TRANSFORMING BIG DATA INTO SMART DATA: Deriving Value via Harnessing Volume, Variety, and Velocity using Semantic Techniques and Technologies

Editor's Notes

1. Starting slide Various Big data problems – Traditional examples vs what we are doing examples. Variety and Velocity than Volume. kHealth problem. People will be interested in Smart Data.Traditional ML techniques, High Performance Computing, Statistics. Human level of Abstraction is Smart data.
2. Note:For images and sources, if not on slides, please see slide notesSome images were taken from the Web Search results and all such images belong to their respective owners, we are grateful to the owners for usefulness of these images in our context.
3. http://www.knowledgeinfusion.com/blog/2011/11/get-your-head-out-of-the-clouds-and-into-big-data/
4. http://www.csc.com/insights/flxwd/78931-big_data_growth_just_beginning_to_explodehttp://www.guardian.co.uk/news/datablog/2012/dec/19/big-data-study-digital-universe-global-volume
5. Types of DataFormats of DataAlso talk about the increase in the platforms that helps generating these data
6. Example high velocity Big Data applications at work:financial services, stock brokerage, weather tracking, movies/entertainment and online retail.Fast data (rate at which data is coming: esp from mobile, social and sensor sources), Rapid changes – in the data content, Stream analysis – to cope with the incoming data for real-time online analytics
7. Source: http://techcrunch.com/2012/10/27/big-data-right-now-five-trendy-open-source-technologies
8. http://radhakrishna.typepad.com/rks_musings/2013/04/big-data-review.htmlGoogle predicted the spread of flu in real time - after analyzing two datasets, a.) 50 million most common terms that Americans type, b.) data on the spread of seasonal flu from public health agency- tested a mammoth of 450 million different mathematical models to test the search terms, comparing their predictions against the actual flu cases- model was tested when H1N1 crisis struck in 2009 and gave more meaningful and valuable real time information than any public health official system (Big Data, Viktor Mayer-Schonberger and Kenneth Cukier, 2013)
9. Better Algorithms Beat More Data — And Here’s Whyhttp://allthingsd.com/20121128/better-algorithms-beat-more-data-and-heres-why/Big Data Cannot Replace Human Judgmenthttp://www.matchcite.com/blog/blog/2012/july/big-data-cannot-replace-human-judgment.aspx**Comments about the articles
10. Better Algorithms Beat More Data — And Here’s Whyhttp://allthingsd.com/20121128/better-algorithms-beat-more-data-and-heres-why/Big Data Cannot Replace Human Judgmenthttp://www.matchcite.com/blog/blog/2012/july/big-data-cannot-replace-human-judgment.aspx**Comments about the articles
11. Better Algorithms Beat More Data — And Here’s Whyhttp://allthingsd.com/20121128/better-algorithms-beat-more-data-and-heres-why/Big Data Cannot Replace Human Judgmenthttp://www.matchcite.com/blog/blog/2012/july/big-data-cannot-replace-human-judgment.aspx**Comments about the articles
12. Top and bottom part of the brain -- http://online.wsj.com/news/articles/SB10001424052702304410204579139423079198270 Top part of the brain is known for generating plansBottom part of the brain deals with current situational awarenessPerception through senses happens in the primitive part of the brain (mostly subconsciously)Machine perception allows us to transform low level sensor observations to higher level abstractions that are directly communicable to the upper part of the brain (non-subconscious)Thus, people can understand/adapt their plan quickly with abstractionsThe left brain here is generating plan of having an apple a day to make a healthy living The right part of the brain identifies an apple through senses
13. Communicating the “abstraction” of less apples at home through “Ambient processing/intelligence”The left/top part of the brain will adapt the plan to shopping for apple soon so that the overall plan of having an apple a day can be achieved
14. Smart data makes sense out of big data – it provides value from harnessing the challenges posed by volume, velocity, variety and veracity of big data, to provide actionable information and improve decision making.
