1 From Sentiment to Emotion Analysis in Social Networks Jie Tang Department of Computer Science and Technology Tsinghua University, China

2 From Info. Space to Social Space Info. Space Social Space Interaction 1.Social Tie & Group 2.Social Influence 3.Collective Intelligence Revolutionary changes…

3 Revolutionary Changes Social Networks Embedding social in search: Google plus FB graph search Bing’s influence Search Human Computation: CAPTCHA + OCR MOOC Duolingo (Machine Translation) Education The Web knows you than yourself: Contextual computing Big data marketing O2O More …...

4 Let us start with sentiment analysis…

5 “Love Obama” I love Obama Obama is great! Obama is fantastic I hate Obama, the worst president ever He cannot be the next president! No Obama in 2012! Positive Negative

6 Homophily Homophily —“birds of a feather flock together” –A user in the social network tends to be similar to their connected neighbors. Originated from different mechanisms –Influence Indicates people tend to follow the behaviors of their friends –Selection Indicates people tend to create relationships with other people who are already similar to them –Confounding variables Other unknown variables exist, which may cause friends to behave similarly with one another.

7 Twitter Data Twitter –1,414,340 users and 480,435,500 tweets –274,644,047 t-follow edges and edges [1] Chenhao Tan, Lillian Lee, Jie Tang, Long Jiang, Ming Zhou, and Ping Li. User-level sentiment analysis incorporating social networks. In KDD’11, pages 1397–1405, 2011.

8 Influence Shared sentiment conditioned on type of connection.

9 Selection Connectedness conditioned on labels

10 One question: what drives users’ sentiments?

11 Sentiment vs. Emotion Charles Darwin: –Emotion serves as a purpose for humans in aiding their survival during the evolution. [1] Emotion is the driving force of user’s sentiments… Emotion stimulates the mind 3000 times quicker than rational thought! [1] Charles Darwin. The Expression of Emotions in Man and Animals. John Murray, 1872.

12 Happy System Can we predict users’ emotion?

13 Observations (cont.) Location correlation (Red-happy) Activity correlation Karaoke ? ? ? ? ? GYM Dorm The Old Summer Palace Classroom

14 Observations (a) Social correlation (a) Implicit groups by emotions (c) Calling (SMS) correlation

15 Observations (cont.) Temporal correlation

16 Methodologies

17 MoodCast: Dynamic Continuous Factor Graph Model Our solution 1. We directly define continuous feature function; 2. Use Metropolis-Hasting algorithm to learn the factor graph model. [1] Jie Tang, Yuan Zhang, Jimeng Sun, Jinghai Rao, Wenjing Yu, Yiran Chen, and ACM Fong. Quantitative Study of Individual Emotional States in Social Networks. IEEE TAC, 2012, Volume 3, Issue 2, Pages

18 Problem Formulation G t =(V, E t, X t, Y t ) Attributes: - Location: Lab - Activity: Working Emotion: Sad Learning Task: Time t Time t-1, t-2…

19 Dynamic Continuous Factor Graph Model Time t’ Time t : Binary function

20 Learning with Factor Graphs Temporal Social Attribute y3y3 y4y4 y5y5 y2y2 y1y1 y'3y'3

21 MH-based Learning algorithm [1] Yuan Zhang, Jie Tang, Jimeng Sun, Yiran Chen, and Jinghai Rao. MoodCast: Emotion Prediction via Dynamic Continuous Factor Graph Model. In ICDM’10. pp

22 Still Challenges Q1: Are there any other social factor that may affect the prediction results? Q2: How to scale up the model to large networks?

23 Q1: Conformity Influence I love Obama Obama is great! Obama is fantastic Positive Negative 2. Individual 3. Group conformity 1. Peer influence [1] Jie Tang, Sen Wu, and Jimeng Sun. Confluence: Conformity Influence in Large Social Networks. In KDD’13, 2013.

24 Conformity Factors Individual conformity Peer conformity Group conformity All actions by user v A specific action performed by user v at time t

25 Q2: Distributed Learning Slave Compute local gradient via random sampling Slave Compute local gradient via random sampling Master Global update Master Global update Graph Partition by Metis Master-Slave Computing Inevitable loss of correlation factors!

