Papers
Topics
Authors
Recent
Gemini 2.5 Flash
Gemini 2.5 Flash
119 tokens/sec
GPT-4o
56 tokens/sec
Gemini 2.5 Pro Pro
43 tokens/sec
o3 Pro
6 tokens/sec
GPT-4.1 Pro
47 tokens/sec
DeepSeek R1 via Azure Pro
28 tokens/sec
2000 character limit reached

Crowdsourcing a Word-Emotion Association Lexicon (1308.6297v1)

Published 28 Aug 2013 in cs.CL

Abstract: Even though considerable attention has been given to the polarity of words (positive and negative) and the creation of large polarity lexicons, research in emotion analysis has had to rely on limited and small emotion lexicons. In this paper we show how the combined strength and wisdom of the crowds can be used to generate a large, high-quality, word-emotion and word-polarity association lexicon quickly and inexpensively. We enumerate the challenges in emotion annotation in a crowdsourcing scenario and propose solutions to address them. Most notably, in addition to questions about emotions associated with terms, we show how the inclusion of a word choice question can discourage malicious data entry, help identify instances where the annotator may not be familiar with the target term (allowing us to reject such annotations), and help obtain annotations at sense level (rather than at word level). We conducted experiments on how to formulate the emotion-annotation questions, and show that asking if a term is associated with an emotion leads to markedly higher inter-annotator agreement than that obtained by asking if a term evokes an emotion.

User Edit Pencil Streamline Icon: https://streamlinehq.com
Authors (2)
  1. Saif M. Mohammad (70 papers)
  2. Peter D. Turney (20 papers)
Citations (2,204)

Summary

Crowdsourcing a Word-Emotion Association Lexicon

The paper, authored by Saif M. Mohammad and Peter D. Turney, elucidates the methodological creation of a comprehensive word-emotion association lexicon through crowdsourcing. Employing Amazon's Mechanical Turk, the researchers constructed an extensive dataset of English terms annotated with their associated emotions and polarities. This endeavor not only addresses the limited availability of emotion lexicons but also pioneers an efficient, scalable approach to such a task.

Methodology

The authors used several measures to ensure the reliability of crowdsourced data. Initially, they selected a diverse set of words from the Macquarie Thesaurus, General Inquirer (GI), and WordNet Affect Lexicon (WAL). These terms covered multiple parts of speech including nouns, verbs, adjectives, and adverbs. Approximately 10,170 terms were included initially.

A critical innovation was implementing a word-choice question preceding emotion annotations to gauge the annotator's understanding of the target term. Incorrect responses led to rejection of the subsequent annotations. They also differentiated between asking if a term "evokes" an emotion versus if it is "associated" with an emotion, finding higher inter-annotator agreement with the latter phrasing.

Results

The final product, referred to as EmoLex, contains annotations for 8,883 terms, classified across eight emotions: joy, sadness, anger, fear, trust, disgust, surprise, and anticipation. Notably, trust and joy emerged as the most frequently associated emotions in the dataset, while surprise had the least frequent associations.

The paper also presented a robust analysis of inter-annotator agreement. For emotions, over 60% of the annotations had unanimous (all five annotators agreed) or near-unanimous (four out of five agreed) consensus. This consensus was slightly lower for polarity annotations, suggesting the inherently subjective nature of emotional and evaluative language.

Implications

This research has several practical and theoretical implications:

  1. Practical Applications: The lexicon generated can be utilized in fields such as customer relationship management, sentiment analysis, human-computer interaction, and NLP applications, allowing for more nuanced emotional understanding and responsiveness.
  2. Evaluation: The lexicon allows for the evaluation and training of emotion detection algorithms, improving automated text analysis systems.
  3. Cross-Language Extension: The methodology paves the way for similar lexicons in other languages, facilitating cross-linguistic and cross-cultural research in emotion representation in text.

Future Work

Key areas for future exploration include expanding the lexicon to cover more terms, applying the methodology to other languages, and utilizing the lexicon in a variety of emotion-sensitive applications. Future research should also consider the integration of advanced techniques such as Maximum Difference Scaling (MaxDiff) to refine the annotation process and potentially improve inter-annotator agreement further.

Conclusion

The paper offers a detailed account of constructing a large-scale, high-quality word-emotion association lexicon using crowdsourcing, setting a precedent for future research in emotion detection and analysis. By leveraging the wisdom of the crowds and implementing rigorous quality control measures, it addresses the crucial need for extensive emotion lexicons in computational linguistics and related applications.