13 th August, EISIC 2013, Uppsala, Sweden Dr. Martin Sykora, Prof. Tom Jackson, Dr. Ann O’Brien and Dr. Suzanne Ela

1. National Security and Social Media Monitoring Extracting the Meaning Of Terse Information in a Visualisation of Emotion 13th August, EISIC 2013, Uppsala, Sweden Dr. Martin Sykora, Prof. Tom Jackson, Dr. Ann O’Brien and Dr. Suzanne Elayan

2. Contents • Introduction • 5 Systems Overview (common featurs.) • EMOTIVE System & Future Work • Conclusions

3. IntroductionSystems Overview EMOTIVE Conclusions • Egyptian activist; “We use Facebook to schedule our protests, Twitter to coordinate and YouTube to tell the world.” (Meier 2011) • Social Media – Polling public opinion: O’Connor et al. 2010, Tumasjan et al. 2010, Cheong et al. 2011, Lansdall-Welfare et al. 2012. • Social Media is first to break-the-news! • - 2008 Mumbai attacks, where individuals on location broke the news via Twitter. • - July 2009 Jakarta bombings, where Twitter broke the news. • - Even earthquakes, ranging from seismic intensity scale 3 or more, were reported quicker by Twitter users as opposed to the relevant Japanese agency. • Commercial Interest: Attensity, Crimson Hexagon, Sysomos, Brandwatch, Vocus, Socialradar, Radian 6, ... • Crisis mapping community: http://crisismappers.net • …in this talk, briefly look at 5 systems: CrisisTracker, Crisees, SensePlace2, Swiftriver(Ushahidi), Twitcident.

4. IntroductionSystems Overview EMOTIVE Conclusions • Prior literature highlights: (Cheong and Lee 2011, Glass and Colbaugh 2012) • importance of gauging public response to terrorism events from social media • ..and specifically highlights importance of automatic sentiment detection in Tweets • Automated monitoring systems (tools & techniques) are necessary to deal with the big data in social streams.

5. IntroductionSystems Overview EMOTIVE Conclusions • Unfortunately (“fortunately”) there is a very wide range of different content and styles of messages communicated on Twitter, which requires selective interpretation of the messages in a National Security setting. • It is apparent (Rogstadius et al. 2011, Johansson et al. 2012, Cheong and Lee 2011) that, • ..to facilitate monitoring of a certain event or entity of interest, • 1-efficient extraction of messages, 2-geo-location, 3-emotion evaluation, 4-clustering and organization of the tweet messages, and an 5-intuitive user-interfaceare necessary. • Hence to facilitate effective national security monitoring tasks: • Keyword / Keyphrase monitoring, first event detection, filtering and extraction • Accurate geo-location detection • Emotion detection and evaluation • Tone of tweet message detection, further semantic enrichment and organisation • User-interface visualisation

6. IntroductionSystems Overview EMOTIVE Conclusions • Topic relevant messages retrieved & spam filtered out: • Manual input; keywords / phrases / #hashtags • keywords based on automated named entity recognition of regularly re-checked trending twitter topics (Cheong & Lee 2011) • Using first story detection (Locality-Sensitive Hashing is popular) (Petrovic et al. 2011) • 2, 3 and 4 further substantially enriches retrieved text messages by providing context through: • extracting location details • extracting communicated emotions / sentiment • extracting various features from the tone of the messages / sem. enrichment • …these 3 steps essentially steps in automated semantic enrichment [22] • real-time visualisation and a faceted user-interface to explore the enriched sparse text message data

7. IntroductionSystems Overview EMOTIVE Conclusions

8. IntroductionSystems Overview EMOTIVE Conclusions

9. IntroductionSystems OverviewEMOTIVE Conclusions • Emotion extraction, prior work: • Notions of affect and sentiment have been rather simplified in current state-of-the-art, often confined to their assumed overall polarity (i.e. positive / negative), Thelwall (2012) • Another problem with polarity-centric sentiment classifiers is that they generally encompass a vague notion of polarity that bundles together emotion, states and opinion • There is no common agreement about which features are the most relevant in the definition of an emotion and which are the relevant emotions and their names, Grassi (2009) • 1-machine learning; 2-lexicon / linguistic analysis; & 3-polarity estimation from term co-occurrence (Thelwall et al. 2012) • Comparison: • de Choudhury and Counts (2012) & Thelwall et al. (2012)

10. IntroductionSystems OverviewEMOTIVE Conclusions • EMOTIVE emotion detection discovers fine-grained explicit emotions in sparse text messages: • Anger, Disgust, Fear, Happiness, Sadness, Surprise(Ekman’s 6 basic emotions) + Shame,and Confusion. • Shame – common on Twitter • Confusion – useful for situational awareness, Oh et al. (2011)

