Finnish sentiment analysis and a tool for social media tracking 10.6.2016 Helsinki
Who we are? Faculty of Business, ICT and Chemical Engineering Program degree: Business information technology Arto Kivinen arto.kivinen@turkuamk.fi Tuomo Helo tuomo.helo@turkuamk.fi Tuukka Ojanen tuukka.ojanen@turkuamk.fi
Why we are doing this? NEMO = Business Value from Negative Emotions http://nemohanke.blogspot.fi/ https://www.talentumshop.fi/negatiiviset-tunteet.html What kind of impact negative information does for business Knowing emotional aspects of social media Calculated positive, neutral and negative classification
Demonstration Sentiment & Data analyzer http://seda.cloudapp.net
Sentiment analysis methods Supervised classification analysis was performed by using multinomial logistic regression, support vector machine, and random forest algorithms Twitter messages were classified into three categories: Negative sentiment, neutral, and positive sentiment The training and test sets were based on more than 9000 manually classified tweets Dictionaries of positive (>1500) and negative (>3000) Finnish words were used The used features included word counts and percentages, but also e.g. link counts and user tag counts
Architecture BigData Social media tracker Third party software MongoDB 3.2 MongoDB 3.2 Futusome API Aggregation MapReduce Crawler Twitter/Facebook… API User HTTP Geocode enrichment Desktop WebSocket NodeJS Sentiment analysis WebSocket Drupal 8 PHP HTTP API R script Optimized in threads Mobile Apache HTTP PostgreSQL
Partners & Use cases Futusome Tampere University of Technology Super Analytics Nooa Säästöpankki Rud Pedersen Skanska S-TOK
Thank you! Questions? Arto Kivinen arto.kivinen@turkuamk.fi Tuomo Helo tuomo.helo@turkuamk.fi Tuukka Ojanen tuukka.ojanen@turkuamk.fi