更是省下了自行搜尋的麻煩，也因此關於美食的評分與評論機制變的更加普及，如：google評論、愛食記、大重點評網、yelp、menu美食誌...等，然而，在眾多被推薦的店家中，該如何快速選擇最想造訪的店家儼然變成了一道難題，此時綜合大眾評價的店家排名即顯得相當重要。 因此本篇論文欲以公認排名推薦較準的社群平台為標準，設計餐飲評分機制學習此社群平台排名的計算方式，日後不再需要透過大眾投票的問卷調查或是美食社群的調查便能找出符合大眾心中的最佳餐廳排行。 本研究利用社群聆聽(social listening)的概念結合情緒分析技術，分析Instagram社群上用戶對於100間店家的評論資料，以公認推薦排名較準的menu美食誌top10餐廳排行榜，作為本研究餐飲評分機制的模型訓練資料，嘗試各種方式預測出趨近於menu美食誌top10的餐廳排名，並進一步探究預測排名準確或不準確之原因，最後驗證排名預測的準確度。

As the continuous development of the Internet, a variety of social media have been generated, such as Facebook, Instagram, PTT, twitter, Dcard and so on. People can give any opinion and discuss everything anywhere and anytime via their cellphones. Therefore, with regard to gastronomes, being able to utilize the public’s opinions to discover a real great delicacy will save their time to look for by themselves and that is the reason why rating systems concerning delicacies, such as reviews of Google Maps, Ifoodie app, Dianping, Yelp, MENU app, have become more and more popular. However, among large amounts of recommended stores, how to rapidly select one store people most want to visit has become a problem. Thus, ranks of stores considering the public’s reviews appear to be very important. Consequently, this thesis will regard ranks of stores people trust as standard ranks to design a rating system in catering industry. In this way, people can easily find out a good rank of stores recommended by the public and they do not have to conduct a survey of the public’s preferences about restaurants or wait for publications of any rank of stores. This thesis will take advantage of the concept of social listening and sentiment analysis to analyze reviews about 100 stores in Instagram. Additionally, this thesis will use ranks of stores of MENU app people highly regard as the training data of the rating system designed by this thesis and try different methods to predict the ranks of stores of MENU app. Furthermore, this thesis will also study why the prediction is accurate or inaccurate and verify the accuracy of the prediction.