15. - HUMAN CENTRIC!!
16. All the data related to human activity, existence and experiencesMore on PCS Computing: http://wiki.knoesis.org/index.php/PCS
17. Information is CREATED by human with the Machinery available – Wikipedia tool, sensors and social networksInformation is STORED in Man+Machine readable format, LODInformation is PROCESSED using the LOD and Human assisted Knowledge-basedHigher level abstraction on info is now consumed in many mechanistic ways (including GIS) to provide EXPERIENCE for humans Example of a human guided modeling and improved performancehttp://research.microsoft.com/en-us/um/people/akapoor/papers/IJCAI%202011a.pdf
18. Actionable information example:In Asthma use case we have a sensor – sensordrone which records luminosity and CO levelsA high correlation between CO level and luminosity is foundThis is an actionable information to the user interpreting it as CO in gush during day time=> Mitigating action can be “closing the window” during day
19. Also, we have weather application which performs abstraction on weather sensory observations to identify blizzard conditions (food for actions!!) :--20,000 weather stations (with ~5 sensors per station)-- Real-Time Feature Streams - live demo: http://knoesis1.wright.edu/EventStreams/ - video demo: https://skydrive.live.com/?cid=77950e284187e848&sc=photos&id=77950E284187E848%21276
20. Lets find it..
21. http://www.huffingtonpost.com/2012/10/30/hurricane-sandy-power-outage-map-infographic_n_2044411.htmlI would like to start with a motivational example here.
22. Fraustino, Julia Daisy, Brooke Liu and Yan Jin. “Social Media Use during Disasters: A Review of the Knowledge Base and Gaps,” Final Report to Human Factors/Behavioral Sciences Division, Science and Technology Directorate, U.S. Department of Homeland Security. College Park, MD: START, 2012. Disaster communication deals with disaster information disseminated to the public by governments, emergency management organizations, and disaster responders as well as disaster information created and shared by journalists and the public. Disaster communication increasingly occurs via social media in addition to more conventional communication modes such as traditional media (e.g., newspaper, TV, radio) and word-of-mouth (e.g., phone call, face-to-face, group). Timely, interactive communication and user-generated content are hallmarks of social media, which include a diverse array of web- and mobile-based tools Disaster communication deals with (1) disaster information disseminated to the public by governments, emergency management organizations, and disaster responders often via traditional and social media; as well as (2) disaster information created and shared by journalists and affected members of the public often through word-of-mouth communication and social media. For information seeking. Disasters often breed high levels of uncertainty among the public (Mitroff, 2004), which prompts them to engage in heightened information seeking, (Boyle, Schmierbach, Armstrong, & McLeod, 2004; Procopio & Procopio, 2007). As expected, information seeking is a primary driver of social media use during routine times and during disasters (Liu et al., in press; PEW Internet, 2011). For timely information. Social media provide real-time disaster information, which no other media can provide (Kavanaugh et al., 2011; Kodrich & Laituri, 2011). Social media can become the primary source of time-sensitive disaster information, especially when official sources provide information too slowly or are unavailable (Spiro et al., 2012). For example, during the 2007 California wildfires, the public turned to social media because they thought journalists and public officials were too slow to provide relevant information about their communities (Sutton, Palen, & Shklovski, 2008). Time-sensitive information provided by social media during disasters is also useful for officials. For example, in an analysis of more than 500 million tweets, Culotta (2010) found Twitter data forecasted future influenza rates with high accuracy during the 2009 pandemic, obtaining a 95% correlation with national health statistics. Notably, the national statistics came from hospital survey reports, which typically had a lag time of one to two weeks for influenza reporting. For unique information. One of the primary reasons the public uses social media during disaster is to obtain unique information (Caplan, Perse, & Gennaria, 2007). Applied to a disaster setting, which is inherently unpredictable and evolving, it follows that individuals turn to whatever source will provide the newest details. Oftentimes, individuals experiencing the event first-hand are on the scene of the disaster and can provide updates more quickly than traditional news sources and disaster response organization. For instance, in the Mumbai terrorist attacks that included multiple coordinated shootings and bombings across two days, laypersons were first to break the news on Twitter (Merrifield & Palenchar, 2012). Research participants report using social media to satisfy their need to have the latest information available during disasters and for information gathering and sharing during disasters (Palen, Starbird, Vieweg, & Hughes, 2010; Vieweg, Hughes, Starbird, & Palen, 2010). For unfiltered information. To obtain crisis information, individuals often communicate with one another via social media rather than seeking a traditional news source or organizational website (Stephens & Malone, 2009). The public check in with social media not only to obtain up-to-date, timely information unavailable elsewhere, but also because they appreciate that information may be unfiltered by traditional media, organizations, or politicians (Liu et al., in press).  