Category-Sensitive Question Routing in Community Question Answering Prepared and Presented by Baichuan Li
Outline Introduction & Motivation Category Sensitive QR Experiments Conclusion 20/4/2019 Paper Presentation 2/24
Introduction Community Question-Answering (CQA) Services 3/24 20/4/2019 Paper Presentation 3/24
Low participation rate Motivation Low participation rate 20/4/2019 Paper Presentation 4/24
Motivation States of tracked question in Yahoo! Answers and Baidu Zhidao within 72 hours Long lag time 20/4/2019 Paper Presentation 5/24
Question Routing Definition Framework Routing a new posted question to users who are most likely to give answers. Framework 20/4/2019 Paper Presentation 6/24
Related work Previous Methods Limitations LMs (Liu et al. 2005, Zhou et al. 2009, Li and King 2010) PLSA (Qu et al. 2009) LDA + LM (Liu et al. 2010) Limitations Based on all previous answered questions Many questions are irrelevant to the routed question 20/4/2019 Paper Presentation 7/24
Category Information Why help? Lower cost Higher precision 8/24 20/4/2019 Paper Presentation 8/24
Category-Sensitive QR Query likelihood language model BCS-QL TCS-QL Translation-based language model BCS-TB TCS-TB 20/4/2019 Paper Presentation 9/24
Query likelihood LM 20/4/2019 Paper Presentation 10/24
BCS-QL 20/4/2019 Paper Presentation 11/24
TCS-QL 20/4/2019 Paper Presentation 12/24
TCS-QL 20/4/2019 Paper Presentation 13/24
Translation-based LM 20/4/2019 Paper Presentation 14/24
BCS-TB & TCS-TB 20/4/2019 Paper Presentation 15/24
Methods Compared LDALM CBLM 20/4/2019 Paper Presentation 16/24
Dataset 20/4/2019 Paper Presentation 17/24
Dataset 20/4/2019 Paper Presentation 18/24
Evaluation Metrics SR(K) MRR Mean QR Time the average of time spent on routing one question 20/4/2019 Paper Presentation 19/24
Experimental Results 20/4/2019 Paper Presentation 20/24
Experimental Results 20/4/2019 Paper Presentation 21/24
Experimental Results 20/4/2019 Paper Presentation 22/24
Conclusion Proposed basic and transferred category- sensitive models Integrated the models with QLLM and TBLM Studied various methods’ performance comparing with previous approaches Future work Better user modeling Personalized top K selection 20/4/2019 Paper Presentation 23/24
THANKS! 20/4/2019 Paper Presentation