在網路資訊量如此豐沛的時代，越來越多店家除了實體店面的經營，更努力在跟上開創網路商店，除了消費者在近年來的習慣正在轉型外，也因為目前疾病肆虐著世界各地，消費者和店家不得不加入網路購物的行列，而外送餐飲平台也在此時此刻成為了人們最常使用的網路商城。比照過去人們最常使用的幾個網路平台，像是拍賣平台、社群平台、影音平台和新聞平台，為了使平台店家端與用戶端都能順利進行，皆會使用推薦系統(Recommendation system)，近年開始，餐飲平台因使用人數上升，也跟上了使用推薦系統的腳步，對於大量的餐廳、餐點、用戶資訊，本論文將會基於人工智慧作法，設計與實作一套適用在餐飲平台的推薦系統。
    在過去經典的推薦系統通常分為三種，基於內容(Content-based)推薦、基於使用者的協同過濾(User-based Collaborative Filtering)推薦、基於項目的協同過濾(Item-based Collaborative Filtering)推薦，這三種推薦方法都是使用者必須已在餐飲平台上有點餐紀錄，才能啟動的，而無使用行為的新使用者則需再另外進行推薦。
考慮現代人的習慣，在點餐之前都會參考網路的推薦文，本論文提出了一種自定義餐廳特徵向量Store2Vec和一種自定義餐點特徵向量Food2Vec，利用每篇推薦文內的所有餐廳或餐點透過類似skip-gram的模型來訓練餐飲特徵向量，藉此找到人們認知中的餐廳與餐廳或餐點與餐點之間的關聯性，而在預測使用者會在未來點哪家餐廳或餐點之前，需要先從使用者的歷史點餐紀錄次數來判斷，此使用者是否為新使用者或舊使用者，若無任何點餐行為之新使用者，將以基於舊使用者的評分紀錄，給予推薦幾項獲得最高評分的餐廳與餐點;反之，將透過KNN(k-nearest neighbors)模型預測舊使用者可能會喜歡的15家餐廳及5樣餐點。最後，為了有效地推薦給使用者符合他們喜好的餐廳及餐點，將會依Top-K推薦排序並再依照每位使用者近期的點餐習慣調整，來提供更好的預測準確度，以上方法能有效地處理資料稀疏、個人化、冷啟動等的問題。

In an era when the amount of Internet information is so abundant, more and more stores are not only operating in physical stores, but also trying harder to keep up with the creation of online stores. In addition to the transformation of consumers’ habits in recent years, it is also because of the current diseases that ravage the world. Everywhere, consumers and stores have to join the ranks of online shopping, and delivery catering platforms have become the most commonly used online shopping malls at this moment. Comparing to several online platforms that people use most often in the past, such as auction platforms, social platforms, audio-visual platforms, and news platforms, in order to make the platform's store side and client-side work smoothly, a recommendation system (Recommendation system) will be used. In recent years, due to the increase in the number of users of catering platforms, they have also kept up with the pace of using recommendation systems. For a large number of restaurants, meals, and user information, this paper will be based on artificial intelligence to design and implement a set suitable for catering The recommendation system of the platform.
In the past, classic recommendation systems were usually divided into three types: Content-based recommendation, User-based Collaborative Filtering recommendation, and Item-based Collaborative Filtering recommendation. The three recommendation methods can be activated only after the user has ordered a meal record on the catering platform, and new users who have no usage behavior need to make additional recommendations.
Considering the habits of modern people, they will refer to the recommended articles on the Internet before ordering food. This paper proposes a custom restaurant feature vector Store2Vec and a custom meal feature vector Food2Vec, using all restaurants or meals in each recommended article. Point trains the restaurant feature vector through a skip-gram-like model to find the correlation between the restaurant and the restaurant or the meal and the meal in people’s perception and predicts which restaurant or meal the user will order in the future. Before ordering, you need to determine whether the user is a new user or an old user from the user’s historical order records. If there is no ordering behavior, the new user will be based on the old user’s score record, to recommend several restaurants and meals with the highest scores; on the contrary, the KNN (k-nearest neighbors) model will be used to predict 15 restaurants and 5 meals that old users may like. Finally, in order to effectively recommend to users, the restaurants and meals that meet their preferences, they will be ranked according to Top-K recommendations and adjusted according to each user’s recent ordering habits to provide better prediction accuracy. The method can effectively deal with the problems of data sparseness, personalization, cold start and so on.

