Recommender Systems Based on TensorFlow and Spark - A Case Study of the Net Buy and Sell Amounts of Foreign Investor

• Main Authors: KE, YUN-TING, 柯勻婷
• Other Authors: CHANG, YI-PING
• Format: Others
• Language: zh-TW
• Published: 2018
• Online Access: http://ndltd.ncl.edu.tw/handle/7d8224

碩士 === 東吳大學 === 財務工程與精算數學系 === 106 === Since the concept of Recommender System realeased, it was used widely in each area such as movies, restaurants, musics, games, et cetera. Finding out the users’ preferences by using the item scoring from the users, Recommender System can recommend the items which the users may be fond of. As FinTech expands rapidly in recent years, many financial issues make use of the method of Machine Learning analysis frequently; moreover, Recommender System is the method often applied in Machine Learning as well. In this case study, the author use TensorFlow developed by Google and Spark based on in-memory computing, through the Latent Factor Models of Collaborative Filtering Recommendation of the Recommneder System to predicts the proportion of Net Buy and Sell in Institutional investors. This case study collects the information through the Daily Trading of Institutional investors and Daily Quotes on the website of Taiwan Stock Exchange Corporation. By using the datas from May 2, 2012 to December 31, 2016 as the training datas, the author predict the scale of Net Foreign Buy and Sell of trading day from January 1, 2017 to December 15, 2017. The results revealed that Spark possesses better predict scores; however, it is lack of the predictable abilities in the situation of extreme high or extreme low in Net Buy and Sell. On the contrary, the second model transfromed from TensorFlow become more complex with more features. In this way, it is more capable to predict than the basic one.