The evolution of large vision-language models (LVLMs) has shed light on the development of many fields, particularly for multimodal recommendation. While LVLMs offer an integrated understanding of textual and visual information of items from user interactions, their deployment in this domain remains limited due to inherent complexities. First, LVLMs are trained from enormous general datasets and lack knowledge of personalized user preferences. Second, LVLMs struggle with multiple image processing, especially with discrete, noisy, and redundant images in recommendation scenarios. To address these issues, we introduce a new reasoning strategy called Visual-Summary Thought (VST) for Multimodal Recommendation. This approach begins by prompting LVLMs to generate textual summaries of item images, which serve as contextual information. These summaries are then combined with item titles to enhance the representation of sequential interactions and improve the ranking of candidates. Our experiments, conducted across four datasets using three different LVLMs: GPT4-V, LLaVA-7b, and LLaVA-13b validate the effectiveness of VST.

Recommender systems (RecSys) aim to predict users’ preferences based on historical interactions and content profiles, and they are vital components of many online services. However, the strict cold-start (SCS) issue, i.e., users/items have no prior interactions, poses significant challenges for RecSys. The existing methods seek to transfer content knowledge, collaborative filtering (CF) knowledge, or combine the two from the warm-start scenario towards the (strict) cold-start scenarios. However, these approaches either ignore the available information or model the information in rough manners such that the two types of knowledge interfere with each other, leading to ineffective and uncontrolled knowledge transfer. In this work, we propose a novel dual-teacher knowledge distillation (DTKD) framework that simultaneously and effectively transfers both content and CF knowledge. The proposed DTKD framework contains two teachers, one for each knowledge type, that is specifically designed according to the characteristics of the content and CF data to distill the knowledge fully. Soft scoring is calculated during the distillation to denoise and augment the original hard-labeled interactions. A knowledge fusion module is then proposed to collect the consensus of the two teachers’ opinions. Finally, DTKD transfers both content and CF knowledge into a student module that learns the shared viewpoints of the teachers. We conduct extensive experiments on real-world datasets under the warm-start as well as three different SCS settings (i.e., strict cold users, strict cold items, and strict cold users & items). Experimental results show that DTKD outperforms strong baselines by large margins under all settings, especially the SCS ones.

Recommendation systems suffer in the strict cold-start (SCS) scenario, where the user-item interactions are entirely unavailable. The well-established, dominating identity (ID)-based approaches completely fail to work. Cold-start recommenders, on the other hand, leverage item contents (brand, title, descriptions, etc.) to map the new items to the existing ones. However, the existing SCS recommenders explore item contents in coarse-grained manners that introduce noise or information loss. Moreover, informative data sources other than item contents, such as users’ purchase sequences and review texts, are largely ignored. In this work, we explore the role of the fine-grained item attributes in bridging the gaps between the existing and the SCS items and pre-train a knowledgeable item-attribute graph for SCS item recommendation. Our proposed framework, ColdGPT, models item-attribute correlations into an item-attribute graph by extracting fine-grained attributes from item contents. ColdGPT then transfers knowledge into the item-attribute graph from various available data sources, i.e., item contents, historical purchase sequences, and review texts of the existing items, via multi-task learning. To facilitate the positive transfer, ColdGPT designs specific submodules according to the natural forms of the data sources and proposes to coordinate the multiple pre-training tasks via unified alignment-and-uniformity losses. Our pre-trained item-attribute graph acts as an implicit, extendable item embedding matrix, which enables the SCS item embeddings to be easily acquired by inserting these items into the item-attribute graph and propagating their attributes’ embeddings. We carefully process three public datasets, i.e., Yelp, Amazon-home, and Amazon-sports, to guarantee the SCS setting for evaluation. Extensive experiments show that ColdGPT consistently outperforms the existing SCS recommenders by large margins and even surpasses models that are pre-trained on 75 - 224 times more, cross-domain data on two out of four datasets. Our code and pre-processed datasets for SCS evaluations are publicly available to help future SCS studies.

