1. Tweet Emotion Mapping: Understanding US Emotions in Time and Space
Romita Banerjee*, Karima Elgarroussi*, Sujing Wang!, Yongli Zhang*, and Christoph F. Eick*
*Department of Computer Science, University of Houston, Houston, Texas
!Department of Computer Science, Lamar University, Beaumont, Texas
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

2. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

3. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

4. Source: https://github.com/pubnub/tweet-emotion
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

5.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

6. UN World Happiness Report
UN World Happiness Report ranks 156 countries by their happiness levels.
Citizens are asked to fill out questionnaires that inquire six key variables that are believed to play a key role for measuring happiness: well-being: income, healthy life expectancy, social support, freedom, trust, and generosity.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

7. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

8. Importance of Emotion Analysis
Emotions are an integral part of human nature and it influences how we think or behave.
Emotions can be expressed by facial expressions, actions or electronically like using emojis, likes and dislikes, and tweets.
Assessing the emotional well being of population is one of the core aspect of many fields like medicine, psychology, etc.
It can help in evaluating how a population feels after a event.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

9. Research Focus
Our paper presents a new framework for tweet emotion mapping and emotion change analysis.
It focuses on the development of data analysis framework that measures and summarizes the emotional well-being of population as well as how it evolves over time.
Ultimately, we plan to provide storytelling capabilities that tells the story of emotion evolution of a region.
As Twitter is one of the most popular social media platform, we are using tweets as our primary knowledge source.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

10. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

11. Goals of the Research Project
Identify the regions with highly positive and highly negative emotions.
To monitor the change of emotions of the previously identified regions over time.
To provide storytelling capabilities that demonstrates the emotional evolution of a region using animations and graphics.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

12. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

13. Emotion Mapping and Change Analysis System
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

14. System Architecture
The entire system can be broadly divided into three parts:
Frontend: Spatial Clustering Component
Backend: Change Analysis Framework
Data Storytelling
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

15. Change Analysis Framework
Monitors change by comparing sets of polygons which represent spatial clusters for a particular time window/batch.
Each spatial cluster belongs to a single batch.
Change of emotions is then analyzed with respect to a set of unary and binary change predicates that are evaluated with respect to the set of spatial clusters.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

16. Change Analysis Framework
Input: Sequence of Sets of Spatial Clusters and their statistical summaries.
Output: An emotion change graph is obtained whose nodes are spatial clusters and whose edges capture different types of temporal relationships between spatial clusters.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

17. Importance of Data Storytelling
Good stories are always engaging.
Successful companies have long been crafting ‘authentic’ stories to sell their products, entertain, and build brands.
Data storytelling should make a point; to offer direction or to sell an audience on a course of action.
As data gets bigger and more complex, being able to tell a compelling story becomes more important than ever.
For that reason, good storytelling is becoming an ever more vital component.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

18. Data Storytelling
Data storytelling is a combination of three key elements: data, visuals, and narrative.
When narrative is coupled with data: it helps to explain to your audience what’s happening in the data and why a particular insight is important. 
When visuals are applied to data, they can enlighten the audience to insights that they wouldn’t see without charts or graphs.
When narrative and visuals are merged together, they can engage or even entertain an audience. 
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

19. Source: https://www. forbes
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

20. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

21. AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

22. AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

23. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

24. Emotion-Weighted Density Estimation
Input: Given a set of objects O (<location>,<emotion-value>).
Output: A 2-dimensional continuous density function O(v) that takes negative and positive values at location (v).
Traditional density estimation techniques consider only the spatial dimension of data points ignoring non-spatial information.
Our approach differs from the traditional density estimation by additionally considering a non-spatial, continuous variable of interest in its influence function.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

25. Emotion-Weighted Density Estimation
The influence of object “o” on “v” is measured as the product of emotional score and a Gaussian kernel density function:
𝑓 𝐼𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 (𝑣,𝑜) = 𝐸(𝑜)∗ 𝑒 − 𝑑(𝑣,𝑜) 2 2 𝜎 2
The accumulated influence of all data objects o∈O on a “query” point v is used to define a density function 𝜓 𝑂 (v) as follows:
𝜓 𝑂 𝑣 = 𝑜𝑒𝑂 𝑓 𝐼𝑛𝑓𝑙𝑢𝑒𝑛𝑐𝑒 𝑣,𝑜
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

26. Emotion-Weighted Density Estimation
The influence of tweets far away from the query point is less than the influence tweets nearby.
Very positive or negative emotion tweets, having very high/low emotion values, have a stronger influence than tweets whose emotion values close to 0.
If O(v)0, there are two possible scenarios:
There is a balance of negative and positive emotions in query point.
The query point is in a very sparse area with respect to tweets and therefore has a low density.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

27. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

28. Overview
Input: location and time were and when the tweets were posted and an emotion assessment score in [-1, +1], with +1 denoting a very positive emotion and -1 a very negative emotion.
Output: Spatial cluster polygons such that all values of the density inside the polygon are either above 1 (positive emotion cluster) or below 2 (negative emotion cluster).
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

29. Preprocessing Data twarc VADER
Date, location and the text of each tweet
Tweet Ids
Emotional Score
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

30. Obtaining Spatial Clusters
Divide the batches of data into a number of grids. The batches can be daily, weekly or monthly.
Using the spatial density function, we calculate the density values for all the grid intersection points.
Using the product of the density values with the emotional scores for all intersection points as an input, we compute the contour lines for the density thresholds.
Using the obtained contour lines, create spatial clusters which in our approach are polygons.
Challenges of creating spatial clusters include:
creating closed contour lines for open contour lines that lie on the boundary of the observation area
distinguishing between holes and spatial clusters
removing small and insignificant clusters.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

31. AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

32. Positive Emotion Clusters in the New York City/Long Island Area on June 1 and on 2, 2014
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

33. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

34. Conclusion
We introduced a novel spatio-temporal framework creates positive and negative emotion spatial clusters and summarizes the change of emotions in time.
The framework creates such clusters by using contouring algorithm that operates on an emotion-weighted density function.
As per our knowledge our density estimation approach is unique in the sense that it considers a non-spatial, continuous variable of interest in the influence function.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

35. Outline Related Work Introduction Goals System Architecture
Spatial Clustering Component
Emotion Weighted Density Estimation
Spatial Clustering Approach
Conclusion
Future Work
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

36. Future Work
Conducting a thorough experimental evaluation of the Frontend system.
Enhancement of the implementation of the Spatial Clustering Component including:
Tools to preselect critical input parameters such as bandwidth, grid sizes and density thresholds
Make the implementation more scalable to be able to deal with a large number of tweets
Addressing the challenges faced during spatial cluster creation
Design and implementation of the Data Storytelling Component and the implementation of the Change Analysis Component.
Contributing to the next United Nations’ Happiness Report.
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston

37. Thank You!
AIKE 2018
Data Analysis and Intelligent Systems (DAIS) Lab at University of Houston