Generalized zero-shot learning (GZSL) has achieved significant progress, with many efforts dedicated to over-coming the problems of visual-semantic domain gap and seen-unseen bias. However, most existing methods directly use feature extraction models trained on ImageNet alone, ignoring cross-dataset bias between GZSL benchmarks. Such a inevitably results in poor-quality visual features for tasks, which potentially limits recognition performance both seen unseen classes. In this paper, we...

Recommender systems usually amplify the biases in data. The model learned from historical interactions with imbalanced item distribution will imbalance by over-recommending items majority groups. Addressing this issue is essential for a healthy ecosystem of recommendation long run. Existing work applies bias control to ranking targets (e.g., calibration, fairness, and diversity), but ignores true reason amplification trades off accuracy.

Large Language Models (LLMs) have demonstrated remarkable performance across diverse domains, thereby prompting researchers to explore their potential for use in recommendation systems. Initial attempts leveraged the exceptional capabilities of LLMs, such as rich knowledge and strong generalization through In-context Learning, which involves phrasing task prompts. Nevertheless, LLMs tasks remains suboptimal due a substantial disparity between training tasks, well inadequate data during...

Generative models such as Adversarial Networks (GANs) and Variational Auto-Encoders (VAEs) are widely utilized to model the generative process of user interactions. However, they suffer from intrinsic limitations instability GANs restricted representation ability VAEs. Such hinder accurate modeling complex interaction generation procedure, noisy interactions caused by various interference factors. In light impressive advantages Diffusion Models (DMs) over traditional in image synthesis, we...

The remarkable achievements of Large Language Models (LLMs) have led to the emergence a novel recommendation paradigm — Recommendation via LLM (RecLLM). Nevertheless, it is important note that LLMs may contain social prejudices, and therefore, fairness recommendations made by RecLLM requires further investigation. To avoid potential risks RecLLM, imperative evaluate with respect various sensitive attributes on user side. Due differences between traditional paradigm, problematic directly use...

Recommender systems have become crucial in information filtering nowadays. Existing recommender extract user preferences based on the correlation data, such as behavioral collaborative filtering, feature-feature, or feature-behavior click-through rate prediction. However, unfortunately, real world is driven by causality , not just correlation, and does imply causation. For instance, might recommend a battery charger to after buying phone, where latter can serve cause of former; causal...

Modern recommender systems learn user representations from historical interactions, which suffer the problem of feature shifts, such as an income increase. Historical interactions will inject out-of-date information into representation in conflict with latest feature, leading to improper recommendations. In this work, we consider Out-Of-Distribution (OOD) recommendation OOD environment shifts. To pursue high fidelity, set additional objectives for learning as: 1) strong generalization and 2)...

On the shoulders of textual dialog systems, multimodal ones, recently have engaged increasing attention, especially in retail domain. Despite commercial value they still suffer from following challenges: 1) automatically generate right responses appropriate medium forms; 2) jointly consider visual cues and side information while selecting product images; 3) guide response generation with multi-faceted heterogeneous knowledge. To address aforementioned issues, we present a Multimodal diAloG...

Online micro-video recommender systems aim to address the information explosion of micro-videos and make personalized recommendation for users. However, existing methods still have some limitations in learning representative user interests, since multi-scale time effects, interest group modeling, false positive interactions are not taken into consideration. In view this, we propose an end-to-end Multi-scale Time-aware Interest modeling Network (MTIN). particular, first present routing...

Conversational image search, a revolutionary search mode, is able to interactively induce the user response clarify their intents step by step. Several efforts have been dedicated conversation part, namely automatically asking right question at time for preference elicitation, while few studies focus on part given well-prepared conversational query. In this paper, we work towards which much difficult compared traditional task, due following challenges: 1) understanding complex from...

Existing studies on multimodal sentiment analysis heavily rely textual modality and unavoidably induce the spurious correlations between words labels. This greatly hinders model generalization ability. To address this problem, we define task of out-of-distribution (OOD) analysis. aims to estimate mitigate bad effect for strong OOD generalization. end, embrace causal inference, which inspects relationships via a graph. From graph, find that are attributed direct prediction while indirect one...

