Social media memes are a challenging domain for hate detection because they intertwine visual and textual cues into culturally nuanced messages. To tackle these challenges, we introduce TRACE, a hierarchical multimodal framework that leverages visually grounded context augmentation, along with a novel caption-scoring network to emphasize hate-relevant content, and parameter-efficient fine-tuning of CLIP’s text encoder. Our experiments demonstrate that selectively fine-tuning deeper text encoder layers significantly enhances performance compared to simpler projection-layer fine-tuning methods. Specifically, our framework achieves state-of-the-art accuracy (0.807) and F1-score (0.806) on the widely-used Hateful Memes dataset, matching the performance of considerably larger models while maintaining efficiency. Moreover, it achieves superior generalization on the MultiOFF offensive meme dataset (F1-score 0.673), highlighting robustness across meme categories. Additional analyses confirm that robust visual grounding and nuanced text representations significantly reduce errors caused by benign confounders. We publicly release our code to facilitate future research.

Publisher Copyright: © 2026 Owner/Author.

Linearization of attention using various kernel approximation and kernel learning techniques has shown promise. Past methods used a subset of combinations of component functions and weight matrices within the random feature paradigm. We identify the need for a systematic comparison of different combinations of weight matrices and component functions for attention learning in Transformer. Hence, we introduce Spectraformer , a unified framework for approximating and learning the kernel function in the attention mechanism of the Transformer. Our empirical results demonstrate, for the first time, that a random feature-based approach can achieve performance comparable to top-performing sparse and low-rank methods on the challenging Long-Range Arena benchmark. Thus, we establish a new state-of-the-art for random feature-based efficient Transformers. The framework also produces many variants that offer different advantages in accuracy, training time, and memory consumption. Our code is available at: https://github.com/cruiseresearchgroup/spectraformer .

Collaborative Filtering (CF) is a class of methods widely used to support high-quality Recommender Systems (RSs) across several industries [6]. Studies have uncovered distinct advantages and limitations of CF in many real-world applications [5, 9]. Besides the inability to address the cold-start problem, sensitivity to data sparsity is among the main limitations recurrently associated with this class of RSs. Past work has extensively demonstrated that data sparsity critically impacts CF accuracy [2, 3, 4]. The proposed talk revisits the relation between data sparsity and CF from a new perspective, evincing that the former also impacts the fairness of recommendations. In particular, data sparsity might lead to unfair bias in domains where the volume of activity strongly correlates with personal characteristics that are protected by law (i.e., protected attributes). This concern is critical for RSs deployed in domains such as the recruitment domain, where RSs have been reported to automate or facilitate discriminatory behaviour [7]. Our work at SEEK deals with recommender algorithms that recommend jobs to candidates via SEEK’s multiple channels. While this talk focuses on our perspective of the problem in the job recommendation domain, the discussion is relevant to many other domains where recommenders potentially have a social or economic impact on the lives of individuals and groups.

‘Watch The Flu’ is a tool that monitors tweets posted in Australia for symptoms of influenza. The tool is a unique combination of two areas of artificial intelligence: natural language processing and time series monitoring, in order to assist public health surveillance. Using a real-time data pipeline, it deploys a web-based dashboard for visual analysis, and sends out emails to a set of users when an outbreak is detected. We expect that the tool will assist public health experts with their decision-making for disease outbreaks, by providing them insights from social media.

Cross-domain sentiment analysis (CDSA) helps to address the problem of data scarcity in scenarios where labelled data for a domain (known as the target domain) is unavailable or insufficient. However, the decision to choose a domain (known as the source domain) to leverage from is, at best, intuitive. In this paper, we investigate text similarity metrics to facilitate source domain selection for CDSA. We report results on 20 domains (all possible pairs) using 11 similarity metrics. Specifically, we compare CDSA performance with these metrics for different domain-pairs to enable the selection of a suitable source domain, given a target domain. These metrics include two novel metrics for evaluating domain adaptability to help source domain selection of labelled data and utilize word and sentence-based embeddings as metrics for unlabelled data. The goal of our experiments is a recommendation chart that gives the K best source domains for CDSA for a given target domain. We show that the best K source domains returned by our similarity metrics have a precision of over 50%, for varying values of K.

