Sentiment analysis under temporal shift

Jan Lukes; Anders Søgaard

Sentiment analysis under temporal shift

Abstract

Sentiment analysis models often rely ontraining data that is several years old. Inthis paper, we show that lexical featureschange polarity over time, leading to degradingperformance. This effect is particularlystrong in sparse models relyingonly on highly predictive features. Usingpredictive feature selection, we are able tosignificantly improve the accuracy of suchmodels over time.

Original language	English
Title of host publication	Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis
Publisher	Association for Computational Linguistics
Publication date	2018
Pages	65–71
Publication status	Published - 2018
Event	9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis - Brussels, Belgium Duration: 31 Oct 2018 → 31 Oct 2018

Workshop

Workshop	9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis
Country/Territory	Belgium
City	Brussels
Period	31/10/2018 → 31/10/2018

Cite this

@inproceedings{ac0f745e21b144e39685eeb219c9f15f,

title = "Sentiment analysis under temporal shift",

abstract = "Sentiment analysis models often rely ontraining data that is several years old. Inthis paper, we show that lexical featureschange polarity over time, leading to degradingperformance. This effect is particularlystrong in sparse models relyingonly on highly predictive features. Usingpredictive feature selection, we are able tosignificantly improve the accuracy of suchmodels over time.",

author = "Jan Lukes and Anders S{\o}gaard",

year = "2018",

language = "English",

pages = "65–71",

booktitle = "Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis",

publisher = "Association for Computational Linguistics",

note = "9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis ; Conference date: 31-10-2018 Through 31-10-2018",

}

TY - GEN

T1 - Sentiment analysis under temporal shift

AU - Lukes, Jan

AU - Søgaard, Anders

PY - 2018

Y1 - 2018

N2 - Sentiment analysis models often rely ontraining data that is several years old. Inthis paper, we show that lexical featureschange polarity over time, leading to degradingperformance. This effect is particularlystrong in sparse models relyingonly on highly predictive features. Usingpredictive feature selection, we are able tosignificantly improve the accuracy of suchmodels over time.

AB - Sentiment analysis models often rely ontraining data that is several years old. Inthis paper, we show that lexical featureschange polarity over time, leading to degradingperformance. This effect is particularlystrong in sparse models relyingonly on highly predictive features. Usingpredictive feature selection, we are able tosignificantly improve the accuracy of suchmodels over time.

M3 - Article in proceedings

SP - 65

EP - 71

BT - Proceedings of the 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis

PB - Association for Computational Linguistics

T2 - 9th Workshop on Computational Approaches to Subjectivity, Sentiment and Social Media Analysis

Y2 - 31 October 2018 through 31 October 2018

ER -