The escalating energy consumption and carbon footprint of training large-scale Web AI models pose urgent challenges for sustainable development. Dataset Distillation (DD) offers a promising avenue for green AI by compressing large datasets into small synthetic ones for efficient training. However, most existing DD methods overfit to the inductive biases of specific source architectures (e.g., CNNs or ViTs), resulting in poor cross-model generalization. This limitation necessitates redundant re-distillation processes for different architectures, severely undermining the energy-saving potential of DD. To address this, we introduce Adversarial Mutual Information Distillation (AMID), a rigorous framework designed to create highly reusable and robust synthetic datasets. From an information-theoretic perspective, we cast model-agnosticism as minimizing the mutual information (MI) between the synthetic data and the specific identity of the distillation model. We convert this intractable objective into a tractable two-player adversarial game, which unifies knowledge preservation with adversarial unlearning of architectural bias. Extensive experiments on CIFAR-10 and Tiny ImageNet demonstrate that AMID achieves state-of-the-art cross-architecture generalization across diverse CNNs and ViTs. Crucially, our analysis confirms that AMID significantly reduces the computational overhead and CO2 emissions of downstream training while maintaining robust performance, paving the way for energy-efficient, transferable, and sustainable Web AI ecosystems.

The emergence of social media has led memes to become a powerful mode of communication, blending text, images, and emojis. However, this surge in meme usage has also seen a rise in offensive material. With manual content moderation proving impractical due to the sheer volume of data, there's a pressing need for automated methods to identify harmful memes. Yet, existing research predominantly targets high-resource languages such as English, neglecting low-resource ones like Nepali. To bridge this gap, we introduce the first Nepali meme dataset annotated for hate speech and sentiment. Our contributions are threefold: (1) We create and release NeMeme, a unique dataset featuring Nepali and code-mixed Nepali memes (combining Nepali and English). (2) We evaluate NeMeme using cutting-edge unimodal and multimodal models to establish initial performance benchmarks. (3) We introduce MemeNePAL, a novel multimodal framework employing prompt-assisted learning to effectively categorize Nepali memes. MemeNePAL overcomes the shortcomings of prior state-of-the-art (SOTA) techniques, which were designed for high-resource languages and struggle with Nepali's linguistic differences and cultural subtleties. This work not only promotes inclusivity in content moderation research but also aligns with UN Sustainable Development Goals such as promoting well-being, reducing inequalities, and fostering peace. We adhere to FAIR principles by making the dataset publicly available.

Recently, frequency transformation (FT) has been increasingly incorporated into deep learning models to significantly enhance state-of-the-art accuracy and efficiency in time series analysis. The advantages of FT, such as high efficiency and a global view, have been rapidly explored and exploited in various time series tasks and applications, demonstrating the promising potential of FT as a new deep learning paradigm for time series analysis. Despite the growing attention and the proliferation of research in this emerging field, there is currently a lack of a systematic review and in-depth analysis of deep learning-based time series models with FT. It is also unclear why FT can enhance time series analysis and what its limitations are in the field. To address these gaps, we present a comprehensive review that systematically investigates and summarizes the recent research advancements in deep learning-based time series analysis with FT. Specifically, we explore the primary approaches used in current models that incorporate FT, the types of neural networks that leverage FT, and the representative FT-equipped models in deep time series analysis. We propose a novel taxonomy to categorize the existing methods in this field, providing a structured overview of the diverse approaches employed in incorporating FT into deep learning models for time series analysis. Finally, we highlight the advantages and limitations of FT for time series modeling and identify potential future research directions that can further contribute to the community of time series analysis.

Medication recommendation aims to suggest personalized drug combinations to patients based on their longitudinal medical histories stored in electronic health record (EHR) datasets. Patients' Dynamic Treatment Regimes (DTRs) determine how patients' drug combinations change along with the evolution of disease treatment. DTRs are effective for comprehending disease-treatment dynamics and for recommending a timely and personalized combination of medications for patients. However, existing medication recommender systems (MRSs) overlook the multiple treatment pathways generated by the intervention of DTRs and can only recommend a single treatment paradigm, ignoring the fact that patients may be at different treatment stages and thus require different treatment regime. Such disregard leads to a significant limitation in recommending personalized medication combinations tailored to different treatment stages, yielding greatly compromised accuracy and applicability of MRSs. Moreover, existing methods often overlook the time interval information over patients' successive visits, which is critical to indicate patients' treatment evolution. To address these significant gaps, we propose a Time-aware Medication Recommendation Framework via Intervention of Dynamic Treatment Regimes, called MR-DTR. To explicitly illustrate the intervention processes of DTRs on similar patients, we employ a co-guided graph to connect various patient sequences. In addition, to fully utilize the time interval information, we design a time-aware guidance mechanism dedicated to the co-guided graph to efficiently learn medication representation using the patient's guidance information. We also introduce relative time intervals in the encoder to act as positional information. Extensive experiments on two real-world datasets demonstrate that MR-DTR surpasses state-of-the-art models in terms of recommendation performance. Our code is available at: https://github.com/liyifo/MR-DTR.

