With the development of the Internet and the explosive expansion of information volume, it becomes arduous for users to make a choice in the face of overwhelming information. This makes the recommendation system an effective solution for the problem of information overload. Recommendation systems can acquire users' personalized preferences by comprehending their interactive behavior, thereby providing accurate recommendations. However, recommendation systems invariably encounter issues such as data sparsity and cold start. The introduction of auxiliary information can effectively alleviate these problems. Meanwhile, the employment of attention mechanisms can assist the system in more accurately understand user behavior and preference information, thus providing more precise personalized recommendations. In this paper, the research progress of knowledge graphs and attention mechanisms in recommendation systems is explored through top journals and several Chinese papers.

We introduce Presto, a novel video diffusion model designed to generate 15-second videos with long-range coherence and rich content. Extending video generation methods to maintain scenario diversity over long durations presents significant challenges. To address this, we propose a Segmented Cross-Attention (SCA) strategy, which splits hidden states into segments along the temporal dimension, allowing each segment to cross-attend to a corresponding sub-caption. SCA requires no additional parameters, enabling seamless incorporation into current DiT-based architectures. To facilitate high-quality long video generation, we build the LongTake-HD dataset, consisting of 261k content-rich videos with scenario coherence, annotated with an overall video caption and five progressive sub-captions. Experiments show that our Presto achieves 78.5% on the VBench Semantic Score and 100% on the Dynamic Degree, outperforming existing state-of-the-art video generation methods. This demonstrates that our proposed Presto significantly enhances content richness, maintains long-range coherence, and captures intricate textual details. More details are displayed on our project page: presto-video.github.io.

Chinese characters embody a uniquely tight coupling between visual form and semantics that mainstream NLP models largely ignore. We introduce Radical-Graphormer, a heterogeneous-graph variational auto-encoder that captures radical-level geometry and semantics in a unified latent space. Leveraging HanziFormSem—a newly 1,832-character corpus aligning Ideographic Description Sequences with multi-source meanings—the model tackles two dual tasks: (i) predicting semantics from structure and (ii) synthesizing IDS glyphs from semantic vectors. It surpasses strong glyph-aware baselines with a Macro-F1 of 0.81 on semantic classification, and produces novel characters with 95 % structural validity and a 4.1 / 5.0 human plausibility score. These findings confirm that fine-grained layout is a powerful predictor of meaning and that controllable, semantics-driven character creation is feasible. The framework’s design principles—heterogeneous graphs, contrastive alignment, and grammar-constrained decoding—are readily transferable to other logographic or iconographic writing systems, opening new avenues for AI-assisted design and digital humanities.

Large language models have demonstrated impressive fluency in artistic description, yet distinguishing semantic abstraction and creativity between large-language-model-generated and human texts remains a challenge. This study systematically compares human and large-language-model-generated calligraphy descriptions through multi-dimensional analyses, including word vectors, syntax, sentiment, metaphor detection, and cross-modal alignment. An optimised BERT-based classifier and BLIP2 closed-loop experiments are proposed to capture stylistic and semantic distinctions. Results show that AI-generated texts exhibit higher syntactic complexity (Avg. sentence length: 18.2 vs. 13.0) and lexical diversity (10.07 vs. 5.26), but lag in creativity (intra-group similarity: 0.781 vs. 0.505) and semantic reduction in cross-modal tasks (cycle accuracy: 21.7% vs. 35.3%). While AI excels in structured expression and consistency, it remains constrained by training patterns. In contrast, human texts demonstrate greater originality via metaphorical and contextual richness. To enhance classification, a whole-word-masked BERT classifier with gated mean pooling and contrastive loss is introduced. Cross-modal experiments using CLIP and BLIP2 further reveal differences in image-text alignment and reconstructability. This study provides a new benchmark and methodological toolkit for analysing generative semantics in multimodal cultural contexts.

Story visualization aims to create visually compelling images or videos corresponding to textual narratives. Despite recent advances in diffusion models yielding promising results, existing methods still struggle to create a coherent sequence of subject-consistent frames based solely on a story. To this end, we propose DreamStory, an automatic open-domain story visualization framework by leveraging the LLMs and a novel multi-subject consistent diffusion model. DreamStory consists of (1) an LLM acting as a story director and (2) an innovative Multi-Subject consistent Diffusion model (MSD) for generating consistent multi-subject across the images. First, DreamStory employs the LLM to generate descriptive prompts for subjects and scenes aligned with the story, annotating each scene's subjects for subsequent subject-consistent generation. Second, DreamStory utilizes these detailed subject descriptions to create portraits of the subjects, with these portraits and their corresponding textual information serving as multimodal anchors (guidance). Finally, the MSD uses these multimodal anchors to generate story scenes with consistent multi-subject. Specifically, the MSD includes Masked Mutual Self-Attention (MMSA) and Masked Mutual Cross-Attention (MMCA) modules. MMSA module ensures detailed appearance consistency with reference images, while MMCA captures key attributes of subjects from their reference text to ensure semantic consistency. Both modules employ masking mechanisms to restrict each scene's subjects to referencing the multimodal information of the corresponding subject, effectively preventing blending between multiple subjects. To validate our approach and promote progress in story visualization, we established a benchmark, DS-500, which can assess the overall performance of the story visualization framework, subject-identification accuracy, and the consistency of the generation model. Extensive experiments validate the effectiveness of DreamStory in both subjective and objective evaluations.

