First Steps towards Implementing a Sahana Eden Social Media Dashboard Irina Temnikova¹, Dogan Biyikli², Francis Boon³ Bulgarian Academy of Sciences¹, Independent Consultant², Sahana Software Foundation³ 1
Structure of the Presentation 1. Work and team presentation 2. Idea 3. Motivations 4. Implementation 5. Results 6. Future work 7. References 2
1. Work and team presentation 1. Developed during the first h4d2 hackathon 2. September 2012 3. Include language technologies in a real application (trying to help) 4. Work done in two days 5. Team members: - crisis management systems developer - emergency management practitioner - computational linguist 3
2. Idea 1. Beginning stage 2. SMS, e-mail, Google Talk, Twitter integration into Sahana Eden 3. Messages filtering by topics using language technologies 4. Messages prioritization 5. To be available for disaster managers via Sahana Eden 4
2. Sahana Eden 1. Open-source software platform 2. Developed by the Sahana Foundation 3. Aim: provide information management solutions for disaster management 4. Highly configurable and easy to integrate with other systems even during a crisis 5. Used for/by: Haiti Earthquake, Wildfires in Chile, Asian Disaster Preparedness Centre (ADPC) 5
3. Motivations 1. Need of intelligence on emergencies for decision-taking on response 2. Need of additional information from unconventional sources 3. Current social media tracking initiatives mostly manual 4. Developed tools not publicly available 6
3. Related Work Related tools: - Twitcident (Abel, F. et al., 2012) - project EPIC, Univ. of Colorado (Verma S. et al., 2011) - Univ. of Sheffield (Varga, A. et al., 2012) - StandBy Task Force? Other? 7
4. Implementation Sahana Eden: 1. Implemented in Python 2. Uses Web2Py 3. Built around a Rapid Application Development (RAD) framework 8
4. Implementation 1. User requirements 2. Message filtering 3. Message prioritisation 9
4. User Requirements 1. Developed on the basis of practical experience and reference works (Bruinsma, 2010; National Incident Management System, 2008) 2. Workflows showing decision taking steps + necessary information 3. Flexible to reflect different types of, levels of complexity of incidents and needed resources 10
4. User Requirements Three types of accidents (human-made/natural/both): 11
4. Message Filtering (Twitter) Hashtags often not used literally ( my exam was a #disaster ) Twitter-specific features (hashtags) +linguistic features (keywords) 12
4. Message Filtering (Twitter) Coverage problem Extracting keywords' synonyms Use: Python Natural Language ToolKit (NLTK) (Bird, S. et al., 2009) WordNet (Fellbaum, 1998) Earthquake earthquake, quake, temblor and seism 13
4. Message Filtering (Twitter) 14
4. Message Prioritisation (Twitter) type (SMS or tweet ) sender (trusted or not trusted ) location (mentioned or not mentioned ) mode (text+image or text ) 15
5. Results 16
6. Future Work WordNet: Part-of-speech ambiguity (fire: to fire/fervor/ardor) More tweets filtering: number of re-tweets, number of followers and their growth over time (Starbird, K. et al., 2012) Messages categorized by types of events Proper evaluation 17
7. References Abel, F. et al. (2012) Twitcident: Fighting Fire with Information from Social Web Stream. In Proceedings of International Conference on Hypertext and Social Media, Milwaukee, USA. Bird, S.et al. (2009) Natural Language Processing with Python - Analyzing Text with the Natural Language Toolkit. O'Reilly Media. Bruinsma, G. W. J. (2010). Adaptive workflow simulation of emergency response. University of Twente. Fellbaum, C. (1998) WordNet: an Electronic Lexical Database, MIT Press. National Incident Management System. (2008) U.S. Homeland Security, December 2008. Starbird, K. et al. (2012) Learning from the Crowd: Collaborative Filtering Techniques for Identifying On-the-Ground Twitterers during Mass Disruptions. ISCRAM 2012. Varga, A. et al. (2012). Exploring the Similarity between Social Knowledge Sources and Twitter for Cross-domain Topic Classification. In proceedings of Knowledge Extraction and Consolidation from Social Media (KECSM2012). Verma, S. et al. (2011) Natural Language Processing to the Rescue: Extracting 'Situational Awareness' Tweets During Mass Emergency. Fifth International AAAI Conference on Weblogs and Social Media. 18
Contacts Irina Temnikova http://pers-www.wlv.ac.uk/~in0290 irina.temnikova@gmail.com Dogan Biyikli dogan@pericula.org Francis Boon fran@sahanafoundation.org 19
Thank you Any suggestions? 20