GNNs have been widely used in deep learning on graphs. They learn effective node representations. However, most methods ignore the heterogeneity. Methods designed for heterogeneous graphs, on the other hand, fail to learn complex semantic representations because they only use meta-paths instead of meta-graphs. Furthermore, they cannot fully capture the content-based correlations, as they either do not use the self-attention mechanism or only use it to consider the immediate neighbors of each node, ignoring the higher-order neighbors. We propose a novel Higher-order Attribute-Enhancing (HAE) framework enhancing node embedding in a layer-by-layer manner. Under the HAE framework, we propose a Higher-order Attribute-Enhancing GNN (HAE\textsubscript{GNN}) for heterogeneous network embeding. HAE\textsubscript{GNN} simultaneously incorporates meta-paths and meta-graphs for rich, heterogeneous semantics, and leverages the self-attention mechanism to explore content-based nodes' interactions. The unique higher-order architecture of HAE\textsubscript{GNN} allows examining the first-order as well as higher-order neighborhoods. Moreover, HAE\textsubscript{GNN} shows good explainability as it learns the importances of different meta-paths and meta-graphs. HAE\textsubscript{GNN} is also memory-efficient, for it avoids per meta-path based matrix calculation. Experimental results not only show HAE\textsubscript{GNN}'s superior performance against the state-of-the-art methods in node classification, node clustering, and visualization, but also demonstrate its superiorities in terms of memory efficiency and explainability.

State-of-the-art Graph Neural Networks (GNNs) have achieved tremendous success in social event detection tasks when restricted to a closed set of events. However, considering the large amount of data needed for training and the limited ability of a neural network in handling previously unknown data, it is hard for existing GNN-based methods to operate in an open set setting. To address this problem, we design a Quality-aware Self-improving Graph Neural Network (QSGNN) which extends the knowledge from known to unknown by leveraging the best of known samples and reliable knowledge transfer. Specifically, to fully exploit the labeled data, we propose a novel supervised pairwise loss with an additional orthogonal inter-class relation constraint to train the backbone GNN encoder. The learnt, already-known events further serve as strong reference bases for the unknown ones, which greatly prompts knowledge acquisition and transfer. When the model is generalized to unknown data, to ensure the effectiveness and reliability, we further leverage the reference similarity distribution vectors for pseudo pairwise label generation, selection and quality assessment. Following the diversity principle of active learning, our method selects diverse pair samples with the generated pseudo labels to fine-tune the GNN encoder. Besides, we propose a novel quality-guided optimization in which the contributions of pseudo labels are weighted based on consistency. Experimental results validate that our model achieves state-of-the-art results and extends well to unknown events.

Social events provide valuable insights into group social behaviors and public concerns and therefore have many applications in fields such as product recommendation and crisis management. The complexity and streaming nature of social messages make it appealing to address social event detection in an incremental learning setting, where acquiring, preserving, and extending knowledge are major concerns. Most existing methods, including those based on incremental clustering and community detection, learn limited amounts of knowledge as they ignore the rich semantics and structural information contained in social data. Moreover, they cannot memorize previously acquired knowledge. In this paper, we propose a novel Knowledge-Preserving Incremental Heterogeneous Graph Neural Network (KPGNN) for incremental social event detection. To acquire more knowledge, KPGNN models complex social messages into unified social graphs to facilitate data utilization and explores the expressive power of GNNs for knowledge extraction. To continuously adapt to the incoming data, KPGNN adopts contrastive loss terms that cope with a changing number of event classes. It also leverages the inductive learning ability of GNNs to efficiently detect events and extends its knowledge from previously unseen data. To deal with large social streams, KPGNN adopts a mini-batch subgraph sampling strategy for scalable training, and periodically removes obsolete data to maintain a dynamic embedding space. KPGNN requires no feature engineering and has few hyperparameters to tune. Extensive experiment results demonstrate the superiority of KPGNN over various baselines.