OpenTalks.AI - Татьяна Любимова, Предсказание спроса и предложения на сырье исходя из неструктурированных текстов

Transcript

1. Actionable dataset creator Predicting supply & demand in commodities based on unstructured texts
2. Why business is looking for sentiment in commodity-related texts? 1
3. • SENTIMENT • SCORING • TEXT LABELLING • CHUNK 2
4. 3 No sign of crude or oil or anything related to commodities
5. So what do we do with all that mess? • do not give sentiment to the text body. • look for specific events, and only then label with sentiment. 4
6. 5 Catch and Classify • define an object and it’s relevance • classify the event • calculate the sentiment
7. People use different words to speak about demand and supply 6
8. People use different words to speak about positive and negative events 7
9. 8 How does the instrument look?
10. 9 Non - predictive Strongest range
11. 10 Sentiment vs. actual price 3 days lagged 2014
13. So what? 12
14. Actionable dataset creator Thank you! tanya@particle.one @wrinklet