Awesome Sentiment Attitude Extraction

A curated list of awesome studes related to sentiment attitude extraction, in which attitude corresponds to the text position conveyed by Subject towards other Object mentioned in text such as: entities, events, etc.

This repository collects works both related to relation extraction and sentiment analysis in which these two domains are inextricably linked, including event factualization as fundamentional studies for sentiment inference, stance detection.

Contributing: Please feel free to make pull requests or contact me [contacts]

Contents

• Related Studies
• Frameworks
• Annotation Schemas
• Papers
  • Large Language Models
    • Fact-Checking Adaptation
    • Chain-of-Thought
    • Conversational Systems
  • Language Models
    • Graph-Based
    • Low Resource Tunings
    • Prompts and Knowledge Examination
    • Architectures
  • Conventional neural-network based Models
  • Conventional Machine Learning Models
  • CRF-based Models
  • Rule-based Verb-applicable Models
  • Subsidiary Studies And Resources
  • Miscellaneous
• Thesises
• Datasets

Related studies

• Natural Language Processing
  • Sentiment Analysis
    • Targeted Setiment Analysis
    • Structured Sentiment Analysis (SemEval Task 10)
    • Aspect-based Sentiment Analysis
    • Hate-speech detection
  • Relation Extraction
  • Stance Detection

Frameworks

• bulk-chain [github]
  • Framework that exploits Chain-of-Thought concept and provides minimalistic solution for zero-shot inferences. For example, you can exploit the concept of aspect-opininon-reason chain from THOR-ISA to adapt it for attitude extraction.
• FaiMA [github]
  • Framework that integrates graph-based models and linguistics, with a core feature aimed at in-context learning for multi-domain SA.
• Reasoning-for-Sentiment-Analysis-Framework [github]
  • This frameworks repesent a reforget 🛠️ version of the THOR-ISA framework:
  • THOR-ISA [github]
    • Propt-based framework for setiment Analysis that based on Chain-of-Though concept for obtaining the result sentiment class out of the LLM system.
• OpenPrompt [github]
  • Enhanced tool for automatic completion of the prompt via the provided resources.
• ChatGPT [site]
  • Conversation system that is trained to follow the instruction in a prompt and provide a detailed response; examples on how it could be adapted reviewed in the following work.
• arekit-prompt-sampler [github] [prompt-engeneering-guide]
  • Sentiment Attitude Extraction sources sampling with language transferring and prompting API for further ChatGPT-alike model requests, powered by AREkit.
• ARElight [github]
  • AREkit-based application for a granular view onto sentiments between entities in a mass-media texts written in Russian
• AREnets [github]
  • Is an OpenNRE like project, but the kernel based on tensorflow library, with implementation of neural networks on top of it, designed for Attitude and Relation Extraction tasks.
• AREkit [github] [research-applicable-paper]
  • Is an open-source and extensible toolkit focused on data preparation for document-level relation extraction organization. It complements the OpenNRE functionality, as in terms of the latter, document-level RE setting is not widely explored (2.4 [paper]).
• DeRE [github] [paper]
  • Is an open-source framework for declaritive relation extraction, and therefore allows to declare your own task (using XML schemas) and apply manually implemented models towards it (using a provided API).
• OpenNRE [github] [paper]
  • Is an open-source and extensible toolkit that provides a unified framework to implement neural models for relation extraction (RE) between named entities.
• DeepPavlov-0.17.0 [docs] [post]
  • Is an entire relation extraction component for DeepPavlov opensource library, proposed by Anastasiia Sedova.
• Others ... [awesome-relation-extraction]

Back to Top

Annotation Schemas

• OpinionML [paper]
• SentiML [paper]
• OpinionMiningML [paper]
• EmotionML [paper]