11. IntroductionSystems OverviewEMOTIVE Conclusions • The ontology contains over 300 emotional terms, with many intensifier, conjunction, negation and interjection words and phrases. It also contains information on the perceived strength of emotions, and some linguistic analysis related information. • Example Emotion Terms from the Ontology: Anger (e.g. enraged, infuriated, peeved, in a tizzy…) Confusion (e.g. chaotic, distracted, perplexed, confuzzled…) Disgust (e.g. appalling, beastly, bullshit, scuzzy…) Fear (e.g. cold feet, goose bumpy, petrified, scary…) Happiness (e.g. blissful, chuffed, delighted, in high spirits…) Sadness (e.g. depressed, devastating, duff, grief stricken…) Shame (e.g. abashment, degrading, hang head in shame, scandalous…) Surprise(e.g. astonished, disbelief, gobsmacked, off guard…)

12. IntroductionSystems OverviewEMOTIVE Conclusions

13. IntroductionSystems OverviewEMOTIVE Conclusions • Study of language performed by an English language and literature PhD level research associate, with training in linguistics and discourse analysis, during a three month time-period. • 600MB of cleaned Tweets on 63 different UK-specific topics / search-terms datasets • Focused on identifying commonly used explicit expressions of emotion • OOV (Out of Vocabulary) terms, Wordnet synset synonym lists of emotional expressions, Dictionary.com, Thesaurus.com, the Oxford English online dictionary, Urbandictionary.com, Internetslang.com… • Emotional terms and activation levels identified and used in work by Choudhury et al. (2012) and lexicon lists of intensifiers, negators & words of basic sentiment used in SentiStrength-2, Thelwall et al. (2012) were also reviewed.

14. IntroductionSystems OverviewEMOTIVE Conclusions • On an initial golden-dataset (annotated by 2 human annotators) of emotive tweets the technique achieved excellent results, F-measure = .962: • This is an extremely high f-measure illustrating the successful nature of the ontology. • To compare our high f-measure to another approach,fine-grained emotion detection from Choudhuryet al. (2012) achieved; .744 / .668 (f-measure), .830 / .658 (precision) and .674 / .680 (recall); direct matching / stemmed matching, respectively. • EMOTIVE’s emotion strength scoring approach was evaluated against SentiStrength-2 (Thelwall et al. 2012): a consistent and statistically significant correlation was found; which indicates that we are measuring in line with a sentiment scoring state-of-the-art system. Recall, precision and f-measure, were computed using an equivalent approach as used in CoNNL-2003 shared task on NER (Tjong et al. 2003).

15. Woolwich Soldier Killing (Lee Rigby) – #woolwich, Anjem Choudary The brutal murder sparked a storm of emotional reactions of Sadness, Disgust and Surprise. At the same time the controversial cleric Anjem Choudary was most often mentioned with extreme emotions of Anger and Disgust. Example reactions to Anjem Choudary .I'm quite angry that Anjem Choudary is on Newsnight tonight - I can only imagine how furious Muslims he falsely claims to speak for must be [anger] .And I'm angry that Anjem Choudary is aloud to preach hate in our towns and city's It's the government we should be angry with not a religion [anger] .Anjem Choudary, gfy. Ruining the 'Choudhary' name for all of us, you complete bastard, it's sickening #woolwich [disgust] .@EDLTrobinson so sad, and so wrong that ANJEM CHOUDARY can get air time saying muslims around the world will call them heroes what a twat. [sadness]

16. Woolwich Soldier Killing (Lee Rigby) – #woolwich .Enjoyed last night's @HyderiCentre event in response to #Woolwich with @cllrjudybest, @jon_bartley, @ihrc, Syed Ammar & Sheikh Panju.. [happiness] .Great @LabourList article from @jonewilson on the town I'm proud to live in. We love #Woolwich http://t.co/W0MXkxIqvm[happiness] .The family of murdered soldier pay tribute. Rebecca Rigby: "I love Lee, I always will and I'm proud to be his wife." #woolwich[happiness] .My heart goes to the soldiers family, friends, the people of #woolwich & all those effected, so pretty much everyone. Terribly sad news. [sadness] .@SkyNewsBreak: Military commanders tell soldiers told not to wear their uniforms in public until further notice #Woolwich" - Sad times :( [sadness] .The fact the soldier was a father upsets me further. Maybe it shouldn't but it does #woolwich [sadness] .#Woolwich Attack: New Shocking Video of Terrorists Charging at Police Car, Getting Shot: http://t.co/icVNVud5ci via @youtube #EDL [surprise] .Utterly astonished to see some videos popping up claiming the #Woolwich attack was a hoax with the media and government colluding! [surprise] .@steveplrose: "Free speech in Britain is threatened by the influence of Muslims in the media" YouGov question. Wow. #woolwich http://t.co/NqFjpDqJpy. [fear] .Following the #Woolwich incident, people in #Britain are anxious. Reports of a man with an axe at #LondonBridge is making people nervous. [fear] .#woolwich absolutely disgusting scene yesterday. Jst so annoying[disgust] .#Woolwich - so awful. Strength to the victim's families.[disgust] .Can't believe the news about yesterday's #woolwich attack. Disgusting. Some people are so sick!! [disgust] .RT @skymartinbrunt: #woolwich She was arrested on Wednesday after apparently asking police for protection when her malicious tweet prompted angry backlash. [anger] Dyka Ayan Hassan from Harrow, 21, arrested for a malicious tweet

