stock price prediction based on social network a survey presented by: chen en
TRANSCRIPT
Stock Price Prediction Based on Social Network
— A survey
Presented by: CHEN En
Outline
IntroductionRelated workMethodologyConclusion
IntroductionRelated workMethodologyConclusion
Outline
Stock price prediction Act of trying to determine the future value of company
stock or other financial instrument trade on financial exchange
Successful prediction could yield significant profit!
Introduction
The efficient-market hypothesis Stock price movement are governed by the random walk
hypothesis Inherently unpredictable
However, the others disagree and possess myriad prediction methods to gain future price information Fundamental analysis - Performance ratio (i.e. P/E ratio) Technical analysis - Charting analysis (i.e. Head and shoulder) Alternative methods - Internet-based data source for prediction
Introduction
IntroductionRelated workMethodologyConclusion
Outline
Traditional investment decision approaches: Capital asset pricing model (CAMP) Arbitrage pricing theory (APT)
Unrealistic and time complexity of the required calculation make them not applicable in real world problem
Current soft computing techniques: Neural network (NN) (A. N. Refenes, M. Azema-Barac, and A. D. Zapranis1993)
Genetic algorithm (GA) (R. Riolo, T. Soule, B. Eorzel2008)
Support Vector Machines (SVM) (G. H. John, P. Miller, and R. Kerber1996)
Because of widely use of the social network, major prediction are based on these public information.
Related work
Why social network? Ubiquitous and important for content sharing
Facebook, Blog, Twitter feeds, etc.
Public information—easily obtained
Behavioral economics demonstrate that emotions can profoundly affect individual behavior and decision-making
Recent research suggests very early acting prediction indicators can be extracted from online social media Online chat activity predicts book sales (Gruhl, D, Guha, R, Kumar, R, Novak,
J2005)
Blog sentiment predicts movie sales (Mishne, G & Glance, N.2006)
Consumer spending indicate disease infection rates (Choi, H & Varian, H.2009)
Related work
IntroductionRelated workMethodologyConclusion
Outline
Analysis of the relation between twitter messages and stock market index Selection of happiness and unhappiness words
Method 1: Twitter message and the stock price
Method 1: Twitter message and the stock price
Analysis of the relation between twitter messages and stock market index Selection of happiness and unhappiness words Evaluating both happiness and unhappiness words in the
same tweet
Where f=frequency of i’th word, Avg_happiness(wordi)=happiness value of word and Avg(T)=average happiness of given tweet
Method 2: Twitter mood predicts the stock price
Analyzing the text content of daily Twitter feeds to find the correlation between stock price and twitter mood Phase 1: Using two mood tracking tools: OpinionFinder &
Google-Profile of Mood states (GPOMs) to extract feature of mood OpinionFinder: Positive vs. nagetive mood GPOMs: Calm, Alert, Sure, Vital, Kind, and Happy
Phase 2: Granger causality analysis to test correlation between Dow Jones Industrial average (DJIA) values and GPOMs and OF values
Phase 3: Deploying a Self-Organizing Fuzzy Neural Network model (non-linear model) to test the hypothesis
Method 2: Twitter mood predicts the stock price
Method 3: Technical analysis with sentiment
Combining technical analysis with sentiment analysis for stock prediction Extract feature (using SentiWordNet):
Time series data (price and volume) source Social network source (on Engadget)
Technical indicators Using a multiple kernel learning framework to learn and
prediction the stock price
Method 3: Technical analysis with sentiment
Technical analysis
Emotion analysis
Outline
IntroductionRelated workMethodologyConclusion
Conclusion
Method 1: It is naïve but useful to predict the stock price index by just using happiness and unhappiness
Method 2: The result showed that changes in the public mood state could indeed be tracked from the content of large-scale Twitter feed using simple text processing techniques.
Method 3: It is considerable to use multiple kernel learning that covers several features.