To determine disaster magnitude. The public uses social media to stay apprised of the extent of a disaster (Liu et al., in press). They may turn to governmental or organizational sources for this information, but research has shown that if the public do not receive the information they desire when they desire it, they, along with others, will fill in the blanks (Stephens & Malone, 2009), which can create rumors and misinformation. On the flipside, when the public believed that officials were not disseminating enough information regarding the size and trajectory of the 2007 California wildfires, they took matters into their own hands, using social media to track fire locations in real-time and notify residents who were potentially in danger (Sutton, Palen, & Shklovski, 2008).  To check in with family and friends. While Americans predominately use social media to connect with family and friends (PEW Internet, 2011), during disasters those connections may shift. For those with family or friends directly involved with the disaster, social media can provide a way to ensure safety, offer support, and receive timely status updates (Procopio & Procopio, 2007; Stephens & Malone, 2009). In a survey of 1,058 Americans, the American Red Cross (2010) found that nearly half of their respondents would use social media to let loved ones know they are safe during disasters. After the 2011 earthquake and tsunami in Japan, the public turned to Twitter, Facebook, Skype, and local Japanese social networks to keep in touch with loved ones while mobile networks were down (Gao, Barbier, & Goolsby, 2011). Researchers also note that disasters may enhance feelings of affection toward family members, and indeed survey participants reported expressing more positive emotions toward their loved ones than usual as a result of the September 11 terrorist attacks, even if they were not directly impacted by the disaster (Fredrickson et al., 2003). Finally, disasters can motivate the public to reconnect with family and friends via social media (Procopio & Procopio, 2009; Semaan & Mark, 2012).  To self-mobilize. During disasters, the public may use social media to organize emergency relief and ongoing assistance efforts from both near and afar. In fact, one research group dubbed those who surge to the forefront of digital and in-person disaster relief efforts as “voluntweeters” (Starbird & Palen, 2011). Other research documents the role of Facebook and Twitter in disaster relief fundraising (Horrigan & Morris, 2005; PEJ, 2010). Research also reveals how social media can help identify and respond to urgent needs after disasters. For example, just two hours after the 2010 Haitian earthquake Tufts University volunteers created Ushahidi-Haiti, a crisis map where disaster survivors and volunteers could send incident reports via text messages and tweets. In less than two weeks, 2,500 incident reports were sent to the map (Gao, Barbier, & Gollsby, 2011).  To maintain a sense of community. During disasters the media in general and social media in particular may provide a unique gratification: sense of community. That is, as the public logs in online to share their feelings and thoughts, they assist each other in creating a sense of security and community, even when scattered across a vast geographical area (Lev-On, 2011; Procopio & Procopio, 2007). As Reynolds and Seeger (2012) observed, social media create communities during disasters that may be temporary or may continue well into the future.  To seek emotional support and healing. Finally, disasters are often inherently tragic, prompting individuals to seek not only information but also human contact, conversation, and emotional care (Sutton et al., 2008). Social media are positioned to facilitate emotional support, allowing individuals to foster virtual communities and relationships, share information and feelings, and even demand resolution (Choi & Lin, 2009; Stephens & Malone, 2009). Indeed, social media in general and blogs in particular are instrumental for providing emotional support during and after disasters (Macias, Hilyard, & Freimuth, 2009; PEJ New Media Index, 2011). Additionally, social media in general and Twitter in particular can aid healing, as research finds during both natural disasters, such as Hurricane Katrina (Procopio & Procopio, 2007), and man-made disasters, such as the July 2011 attacks in Oslo, Norway (Perng et al., 2012).