Trust can be employed for finding reliable information in Online Social Networks (OSNs). Since users in OSNs may intentionally change their behavior over time (in some cases for deceiving other users), modeling (pair-wise) trust relations in such complex environment is a challenging task. However, most of the existing trust prediction approaches assume that trust relations are fixed over time and they fail to capture the dynamic behavior of users in OSNs. In this paper, we propose a dynamic deep trust prediction model. As the impact of incidental emotions on trust has been proven in psychology studies, in this paper, we also study this impact on our trust prediction approach. First, we propose a novel deep structure that incorporates users' emotions and their textual contents in OSNs. Second, we use embeddings to represent the users and their self-descriptions provided. Finally, considering different time windows, we dynamically predict pair-wise trust relations. To evaluate our approach, we collected a large twitter dataset. The evaluation results demonstrate the effectiveness of our approach compared to the state-of-the-art approaches.

Harnessing data from social media to monitor health events is a promising avenue for public health surveillance. A key step is the detection of reports of a disease (referred to as ‘health mention classification’) amongst tweets that mention disease words. Prior work shows that figurative usage of disease words may prove to be challenging for health mention classification. Since the experience of a disease is associated with a negative sentiment, we present a method that utilises sentiment information to improve health mention classification. Specifically, our classifier for health mention classification combines pre-trained contextual word representations with sentiment distributions of words in the tweet. For our experiments, we extend a benchmark dataset of tweets for health mention classification, adding over 14k manually annotated tweets across diseases. We also additionally annotate each tweet with a label that indicates if the disease words are used in a figurative sense. Our classifier outperforms current SOTA approaches in detecting both health-related and figurative tweets that mention disease words. We also show that tweets containing disease words are mentioned figuratively more often than in a health-related context, proving to be challenging for classifiers targeting health-related tweets.

Personal health mention detection deals with predicting whether or not a given sentence is a report of a health condition. Past work mentions errors in this prediction when symptom words, i.e., names of symptoms of interest, are used in a figurative sense. Therefore, we combine a state-of-the-art figurative usage detection with CNN-based personal health mention detection. To do so, we present two methods: a pipeline-based approach and a feature augmentation-based approach. The introduction of figurative usage detection results in an average improvement of 2.21% F-score of personal health mention detection, in the case of the feature augmentation-based approach. This paper demonstrates the promise of using figurative usage detection to improve personal health mention detection.

Customer reviews are now increasingly available on Online Social Networks (OSNs) for a wide range of products and services. Trust in the review's author is a crucial basis for believing in the reliability of reviews generated on such networks. In this context, the main challenge is to predict the unknown trust relationship between two users. Existing trust prediction approaches fail to incorporate textual footprint of users. To address this challenge, we present a deep learning-based graph analytics model to predict trust relations in OSNs. We leverage and extend GraphSAGE, a method for computing node representations in an inductive manner, to develop a deep classifier. We present our experiment with datasets from review websites to train classifiers that predict trust relations between pairs of users, and highlight how our approach significantly improves the quality of predicted trust relations compared to the state-of-the-art approaches.

Sarcasm Suite is a browser-based engine that deploys five of our past papers in sarcasm detection and generation. The sarcasm detection modules use four kinds of incongruity: sentiment incongruity, semantic incongruity, historical context incongruity and conversational context incongruity. The sarcasm generation module is a chatbot that responds sarcastically to user input. With a visually appealing interface that indicates predictions using `faces' of our co-authors from our past papers, Sarcasm Suite is our first demonstration of our work in computational sarcasm.

This paper is a novel study that views sarcasm detection in dialogue as a sequence labeling task, where a dialogue is made up of a sequence of utterances.We create a manuallylabeled dataset of dialogue from TV series 'Friends' annotated with sarcasm.Our goal is to predict sarcasm in each utterance, using sequential nature of a scene.We show performance gain using sequence labeling as compared to classification-based approaches.Our experiments are based on three sets of features, one is derived from information in our dataset, the other two are from past works.Two sequence labeling algorithms (SVM-HMM and SEARN) outperform three classification algorithms (SVM, Naive Bayes) for all these feature sets, with an increase in F-score of around 4%.Our observations highlight the viability of sequence labeling techniques for sarcasm detection of dialogue.

This paper makes a simple increment to state-of-the-art in sarcasm detection research. Existing approaches are unable to capture subtle forms of context incongruity which lies at the heart of sarcasm. We explore if prior work can be enhanced using semantic similarity/discordance between word embeddings. We augment word embedding-based features to four feature sets reported in the past. We also experiment with four types of word embeddings. We observe an improvement in sarcasm detection, irrespective of the word embedding used or the original feature set to which our features are augmented. For example, this augmentation results in an improvement in F-score of around 4\% for three out of these four feature sets, and a minor degradation in case of the fourth, when Word2Vec embeddings are used. Finally, a comparison of the four embeddings shows that Word2Vec and dependency weight-based features outperform LSA and GloVe, in terms of their benefit to sarcasm detection.