A short manual to Stock Owl

Hi,

This is short manual for Stock Owl service. Stock Owl is stock news aggregator who utilises neural network for categorising news sentiment. Scale for sentiments is from -1 (negative) to 1 (positive).

Header
Top left: Shows how many votes you have given in this session. You vote by giving "Negative", "Neutral" or "Positive" vote for each news article. Votes are reseted when you close your browser.

Top center: Shows average sentiment for news for given timeline (see Select timeline selection box) and category (see Select category selection box). If nothing is selected, sentiment is calculated from all news. If i.e. Nokia is selected, sentiment is calculated for Nokia. Others sentiment are calculated from users votes. Owls sentiment is calculated from neural networks classified news. Three different algorithms are used (naive bayes, logistic regression and linear svc) and their average is calculated.

Home - goes to start page
Blog - goes here
TOP - goes to statistics page. You can change the timeline

Select category - you can go to stock specific page
Select timeline - you can change timeline. Calculated average sentiment changes accordingly.

Center
Shows the news (updated every 15 min): header (opens article in new window), content, source, timestamp, which company it handles, how many times it has been clicked
Buttons are used for voting: if you think this article has negative affect on subject, please click on red button etc. This information is used for training the neural network: it should mimic the judgment of masses. Others or neural network sentiment is not shown before voting, because it might affect your sentiment. You can vote only once.

Once you have chosen the company, you can see some statistics. First chart shows price and volume (delayed by 1 day). Second chart shows interest and sentiment daily (updated every 15 min). Third chart is embed from Google (it doesn't always show up). It shows search interest towards company name.

Select timeline - you can change timeline. Calculated average sentiment changes accordingly.

Footer is used for changing pages

Stock Owl

I have finally published beta version of a stock news aggregator called Stock Owl. It collects Finnish economy news from different sources and uses neural network and crowdsourcing to categorise each article's sentiment.

So what it does for you:
- You can read Finnish economic news from one website. It's mobile phone optimised service.
- Artificial intelligence analyses all news sentiments
- Shows other users interest and sentiment about different news
   -> Helps to find out general sentiment about company or economy -> predict market behavior

I suggest following pipeline for measuring and predicting stock market actions with big data:
- Awareness -> Interest -> Action (do nothing, or buy or sell stocks)
   - News (new information) arise awareness and affect emotions; spread and sentiment of news analysed in Stock Owl
   - Investors gather information before decision; they discuss with others, read magazines or use internet search engines. Interest towards company can be measured from search engines search trends.
   - Actions can be measured from stock market; price, option and volume changes. Could higher awareness and interest towards company used for predicting markets?

This is a part of my Economics Master's Thesis. My Masters Thesis hypothesis based on following:
- Humans rationality is bounded; emotions, biases and heuristics (i.e. herding) affect decision making process. Google for Tversky and Kahneman.
- Markets are not always efficient and there are limits to arbitrage
- There are different types of investors: differences in rationality (sources of information and ability/speed to process information), and in criteria for entering and exiting markets
- Todays overwhelming information flow creates poverty of attention
- Keynesian Beauty Contest

Beta version means following:
- There might be bugs in the code
- Neural network needs more human classified data to be accurate classifier (now 90% accuracy based on training)
- Some features are rough or missing
- Language is English for the moment, a translation to Finnish is coming later
- The name of service is not final

You can find Stock Owl from here:

http://ec2-52-59-224-65.eu-central-1.compute.amazonaws.com/

Please share your thoughts and comments with me.

Project plan revision 0

Project Plan revision 1
Date: 25.12.2016
Schedule : 9 months

Resources: me for the moment

Target: see point 0

 My project has been delayed for a year from original plan, aaand finally I have clarified my target!

Plan:

0) Generate target for the project. (Ready: End of this year)

- Done: Find out if collective intelligence, crowds, AI and big data can used for predicting markets

1) Gather theory, articles and empirical evidence (Needed time: 6-7 months)

- Partially done. I have loads of articles.

2) Generate ideas: discuss and write about it (Needed time: 2-3 months)
- Starting. I need volunteers to help me out. I need to create database to train sentiment analysis (neural network)

Hello world

My name is Jarkko Heikura. I'm currently working in a Finnish company and studying economics in University of Tampere. I'm strongly interested in economics, psychology and artificial intelligence. This the project page for my master thesis, which takes around 9 months.

Motivation for the project is following. What is optimal decision and could it be broken to pieces? What does it mean in uncertainty? Could computers and data be harnessed to help people to make better decisions in a complex world? I hope that we'll find some answers and develope applications to these and many more during this project.

Please feel free to contact me for discussion.