26 Random Factor Graphs Master: Optimize with Gradient Descent Slave: Distributedly compute Gradient via LBP Master-Slave Computing Gradients Parameters

27 Model Inference Calculate marginal probability in each subgraph Aggregate the marginal probability and normalize

28 Theoretical Analysis Θ * : Optional parameter of the complete graph Θ: Optional parameter of the subgraphs P s,j : True marginal distributions on the complete graph G * s,j : True marginal distributions on subgraphs Let E s,j = log G * s,j – log P s,j ， we have:

29 Experiments

30 Results for Sentiment Analysis Twitter –1,414,340 users and 480,435,500 tweets –274,644,047 t-follow edges and edges Baseline –SVM Vote Measures –Accuracy and Macro F1

31 Performance

32 Results of Different Learning Algorithms

33

34 Data Set Baseline –SVM –SVM with network features –Naïve Bayes –Naïve Bayes with network features Evaluation Measure: Precision, Recall, F1-Measure #UsersAvg. Links#LabelsOther MSN303.29,869>36,000hr LiveJournal469, ,665,166 Results for Emotion Analysis

35 Performance Result

36 Factor Contributions All factors are important for predicting user emotions Mobile

37 Online Applications: Emotion Analysis on Flickr

38

39  Framework: Images -Aesthetic Effects -Emotions  Model: Factor Graphs for images in Social Networks [1] Jia Jia, Sen Wu, Xiaohui Wang, Peiyun Hu, Lianhong Cai, and Jie Tang. Can We Understand van Gogh’s Mood? Learning to Infer Affects from Images in Social Networks. In ACM Multimedia. pp [2 ] Grand Challenge 2nd Prize Award

40 App1: Emotion Distribution on Flickr Before Thanksgiving 2011 VS During Thanksgiving holiday Happy, Cheerful, and Peaceful 100,000 Images from Flickr

41 App2: Modify Images with Emotional Words Happy Natural Clear Original Image Summer? Autumn? Winter? More than 180 different effects

42 Summary

43 Summary Social networks bring revolutionary changes to the Web and unprecedented opportunities for us Emotion stimulates minds 3000 times faster than rational thoughts! Embedding social theories into sentiment/emotion analysis can benefit many applications

44 Related Publications Chenhao Tan, Lillian Lee, Jie Tang, Long Jiang, Ming Zhou, and Ping Li. User-level sentiment analysis incorporating social networks. In KDD’11, pages 1397–1405, Jie Tang, Yuan Zhang, Jimeng Sun, Jinghai Rao, Wenjing Yu, Yiran Chen, and ACM Fong. Quantitative Study of Individual Emotional States in Social Networks. IEEE Transactions on Affective Computing (TAC), 2012, Volume 3, Issue 2, Pages (Selected as the Spotlight Paper) Yuan Zhang, Jie Tang, Jimeng Sun, Yiran Chen, and Jinghai Rao. MoodCast: Emotion Prediction via Dynamic Continuous Factor Graph Model. In ICDM’10. pp Jia Jia, Sen Wu, Xiaohui Wang, Peiyun Hu, Lianhong Cai, and Jie Tang. Can We Understand van Gogh’s Mood? Learning to Infer Affects from Images in Social Networks. In ACM MM, pages , Xiaohui Wang, Jia Jia, Peiyun Hu, Sen Wu, Lianhong Cai, and Jie Tang. Understanding the Emotional Impact of Images. (Grand Challenge) In ACM MM. pp (Grand Challenge 2nd Prize Award)

45 Thanks you ！ Collaborators: Lillian Lee, Chenhao Tan (Cornell) Ming Zhou, Long Jiang (Microsoft), Yuan Zhang (MIT) Jimeng Sun (IBM), Jinghai Rao (Nokia) Sen Wu, Jia Jia, Xiaohui Wang, Yiran Chen, Wenjing Yu (THU) Jie Tang, KEG, Tsinghua U, Download data & Codes,