Home APIs DataExploration DataMining Showcase Challenges Contact
NLP_Notebook [LIVE]Predicting

Live NLP predictions: Danske Bank stock price and Berlingske news (every 5 min. during open market hours)

Predicting whether a stock price goes up/constant OR down can be derived from correlation between price and media news:

Top NLP model features:

feature                     Coefficient
 grusomt puds               0.674492   
 penge eksperter            0.621060   
 kendt ingen                0.557319   
 england ulækkert           0.551513   
 grimt overnatter           0.507227   
 udfordrer lokker           0.506352   
 helvede secret             0.504128   
 flopper ikonisk            0.486891   
 ledelse aeroporto          0.463352   
 spørgsmålene lykkeforsker  0.459704   

Tail NLP model features:

feature                    Coefficient
 ham britisk              -0.407934   
 bedste verdens           -0.410341   
 lige fødevareskandale    -0.444937   
 grimt når                -0.459368   
 tilbage kaffekonge       -0.464219   
 stræder når              -0.475946   
 undrer hjemløse          -0.475946   
 palæstinensere aeroporto -0.533765   
 arbejdsmønstre klip      -0.546474   
 grusomt pus              -0.573037   

Description: "1" when stock price value rise or stays the same "0" when stock price value decrease. A new model is trained every 5 minutes based on newest data.

Model performance: confusion matrix

Chart

Description: look at off-diagonal values like false positives that should be close to zero.

Same confusion matrix in colors

Chart

Description: look at off-diagonal values like false positives that should be close to zero.