23. http://www.buzzfeed.com/annanorth/how-social-media-is-aiding-the-hurricane-sandy-rec -- Facebook help during Hurricane Sandyhttp://blog.twitter.com/2012/10/hurricane-sandy-resources-on-twitter.html – Twitter page for Hurricane Sandyhttp://www.treehugger.com/culture/12-ways-help-hurricane-sandy-relief-efforts.htmlCategorization of severity based on weather conditions. Actionable information is contextually dependent.
24. http://news.cnet.com/8301-1023_3-57541566-93/report-twitter-hits-half-a-billion-tweets-a-day/http://semiocast.com/en/publications/2012_07_30_Twitter_reaches_half_a_billion_accounts_140m_in_the_USLet me consider one small example of how social data (in turn data) can help people during disasters. Data becomes smart data if it takes recipient into account - context.Sensor data for emergency responders. Who in the population needs immediate attention (1) Location (2) Severity (3) Health Condition Need for abstraction. – Semantic Perception needs abstraction. 90 + Heart Problem  Don’t run out23  Run out
25. http://news.cnet.com/8301-1023_3-57541566-93/report-twitter-hits-half-a-billion-tweets-a-day/http://semiocast.com/en/publications/2012_07_30_Twitter_reaches_half_a_billion_accounts_140m_in_the_UShttp://www.internews.org/sites/default/files/resources/InternewsEurope_Report_Japan_Connecting%20the%20last%20mile%20Japan_2013.pdfLet me consider one small example of how social data (inturn data) can help people during disasters. Data becomes smart data if it takes recipient into account and changes contact accordingly.Sensor data for emergency responders. Who in the population needs immidiate attention (1) Location (2) Severity (3) Health Condition Need for abstraction. – Semantic Perception needs abstraction. 90 + Heart Problem  Don’t run out23  Run out
26. http://news.cnet.com/8301-1023_3-57541566-93/report-twitter-hits-half-a-billion-tweets-a-day/http://semiocast.com/en/publications/2012_07_30_Twitter_reaches_half_a_billion_accounts_140m_in_the_USLet me consider one small example of how social data (inturn data) can help people during disasters. Data becomes smart data if it takes recipient into account and changes contxt accordingly.Sensor data for emergency responders. Who in the population needs immediate attention (1) Location (2) Severity (3) Health Condition Need for abstraction. – Semantic Perception needs abstraction. 90 + Heart Problem  Don’t run out23  Run out
27. http://www.buzzfeed.com/jackstuef/the-man-behind-comfortablysmug-hurricane-sandysDuring the storm last night, user @comfortablysmug was the source of a load of frightening but false information about conditions in New York City that spread wildly on Twitter and onto news broadcasts before Con Ed, the MTA, and Wall Street sources had to take time out of the crisis situation to refute them.
