Please enable JavaScript.

Coggle requires JavaScript to display documents.

Chapter 10 (Mining News Stories to Predict Stock Price Movement (The task,…

- - - - News arrive fairly often and the market responds quickly
      - It's unrealistic to predict what price a stock will have a week from now based on news today
      - Predict same day stock prices based on same day news
    - - It is therefore satisfying to predict the direction of the movement
        
        down
        
        no change
        
        up
      - Can be simplified further into change and no change
    - - Instead it is easier to predict relatively large changes
      - This will make the signal a bit cleaner at the expense of yielding fewer events
      - Makes signal a bit cleaner at the expense of yielding fewer events
      - Deliberately ignore the subtlety of small fluctuations
  - - - Overly permissive
      - Relevance of a stock to a document may not be clear without a careful reading
      - Two streams of data
- - - - In sequence they name unique entities with interesting identifiers
  - - - Though not always optimal
    - - topic layer
        
        First model the set of topics in a corpus separately
        
        Each document constitutes a sequence of words
        
        Instead of the words being used directly by the final classifier, the words map to one or more topics