With the booming development of multimodal data (e.g., image, text) on internet platforms, multimodal sequential recommendation methods continue to emerge. Most existing methods incorporate item modal features as auxiliary information, typically concatenating them to learn unified user representations. However, these methods directly use modal features for representation learning, neglecting the impact of inherent modal noise. We argue that internal-modal noise and cross-modal noise hinder the acquisition of more accurate user representations. To address this problem, we propose SGP4SR - Separated-modality Guided user Preference learning for multimodal Sequential Recommendation. Globally, the user preference modeling is carried out from a separated-modality perspective to alleviate cross-modal noise. Locally, for each individual modality, we use item relationship graphs and user interest centers, aggregated with ID embeddings, to replace direct modal features, thereby mitigating internal-modal noise. Finally, user representations from both separated-modality and multimodal perspectives participate in prediction independently. In experiments conducted on four real-world datasets, our method outperforms state-of-the-art approaches, achieving an average performance improvement of up to 8.84% over the best baseline. The comprehensive experiments further validate the superior noise tolerance and robustness of our method.

Visual Commonsense Reasoning (VCR) is a cognitive task, challenging models to answer visual questions, and to explain the rationale behind their answers. While Large Language Models (LLMs) offer potential for this task, VCR’s complex scenes require specialized approaches to activate their commonsense reasoning abilities, as existing Multimodal LLMs struggle with VCR’s visual events and unique reference tags. To address these challenges, we propose EventLens, which enhances VCR through Event-Aware Pretraining and Cross-modal Linking in Supervised Fine-tuning. First, we introduce a new pretraining stage that emulates human cognitive processes to improve LLM comprehension of complex scenarios. Second, during supervised fine-tuning, we leverage reference tags to explicitly bridge RoI features with text, maintaining semantic integrity across modalities. Additionally, instruct prompts and task-specific adapters help integrate LLMs’ knowledge with new commonsense reasoning. Experimental results demonstrate competitive performance with state-of-the-art methods and ablation studies verify the effectiveness of proposed EventLens.

Set similarity search, as a foundational operation in data processing with diverse applications in different domains, has been extensively studied. However, in the era of big data where sets sizes and quantities are rapidly increasing, set similarity search suffers from significant computational and storage overheads. Additionally, traditional approaches struggle to universally address the search problem across different similarity measures and query types. To tackle these challenges, AI techniques, with their powerful learning capabilities, may provide a viable solution. In this paper, we first propose a multi-task representation learning approach with box embeddings that accurately simulates different similarity measures simultaneously by estimating the overlap and union relationships between set pairs in latent box space. Based on the compressed representations of sets, we then introduce a universal search approach designed to answer various set similarity queries with parallel implementation. Extensive experiments conducted on real-world datasets demonstrate the universality, accuracy and efficiency of the proposed approach, showing that it outperforms competing methods. For reproduction, we release our source code on https://github.com/yangzhong901/MTBUS.

The string similarity search is core functionality in a range of applications, including data cleaning, near-duplicate object detection, and data integration. We study the problem of threshold similarity search with the edit distance, where given a set of strings, a threshold$k$, and a query string$q$, we aim to find all strings in the set whose edit distances to$q$are no larger than$k$. Extensive studies have been proposed for the threshold similarity search problem with the edit distance. However, they suffer from a huge space consumption issue when achieving only an acceptable efficiency, especially for long strings. In this paper, we propose a simple yet small index, called minIL, to eliminate this issue. First, we adopt a minhash family to capture pivot characters and to construct sketch representations for strings. Second, we develop a multi-level inverted index to search sketches with a low space consumption. Finally, we apply a novel learned index technique on top of the index that further improves the query efficiency. Extensive experiments on real-world datasets offer insight into the performance of our method and show that it substantially reduces the index size, and is capable of outperforming the baseline approaches.

Streaming spatial-textual data that contains geographic and textual information, e.g., geo-tagged tweets, has an unprecedented increase in amount. As one of the basic operations, the continuous spatial-textual queries that retrieve real-time results continuously on large-scale spatial-textual streams call for means of efficient distributed processing. However, existing proposals either are spatialaware only, or superficially exploit textual information for pruning. We propose a distributed system, called HASTE, for hybrid and adaptive processing on streaming spatial-textual data. The novelty lies on three aspects: (1) We propose a novel method to reduce the workload beforehand by dividing objects and queries into mutually exclusive types; (2) We develop a novel load partitioning strategy and a novel cost model that consider both spatial and textual properties; (3) We design a multi-level load adjustment strategy that adaptively copes with different degrees of load imbalance. We report on extensive experiments with real-world data that offer insight into the performance of the solution, and show that the solution is capable of outperforming the state-of-the-art proposals.

The set similarity join (SSJ) is core functionality in a range of applications, including data cleaning, near-duplicate object detection, and data integration. Threshold-based SSJ queries return all pairs of sets with similarity no smaller than a given threshold. As results, and their utility, are very sensitive to the choice of threshold value, it is a problem that it is difficult to choose such an appropriate value. Doing so requires prior knowledge of the data, which users often do not have. To avoid this problem, we propose a solution to the top-k overlap set similarity join (TkOSSJ) that returns k pairs of sets with the highest overlap similarities. The state-of-the-art solution disregards the effect of the so-called step size, which is the number of elements accessed in each iteration of the algorithm. This affects its performance negatively. To address this issue, we first propose an algorithm that uses a fixed step size, thus taking advantage of the benefits of a large step size, and then we present an adaptive step size algorithm that is capable of automatically adjusting the step size, thus reducing redundant computations. An extensive empirical study offers insight into the new algorithms and indicates that they are capable of outperforming the state-of-the-art method on real, large-scale data sets.