A Study of Sentiment Analysis Based Text Mining Using Cosmetic Recommendation System for Example

• Main Authors: HSIEH,CHENG-LIN, 謝承霖
• Other Authors: LIN,CHIA-CHEN
• Format: Others
• Language: zh-TW
• Published: 2017
• Online Access: http://ndltd.ncl.edu.tw/handle/85ze7m

碩士 === 靜宜大學 === 資訊管理學系 === 105 === Nowadays, the development of the Internet is rapid, and the Internet contributes to the rapid development of E-commerce. At the same time, the Internet commentary becomes more and more important along with the rapid development of the Internet and the freedom to speech. Due to the development of Internet commentary, lots of consumers are used to collect commentary before they purchase products either from the On-line shops or the physical shops. As the result, the consumers’ commentary plays a crucial role to the enterprises. This research combines CKIP developed by Academia Sinica and techniques of Opinion Analysis, and then analyzes consumers’ commentary by using data mining technique. To improve the precision of our prototype, a cosmetic-based word library was generated in advance with the databases derived from HowNet, NTUSD and on-line cosmetic commentary collected from on-line cosmetic forms. Then, scores also called opinion scores generated after comparing the generated Chinese word segmentation result and cosmetic-based word library. Doctor Pei-Hsuan Wu is invited to serve as our cosmetic consultant and offer us professional cosmetic domain knowledge. Finally, a prototype system called CosDunk is developed and recommends proper cosmetic product(s) to user through a friendly user interface. The user can describe his/her symptoms by selecting the related text description(s) and point out the corresponding position of virtual face offered by the system to interact with our prototype system CosDunk. Once the prototype system CosDunk collects data from user(s), it will list the proper cosmetic products according to the computed opinion scores from the highest one to the lowest one. According to our best knowledge, the proposed system CosDunk is the first system which combines features of products and opinion of on-line cosmetic users’ comments to generate the recommendations.