Back to Top

Papers

Back to Top

Large Language Models

Awesome-LLM list

Fact-Checking Adaptation

NOTE: Requires / Assumes the presence of factual knowledgebase

• Consistent Document-Level Relation Extraction via Counterfactuals [paper] [code]
  • Concept: use factual relations for fictional context construction and LLM validation
  • Ali Modarressi, Abdullatif Köksal, Hinrich Schütze
  • EMNLP-2024, 15th of October 2024
• Learning to Refine with Fine-Grained Natural Language Feedback [paper] [code]
  • Concept: When treating attitudes as facts, we can adopt zero-shot LLM-based fact cheking as: **Detect**-**Critique**-**Refine**
  • Manya Wadhwa, Xinyu Zhao, Junyi Jessy, Li Greg Durrett
  • EMNLP-2024
• Zero-Shot Fact Verification via Natural Logic and Large Language Models [paper] [code]
  • Concept: Use natural logic for proving the fact of attitude presence in a zero-shot learning mode (see code)
  • EMNLP-2024

Chain-of-Thought

• FaiMA: Feature-aware In-context Learning for Multi-domain Aspect-based Sentiment Analysis [paper] [code]
  • Framework that integrates graph-based models and lingustics, with core feature aimed at in-context-learning feature for multi-domain SA; The framework is designed for multidomain datasets; Due to graphs and pairs-generation module, it may find major contribution in **attitude-based** sentiment extraction and target-oriented SA.
  • Songhua Yang, Xinke Jiang, Hanjie Zhao, Wenxuan Zeng, Hongde Liu, Yuxiang Jia
  • LREC-COLING 2024, Long Paper; Submitted 2 Mar. 2024.
• Aspect-Based Sentiment Analysis with Explicit Sentiment Augmentations [paper] [harvard-paper]
  • integrates explicit sentiment augmentations, acted as <<clues>> that augment LLM input context
  • Jihong Ouyang, Zhiyao Yang, Silong Liang, Bing Wang, Yimeng Wang, Ximing Li
  • Arxiv Pre-print, submitted: 18 Dec. 2024
• Sentiment Analysis through LLM Negotiations [paper] [open-review]
  • generator-discriminator of negotiating the result label
  • Xiaofei Sun, Xiaoya Li, Shengyu Zhang, Shuhe Wang, Fei Wu, Jiwei Li, Tianwei Zhang, Guoyin Wang
  • Arxiv Pre-print, submitted: 2024
• Reasoning Implicit Sentiment with Chain-of-Thought Prompting [paper] [code]
  • Sequence of 3 prompts for conversational system, complemented by tge system responses. Reason is to cope with hallucination similar-studies
  • Hao Fei, Bobo Li, Qian Liu, Lidong Bing, Fei Li, Tat-Seng Chua
  • ACL 2023, Short Papers

Conversational Systems

Using Language Models (usually LARGE-sized) in a combination with promts/questions

• Sentiment Analysis in the Era of Large Language Models: A Reality Check [paper]
  • application of the LLM and based on the latter ChatGPT for the variety set of sentiment analysis problems
  • Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Jialin Pan, Lidong Bing
  • arXiv, 24 May 2023
• Is ChatGPT better than Human Annotators? Potential and Limitations of ChatGPT in Explaining Implicit Hate Speech [paper]
  • Huang Fan, Kwak Haewoon, An Jisun
  • Harvard, Februrary, 2023
• How would Stance Detection Techniques Evolve after the Launch of ChatGPT? [paper]
  • Introducing prompt templater which allows to reach state-of-the art with zero-shot learning!
  • Bowen Zhang, Daijun Ding, Liwen Jing
  • Harvard, December, 2022

Language Models

Awesome-LLM list

Graph-Based

• Comparing Graph- and Seq2Seq- based Models Highlights Difficulty in Structured Sentiment Analysis [paper] [code]
  • Gaku Morio, Hiroaki Ozaki, Atsuki Yamaguchi, and Yasuhiro Sogawa
  • ACL-Workshop, 2022
• Enhancing Zero-shot and Few-shot Stance Detection with Commonsense Knowledge Graph [paper]
  • Rui Liu, Zheng Lin, Yutong Tan1, Weiping Wang
  • ACL-IJCNLP 2021