目錄	VI
圖目錄	VIII
表目錄	X
第一章、簡介	1
第二章、相關研究	5
第三章、背景知識	7
3-1推薦系統(Recommendation System, RS)	7
3-2基於內容技術（Content-Based, CB）	7
3-3基於協同過濾技術(Collaborative Filtering, CF)	8
3-4混合式推薦技術(Hybrid Recommendation, HR)	12
3-5冷啟動(Cold start)	15
3-6稀疏性 (sparsity)	15
3-7餘弦相似度(Cosine Similarity)	16
3-8皮爾森相關係數(Pearson Correlation Coefficient)	16
3-9鄰近演算法(K-Nearest Neighbors)	16
第四章、系統架構	18
4-1環境與問題描述	18
4-1-1環境與問題描述	18
4-1-2目標	18
4-2系統架構	20
A.	預處理階段	21
B.	預處理資料	23
C.	預測模型	25
D.	推薦列表	28
第五章、實驗分析	29
5-1實驗環境	29
5-2數據集	29
5-3實驗結果	29
第六章、結論	31
參考文獻	32
附錄-英文論文	34
  
圖目錄
圖1、基於使用者的協同過濾(User-based)	10
圖2、基於項目的協同過濾(Item-based)	11
圖3、混合式推薦技術	13
圖4、設計之推薦系統	19
圖5、系統流程架構圖	20
圖6、收集餐廳資料	22
圖7、收集餐點資料	22
圖8、收集餐廳推薦文	23
圖9、收集餐點推薦文	23
圖10、自定義餐廳特徵向量Store2Vec之算法	24
圖11、自定義餐點特徵向量Food2Vec之算法	25
圖12、訓練KNN餐廳模型-計算餐廳相似度	27
圖13、訓練KNN餐點模型-計算餐點相似度	27
圖14、模型使用期及推薦結果	28
圖15、點餐平臺之推薦系統實作畫面	30
圖16、SVD、Random及本論文算法之RMSE比較圖	30
  
表目錄
表1、相關文獻比較表	6

[1] Shih-Hsiung Lee, Tzu-Yu Chen, Yu-Ting Hsien, Lin-Roung Cao, “A Music Recommendation System for Depression Therapy Based on EEG,” in 2020 IEEE International Conference on Consumer Electronics - Taiwan (ICCE-Taiwan), Taoyuan, Taiwan, Sept. 2020.
[2] Nimish Kapoor, Saurav Vishal, Krishnaveni K. S., “Movie Recommendation System Using NLP Tools,” in 2020 5th International Conference on Communication and Electronics Systems (ICCES),  Coimbatore, India, June 2020.
[3] Jeffrey Lund, Yiu-Kai Ng, “Movie recommendations using the deep learning approach,” in 2018 IEEE International Conference on Information Reuse and Integration (IRI), Salt Lake City, UT, USA, July 2018.
[4] G. Linden, B. Smith, J. York, “Amazon.com recommendations: item-to-item collaborative filtering,” in IEEE Internet Computing, vol. 7, no. 1, pp. 76-80, Jan. 2003.
[5] Raciel Yera Toledo, Ahmad A. Alzahrani, Luis Martínez, “A Food Recommender System Considering Nutritional Information and User Preferences,” in IEEE Access, Vol. 7, pp. 96695 – 96711, July, 2019.
[6] Abinash Pujahari, Vineet Padmanabhan, “An Approach to Content Based Recommender Systems Using Decision List Based Classification with k-DNF Rule Set ,” 2014 International Conference on Information Technology, Bhubaneswar, India, Dec. 2014.
[7] Tanya V. Yadalam, Vaishnavi M. Gowda, Vanditha Shiva Kumar, Disha Girish, Namratha M., “Career Recommendation Systems using Content based Filtering,” 2020 5th International Conference on Communication and Electronics Systems (ICCES), Coimbatore, India, June 2020.
[8] Kriti Bansal, Vinod Jain, “Performance Analysis of Collaborative Filtering based Recommendation System on Similarity Threshold,” 2020 Fourth International Conference on Computing Methodologies and Communication (ICCMC), Erode, India, March 2020.
[9] Badrul Sarwar, George Karypis, Joseph Konstan, and John Riedl, “Item-based Collaborative Filtering RecommendationAlgorithms,” WWW10, May 1-5, 2001, Hong Kong.
[10] Oren Barkan, Noam Koenigstein, “Item2Vec: Neural Item Embedding for Collaborative Filtering,” 2016 IEEE Machine Learning for Signal Processing (MLSP), 13 Sep. 2016.
[11] Hanbyul Bang, Jee-Hyong Lee, “Collective Matrix Factorization Using Tag Embedding for Effective Recommender System,” 2016 Joint 8th International Conference on Soft Computing and Intelligent Systems (SCIS) and 17th International Symposium on Advanced Intelligent Systems (ISIS), Sapporo, Japan, Aug. 2016.
[12] Anand Shanker Tewari, Abhay Kumar, Asim Gopal Barman, “Book recommendation system based on combine features of content based filtering, collaborative filtering and association rule mining,” 2014 IEEE International Advance Computing Conference (IACC), Gurgaon, India, Feb. 2014.