Machine learning for social media analytics
- 1. Machine Learning for Social Media Analytics
- 2. Who I am Yevhen Terpil Data Scientist
- 3. The company YouScan is a social media analytics company. Offices in Kiev and Moscow. Our social media monitoring platform is used by 300+ major brands and agencies, including: Rated #1 social media monitoring platform in CIS in AdIndex Technology Index 2016.
- 4. The product: YouScan.io for social media listening SaaS platform which provides real-time monitoring and analytics of brand mentions on social networks, blogs, forums, review sites and online news. Used by PR, Marketing, Client Service departments to: - manage brand’s online reputation; - discover consumer insights and social media audiences involved in discussions, find brand influencers and detractors; - provide customer support on social media.
- 5. Key product features Real-time social media monitoring YouScan monitors major social networks, as well as blogs, forums, product reviews and online news, collecting and analyzing 50M+ posts daily. Results are available in a slick interface in real-time along with beautiful automatic reports. Unlimited Results Unexpected surge of discussions due to major event? No problems - we know how to handle Big Data, so you don't need to pay extra. All mentions are stored permanently in your account, and there's no limit to the amount of mentions you can monitor with our transparent pricing.
- 6. Key product features In-depth Analytics with AI and NLP • Slice and dice all reports across ANY dimensions (specific sources and their types, authors' gender, location, sentiment, keywords etc) on-the-fly • AI discovery of trending topics helps you stay on top of emerging discussions • Use audience insights report to find brand advocates and detractors, analyze their profiles and better connect to your audience • Many other reports available (sources, sentiment, tags etc)
- 7. Machine learning in YouScan Sentiment analysis Autocategories Text classification Anomaly detection Smart Alerts Trends*
- 8. #SE2017 All Sources, collected by YouScan
- 11. Big data problem #SE2017, All Sources, collected by YouScan anomaly?
- 12. My posts Irrelevant posts Word of mouth News all mentions Subsets of mentions
- 13. Mention classification Irrelevant posts Autotags Smart tags Word of mouth News My posts machine learning custom rules
- 16. Top authors top authors by influence top authors by mentions #SE2017, WOM, collected by YouScan
- 17. Under the hood
- 18. SUPERVISED LEARNING UNSUPERVISED LEARNING labeled data raw data Machine Learning types
- 19. Text classification TF-IDF + SVM Recurrent Neural Nerworks 1. Rule based 2. Lexicon based 3. Classic Machine Learning TF-IDF + Logistic Regression 4. Deep Learning labeled data SUPERVISED LEARNING
- 20. iPhone looks and feels great apple and samsung start new I love mac and cheese 0 1.2 0 … 0 0.3 1.4 10000 TF-IDF + Logistic Regression great iPhone
- 21. Text clustering Word2Vec / FastText / Glove Doc2Vec TF-IDF + SVD FastSent SDAE 3. Text embeddings 2. Word embeddings 1. Classic approach UNSUPERVISED LEARNING raw data
- 22. Word2Vec
- 24. News Active discussion Map of mentions
- 25. #SE2017 mentions #SE2017, Word2Vec, collected by YouScan
- 26. #SE2017 mentions #SE2017, Word2Vec, collected by YouScan Active discussions Ukr language
- 27. SE2017 mentions #SE2017, Word2Vec, collected by YouScan
- 28. LeadScanner
- 29. LeadScanner
- 30. Intent Intelligence Technology Intent Intelligence Technology (IIT) identify social media posts containing explicit (and implicit) consumer intent, wish or request towards specific goods and services. 1. rule based 2. classic ml 3. deep learning iterations
- 31. Intent Intelligence Technology Full stream of posts Neural Network post-processing to improve precision Stream of posts containing intent enriched with intent objects ML-based discovery of new intent patterns Intent patterns and markers set (common & case- specific) Intent matching Linguists Object detection Spam filtering
- 32. Next steps
- 33. Ukrainian Facebook , Top users by engagement Russian users Pro Russian Lviv Army and Volunteers Politicians and journalists Showbiz Dog lovers Audience insights
- 40. Thank you for your attention!