Back to Top

Low Resource Tunings

• Zero-shot Sentiment Analysis in Low-Resource Languages Using a Multilingual Sentiment Lexicon [paper] [code]
  • Fajri Koto, Tilman Beck, Zeerak Talat, Iryna Gurevych, Timothy Baldwin
  • NAACL-2024
• Black-Box Tuning for Language-Model-as-a-Service [paper] [code]
  • Non gradient p-tunes, wrapped in API in order to consider large Pre-Trained models (PTMs) adoptation as Service models
  • Tianxiang Sun, Yunfan Shao, Hong Qian, Xuanjing Huang, Xipeng Qiu
  • Arxiv Pre-print, 2022
• P-Tuning v2: Prompt Tuning Can Be Comparable to Fine-tuning Universally Across Scales and Tasks [paper] [code]
  • Proceeds Prefix-Tuning idea onto multiple layers of LM-model
  • Xiao Liu, Kaixuan Ji, Yicheng Fu, Zhengxiao Du, Zhilin Yang, Jie Tang
  • Dblp Jornal, 2021
• The Power of Scale for Parameter-Efficient Prompt Tuning [paper] [code]
  • Prompt-designing, prompt-tuning comparison studies
  • Brian Lester, Rami Al-Rfou, Noah Constant
  • EMNLP-2021
• GPT Understands, Too [paper] [code]
  • Promt Tuning (p-tuning), i.e. training only promt token embeddings before and after input sequence (x)
  • Xiao Liu, Yanan Zheng, Zhengxiao Du, Ming Ding, Yujie Qian, Zhilin Yang, Jie Tang
  • 2021
• Prefix-Tuning: Optimizing Continuous Prompts for Generation [paper] [code]
  • Training token prefixes for downstream tasks with frozen LM parameters
  • Xiang Lisa Li, Percy Liang
  • ACL/IJCNLP-2021
• Language Models are Few-Shot Learners [paper]
  • Prompt designing. FS, 1S by presenting context as "[input,result] x k-times", where k > 1 (FewShot), k = 1 (OneShot); ZeroShot includes only descriptor of expected result
  • Tom B. Brown, et. al.
  • NeurIPS-2020
• AutoPrompt: Eliciting Knowledge from Language Models with Automatically Generated Prompts [paper] [code]
  • Considering sentiment analysis task as MLM by predicting [MASK]; prompting input (x) with tokens (p1...pk), selected by gradient search (considering that label has corresponding tokens (prompts))
  • Taylor Shin, Yasaman Razeghi, Robert L. Logan IV, Eric Wallace, Sameer Singh
  • EMNLP-2020

Back to Top

Prompts and Knowledge Examination

• Sentiment Analysis in the Era of Large Language Models: A Reality Check [paper]
  • duplicated from the one in conversational systems section
  • Wenxuan Zhang, Yue Deng, Bing Liu, Sinno Jialin Pan, Lidong Bing
  • arXiv, 24 May 2023
• How Can We Know What Language Models Know? [paper] [code]
  • Implemented model LPAQA: Language model Prompt And Query Archive
  • Zhengbao Jiang, Frank F. Xu, Jun Araki, Graham Neubig
  • TACL-2020
• Language Models as Knowledge Bases? [paper] [code]
  • Fabio Petroni, Tim Rocktäschel, Patrick Lewis, Anton Bakhtin, Yuxiang Wu, Alexander H. Miller, Sebastian Riedel
  • EMNLP-2019
• Utilizing BERT for Aspect-Based Sentiment Analysis via Constructing Auxiliary Sentence [paper] [code]
  • Adopting a predefined prompt (QA/NLI formats) as a TextB input part
  • Chi Sun, Luyao Huang, Xipeng Qiu
  • NAACL-HLT 2019

Back to Top

Architectures

• BERT-based models (Encoder Reprsentation From Transorfmers) [papers]
  • Considering BERT model as classifier
  • Joohong Lee, Awesome Relation Extraction
• GPT-based (Encoder Reprsentation From Transorfmers) [papers]
  • Considering GPT model competed for classification task
  • Joohong Lee, Awesome Relation Extraction
• Comparing Graph- and Seq2Seq- based Models Highlights Difficulty in Structured Sentiment Analysis [paper] [code]
  • T5 and mT5 finetunnning, i.e. Text-To-Text Transfer Transoformer application
  • Gaku Morio, Hiroaki Ozaki, Atsuki Yamaguchi, and Yasuhiro Sogawa
  • ACL-Workshop, 2022

Back to Top

Conventional Neural-network based Models

In this section we consider neural-network models based on convolutional, recurrent, recursive architectures.