28. Although we face challenges like these with data everytime. The most important thing is what you aim to do with the data. I mean what value do you intend to provide from the data
29. http://www.wired.com/insights/2013/04/big-data-fast-data-smart-data/
30. http://www.wired.com/insights/2013/04/big-data-fast-data-smart-data/
31. -- Contextual Questioning – Potential Information needed from Humans
32. "2600 BC – Imhotep wrote texts on ancient Egyptian medicine describing diagnosis and treatment of 200 diseases in 3rd dynasty Egypt.”Sir William Osler, 1st Baronet, was a Canadian physician and one of the four founding professors of Johns Hopkins Hospital. He was called the father of modern medicine. Sir William Osler called Imhotep as the true father of medicine.Observations related to human body was quite limited Initially, doctors communicated with patients asking for their symptoms (subjective)Laennec’s [Rene TheophileHyacintheLaënnec, a French Physician] stethoscope was the fist peek into the observations of human body (objective)Now, there are petabytes of data being generated for observations of human body
33. Larry Smarr is a professor at the University of California, San DiegoAnd he was diagnosed with crohn's diseaseWhat’s interesting about this case is that Larry diagnosed himselfHe is a pioneer in the area of Quantified-Self, which uses sensors to monitor physiological symptomsThrough this process he discovered inflammation, which led him to discovery of Crones DiseaseThis type of self-tracking is becoming more and more common
34. - With this ability,many problems could be solved- For example: we could help solve health problems (before they become serious health problems) through monitoring symptoms and real-time sense making, acting as an early warning system to detect problematic health conditions
35. ADHF – Acute Decompensated Heart Failure
36. 1)www.pollen.com(For pollen levels)2)http://www.airnow.gov/(For air quality levels)3)http://www.weatherforyou.com/(For temperature and humidity)
37. Data overload in the context of asthma
38. “
39. AmitSheth, Pramod Anantharam, Cory Henson, 'Physical-Cyber-Social Computing: An Early 21st Century Approach,' IEEE Intelligent Systems, vol. 28, no. 1, pp. 78-82, Jan.-Feb., 2013.
40. Research on Asthma has three phases Data collection: what signals to collect?Analysis: what analysis to be done?Actionable information: what action to recommend?In the next slide, we take a peek into the analysis that we do for Asthma
41. What is the current state of a person/patient? => Summarizing all the observations (sensor and personal) into a single score indicating health of a personInstead of presenting all the raw data (often to much e.g., Asthma application we have developed collects CO, temperature, and humidity every 10 seconds resulting in 8,640 observations/day) which may not be comprehensible to the patient, we empower them by providing actionable summaries.
42. What is the likely state of the person in future? => Given the current state and the changing environmental conditions, estimate the state of the person by summarizing it into a number which is actionable. For example, vulnerability score for a person with Asthma is computed with environmental factors (pollen, air quality, external temperature and humidity) and current state of the patient. Intuitively, a person with well controlled asthma should have a lower vulnerability score than a person with poorly controlled asthma both being in a poor environmental state.
43. In the absence of declarative knowledge in a domain, we resort to statistical approaches to glean insights from dataEven if there is declarative knowledge of a domain, it may have to be personalizedThe CO level may be related to the luminosity as observed by the sensordrone – as it gets brighter the CO level also increases => high CO level in daytime If such an insight is provided to a person, the interpretation can be:Some activity inside the house leads to high CO levelsOutside activity leads to high CO levels inside the houseSince the person knows that he/she is absent in the house during mornings, it has to be something from outside.- Person narrows down to a possible opened window at home (forgot to close more often)
44. There are two components in making sense of Health Signals:Health signal extraction – processing, aggregating, and abstracting from raw sensor/textual data to create human intelligible abstractionsHealth signal understanding – derive (1) connections between abstractions and (2) Action recommendation:ContinueContact nurseContact doctor
45. Only score based structure extraction is presented here. Other popular structure extraction techniques include constraint based approaches which finds independences between random variables X1, …, XnI-Map => different structures result in the same loglikelihood score. Thus recovering the original structure of the graph generating data using data alone is considered impossible! We go the the rescue of declarative knowledge to: (1) choose promising structures and (2) to break ties when two structure results in the same score
46. Massive amount of data is collected by sensors and mobile devices yet patients and doctors care about “actionable” information.This data has all the four Vs of big data and we used knowledge enabled techniques to transform it into valueIn the context of PD, we analyzed massive amount of sensor data collected by sensors on a smartphones to understand detection and characterization of PD severity.
