Keyword-based Recommendation Service using Implicit Feedback on Hadoop

• Main Authors: TSAI, JIA-HAN, 蔡佳翰
• Other Authors: HSIEH, MENG-YEN
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/ngn679

碩士 === 靜宜大學 === 資訊工程學系 === 105 === The recommendation system provides great convenience for the use of information. Whether you are shopping, watching movies or reading articles, we can use the recommended system. However, when a new user(item) joins the system, it may cause recommendation difficulties due to the lack of its relation information, in this situation, it called Cold-Start problem. The general way to solve the Cold-Start problem is to carry out the information acquisition stage before the recommendation process. The method of collecting information is divided into explicit feedback and implicit feedback. Past research focused on explicit feedback. That is, it requires the user to provide information files in order to analyze their preferences, but this will reduce the system's friendliness. On the other hand, the implicit feedback that the system actively collects, is the user's behavior records to be used as recommended information. The implicit feedback may skip the step of providing the initial information by the user, hence the complexity of system usage can be reduced. In this paper, mobile APP's records will be used as implicit feedback. In our recommendation system, movies and mobile APP's information are collected by our crawlers, then it calculates the similarity between the movies and APP's information. The similarity is then used as a reference to analyze the user's preferences in a Matrix Factorization, and to further solve the cold start problem. Consequently, we implement a movie recommendation system utilizing a collaborative filter to offer keyword-base recommended services by user implicit feedback in a Hadoop cluster