Presentation is loading. Please wait.

Presentation is loading. Please wait.

Question Identification on Twitter Baichuan Li, Xiance Si, Michael R. Lyu, Irwin King, and Edward Y. Chang 10/9/20151.

Similar presentations


Presentation on theme: "Question Identification on Twitter Baichuan Li, Xiance Si, Michael R. Lyu, Irwin King, and Edward Y. Chang 10/9/20151."— Presentation transcript:

1 Question Identification on Twitter Baichuan Li, Xiance Si, Michael R. Lyu, Irwin King, and Edward Y. Chang 10/9/20151

2 Agenda Background Two-phase Classification Experiments Conclusion 10/9/20152

3 Background 10/9/20153

4 4

5 Two Challenges 140 characters Special features 10/9/20155

6 Two-phase Classification Interrogative Tweet Detection – Tweets which contain question sentences Qweet Extraction – Interrogative tweets which require some information or help and thus need to be answered Interrogative Tweet Detection Tweets Qweet Extraction Qweets Interrogative Tweets 10/9/20156

7 Interrogative Tweet Detection Rule-based Approach – Question marks – 5W1H words and Refined 5W1H words – Heuristic Rules (Efron and Winget, 2010) Learning-based Approach – Frequent question patterns mining (Pei et al., 2001) + One-class SVM (Schölkopf et al., 2001) – Over 850,000 QA pairs in community question answering (CQA) portals were used 10/9/20157

8 Qweet Extraction Types of Interrogative Tweets 10/9/20158

9 Qweet Extraction Types of Interrogative Tweets 10/9/20159

10 Qweet Extraction Types of Interrogative Tweets 10/9/201510

11 Qweet Extraction Feature Extraction 10/9/201511

12 Experiments Data Set 10/9/201512

13 Results: Interrogative Tweet Detection Heuristics – H1: Must appear at the beginning of one sentence – H2: Add auxiliary words to the original 5W1H words “what” -> “what is” and “what are” 10/9/201513

14 Results: Qweet Extraction Context features are of great importance in distinguishing qweets from non-qweets Tweet-specific features also help in qweet identification 10/9/201514

15 Conclusion First Attempt in discovering questions from tweets automatically Two-phase classification – Interrogative Tweet Detection – Qweet Extraction Limitations and future work – Tweets containing rhetorical questions and complicated self-ask-self-answer sentences – Real-time clustering (Ahmed et al., 2011) – Question analysis and classification 10/9/201515

16 Thank You! Q&A 10/9/201516


Download ppt "Question Identification on Twitter Baichuan Li, Xiance Si, Michael R. Lyu, Irwin King, and Edward Y. Chang 10/9/20151."

Similar presentations


Ads by Google