Social Media Customer Intelligence: Data Network Analytics meets Text Mining

Rosaria Silipo, Data Scientist @Knime, talk at Data Science London @ds_ldn

Transcript

1. Creating Customer Intelligence from Social Media Data Network Analytics meets Text Mining with KNIME Dr. Rosaria Silipo
2. Social Media Data Water Water Everywhere, and not a drop to drink 2
3. Social Media Data Water Water Everywhere, and not a drop to drink What companies do with it: • Download and keep • Topic [Shift] Detection (email content routing, detect market interest shift, clinical studies, query non structured DBs) • Sentiment Analysis (marketing, polls, elections) • Connection Analysis (influencers, risk analysis) • .... 3
4. Social Media Data Water Water Everywhere, and not a drop to drink The Analysis Tools: • Web Crawlers • Visual Exploration • Topic Detection (NLP, Ontologies) • Sentiment Score (NLP) • Influence Score (Network Analytics) • Predictive Analytics (?) 4
5. Case Study Example: Slashdot Data 5 Basic Numbers: • 24532 users • 491 threads with • 15 – 843 responses • 12 – 507 users • 113505 posts • 60 main topics • Selected Topic: Politics Post Comments
6. Case Study Example: Slashdot • Very rich data sources about customers ! • We want to establish: • How users feel about the discussed topic • Whether it matters how users feel • A more general abstraction of the results 6
7. Sentiment Analysis Remove anonymous users, group by PostID Words Tagging Positive words Negative words MPQA Corpus BoW, Entity Filter, Word Frequency, Attitude Calculation by Document User Bins Word cloud for selected users Total Attitude by User
8. Slashdot – Sentiment Analysis • 16016 positive users • 7107 negative users • Most positive user: dada21 (2838 positive/1725 negative words) • Most negative user: pNutz (43 positive/109 negative words) • Which Topics have positive users in common ? – Government – People – Law/s – Money – Market – Parties
9. Slashdot – Text Mining Most Negative User pNutz
10. Slashdot – Text Mining Most Positive User dada21
11. 11 Network Creation User1 User2 User3 User6 User4 User5
12. Topic Graphs 12
13. 13 Topic Graphs
14. Topic Graph: NASA 14
15. Topic Graph: Sci-Fi 15
16. Hubs & Authorities 16 • Hubs = Follower • Authorities = Leader Filtering anonymous users and creating network Centrality index to define hub weight and authority weight Users with hub and authority weights and other features
17. Hubs & Authorities 17 dada21 Doc Ruby Carl Bialik from the WSJ pNutz 99BottlesOfBeerInMyF Tube Steak
18. 18 KNIME: Bringing it all together Network Analysis Text Analysis Users with hub and authority weights and other features Users bins: positive, negative, neutral
19. 19 Carl Bialik from the WSJ dada21 Doc Ruby 99BottlesOfBeerInMyF WebHosting Guy pNutz Tube Steak Catbeller
20. What we have found ... - The positive leaders - The neutral leaders - The negative leaders - The inactive users 20 What identifies each group? How do I identify a new user? How do I handle each user?
21. Why Clustering? - No a priori knowledge (not even on a subset of users) - Prediction and interpretation capabilities required 21 k-Means algorithm
22. Re-sampling the Training Set 23 k = 10
23. The k-Means Clusters 24
24. Additional Discoveries • There are only very few real leaders! Authority and hub scores identify active participants rather than leaders. • Superfans can be found in cluster_3 • Negative and (sigh!) active users are collected in cluster_1. • Neutral users are usually inactive (cluster_2, cluster_7, and cluster_8) • Positive users with different degrees of activity are scattered across the remaining clusters. 25
25. The k-Means Clusters 26 Superfans Negative users Neutral users Fans
26. The operational Workflow 27 Pre-processing Cluster Extraction Assignment of new data
27. Notes • MPQA Corpus: publicly available Subjectivity Lexicon (http://www.cs.pitt.edu/mpqa/lexicons.html) • User Characterization is Sum -> Mean • NLP: No sentence splitting, no negation identification. • For a more refined syntaxis-based sentiment analysis -> „External Tool“ node 28
28. External Tool Node The „External Tool“ node executes any external program from command line 1. Writes input data to an input file 2. Calls Tool to run on input file and command line options and to write results to output file 3. Reads output file and presents data at output port 29
29. Alternative Sentiment Analysis Free non-interactive Command Line running Tools for Sentiment Analysis not found SentiStrength v2.2 (still interactive) 30 External Tool and Generic Web Service Client
30. Web Crawling Workflow 31 Community Web Crawler Node XML Parsing Nodes
31. Next Steps - Integrate topic information - Integrate user demographic and behavioural information - Discover [time series] patterns for early detection of negative users and superfans - Try other techniques, maybe even on manually segmented data, to discover new user segments 32
32. Where do I find more? Whitepaper: rosariasilipo@yahoo.com Complete Workflows + Data: www.knime.com - text mining - network mining - combined analysis (note the above 3 process huge data and require 16G memory) – clustering Open Source Software: KNIME www.knime.com 33