Increasing the trading prediction by mining aggregated human texting messages
summary
Claims that mining aggregated social media and news titles improves up/down stock prediction and shows modest accuracy gains but hides exact numbers; uses word2vec embeddings (skip-gram/continuous bag), TF-IDF and PCA for features, then feeds vectors into a TensorFlow model combining CNN + logistic regression layers to predict 3-way (up/flat/down) outcomes from S&P/Yahoo/CNN and Twitter-like sources; focuses on title-only data in English and Chinese, notes heavy noise and missing days and says the prototype improves results without publishing full metrics.
AI-written summary · about
chapters (18)
- 0:00Introduction
- 1:05Research motivation
- 2:32Literature review
- 3:40Data collection
- 5:33Data sources
- 7:23Text mining techniques
- 9:16Word2Vec model
- 11:18Emotional analysis
- 13:42Model training
- 15:43Stock prediction
- 16:47Neural network
- 18:39Model evaluation
- 20:32Handling noise
- 21:35Q&A session part one
- 23:24Q&A session part two
- 24:53Future steps
- 26:45Stop words
- 28:41Q&A session part three and conclusion
related talks
35mCan Memes Predict Markets? Using Social Media Sentiment for Stock Forecasting
30mPredictive trading
01h 26mImplementing a highly scalable Stock prediction system with R, GemFire and Spring XD
30mHow to build a stock market money printing machine with AI
08mQ n A with Dr Arun Verma
30mMachine Learning: Beyond the Tutorial
01h 05mSocial Text, Sentiment, and Tone Analysis
45mDeep Learning for Natural Language Processing