Mobile Data Stream Mining (Advances)

Download Mobile Data Stream Mining (Advances)

Post on 28-Jul-2015

134 views

Category:

Data & Analytics

5 download

Embed Size (px)

TRANSCRIPT

1. What is new? STAR MARS MSA PDM Mobile Data Stream Mining (Advances) Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University 21 May 2015 Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 2. What is new? STAR MARS MSA PDM 1 What is new? 2 STAR 3 MARS 4 MSA 5 PDM Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 3. What is new? STAR MARS MSA PDM 1 What is new? 2 STAR 3 MARS 4 MSA 5 PDM Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 4. What is new? STAR MARS MSA PDM Recent Advances Smartphones have become more ubiquitous and powerful Streaming sensory data led to the Internet of Things (IoT) Human activity recognition has emerged as an important application for mobile users (STAR and MARS) With reliance on social media through smartphones, applications like mobile sentiment analysis have emerged (MSA) Distributed and autonomous mobile computing has become a necessity (PDM) Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 5. What is new? STAR MARS MSA PDM 1 What is new? 2 STAR 3 MARS 4 MSA 5 PDM Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 6. What is new? STAR MARS MSA PDM STAR: Overview It stands for STream learning for mobile Activity Recognition It provides Dynamic incremental learning from evolving data stream Eective active learning with lowest cost Mobile real-time AR application Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 7. What is new? STAR MARS MSA PDM STAR System Architecture Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 8. What is new? STAR MARS MSA PDM Modelling Component of STAR Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 9. What is new? STAR MARS MSA PDM Recognition Component of STAR Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 10. What is new? STAR MARS MSA PDM Four Measures for Activity Recognition Distance Gravity Density Deviation Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 11. What is new? STAR MARS MSA PDM Adaptation Component of STAR Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 12. What is new? STAR MARS MSA PDM OPPORTUNITY Dataset Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 13. What is new? STAR MARS MSA PDM STAR Performance Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 14. What is new? STAR MARS MSA PDM 1 What is new? 2 STAR 3 MARS 4 MSA 5 PDM Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 15. What is new? STAR MARS MSA PDM MARS: Mobile Activity Recognition System Unlike earlier approaches, the classier is built/updated on-board the mobile device itself utilising data stream mining techniques. The advantages of on-board data stream mining for mobile activity recognition are: personalisation of models built to individual users; increased privacy as the data is not sent to an external site; and adaptation of the model as the users activity prole changes. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 16. What is new? STAR MARS MSA PDM MARS Training Process Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 17. What is new? STAR MARS MSA PDM MARS Performance We used incremental Naive Bayes classication The OPPORTUNITY dataset has been used Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 18. What is new? STAR MARS MSA PDM MARS Interface Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 19. What is new? STAR MARS MSA PDM 1 What is new? 2 STAR 3 MARS 4 MSA 5 PDM Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 20. What is new? STAR MARS MSA PDM Mobile Sentiment Analysis It is based on the SentiCorr system which performs multi-lingual sentiment analysis of personal correspondence, and correlate the inferred sentiment to stress level. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 21. What is new? STAR MARS MSA PDM Mobile Sentiment Analysis: A Wide Media Coverage A wide media coverage including articles in the BBC, CNN, the Independent, the DailyMail, Los Angeles Times, The Age, Computer Weekly, Times Higher Education, and several radio interviews. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 22. What is new? STAR MARS MSA PDM MSA Workow SentiCorr achieves sentiment classication at the sentence level by using POS (Position Of Speech) tagging to identify the types of the words in the sentence; the subjectivity detection stage then uses the POS tags to identify opinion lexicon and hence if the sentence is subjective or objective; the polarity detection stage also utilises the POS tags to search for patterns in the sentence that indicate positive or negative expressions. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 23. What is new? STAR MARS MSA PDM Subjectivity Detection The principle employed for subjectivity detection is boosting, by use of the Ada- Boost (adaptive boosting algorithm) Features utilised are POS tags, pre-dened lexicons that contain positive, negative and negation words, the presence of exactly one positive word, the presence of multiple positive words, the presence of exactly one negative word and the presence of multiple negative words, and whenever a positive or negative word is directly preceded by a word from the negation list, its polarity is ipped. