Categorization of tweets for damages: infrastructure and human damage assessment using fine-tuned BERT model.

Identification of infrastructure and human damage assessment tweets is beneficial to disaster management organizations as well as victims during a disaster. Most of the prior works focused on the detection of informative/situational tweets, and infrastructure damage, only one focused on human damage. This study presents a novel approach for detecting damage assessment tweets involving infrastructure and human damages. We investigated the potential of the Bidirectional Encoder Representations from Transformer (BERT) model to learn universal contextualized representations targeting to demonstrate its effectiveness for binary and multi-class classification of disaster damage assessment tweets. The objective is to exploit a pre-trained BERT as a transfer learning mechanism after fine-tuning important hyper-parameters on the CrisisMMD dataset containing seven disasters. The effectiveness of fine-tuned BERT is compared with five benchmarks and nine comparable models by conducting exhaustive experiments. The findings show that the fine-tuned BERT outperformed all benchmarks and comparable models and achieved state-of-the-art performance by demonstrating up to 95.12% macro-f1-score, and 88% macro-f1-score for binary and multi-class classification. Specifically, the improvement in the classification of human damage is promising.

How to detect propaganda from social media? Exploitation of semantic and fine-tuned language models.

Online propaganda is a mechanism to influence the opinions of social media users. It is a growing menace to public health, democratic institutions, and public society. The present study proposes a propaganda detection framework as a binary classification model based on a news repository. Several feature models are explored to develop a robust model such as part-of-speech, LIWC, word uni-gram, Embeddings from Language Models (ELMo), FastText, word2vec, latent semantic analysis (LSA), and char tri-gram feature models. Moreover, fine-tuning of the BERT is also performed. Three oversampling methods are investigated to handle the imbalance status of the Qprop dataset. SMOTE Edited Nearest Neighbors (ENN) presented the best results. The fine-tuning of BERT revealed that the BERT-320 sequence length is the best model. As a standalone model, the char tri-gram presented superior performance as compared to other features. The robust performance is observed against the combination of char tri-gram + BERT and char tri-gram + word2vec and they outperformed the two state-of-the-art baselines. In contrast to prior approaches, the addition of feature selection further improves the performance and achieved more than 97.60% recall, f1-score, and AUC on the dev and test part of the dataset. The findings of the present study can be used to organize news articles for various public news websites.

Rumour identification on Twitter as a function of novel textual and language-context features.

Social microblogs are one of the popular platforms for information spreading. However, with several advantages, these platforms are being used for spreading rumours. At present, the majority of existing approaches identify rumours at the topic level instead of at the tweet/post level. Moreover, prior studies used the sentiment and linguistic features for rumours identification without considering discrete positive and negative emotions and effective part-of-speech features in content-based approaches. Similarly, the majority of prior studies used content-based approaches for feature generation, and recent context-based approaches were not explored. To cope with these challenges, a robust framework for rumour detection at the tweet level is designed in this paper. The model used word2vec embeddings and bidirectional encoder representations from transformers method (BERT) from context-based and discrete emotions, linguistic, and metadata characteristics from content-based approaches. According to our knowledge, we are the first ones who used these features for rumour identification at the tweet/post level. The framework is tested on four real-life twitter microblog datasets. The results show that the detection model is capable of detecting 97%, 86%, 85%, and 80% of rumours on four datasets respectively. In addition, the proposed framework outperformed the three latest state-of-the-art baselines. BERT model presented the best performance among context-based approaches, and linguistic features are best performing among content-based approaches as a stand-alone model. Moreover, the utilization of two-step feature selection further improves the detection model performance.

Identification of offensive language in Urdu using semantic and embedding models.

Automatic identification of offensive/abusive language is very necessary to get rid of unwanted behavior. However, it is more challenging to generalize the solution due to the different grammatical structures and vocabulary of each language. Most of the prior work targeted western languages, however, one study targeted a low-resource language (Urdu). The prior study used basic linguistic features and a small dataset. This study designed a new dataset (collected from popular Pakistani Facebook pages) containing 7,500 posts for offensive language detection in Urdu. The proposed methodology used four types of feature engineering models: three are frequency-based and the fourth one is the embedding model. Frequency-based are either determined by the term frequency-inverse document frequency (TF-IDF) or bag-of-words or word n-gram feature vectors. The fourth is generated by the word2vec model, trained on the Urdu embeddings using a corpus of 196,226 Facebook posts. The experiments demonstrate that the stacking-based ensemble model with word2vec shows the best performance as a standalone model by achieving 88.27% accuracy. In addition, the wrapper-based feature selection method further improves performance. The hybrid combination of TF-IDF, bag-of-words, and word2vec feature models achieved 90% accuracy and 97% AUC. In addition, it outperformed the baseline with an improvement of 3.55% in accuracy, 3.68% in the recall, 3.60% in f1-measure, 3.67% in precision, and 2.71% in AUC. The findings of this research provide practical implications for commercial applications and future research.