Mental health concerns are a critical issue in today’s digital age, posing a threat to both individual and societal well-being and making the identification of at-risk individuals crucial. Analyzing an individual’s social media post history can offer insights into their mental health state and help identify the presence of suicidal ideation. However, the complexity of linguistic and temporal data, along with sparsity and time irregularities, poses a formidable challenge in machine learning. Previous methods in this domain either rely on Euclidean space for processing which does not adequately model the power-law properties of social media posts, or lose information due to the discretization of the time axis. To address these challenges, we propose a novel framework, THYMES, which leverages pre-trained encoders and a rich representation learning paradigm with hyperbolic learning to model power-law features for enhanced sequence modeling. We perform experiments on two datasets and demonstrate that THYMES outperforms previously proposed methods while maintaining classification fairness under heavy data imbalances. Additionally, we qualitatively analyze commonly misclassified samples to reveal the shortcomings of models in this domain.

The rising rate of mental health issues in the digital age underscores the critical need for proactive interventions to assess an individual’s well-being. This problem is further exacerbated by the social stigma surrounding the subject, which suppresses the willingness of victims to seek help. Social media can serve as an outlet for such individuals to express their negative emotions or thoughts of self-harm. The social media account of an individual can offer a plethora of valuable information that can be used to predict their mental health. By unifying principles of robust classifier training and selective classification, we propose a novel framework, SAFENet, to predict the suicide risk of users by using their historical social media posts. When the confidence of prediction is low or the individual is classified as a high-risk user, SAFENet delegates the analysis of the posts to a human evaluator for further intervention. Our experiments show that SAFENet outperforms existing state-of-the-art frameworks. We further qualitatively analyze predictions from SAFENet and demonstrate that it performs robustly on difficult samples that may cause contemporary methods to make errors. Our system addresses the urgent need for efficient and effective mental health intervention in the digital era.

The proliferation of social media platforms has fueled the rapid dissemination of fake news, posing threats to our real-life society. Existing methods use multimodal data or contextual information to enhance the detection of fake news by analyzing news content and/or its social context. However, these methods often overlook essential textual news content (articles) and heavily rely on sequential modeling and global attention to extract semantic information. These existing methods fail to handle the complex, subtle twists1 in news articles, such as syntax-semantics mismatches and prior biases, leading to lower performance and potential failure when modalities or social context are missing. To bridge these significant gaps, we propose a novel multi-hop syntax aware fake news detection (MSynFD) method, which incorporates complementary syntax information to deal with subtle twists in fake news. Specifically, we introduce a syntactical dependency graph and design a multi-hop subgraph aggregation mechanism to capture multi-hop syntax. It extends the effect of word perception, leading to effective noise filtering and adjacent relation enhancement. Subsequently, a sequential relative position-aware Transformer is designed to capture the sequential information, together with an elaborate keyword debiasing module to mitigate the prior bias. Extensive experimental results on two public benchmark datasets verify the effectiveness and superior performance of our proposed MSynFD over state-of-the-art detection models.

Multivariate time series (MTS) forecasting is crucial in many real-world applications. To achieve accurate MTS forecasting, it is essential to simultaneously consider both intra- and inter-series relationships among time series data. However, previous work has typically modeled intra- and inter-series relationships separately and has disregarded multi-order interactions present within and between time series data, which can seriously degrade forecasting accuracy. In this article, we reexamine intra- and inter-series relationships from the perspective of mutual information and accordingly construct a comprehensive relationship learning mechanism tailored to simultaneously capture the intricate multi-order intra- and inter-series couplings. Based on the mechanism, we propose a novel deep coupling network for MTS forecasting, named DeepCN, which consists of a coupling mechanism dedicated to explicitly exploring the multi-order intra- and inter-series relationships among time series data concurrently, a coupled variable representation module aimed at encoding diverse variable patterns, and an inference module facilitating predictions through one forward step. Extensive experiments conducted on seven real-world datasets demonstrate that our proposed DeepCN achieves superior performance compared with the state-of-the-art baselines.

