Structured Models for Fine-to-Coarse Sentiment Analysis

Auteur-es

McDonald, Ryan; Hannan, Kerry; Neylon, Tyler; Wells, Mike; Reynar, Jeff

Nombre Auteurs

5

Titre

Structured Models for Fine-to-Coarse Sentiment Analysis

Année de publication

2007

Référence (APA)

McDonald, R., Hannan, K., Neylon, T., Wells, M., & Reynar, J. (2007). Structured Models for Fine-to-Coarse Sentiment Analysis. Proceedings of the 45th Annual Meeting of the Association of Computational Linguistics, 432‑439. https://aclanthology.org/P07-1055

Mots-clés

ND

URL

https://aclanthology.org/P07-1055

Accessibilité de l'article

Open access

Champ

Natural Language Processing

Type contenu (théorique Applicative méthodologique)

Applicative

Méthode

We describe a simple structured model that jointly learns and infers sentiment on different levels of granularity

Cas d'usage

Online customer reviews

Objectifs de l'article

In this paper we investigate a structured model for jointly classifying the sentiment of text at varying levels of granularity.
In this paper we have investigated the use of a global structured model that learns to predict sentiment on different levels of granularity for a text.

Question(s) de recherche/Hypothèses/conclusion

Research question(s) : Extracting sentiment from text is a challenging problem with applications throughout Natural Language Processing and Information Retrieval. [...] The ability to classify sentiment on multiple levels is important since different applications have different needs. [...] We call the problem of identifying the sentiment of the document and of all its subcomponents, whether at the paragraph, sentence, phrase or word level, fine-to-coarse sentiment analysis.

Hypothesis(es) : The simplest approach to fine-to-coarse sentiment analysis would be to create a separate system for each level of granularity. There are, however, obvious advantages to building a single model that classifies each level in tandem. [...]
This work focuses on models that jointly classify sentiment on multiple levels of granularity.

Conclusion(s) : We described a simple model for sentence-document analysis and showed that inference in it is tractable. Experiments show that this model obtains higher accuracy than classifiers trained in isolation as well as cascaded systems that pass information from one level to another at test time. Furthermore, extensions to the sentence-document model were discussed and it was argued that a nested hierarchical structure would be beneficial since it would allow for efficient inference algorithms.

Cadre théorique/Auteur.es

Sentiment analysis (Pang et al., 2002 ; Turney, 2002 ; Pang et Lee, 2004 ; Choi et al., 2005 ; Choi et al., 2006 ; Mao et Lebanon, 2006 ; Thomas et al., 2006)

Structured learning algorithms (Rabiner, 1989 ; Lafferty et al., 2001 ; Sutton et McCallum, 2006 ; Taskar et al., 2003 ; Tsochantaridis et al., 2004 ; McDonald et al., 2005 ; Daumé III et al., 2006 ; Taskar et al., 2004 ; Liang et al, 2006)

Structured models used for sentiment analysis (Choi et al., 2005 ; Choi et al., 2006 ; Mao et Lebanon, 2006)

Cascaded models for fine-to-coarse sentiment analysis (Pang et Lee, 2004)

Learning and/or predicting multiple outputs jointly (Miller et al., 2000 ; Roth et Yih, 2004 ; Sutton et al., 2004 ; Popescu et Etzioni, 2005)

Concepts clés

Sentiment analysis

Données collectées (type source)

To test the model we compiled a corpus of online product reviews from three domains: car seats for children, fitness equipment, and Mp3 players.

We discarded duplicate reviews, reviews with insufficient text, and spam. All reviews were labeled by online customers as having a positive or negative polarity on the document level, i.e., Y(d) = {pos, neg}. Each review was then split into sentences and every sentence annotated by a single annotator as either being positive, negative or neutral, i.e., Y(s) = {pos, neg, neu}. All sentences were annotated based on their context within the document.

Définition des émotions

No definition

Negative, neutral, positive labeling

Ampleur expérimentation (volume de comptes)

Corpus of 600 online product reviews, 554 kept

Car seats : 178 reviews, 1179 sentences
Fitness equipment : 189 reviews, 1574 sentences
Mp3 players : 187 reviews, 1163 sentences
TOTAL : 554 reviews, 3916 sentences

Technologies associées

Standard sequence classification techniques using constrained Viterbi
Structured models

Mention de l'éthique

ND

Finalité communicationnelle

One interesting application of sentence level sentiment analysis is summarizing product reviews on retail websites like Amazon.com or review aggregators like Yelp.com.

Extracting sentiment from text is a challenging problem with applications throughout Natural Language Processing and Information Retrieval.

Résumé

In this paper we investigate a structured model for jointly classifying the sentiment of text at varying levels of granularity. Inference in the model is based on standard sequence classification techniques using constrained Viterbi to ensure consistent solutions. The primary advantage of such a model is that it allows classification decisions from one level in the text to influence decisions at another. Experiments show that this method can significantly reduce classification error relative to models trained in isolation.