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 24. What is new? STAR MARS MSA PDM Polarity Detection The principle employed for polarity detection is RBEM which uses rules to dene an emissive model. Eight rules are applied (e.g., for each positive pattern an emission value is calculated based on the distance of the elements in the sentence from the centre of the positive pattern) Once the rules have been applied, every element of the sentence has an emission value and the nal polarity of the message is calculated by summing the emission values for each element. If the nal polarity of the sentence is greater than zero, the sentence is positive; if it is less than zero the sentence is negative, if it is zero the polarity of the sentence is unknown due to insucient patterns in the sentence model. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 25. What is new? STAR MARS MSA PDM MSA Performance Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 26. What is new? STAR MARS MSA PDM 1 What is new? 2 STAR 3 MARS 4 MSA 5 PDM Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 27. What is new? STAR MARS MSA PDM Pocket Data Mining Pocket Data Mining (PDM) is the term we coined in 2010 to describe the collaborative mining of streaming data in mobile and distributed computing environments. Three technological enablers: data stream mining; mobile software agents; and programming for small devices. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 28. What is new? STAR MARS MSA PDM What is a Mobile Agent? A software program Moves from machine to machine under its own control Suspends execution at any point in time, transport itself to a new machine and resume execution Once created, a mobile agent autonomously decides which locations to visit and what instructions to perform Continuous interaction with the agents originating source is not required Implicitly specied through the agent code Specied through a run-time modiable itinerary Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 29. What is new? STAR MARS MSA PDM PDM Agents and Architecture PDM Agents (Mobile) agent miners (AM): are either distributed over the network when the mining task is initiated or are already located on the mobile device. Mobile data stream mining Mobile agent resource discoverers (MRD): are used to explore the available resources. Mobile cloud Mobile agent decision makers (MADM): roam the network consulting the mobile agent miners to collaborate in reaching the nal decision. Ensemble learning PDM Architecture Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 30. What is new? STAR MARS MSA PDM PDM Flowchart Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 31. What is new? STAR MARS MSA PDM Simple Weighted Majority Voting of the MADM Y = 1.75 (0.55+0.65+0.55) X = 1.80 (0.95+0.85) Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 32. What is new? STAR MARS MSA PDM PDM Performance Each AM has access to 20%, 30%, or 40% of the features (random vertical partitioning). Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 33. What is new? STAR MARS MSA PDM PDM Performance Hoeding Trees Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 34. What is new? STAR MARS MSA PDM PDM Performance Naive Bayes Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 35. What is new? STAR MARS MSA PDM PDM Performance: HT and NB (Heterogeneous) Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 36. What is new? STAR MARS MSA PDM Model Selection for MADM (Coll-Stream) It addresses concept drift issues The Coll-Stream is a selection method that partitions the instance space X into a set of regions R. For each region, an estimate of the models accuracy is maintained over a sliding window. This estimated value is updated incrementally as new labelled records are observed in the data stream or new models are available. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 37. What is new? STAR MARS MSA PDM Coll-Stream Performance Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 38. What is new? STAR MARS MSA PDM PDM Demo Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 39. What is new? STAR MARS MSA PDM PDM Book Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 40. What is new? STAR MARS MSA PDM Summary Smartphones and tablets have become ubiquitous computing devices Human activity recognition can serve many important applications in the era of IoT MARS and STARS are two systems that provide eective and ecient activity recognition PDM is a framework for distributed data stream mining in the mobile environment, potentially serving a large number of applications The eld is opening up for more contributions Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 41. What is new? STAR MARS MSA PDM Some References Gaber, M. M., Gomes, J. B., & Stahl, F. (2014). Pocket data mining. Springer. Abdallah, Z. S., Gaber, M. M., Srinivasan, B., & Krishnaswamy, S. (2015). Adaptive mobile activity recognition system with evolving data streams. Neurocomputing, 150, 304-317. Gomes, J. B., Gaber, M. M., Sousa, P. A., & Menasalvas, E. (2013). Collaborative data stream mining in ubiquitous environments using dynamic classier selection. International Journal of Information Technology & Decision Making, 12(06), 1287-1308. Chambers, L., Tromp, E., Pechenizkiy, M., & Gaber, M. (2012, September). Mobile sentiment analysis. In Proceedings of the 16th International Conference on Knowledge-Based and Intelligent Information & Engineering Systems. Gomes, J. B., Krishnaswamy, S., Gaber, M. M., Sousa, P. A., & Menasalvas, E. (2012). Mobile activity recognition using ubiquitous data stream mining (pp. 130-141). DaWaK. Springer Berlin Heidelberg. Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances) 42. What is new? STAR MARS MSA PDM Acknowledgements Dr Frederic Stahl Dr Joao Gomes Dr Zahraa Said Abdallah Dr Mykola Pechenizkiy Lorraine Chambers Erik Tromp Prof. Philip Yu Prof. Max Bramer Prof. Ernestina Menasalvas Dr Mohamed Medhat Gaber Reader, School of Computing Science and Digital Media Robert Gordon University Mobile Data Stream Mining (Advances)...