Industrial data collection is the foundation for implementing enterprise digitization, which is of great significance to the development of intelligent manufacturing. However, Industrial Data Collection Standards (IDCS) are primarily published in paper or PDF formats, which makes it challenging to associate and reuse knowledge. This brings difficulties to data collection and management of heterogeneous devices, thus affecting the adaptive collection of industrial data. For this reason, this paper introduces SMART into industrial data collection and proposes the construction and application of a knowledge graph-based SMART Model. First, ontology semantic reasoning and natural language processing techniques are utilized to develop an ontology model for industrial data collection and extract fine-grained IDCS knowledge. Then, the knowledge is integrated according to the standard primitive structure and conceptual reasoning rules by constructing a standard association model to form the Industrial Data Collection Standards Knowledge Graph (IDCS-KG). Finally, the SMART competence level is quantitatively assessed based on the completion degree of each operation within the SMART Model. An experimental case study demonstrates that the SMART Model can collect intelligent adaptive industrial data through the reasoning and analysis of equipment adaptation protocols and data quality management strategies.

Massive high-dimensional data generated by the Internet typically contains sensitive privacy information. Protecting data privacy while maintaining utility has become a pressing challenge. We propose a novel high-dimensional temporal data publishing method leveraging dynamic Bayesian networks and differential privacy. Initially, a dynamic Bayesian network is constructed, utilizing mutual information filtering of data. Subsequently, we calculate the Coherent Neighborhood Propinquity for each node within the network to determine edge sensitivity and establish a privacy budget. Noise is then strategically added to attribute data in accordance with the sensitivity and privacy budget requirements, ensuring the dataset complies with ε-differential privacy standards. Experimental results demonstrate that the data availability performance of the SMAP dataset (Soil Moisture Active Passive) surpasses that of competing algorithms while providing an equivalent level of privacy protection. Hence, our method significantly enhances data availability without compromising differential privacy protection.

Industry 4.0 brings new features to the manufacturing industry, including informatization, intelligence, and higher integration. Complex interrelationships among components within the industrial Cyber-Physical System (iCPS) further increase the automation system design and development difficulty. Therefore, integrated design models for contemporary industrial systems should ensure flexibility and interoperability to accommodate highly integrated systems and rapidly changing requirements. This paper proposes a generic process modeling method based on Process-Oriented Models (POM) for automatic code generation. Process-Oriented Models can be regarded as a semantic set of operations encapsulating the process and corresponding attributes in the manufacturing process. Each operation is executable. The execution results can be used for model optimization to achieve the optimization process. These models with parameters can be further converted into modular code automatically according to pre-defined mapping rules. A process manufacturing case study proves the proposed method can achieve complete and correct process modeling of automation systems. Industrial software development based on Process-Oriented Models can significantly increase the efficiency and accuracy of software development of industrial Cyber-Physical Systems.

Traditional video instance segmentation (VIS) aims to detect, segment, and track object instances from a known class set in videos. In real-world applications, however, video instance segmentation typically need to cope with novel-class instances and to fast adapt with a few labeled videos. In this work, we aim to tackle the task of few-shot video instance segmentation (FVIS), which is challenging due to large variations in object appearance and motion. We propose a robust temporally coherent strategy, termed as VTFA, based on a two-stage fine-tuning approach. VTFA enforces the instance segmentation of novel classes to be temporally smooth and reduces the classification bias between novel and base classes. The proposed Memory-aware Temporal Context Encoding Module (MTCE) in VTFA encodes the temporal context information, which contributes to the consistency in the final predictions. We also propose a loss named Instance-level Pair-wise Contrastive (IPC) Loss on both the novel and base classes to enhance the robustness of instance classification. To validate our method, we develop a YouTube-VIS-FS benchmark to compare our method with several baselines. The experimental evaluation shows that our strategy is superior or competitive to those strong baselines.

Establishing semantic correspondence is a core problem in computer vision and remains challenging due to large intra-class variations and lack of annotated data. In this paper, we aim to incorporate global semantic context in a flexible manner to overcome the limitations of prior work that relies on local semantic representations. To this end, we first propose a context-aware semantic representation that incorporates spatial layout for robust matching against local ambiguities. We then develop a novel dynamic fusion strategy based on attention mechanism to weave the advantages of both local and context features by integrating semantic cues from multiple scales. We instantiate our strategy by designing an end-to-end learnable deep network, named as Dynamic Context Correspondence Network (DCCNet). To train the network, we adopt a multi-auxiliary task loss to improve the efficiency of our weakly-supervised learning procedure. Our approach achieves superior or competitive performance over previous methods on several challenging datasets, including PF-Pascal, PF-Willow, and TSS, demonstrating its effectiveness and generality.

Entity search over text corpora is not geared for relationship queries where answers are tuples of related entities and where a query often requires joining cues from multiple documents. With large knowledge graphs, structured querying on their relational facts is an alternative, but often suffers from poor recall because of mismatches between user queries and the knowledge graph or because of weakly populated relations.

In this paper, we analyse the data access characteristics of a typical XML information retrieval system and propose a new query aware buffer replacement algorithm based on prediction of Minimum Reuse Distance (MRD for short). The algorithm predicts an object's next reference distance according to the retrieval system's running status and replaces the objects that have maximum reuse distances. The factors considered in the replacement algorithm include the access frequency, creation cost, and size of objects, as well as the queries being executed. By taking into account the queries currently running or queuing in the system, MRD algorithm can predict more accurately the reuse distances of index data objects.