• No Permanent Friends or Enemies: Tracking Relationships between Nations from News [paper]
  • Xiaochuang Han, Eunsol Choi, Chenhao Tan
  • NAACL-HLT 2019
• Neural networks for open domain targeted sentiment [paper]
  • Meishan Zhang, Yue Zhang, Duy-Tin Vo
  • ACL 2015

Back to Top

Conventional Machine Learning Models

• Document-level Sentiment Inference with Social, Faction, and Discourse Context [paper]
  • Eunsol Choi, Hannah Rashkin, Luke Zettlemoyer, Yejin Choi
  • ACL-2016
• Sentiment Analysis: Capturing Favorability Using Natural Language Processing [paper]
  • it is originally called favorability analysis, semantic establishment between sentiment and subject
  • Tetsuya Nasukawa, Jeonghee Yi
  • K-CAP-2003 (ACM)

CRF-based Models

• Open Domain Targeted Sentiment [paper]
  • Margaret Mitchell, Jacqueline Aguilar, Theresa Wilson, Benjamin Van Durme
  • ACL 2013

Rule-based Verb-applicable Models

• Stance detection in Facebook posts of a German right-wing party [paper]
  • Manfred Klenner, Don Tuggener, Simon Clematide
  • Verb-usages form
  • ACL 2017 (2nd Workshop on Linking Models of Lexical, Sentential and Discourse-level Semantics)
• An object-oriented model of role framing and attitude prediction [paper]
  • Object-oriented model
  • ACL 2017 (2nd Workshop on Linking Models of Lexical, Sentential and Discourse-level Semantics)
• Joint Prediction for Entity/Event-Level Sentiment Analysis using Probabilistic Soft Logic Models [paper]
  • Lingjia Deng, Janyce Wiebe
  • EMNLP 2015
• FactBank: a corpus annotated with event factuality [paper]
  • Roser Saurí, James Pustejovsky
  • 2009

Back to Top

Subsidiary Studies and Resources

• RIVETER Measuring Power and Social Dynamics Between Entities [paper]
  • Maria Antoniak, Anjalie Field, Jimin Mun, Melanie Walsh, Lauren F. Klein, Maarten Sap
  • ACL-2023
• Multilingual Connotation Frames: A Case Study on Social Media for Targeted Sentiment Analysis and Forecast [paper] [resources]
  • Hannah Rashkin, Eric Bell, Yejin Choi, Svitlana Volkova
  • ACL-2017
• Learning Lexico-Functional Patterns for First-Person Affect [paper]
  • Lena Reed, Jiaqi Wu, Shereen Oraby
  • ACL-2017
• Understanding Abuse: A Typology of Abusive Language Detection Subtasks [paper]
  • Zeerak Waseem, Thomas Davidson, Dana Warmsley, Ingmar Weber
  • ACL-2017
• Connotation Frames: A Data-Driven Investigation [paper]
  • Hannah Rashkin, Sameer Singh, Yejin Choi
  • ACL-2016
• Do Characters Abuse More Than Words? [paper]
  • Yashar Mehdad, Joel Tetreault
  • SIGDIAL-2016

Back to Top

Miscellaneous

• Verifying the robustness of opinion inference [paper]
  • Josef Ruppenhofer, Jasper Brandes
  • KONVENS 2016

Back to Top

Thesises

• Mitigation of Gender Bias in Text using Unsupervised Controllable Rewriting [master-thesis]
  • Maja Brinkmann
  • Paderborn University, 2022
    • Connotation Frames (2.1.3.)
    • Connotational Frames and Lexicon (3.1.1.)

Back to Top

Datasets

• NOW (2010 -- present) [site] -- News on the Web Corpus.
  • Contains data from online magazines and newspapers in 20 different English-speaking countries from 2010 to the current time. (Raw texts only).
• MPQA-3.0, (2015) [site] [paper]
• SNLI [site] [paper] -- Stanford Natural Language Inference
  • 570k human-written English sentence pairs manually labeled for balanced classification with the labels entailment, contradiction, and neutral
• FactBank 2009, [paper] -- a corpus annotated with event factuality
  • Consists of 208 documents and contains a total of 9,488, including TimeBank data; manually annotated events.
• TimeBank, 2003 [site] [paper]
  • Annotated to indicate events, times, and temporal relations

Back to Top