17. IntroductionSystems OverviewFuture Work Conclusions Funding – ReDites (Real Time, Detection, Tracking, Monitoring and Interpretation of Events in Social Media) Event detection & tracking Location profiling EMOTIVE emotions extraction Author profiling (tone of messages; first hand, bots, astroturfing, social network analysis) Event exploration and event summarisation / notification interface

18. IntroductionSystem Overview Interpretation Conclusions • Powerful (fine-grained) Emotion extraction from Tweets is mostly missing in these types of Systems, although prior work found it to be of significant importance! • Social media monitoring system – EMOTIVE • Excellent F-measure achieved & evaluated against two other systems • Aiding analysts to interpret live events • Learning and predicting from previous datasets • ReDites to tackle the highlighted System Elements – Deliver a Demonstrator for National Security Monitoring

19. Thanks ReDites

20. References Cheong M. and Lee V. C. S., 2011. A microblogging-based approach to terrorism informatics: Exploration and chronicling civilian sentiment and response to terrorism events via Twitter, Journal of Information Systems Frontiers – Springer 13, pp. 45-59 ChoudhuryM. and Counts S., 2012. The Nature of Emotional Expression in Social Media: Measurement, Inference and Utility, Technical Report: Microsoft. Drummond T., 2004. Vocabulary of Emotions [Online], North Seattle Community College, [last viewed 9.1.2012]. Available from http://www.sba.pdx.edu/faculty/mblake/448/FeelingsList.pdf Ekman P., 1994. All emotions are basic. The nature of emotion: Fundamental questions 15-19. Gimpel K., Schneider N., O'Connor B., Das D., Mills D., Eisenstein J., Heilman M., Yogatama D., Flanigan J. and Smith N., 2010. Part-of-speech tagging for twitter: Annotation, features, and experiments, Technical Report. Glass K. and Colbaugh R., 2012. Estimating the sentiment of social media content for security informatics applications, Security Informatics 1, pp. 1-16 GrassiM., 2009. Developing HEO human emotions ontology, Biometric ID Management and Multimodal Communication, Springer Berlin Heidelberg, pp. 244-251 Izard C. E., 2009. Emotion theory and research: Highlights, unanswered questions, and emerging issues. Annual Review of Psychology 60, 1-25. O'Connor B., Krieger M. and Ahn D., 2010. TweetMotif: Exploratory Search and Topic Summarization for Twitter, Proceedings of the International AAAI Conference on Weblogs and Social Media, Washington DC (USA) Oh O., Agrawal M. and Rao H., 2011. Information control and terrorism: Tracking the Mumbai terrorist attack through twitter, Information Systems Frontiers 13, pp. 33-43 Potts C., 2011. Potts Twitter-aware Tokeniser - http://sentiment.christopherpotts.net/code-data/happyfuntokenizing.py, [last viewed 29.3.2013] Plutchik R., 1980. Emotion: A Psychoevolutionary Synthesis. Longman Higher Education. Ritter A., Clark S., Mausam and Etzioni O., 2011. Named Entity Recognition in Tweets: An Experimental Study, Proceedings of Conference on Empirical Methods in Natural Language Processing, Edinburgh (UK) Sykora M., Jackson T. W., O’Brien A. and Elayan S., 2013. EMOTIVE Ontology: Extracting Fine-Grained Emotions from Terse, Informal Messages, IADIS International Conference on Intelligent Systems and Agents 2013, Prague (Czech Republic). Thelwall M., Buckley K. and Paltoglou G., 2012. Sentiment Strength Detection for the Social Web, Journal of the American Society for Information Science and Technology 63, pp. 163-173 Tjong E. F., Sang K., Meulder F. D., Introduction to the CoNLL-2003 Shared Task: Language-Independent Named Entity Recognition, Proceedings of the ACL seventh conference on Natural Language Learning.