Patent documents in the patent database (PatDB) are crucial for research, development, and innovation as they contain valuable technical information. However, PatDB presents a multifaceted challenge in comparison to publicly available preprocessed databases due to the intricate nature of patent text and the inherent sparsity within the patent citation network. Although patent text analysis and citation analysis bring new opportunities to explore patent data mining, no existing work exploits the complementation of them. To this end, we propose a joint semantic-topological evolutionary graph learning approach (PatSTEG) to model the formation dynamics of patent citation networks. More specifically, we first create a real-world dataset of Chinese patents named CNPat, and leveraging its patent texts and citations to construct a patent citation network. Then, PatSTEG is modeled to study the evolutionary dynamics of patent citation formation by jointly considering the semantic and topological information. Extensive experiments are conducted on both CNPat and public datasets to prove the superiority of PatSTEG over other state-of-the-art methods. All the results provide valuable references for patent literature research and technical exploration.

Fashion recommendation (FR) has received increasing attention in the research of new types of recommender systems. Existing fashion recommender systems (FRSs) typically focus on clothing item suggestions for users in three scenarios: 1) how to best recommend fashion items preferred by users; 2) how to best compose a complete outfit, and 3) how to best complete a clothing ensemble. However, current FRSs often overlook an important aspect when making FR, that is, the compatibility of the clothing item or outfit recommendations is highly dependent on the scene context. To this end, we propose the scene-aware fashion recommender system (SAFRS), which uncovers a hitherto unexplored avenue where scene information is taken into account when constructing the FR model. More specifically, our SAFRS addresses this problem by encoding scene and outfit information in separation attention encoders and then fusing the resulting feature embeddings via a novel scene-aware compatibility score function. Extensive qualitative and quantitative experiments are conducted to show that our SAFRS model outperforms all baselines for every evaluated metric.

Users’ purchase behaviors are complex and dynamic, which are usually driven by various personal demands evolving with time. According to psychology and economic theories, user demands can be satisfied with a sequence of purchase behaviors, resulting in a basket of items. However, most of the existing works simply predict the next basket from a shallow perspective of (purchase) sequence data modeling without deep insight into the underlying factors which drive user purchase behaviors. In fact, filling a basket with multiple items is a process to incrementally satisfy a user's demand. Therefore, the key challenges to predict a user's next basket lie in (1) how to track the changes of the user's demand, and (2) how to satisfy her demand at a given moment. To this end, we propose an Evolving DEmand SAtisfaction (EvoDESA) model to model a user's demand evolution for next-basket prediction. In EvoDESA, a demand evolution module learns the dynamics of user demand over a sequence of basket-purchase behaviors. Then, a next-basket planning module effectively packs an optimal combination of items to best satisfy the user's current demand. Extensive experiments on three real-world transaction datasets demonstrate the considerable superiority of EvoDESA over the state-of-the-art approaches.

In recent years, sequential recommender systems (SRSs) and session-based recommender systems (SBRSs) have emerged as a new paradigm of RSs to capture users' short-term but dynamic preferences for enabling more timely and accurate recommendations. Although SRSs and SBRSs have been extensively studied, there are many inconsistencies in this area caused by the diverse descriptions, settings, assumptions and application domains. There is no work to provide a unified framework and problem statement to remove the commonly existing and various inconsistencies in the area of SR/SBR. There is a lack of work to provide a comprehensive and systematic demonstration of the data characteristics, key challenges, most representative and state-of-the-art approaches, typical real- world applications and important future research directions in the area. This work aims to fill in these gaps so as to facilitate further research in this exciting and vibrant area.

