Figure - uploaded by Abhishek Dhankar
Content may be subject to copyright.
Text Entailment Task Labels in Terms of Original FACTIFY Task Labels

Text Entailment Task Labels in Terms of Original FACTIFY Task Labels

Source publication
Conference Paper
Full-text available
Identifying fake news is a very difficult task, especially when considering the multiple modes of conveying information through text, image, video and/or audio. We attempted to tackle the problem of automated misinformation/disinformation detection in multi-modal news sources (including text and images) through our simple, yet effective, approach i...

Context in source publication

Context 1
... can be clearly seen that the shared task can now be broken down into two sub-tasks; namely, text entailment and image entailment, where text entailment consists of classes í µí²¯ _0, í µí²¯ _1 and í µí²¯ _2, and image entailment consists of classes ℐ_0, ℐ_1, ℐ_2. These new classes are the combination of the original class labels as shown in Table 2 and Table 3 for the text entailment and image entailment tasks respectively. Once the dataset is rearranged according to the sub-task labels, we end up with one dataset for each sub-task. ...

Similar publications

Full-text available
Medical interventions may be more effective in some types of individuals than others and identifying characteristics that modify the effectiveness of an intervention is a cornerstone of precision or stratified medicine. The opportunity for detailed examination of treatment-covariate interactions can be an important driver for undertaking an individ...


... UofA-Truth [48] break down the task into two sub-tasks: text entailment and image entailment. For text entailment, they obtain claim text and document text embeddings by passing the text through sentence BERT [49]. ...
Full-text available
Fake news can spread quickly on social media and it is important to detect it before it creates lot of damage. Automatic fact/claim verification has recently become a topic of interest among diverse research communities. We present the findings of the Factify shared task, which aims undertake multi-modal fact verification, organized as a part of the De-Factify workshop at AAAI'22. The task is modeled as a multi-modal entailment task, where each input needs to be classified into one of 5 classes based on entailment and modality. A total of 64 teams participated in the Factify shared task, and of them, 9 teams submitted their predictions on test set. The most successful models were BigBird or other variations of BERT. The highest F1 score averaged across all the classes was 76.82%.