Stance detection on social media：State of the art and trends论文笔记

Stance detection on social media：State of the art and trends论文笔记（pdf下载）

·Year:15 Apr 2021
·Authors:Abeer AlDayel, Walid Magdy
·Journal Name:Information Processing & Management

0.Key Words：

·Stance detection
·Stance
·Social media
·Stance classification

1.Research Objective：

·to survey the
work on stance detection across those communities and present an exhaustive review of stance
detection techniques on social media
·to point out the features and models that work best
·to explore the emerging trends
and different applications of stance detection on social media
·to highlights the possible future directions for stance detection on social media

2.Background And Problem Statement

Stance detection on social media is an emerging opinion mining paradigm for various social and political applications in which sentiment analysis may be sub-optimal. There has been a growing research interest for developing effective methods for stance detection methods varying among multiple communities including natural language processing, web science, and social computing, where each modeled stance detection in different ways.

3.Contents

·Three different formulas for sentiment analysis(T:text,U:user,G:target):
1.

2.

3.

·Level of stance detection:
1.The statement level
2.The user level
·Classification of stance detection according to target:
1.Target-specific stance detection
2.Multi-related-target stance detection: The memory-based model of Wei,Lin,and Mao provides the current state-of-the-art result so far on the multi-target benchmark dataset
3.Claim-based stance detection(C:claim):

·Another classification of stance detection,namely stance detection and stance prediction:
1.Stance detection for expressed views
2.Stance prediction of unexpressed views:micro level and macro level
·Features used for stance detection:
1.Content features(linguistic features and users’ vocabulary):

In the SemEval-2016 stance detection task, the use of word and character n-gram modeling obtained the best f-score among the other
participating systems
2.Network features(users’ behavioral data and users’ meta-data attributes)
The similarity between users is considered a core property that helps in inferring stances
·Machine learning methods for stance detection:
1.Supervised learning( the most common approach)
2.Unconstrained supervised learning(to address the scarcity of the labeled data for each target in stance detection tasks)(many studies have incorporated the concept of transfer learning and achieved a good effect)
3.Unsupervised learning(The study by Darwish et al.(2020) used a clustering technique to create an initial set of stance partition for annotation. They used unlabeled tweets related tothree topics, Kavanaugh, Trump, and Erdogan. Their findings showed that using retweets as a feature provided the best performance score upon implementing clustering algorithm (DBSCAN), which surpassed the supervised method when using the fast-text and SVM
models)
·Stance detection applications:
1.Analytiical studies
2.Applications related to social-based phenomena
3.Veracity checking applications
·Possible directions for moving forward:
1.The transfer learning techniques for stance detection need a further enhancement
2.There is a real need for new benchmark datasets for stance detection that provide a sufficient amount of data to train and test the stance detection models
3.Creating more benchmark datasets of multiple languages would be of interest
Privacy measures could be developed to better preserve the privacy of the social media users

4.Conclusion

·The relationship of sentiment and stance are orthogonal
·Individuals with the same position tend to use the same choice of words to express their opinions
·Using sentiment analyzer for stance detection is sub-optimal
·Network features almost always provide the best performance in comparison with content features in all datasets in this paper
·However,sometimes it might be better to combine both network and content feature
·The use of network features also has been beneficial in detecting social media users’ stances and future attitudes in the aftermath of an event
·Supervised methods using SVM were found to be effective for different datasets
·The recent attempts to apply transfer learning and unsupervised learning for stance detection have promising results and are expected to be one of the main research direction in this area in the future

5.Notes

·Stance detection and sentiment analysis are two confusing concepts
·Sentiment analysis mainly uses textual attributes, while stance detection mainly be achieved by utilizing non-textual features, such as network and contextual features
·Defining the level of stance detection is essential, since it affects the task itself
·Interaction elements are used to define similarities between users: 1. retweets relationships 2.user tags
·Method of position detection:
1.Machine learning based approach(SVM,Logistic regression,Naive Bayes,Decision Tree,ANN,MLP…)
2.Deep learning based approach(LSTM,RNN,GRU,CNN,attention)
ensemble learning approach(random forest,Proprietary ensemble learners,boosting and bagging)

6.References

N/A

7.Summary

N/A