Automatic Verification of Stock Recommendations in Investment Social Webs

• Main Authors: Yu-Jhan Dai, 戴鈺展
• Other Authors: Sheng-Luen Chung
• Format: Others
• Language: zh-TW
• Published: 2016
• Online Access: http://ndltd.ncl.edu.tw/handle/3g7y8u

碩士 === 國立臺灣科技大學 === 電機工程系 === 104 === Veracity is the last V words attributed to the nature of big data in addition to volume, velocity and variety; veracity may be the most critical factor weighing the value of data at hand. For example, there are numerous stock pick recommendation postings commonly seen in investment-oriented social webs. Bloggers or professionals alike make sell or short recommendations, most time even mingled with view-conflicting postings. However, there are few, if any, follow-ups to check their validity not only in practice, but also in research literature.This study aims to track and validate Taiwan stock recommendations from two major investment social webs of PTT-stock and cnYES-Foreign investment. In particular, we are to track short and long term yields performance of each recommendation. With reference to the Oraclet framework in verification of prediction document, this study is to extract from each recommendation three components of: preface, charge, and prognostication, which is called registration process,followed by a verification process to validate the associated performance. In doing so, ER-diagram analysis is performed and a relational database implemented accordingly to accommodate the process involved. Crawlers are designed to pull stock recommendations from the two social webs, followed by the registration and verification processes on each recommendation. Meanwhile, to speed up, the system is constructed in the framework of cloud computing, the two processes of registration and verification are run on Spark.For demonstration, the implemented Automatic Verification of Stock Recommendation (AVSR) systemkeeps registering and verifying all recommendation postings from the two investment social webs since the year of 2011; each and every of more than ten thousand recommendations have been verified.The system is almost real-time, recommendation postings are processed at every hour. Verification results and pending oraclets are aligned in time axis for easy reviews. Recommendations of similar or conflicting views are compiled in batch for comparison. Performance histories of all bloggers or professionals appearing in these webs are also tracked.