Roses are Red, Violets are Blue: Detection of Valid Sentiment-Target Pairs Svitlana Vakulenko Albert Weichselbraun Arno Scharl MODUL University Vienna University of Applied Sciences Chur International Conference on Web Intelligence October 13–16, 2016 in Omaha, Nebraska, USA
15
Embed
Roses are Red, Violets are Blue: Detection of Valid Sentiment-Target Pairs
This document is posted to help you gain knowledge. Please leave a comment to let me know what you think about it! Share it to your friends and learn new things together.
Transcript
Roses are Red, Violets are Blue:Detection of Valid Sentiment-Target Pairs
Svitlana Vakulenko Albert Weichselbraun Arno Scharl
MODUL University ViennaUniversity of Applied Sciences Chur
International Conference on Web IntelligenceOctober 13–16, 2016 in Omaha, Nebraska, USA
Motivation
Roses are red, violets are blue
T1 S1 T2 S2
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 2 / 15
Application: Sentiment Analysis
Fine-grained sentiment analysis in product reviews
Example
The design is outstanding, but the sound quality is poor
Brand monitoring on-line (social) media
Example
[Apple, iPad , iPhone,MacBook,TimCook, ...]The weather was disappointing during Tim Cook’s visit to India.
Stock prediction using Twitter sentiment analysis
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 3 / 15
Problem Statement
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 4 / 15
Related Work
Sequence Labeling (Joint Model)[Yang and Cardie, 2013, Deng and Wiebe, 2015]
Classification (Sentiment/Target/Relation)[Kessler et al., 2010]
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 5 / 15
Feature Engineering
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 6 / 15
Feature Groups
Sentiment/Target (ST Penn: [VBP,NN])
Lexical Path (L Penn: [TO,VB,DT]; L Dist: 3)
Dependency Path (D Penn: [TO,VB]; D Rels: [OPRD,IM,OBJ])
I like to drive the car
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 7 / 15
Deng, L. and Wiebe, J. (2015).Joint prediction for entity/event-level sentiment analysis usingprobabilistic soft logic models.In Proceedings of the Conference on Empirical Methods in NaturalLanguage Processing (EMNLP 15).
Kessler, J. S., Eckert, M., Clark, L., and Nicolov, N. (2010).The ICWSM 2010 JDPA sentiment corpus for the automotivedomain.In Proceedings of the 4th International AAAI Conference on Weblogsand Social Media Data Workshop Challenge (ICWSM-DWC 2010).
Yang, B. and Cardie, C. (2013).Joint inference for fine-grained opinion extraction.In Proceedings of the 51st Annual Meeting of the Association forComputational Linguistics (ACL 13), pages 1640–1649.
Vakulenko et al. (MODUL University Vienna) Detection of Valid Sentiment-Target Pairs WI’16 in Omaha, Nebraska, USA 15 / 15