47. Main idea: Prior knowledge of PD was used to facilitate its detection from massive sensor data by reducing the search spaceDetails:Declarative knowledge of PD includes PD severity and their symptoms as shown in the logical rule aboveEach PD severity level is a conjunction of a set of PD symptomsEach symptom was mapped to its manifestation in sensor observationsThe availability of declarative knowledge significantly improved the analytics by aiding feature selection processThe graphs above contrasts the physical movements and voice of two control group members and two PD patients
48. sense making based on human cognitive models
49. perception cycle contains two primary phasesexplanationtranslating low-level signals into high-level abstractions inference to the best explanationdiscriminationfocusing attention on those properties that will help distinguish between multiple possible explanationsused to intelligently task sensors and collect additional observations (rather than brute force approach of blindly collecting all observations)
50. perception cycle contains two primary phasesexplanationtranslating low-level signals into high-level abstractions inference to the best explanationdiscriminationfocusing attention on those properties that will help distinguish between multiple possible explanationsused to intelligently task sensors and collect additional observations (rather than brute force approach of blindly collecting all observations)
51. A single-feature (disease) assumption means that all the observed properties (symptoms) must be explained by a single feature.i.e., this framework is not expressive enough to model comorbidity where there may be more than one feature (disease) co-existing For example, if there are two diseases causing disjoint symptoms, and all the symptoms of both the diseases are observed, then this framework will not be able to find the coverage and returns no diseases.Parsimony criteria is single feature assumption to choose from among multiple explanationsNot true: if multiple disease account for single property…Rewrite with more relaxed parcimony criteria (complex, cannot be modeled in OWL)Make KB more intelligent: create an individual that represents the two disease which together explain a symptom
52. perception cycle contains two primary phasesexplanationtranslating low-level signals into high-level abstractions inference to the best explanationdiscriminationfocusing attention on those properties that will help distinguish between multiple possible explanationsused to intelligently task sensors and collect additional observations (rather than brute force approach of blindly collecting all observations)
53. So check galvanic skin response sensor
54. Intelligence distributed at the edge of the networkRequires resource-constrained devices (mobile phones, gateway notes, etc.) to be able to utilize SW technologies
55. Intelligence distributed at the edge of the networkRequires resource-constrained devices (mobile phones, gateway notes, etc.) to be able to utilize SW technologiesHenson et al. 'An Efficient Bit Vector Approach to Semantics-based Machine Perception in Resource-Constrained Devices, ISWC 2012.
56. compute machine perception inferences -- i.e., explanation and discrimination -- of high-complexity on a resource-constrained devices in milisecondsDifference between the other systems and what this system provides
57. Intelligence at the age. Shipping computation and domain models to the edge (Distributed)
58. http://www.adrants.com/2013/09/what-is-social-good-and-how-can-brands.php
59. http://www.guardian.co.uk/news/datablog/2012/oct/31/twitter-sandy-floodinghttp://www.huffingtonpost.com/2012/11/02/twitter-hurricane-sandy_n_2066281.htmlhttp://mashable.com/2012/10/31/hurricane-sandy-facebook/We in our lab have quite a bit of Social Data Research going on. So I would like to focus on the use of social networks during these disasters/crisis.Twitter and Facebook are massively used during disasters. During Hurricane Sandy there were …Not only this a major outbreak of tweets were during Japan earthquake which crossed more that 2000 tweets/sec.So why do people intend to use social networks to this extent during disasters.