Short-head and long-tail distributed data are widely observed in the real world. The same is true of recommender systems (RSs), where a small number of popular items dominate the choices and feedback data while the rest only account for a small amount of feedback. As a result, most RS methods tend to learn user preferences from popular items since they account for most data. However, recent research in e-commerce and marketing has shown that future businesses will obtain greater profit from long-tail selling. Yet, although the number of long-tail items and users is much larger than that of short-head items and users, in reality, the amount of data associated with long-tail items and users is much less. As a result, user preferences tend to be popularity-biased. Furthermore, insufficient data makes long-tail items and users more vulnerable to shilling attack. To improve the quality of recommendations for items and users in the tail of distribution, we propose a coupled regularization approach that consists of two latent factor models: C-HMF, for enhancing credibility, and S-HMF, for emphasizing specialty on user choices. Specifically, the estimates learned from C-HMF and S-HMF recurrently serve as the empirical priors to regularize one another. Such coupled regularization leads to the comprehensive effects of final estimates, which produce more qualitative predictions for both tail users and tail items. To assess the effectiveness of our model, we conduct empirical evaluations on large real-world datasets with various metrics. The results prove that our approach significantly outperforms the compared methods.

In the real-world environment, users have sufficient experience in their focused domains but lack experience in other domains. Recommender systems are very helpful for recommending potentially desirable items to users in unfamiliar domains, and cross-domain collaborative filtering is therefore an important emerging research topic. However, it is inevitable that the cold-start issue will be encountered in unfamiliar domains due to the lack of feedback data. The Bayesian approach shows that priors play an important role when there are insufficient data, which implies that recommendation performance can be significantly improved in cold-start domains if informative priors can be provided. Based on this idea, we propose a Weighted Irregular Tensor Factorization (WITF) model to leverage multi-domain feedback data across all users to learn the cross-domain priors w.r.t. both users and items. The features learned from WITF serve as the informative priors on the latent factors of users and items in terms of weighted matrix factorization models. Moreover, WITF is a unified framework for dealing with both explicit feedback and implicit feedback. To prove the effectiveness of our approach, we studied three typical real-world cases in which a collection of empirical evaluations were conducted on real-world datasets to compare the performance of our model and other state-of-the-art approaches. The results show the superiority of our model over comparison models.

Document similarity analysis is increasingly critical since roughly 80% of big data is unstructured. Accordingly, semantic couplings (relatedness) have been recognized valuable for capturing the relationships between terms (words or phrases). Existing work focuses more on explicit relatedness, with respective models built. In this paper, we propose a comprehensive semantic similarity measure: Semantic Coupling Similarity (SCS), which (1) captures intra-term pair couplings within term pairs represented by patterns of explicit term co-occurrences in a document set, (2) extracts inter-term pair couplings between term pairs indicated by implicit couplings between term pairs through indirectly linked terms and paths between terms after term connections are converted to a graph presentation; and (3) semantic coupling similarity, integrating intra- and inter-term pair couplings towards a comprehensive capturing of explicit and implicit couplings between terms across documents. SCS caters for both synonymy and polysemy, and outperforms baseline methods consistently on all real data sets.

Choice modeling (CM) aims to describe and predict choices according to attributes of subjects and options. If we presume each choice making as the formation of link between subjects and options, immediately CM can be bridged to link analysis and prediction (LAP) problem. However, such a mapping is often not trivial and straightforward. In LAP problems, the only available observations are links among objects but their attributes are often inaccessible. Therefore, we extend CM into a latent feature space to avoid the need of explicit attributes. Moreover, LAP is usually based on binary linkage assumption that models observed links as positive instances and unobserved links as negative instances. Instead, we use a weaker assumption that treats unobserved links as pseudo negative instances. Furthermore, most subjects or options may be quite heterogeneous due to the long-tail distribution, which is failed to capture by conventional LAP approaches. To address above challenges, we propose a Bayesian heteroskedastic choice model to represent the non-identically distributed linkages in the LAP problems. Finally, the empirical evaluation on real-world datasets proves the superiority of our approach.

Collaborative filtering (CF) is a major technique in recommender systems to help users find their potentially desired items. Since the data sparsity problem is quite commonly encountered in real-world scenarios, Cross-Domain Collaborative Filtering (CDCF) hence is becoming an emerging research topic in recent years. However, due to the lack of sufficient dense explicit feedbacks and even no feedback available in users' uninvolved domains, current CDCF approaches may not perform satisfactorily in user preference prediction. In this paper, we propose a generalized Cross Domain Triadic Factorization (CDTF) model over the triadic relation user-item-domain, which can better capture the interactions between domain-specific user factors and item factors. In particular, we devise two CDTF algorithms to leverage user explicit and implicit feedbacks respectively, along with a genetic algorithm based weight parameters tuning algorithm to trade off influence among domains optimally. Finally, we conduct experiments to evaluate our models and compare with other state-of-the-art models by using two real world datasets. The results show the superiority of our models against other comparative models.