Stance Classification in Out-of-Domain Rumours: A Case Study Around Mental Health Disorders

Aker, Ahmet; Zubiaga, Arkaitz; Bontcheva, Kalina; Kolliakou, Anna; Procter, Rob; Liakata, Maria

doi:10.1007/978-3-319-67256-4_6

Cited by 10 publications

(9 citation statements)

References 15 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Additionally, the current credibility prediction models face difficulty when they are applied to different events (Boididou et al, 2014;Aker et al, 2017), as the performance accuracy is overestimated. Our results found that there are significant differences of content of the same credibility level and topic when generated from different locations.…”

Section: Discussionmentioning

confidence: 99%

Location impact on source and linguistic features for information credibility of social media

Aladhadh

Zhang

Sanderson

2019

OIR

View full text Add to dashboard Cite

Purpose Social media platforms provide a source of information about events. However, this information may not be credible, and the distance between an information source and the event may impact on that credibility. Therefore, the purpose of this paper is to address an understanding of the relationship between sources, physical distance from that event and the impact on credibility in social media. Design/methodology/approach In this paper, the authors focus on the impact of location on the distribution of content sources (informativeness and source) for different events, and identify the semantic features of the sources and the content of different credibility levels. Findings The study found that source location impacts on the number of sources across different events. Location also impacts on the proportion of semantic features in social media content. Research limitations/implications This study illustrated the influence of location on credibility in social media. The study provided an overview of the relationship between content types including semantic features, the source and event locations. However, the authors will include the findings of this study to build the credibility model in the future research. Practical implications The results of this study provide a new understanding of reasons behind the overestimation problem in current credibility models when applied to different domains: such models need to be trained on data from the same place of event, as that can make the model more stable. Originality/value This study investigates several events – including crisis, politics and entertainment – with steady methodology. This gives new insights about the distribution of sources, credibility and other information types within and outside the country of an event. Also, this study used the power of location to find alternative approaches to assess credibility in social media.

show abstract

Section: Discussionmentioning

confidence: 99%

Location impact on source and linguistic features for information credibility of social media

Aladhadh

Zhang

Sanderson

2019

OIR

View full text Add to dashboard Cite

show abstract

“…In total we included the following features, which turned out to be useful in prior research in the area of stance detection in Twitter communication [53] and were already annotated in the data set: for the authorrelated features we used authors Twitter account description, length of the account description, and role (refers to the relation between follower and followee number), for the message-related features we took URL included, location included, person included, date included, negation included, Google bad word included (using a dictionary from Google to check if the tweet contains slang words), geo information enabled, average word length, and for the meta-informational features we comprised originality (refers to the number of tweets of a user), number of followers, engagement (refers to the number of tweets related to user account age) and sentiment (describes on a scale ranging from positive to negative the valence of the tweet with an assigned value between 0…”

Section: Insert Figure 1 About Herementioning

confidence: 99%

The Impact of Twitter Features on Credibility Ratings - An Explorative Examination Combining Psychological Measurements and Feature Based Selection Methods

Meinert¹,

Aker²,

Krämer³

2019

Proceedings of the Annual Hawaii International Conference on System Sciences

Self Cite

View full text Add to dashboard Cite

In a post-truth age determined by Social Media channels providing large amounts of information of questionable credibility while at the same time people increasingly tend to rely on online information, the ability to detect whether content is believable is developing into an important challenge. Most of the work in that field suggested automated approaches to perform binary classification to determine information veracity. Recipients´ perspectives and multidimensional psychological credibility measurements have rarely been considered. To fill this gap and gain more insights into the impact of a tweet´s features on perceived credibility, we conducted a survey asking participants (N=2626) to rate the credibility of crisis-related tweets. The resulting 24.823 ratings were used for an explorative feature selection analysis revealing that mostly meta-related features like the number of followers of the author, the count of tweets produced and the ratio of tweet number and days since account creation affect credibility judgments.

show abstract

“…PHEME5+Aug+boston is PHEME5+Aug combined with the "bostonbombings". We employ Kochkina et al [4]'s method as a SoA baseline model of rumor detection with three modifications 9 . In their model, source tweets and replies are represented as 300dimensional Word2Vec word embeddings pre-trained on the Google News data set 10 .…”

Section: Rumor Detectionmentioning

confidence: 99%

“…This restricts a further exploration of NNs for representation learning through many layers of nonlinear processing units and different levels of abstraction [8], which results in overfitting and generalization concerns. The scarcity of labeled data is a major challenge facing for the research of rumors in social media [9]. Another problem is that publicly available data sets for rumor-related tasks suffer from imbalanced class distributions [10,4].…”

Section: Introductionmentioning

confidence: 99%

Neural language model based training data augmentation for weakly supervised early rumor detection

Han

Gao

Ciravegna

2019

Proceedings of the 2019 IEEE/ACM International Conference on Advances in Social Networks Analysis and Mining

View full text Add to dashboard Cite

The scarcity and class imbalance of training data are known issues in current rumor detection tasks. We propose a straight-forward and general-purpose data augmentation technique which is beneficial to early rumor detection relying on event propagation patterns. The key idea is to exploit massive unlabeled event data sets on social media to augment limited labeled rumor source tweets. This work is based on rumor spreading patterns revealed by recent rumor studies and semantic relatedness between labeled and unlabeled data. A state-ofthe-art neural language model (NLM) and large credibilityfocused Twitter corpora are employed to learn context-sensitive representations of rumor tweets. Six different real-world events based on three publicly available rumor datasets are employed in our experiments to provide a comparative evaluation of the effectiveness of the method. The results show that our method can expand the size of an existing rumor data set nearly by 200% and corresponding social context (i.e., conversational threads) by 100% with reasonable quality. Preliminary experiments with a state-of-the-art deep learning-based rumor detection model show that augmented data can alleviate over-fitting and class imbalance caused by limited train data and can help to train complex neural networks (NNs). With augmented data, the performance of rumor detection can be improved by 12.1% in terms of F-score. Our experiments also indicate that augmented training data can help to generalize rumor detection models on unseen rumors.

show abstract

Stance Classification in Out-of-Domain Rumours: A Case Study Around Mental Health Disorders

Cited by 10 publications

References 15 publications

Location impact on source and linguistic features for information credibility of social media

Location impact on source and linguistic features for information credibility of social media

The Impact of Twitter Features on Credibility Ratings - An Explorative Examination Combining Psychological Measurements and Feature Based Selection Methods

Neural language model based training data augmentation for weakly supervised early rumor detection

Contact Info

Product

Resources

About