Representations of emotion concepts: Comparison across pairwise, appraisal feature-based, and word embedding-based similarity spaces

This repository contains code and data for the publication "Representations of emotion concepts: Comparison across pairwise, appraisal feature-based, and word embedding-based similarity spaces" by Kwon, M., Wager, T., & Phillips, J. (2022), published in the Proceedings of the Annual Meeting of the Cognitive Science Society, 44(44) and can be found at https://escholarship.org/uc/item/8vj3d366. All code is written in R. Updates We have updated our analyses based on feedback received since the annual meeting. The changes do not alter our main findings or conclusion. The included analysis script (`EMOCON_cogsci2022_revised.Rmd`) reflects these updates: Additional emotion concept pairs: Pairs involving `comfortableness`, `gratefulness`, `relaxedness`, `romanticness`, `sereneness`, `protectiveness` were added to the analysis, which were omitted in the previous version. The revised script includes all pairs and reflects changes in the feature-based similarity matrix, correlation between the similarity measures, loading values from principle component analysis on appraisal features, regression coefficients from multiple regression with the affective features components from PCA, and difference scores of the affective feature components. Scaling before PCA: PCA was rerun after rescaling data. `script` Scripts used for analysis are included in a R markdown file. `data` Data for appraisal feature rating, pairwise similarity rating, word embedding from word2vec models trained on Google news and Wikipedia are included. References for these data are listed below and also in the paper. Word embeddings from GPT3 can be accessed via [OpenAI](https://openai.com/)'s API, with relevant documentation on how-to found [here](https://beta.openai.com/docs/guides/embeddings/what-are-embeddings). Word embedding from W2V model trained on Google news: Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., & Dean, J. (2013). Distributed Representations of Words and Phrases and their Compositionality. Advances in Neural Information Processing Systems, 26. Word embeddings from W2v Model trained on Wikipedia: Fares, M., Kutuzov, A., Oepen, S., & Velldal, E. (2017). Word vectors, reuse, and replicability: Towards a community repository of large-text resources. Proceedings of the 21st Nordic Conference on Computational Linguistics, 271–276.

本仓库包含Kwon、M.、Wager、T.与Phillips、J.于2022年发表的论文《情感概念表征：成对比较、基于评价特征与基于词嵌入的相似性空间对比》（原英文标题：Representations of emotion concepts: Comparison across pairwise, appraisal feature-based, and word embedding-based similarity spaces）的配套代码与数据。该论文刊载于《认知科学学会年会会议录》（Proceedings of the Annual Meeting of the Cognitive Science Society）第44卷第44期，原文可通过https://escholarship.org/uc/item/8vj3d366获取。所有代码均采用R语言编写。

更新说明：本团队基于年会后收到的反馈更新了分析内容，此次调整未改变核心研究发现与结论。本次提供的分析脚本`EMOCON_cogsci2022_revised.Rmd`已纳入所有更新：
1. 新增情感概念对：本次分析新增了涉及`comfortableness`（舒适感）、`gratefulness`（感激感）、`relaxedness`（放松感）、`romanticness`（浪漫感）、`sereneness`（宁静感）、`protectiveness`（保护感）的情感概念对，此类概念对在先前版本中被遗漏。
2. 修订后的脚本覆盖全部情感概念对，并更新了以下内容：基于评价特征的相似性矩阵、相似性度量间的相关性、评价特征主成分分析（PCA，Principal Component Analysis）的载荷值、基于PCA得到的情感特征成分进行多元回归的回归系数，以及情感特征成分的差异分数。
3. 主成分分析前的数据标准化：已对数据进行标准化处理后重新运行主成分分析。

脚本说明：分析所用脚本包含于R Markdown文件中。

数据说明：本仓库包含三类数据：评价特征评分数据、成对相似性评分数据，以及基于谷歌新闻（Google News）与维基百科（Wikipedia）语料训练的word2vec（W2V）模型生成的词嵌入（word embedding）数据。上述数据的参考文献已列于下文及论文正文中。
基于GPT3的词嵌入可通过OpenAI（开放人工智能公司）的应用程序编程接口（API）获取，相关使用指南可参见https://beta.openai.com/docs/guides/embeddings/what-are-embeddings。

各数据集参考文献如下：
1. 基于谷歌新闻语料训练的W2V模型生成的词嵌入：Mikolov, T., Sutskever, I., Chen, K., Corrado, G. S., & Dean, J. (2013). 《分布式词与短语表征及其组合性》（原英文标题：Distributed Representations of Words and Phrases and their Compositionality），刊载于《神经信息处理系统进展》（Advances in Neural Information Processing Systems）第26卷。
2. 基于维基百科语料训练的W2V模型生成的词嵌入：Fares, M., Kutuzov, A., Oepen, S., & Velldal, E. (2017). 《词向量、复用与可复现性：构建大型文本资源社区仓库》（原英文标题：Word vectors, reuse, and replicability: Towards a community repository of large-text resources），刊载于第21届北欧计算语言学会议会议录，第271–276页。