Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Search
A novel stock recommendation system using Guba sentiment analysis
Sun Y., Fang M., Wang X. Personal and Ubiquitous Computing22 (3):575-587,2018.Type:Article
Date Reviewed: Sep 5 2018

It’s pretty obvious that if you know what investors are thinking, you can predict what stocks they’ll buy and sell. Stock scams have operated for years on the complement of this thought, that is, attempt to change investors’ opinions to match the stocks that the influencers are peddling. So it’s not the idea that’s novel, it’s the implementation. And there have always been multiple problems in implementing processes to learn what investors are thinking.

First, who exactly are the investors, the voices that matter, and what are the right opinions? Second, how do you find out what they’re thinking without influencing or even coercing their expressed opinion to deviate from their true opinion (the Heisenberg uncertainty principle)? And, in China, how do you determine stock market opinions using an ancient language that has no idioms for stock entities and transactions?

Many threads have come together to make this research possible, and especially its positive outcome. Guba, a social forum specifically serving professional Chinese stock traders, has reached a significant level of posting and a critical mass of influence. Guba members are inventing and evolving the Chinese vocabulary for stock trading. Guba posts are also characterized by: (1) being brief and (2) including emotional content (“It’s over!” and “Keep going!”). These concise expressions of sentiment are perfectly aligned with the technology of sentiment analysis, which has only recently coalesced into stable and reliable toolkits. It’s a perfect storm of Chinese stock sentiment meeting newly minted analysis tools.

The researchers ran seven years of stored Guba dialog through their tools and compared it to the actual stock market history for the same period. They found strong correlation. They then created a stock recommendation system based on their sentiment analysis and ran it for the same seven-year period. Their recommendations outperformed the benchmark CSI300 by a multiple of eight. You could tweak the finer points of their recommendation model to make the outcome go higher or lower, but the main point is that automated sentiment analysis provided a real advantage in predicting a stock market.

The authors reveal their hypotheses and validation techniques, as well as finer points such as parts of their lexicon, GubaLex (which replaced historically significant Chinese lexicons due to their lack of stock market vocabulary), which they plan on maintaining for the foreseeable future. The paper is an effective blueprint for duplicating their work.

I’m sure professional trading companies around the world are already employing similar technology in-house, but I’m not aware of any such service for independent investors. I believe this paper is a harbinger of such services.

Reviewer:  Bayard Kohlhepp Review #: CR146233 (1811-0588)
Bookmark and Share
  Reviewer Selected
Featured Reviewer
 
 
Information Filtering (H.3.3 ... )
 
 
Financial (J.1 ... )
 
Would you recommend this review?
yes
no
Other reviews under "Information Filtering": Date
An information push-delivery system design for personal information service on the Internet
Chen C., Tai W. Information Processing and Management: an International Journal 39(6): 873-888, 2003. Type: Article
Mar 3 2004
Visualization schemes for domain novices exploring a topic space: the navigation classification scheme
Leide J., Large A., Beheshti J., Brooks M., Cole C. Information Processing and Management: an International Journal 39(6): 923-940, 2003. Type: Article
Feb 2 2004
Relevance as resonance: a new theoretical perspective and a practical utilization in information filtering
Brouard C., Nie J. Information Processing and Management: an International Journal 40(1): 1-19, 2004. Type: Article
Jun 29 2004
more...

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 1999-2024 ThinkLoud®
Terms of Use
| Privacy Policy