60. http://www.flickr.com/photos/twitteroffice/5897088517/sizes/o/in/photostream/http://bayarea.sbnation.com/49ers/2013/2/3/3947738/super-bowl-prop-bets-2013-twitterhttp://bayarea.sbnation.com/49ers/2013/2/3/3947738/super-bowl-prop-bets-2013-twitterhttp://expandedramblings.com/index.php/march-2013-by-the-numbers-a-few-amazing-twitter-stats/
61. Much of the early work in Big data is being done with focusing on uni-directional among XYZ.
62. http://semanticweb.com/picking-the-president-twindex-twitris-track-social-media-electorate_b31249http://semanticweb.com/election-2012-the-semantic-recap_b33278
63. http://knoesis.wright.edu/library/resource.php?id=1787
64. Categorization of severity based on weather conditions. Actionable information is contextually dependent.
65. - 1 (+half) minuteAlright, so let’s motivate by this situation during emergency - Various actors: resource seekers, responder teams, resource providers at remote siteAnd - each of these actor groups have questions --- - needs - providers - responders: wondering!Here we have social network to connect these actors and bridge the gap for communication platformBut it’s potential use is yet to be realized for effective help
66. Talk about what kind of smart data we provide that helps the actions of crisis response coordination.
67. Source: Purohit et. al 2013 (https://docs.google.com/a/knoesis.org/document/d/1aBJ2egHICUwaWxR8jOoTIUfEYj1QAnUt0q7haIKoYGY/edit# , http://www.knoesis.org/library/resource.php?id=1865)
68. http://twitris.knoesis.org/oklahomatornado
69. (It is real-time widget for monitoring of needs, so will not be active after the event has passed) http://twitris.knoesis.org/oklahomatornado
70. Highly rich interface for response team
71. Definition of the event US Elections and some changes/subevents --- Primaries --- Debates -- People/Places/Organizations involved in the eventArab Spring -- Subevents during those -- Egypt protests
72. Explain about continuous semantics
73. Pucher, J., Korattyswaroopam, N., & Ittyerah, N. (2004). The crisis of public transport in India: Overwhelming needs but limited resources. Journal of Public Transportation, 7(4), 1-30.
74. Twitter as a source of real-time informationThere are over 200 million users generating 500 million tweets / dayTwitter as a source of events in a cityCitizens use twitter to express their concerns of city infrastructure that impacts their life
75. The red-tweets are the tweets that are related to city infrastructure e.g., trafficThere are two steps in converting raw tweets from a city to city related events:City event annotation: sequence labeling technique to spot location and event termsCity event extraction: aggregating all the location + event terms to derive eventsTo do this aggregation, we follow some principles that characterize city events
76. CRF assigns a tag to each tokenGlobal normalization is the argmax termRHS is just a regression based implementation of linear chain (potentials defined only over adjacent tags) CRFLingPipe implementation of CRF is used in our experiments
77. There principles characterize city events
78. localized event detection strategy, city a composition of smaller geographical unitsWe call these geographical units as grids Geohash provides us a way of compartmentalizing a city into uniquely addressable gridsDistance computed using the formula:dlon = lon2 - lon1 dlat = lat2 - lat1 a = (sin(dlat/2))^2 + cos(lat1) * cos(lat2) * (sin(dlon/2))^2 c = 2 * atan2( sqrt(a), sqrt(1-a) ) d = R * c (where R is the radius of the Earth)Found the box for the tweet!37.7545166015625, -122.42065429687537.7545166015625, -122.4096679687537.7490234375, -122.4096679687537.7490234375, -122.420654296875
79. These algorithms take the annotated tweets as input and then emit events with their metadata
80. Now that we have presented the (1) event extraction and (2) event aggregation algorithms, how well are we doing?We evaluate both the componentsThe ground truth are the events reported on 511.orgWe compare the events we extract from tweets with the 511.org events
81. We evaluate the extracted events based on there orthogonal metrics.We compare (1) events extracted from tweets using our algorithms and (2) 511.org events Complementary events – the events from (1) and (2) may complement each other i.e., one providing a different view from the otherCorroborative events – the events from (1) and (2) may support each other i.e., redundant eventsTimeliness – the events were reported on (1) before it was reported on (2)
82. Next few slides give examples of the evaluation metric
83. The record of 511.org may have its own timestamp which may be before tweets
84. More at: http://wiki.knoesis.org/index.php/PCSAnd http://knoesis.org/projects/ssw/