The past decade has witnessed the boom of human-machine interactions, particularly via dialog systems. In this paper, we study task response generation in open-domain multi-turn Many research efforts have been dedicated to building intelligent systems, yet few shed light on deepening or widening chatting topics a conversational session, which would attract users talk more. To end, paper presents novel deep scheme consisting three channels, namely global, wide, and ones. global channel...

As an intelligent way to interact with computers, the dialog system has been catching more and attention. However, most research efforts only focus on text-based systems, completely ignoring rich semantics conveyed by visual cues. Indeed, desire for multimodal task-oriented systems is growing rapid expansion of many domains, such as online retailing travel. Besides, few work considers hierarchical product taxonomy users' attention products explicitly. The fact that users tend express their...

The ubiquity of implicit feedback makes them the default choice to build online recommender systems. While large volume alleviates data sparsity issue, downside is that they are not as clean in reflecting actual satisfaction users. For example, E-commerce, a portion clicks do translate purchases, and many purchases end up with negative reviews. As such, it critical importance account for inevitable noises training. However, little work on recommendation has taken noisy nature into...

Recommender systems usually face the issue of filter bubbles: overrecommending homogeneous items based on user features and historical interactions. Filter bubbles will grow along feedback loop inadvertently narrow interests. Existing work mitigates by incorporating objectives apart from accuracy such as diversity fairness. However, they typically sacrifice accuracy, hurting model fidelity experience. Worse still, users have to passively accept recommendation strategy influence system in an...

Most recommender systems evaluate model performance offline through either: 1) normal biased test on factual interactions; or 2) debiased with records from the randomized controlled trial. In fact, both tests only reflect part of whole picture: interactions are collected recommendation policy, fitting them better implies benefiting platform higher click conversion rate; in contrast, eliminates system-induced biases and thus is more reflective user true preference. Nevertheless, we find that...

Hypergraph Convolutional Network (HCN) has be-come a proper choice for capturing high-order relationships. Existing HCN methods are tailored static hypergraphs, which unsuitable the dynamic evolution in real-world scenarios. In this paper, we explore based on attention mechanism (DyHCN) time series prediction. It not only effectively exploits spatial and temporal relationships hypergraph, but also continuously aggregates cues of time-varying hypergraphs with global local embeddings....

Data-driven recommender systems have demonstrated great success in various Web applications owing to the extraordinary ability of machine learning models recognize patterns (ie correlation) from users' behaviors. However, they still suffer several issues such as biases and unfairness due spurious correlations. Considering causal mechanism behind data can avoid influences In this light, embracing modeling is an exciting promising direction.

Multimodal information (e.g., visual, acoustic, and textual) has been widely used to enhance representation learning for micro-video recommendation. For integrating multimodal into a joint of micro-video, fusion plays vital role in the existing recommendation approaches. However, static previous studies is insufficient model various relationships among different micro-videos. In this article, we develop novel meta-learning-based framework called Meta Fusion (MetaMMF), which dynamically...

Existing work on Multimodal Sentiment Analysis (MSA) utilizes multimodal information for prediction yet unavoidably suffers from fitting the spurious correlations between features and sentiment labels. For example, if most videos with a blue background have positive labels in dataset, model will rely such prediction, while "blue background'' is not sentiment-related feature. To address this problem, we define general debiasing MSA task, which aims to enhance Out-Of-Distribution (OOD)...

Climate Change EconomicsAccepted Papers No AccessTemperature Variation, Health, and Private Medical Costs: Evidence from ChinaMingyang Zhang, Wenjie Wang, Rui Hu, Zhiqiang Cheng, Jia Li, Xiaoxiao ZhangMingyang Wang Search for more papers by this author , Hu Cheng Li Zhang https://doi.org/10.1142/S2010007825500010Cited by:0 (Source: Crossref) Next AboutFiguresReferencesRelatedDetailsPDF/EPUB ToolsAdd to favoritesDownload CitationsTrack CitationsRecommend Library ShareShare...