Repetitive indoor layouts frequently cause spatial disorientation. Current navigation systems typically intervene reactively with generic instructions, lacking insight into the environmental root causes or the user’s cognitive state. To address this problem, we conducted a VR experiment (N=40) systematically manipulating geometric symmetry and feature similarity while capturing multimodal behaviors. Results reveal a functional separation: geometric symmetry primarily drives exploratory body rotation, whereas feature similarity determines navigation outcomes. Critically, simultaneous cue failure triggers a performance collapse—increasing mean hesitation duration by 370%—and forces users to switch from active reorientation (scanning via body rotation) to locomotor compensation (e.g., wall-following) based on a dynamic cost-benefit trade-off. Leveraging these patterns, our CNN-BiLSTM model detects the behaviorally defined getting lost state with >90% agreement with heuristic labels. We contribute design principles for dual context-awareness systems. By integrating environment context (geometric or featural ambiguity) and user context (cognitive state), systems can deploy content-adaptive aid—specifically orienting or discriminating aids—to dynamically balance navigation efficiency with active spatial cognition.

Multimodal Sentiment Analysis (MSA) with incomplete data has gained significant attention recently.Existing studies focus on optimizing model structures to handle modality missingness, but models still face challenges in robustness when dealing with uncertain missingness.To this end, we propose a data-centric robust multimodal sentiment analysis method, Proxy-Driven Robust Multimodal Fusion (P-RMF).First, we map unimodal data to the latent space of Gaussian distributions to capture core features and structure, thereby learn stable modality representation.Then, we combine the quantified modality intrinsic uncertainty to learn stable multimodal joint representation (i.e., proxy modality), which is further enhanced through multi-layer dynamic cross-modal injection to increase its diversity.Extensive experimental results show that P-RMF outperforms existing models in noise resistance and achieves state-of-the-art performance on multiple benchmark datasets.

Existing approaches to leveraging knowledge graphs in large language models often lack explicit structural modeling, which can lead to hallucinations and unstable reasoning over graph data. To address this, we propose SAKG, a structure-aware prompting framework designed to enhance the alignment between knowledge graph representations and language model inference. SAKG employs a hierarchical prompting strategy that integrates explicit task instructions, structural embeddings, and selectively filtered neighbor context. In particular, we introduce a progressive neighbor selection mechanism that combines relation co-occurrence statistics with embedding-based semantic similarity, ensuring that only informative and relevant neighbors are included in the prompt. This design enables the model to better capture relational semantics, structural dependencies, and contextual cues within the graph. Experimental results on multiple benchmarks demonstrate that SAKG consistently improves the effectiveness and factual consistency of knowledge graph reasoning with large language models.

Existing multimodal knowledge graph completion methods often overlook triple correlations between relations and images, limiting the expressiveness of multimodal embeddings. In this paper, we categorize triple correlations into Intra-triple Correlations (IaC) and Inter-triple Correlations (IeC), and propose a method called Relation-Sensitive Visual Aggregation (RSVA) to explicitly model them. Specifically, RSVA consists of two modules. The first is the Visual Semantic Aggregation Module, aggregating visual features for the central entity considering IaC. The second is the Contextual Neighbor Aggregation Module, capturing IeC by aggregating visual semantics from neighboring entities. In link prediction experiments, RSVA demonstrates the effect of IaC and IeC on the embeddings of central entities and achieves improved performance compared with previous approaches. These results demonstrate the effectiveness of RSVA, indicating that explicitly modeling the latent correlations between relations and images can enhance the representational capability of multimodal knowledge graphs.

As large language models gain prominence, there is increasing concern about the potential biases they may perpetuate. While various biases have been studied, ageism in language models remains underexplored. According to the World Health Organization, ageism can significantly impact the physical and mental well-being of older adults, an impact that could grow as the global aging population increases. To address this research gap, we developed AgeismSet, a comprehensive Chinese dataset comprising 6,444 sentences, enhanced with neutral impression options to provide a balanced evaluation framework. Our study then used AgeismSet to investigate ageism in large language models across cognitive, affective, and behavioral dimensions, using AgeismSet to evaluate models such as GPT-4, GPT-3.5, GLM-3-Turbo, ERNIE Bot, Gemini Pro, and DeepSeek-V3. Our findings, quantified by the Ageism Score (AS), reveal that while some models perform well, there is considerable room for improvement in mitigating ageism. This work underscores the necessity for targeted interventions to ensure more equitable AI systems.

Large language models (LLMs) encounter challenges when addressing few-shot nested named entity recognition (NER) tasks. Traditional LLM-based approaches typically either prompt the model to generate entity words or types in sentences based on entity categories or word spans, or directly extract all entities of specific types present in the sentences. These methods often suffer from issues such as low query efficiency or suboptimal accuracy. This paper introduces an innovative framework, SynNER, which synergizes small and large language models to address these limitations. Initially, a small language model identifies low-confidence word spans, which are then refined and refined by a large language model. To simultaneously ensure recognition accuracy and improve the query efficiency of the LLM, we propose a Batch-Prompt strategy and an Entity Indexing method. These techniques enable the LLM to process multiple test instances simultaneously while maintaining precise correction results. Experimental results demonstrate that our method achieves significant performance gains on benchmark datasets, offering a cost-effective solution for few-shot nested NER tasks.

Large language models (LLMs) encounter challenges when addressing few-shot nested named entity recognition (NER) tasks. Traditional LLM-based approaches typically either prompt the model to generate entity words or types in sentences based on entity categories or word spans, or directly extract all entities of specific types present in the sentences. These methods often suffer from issues such as low query efficiency or suboptimal accuracy. This paper introduces an innovative framework, SynNER, which synergizes small and large language models to address these limitations. Initially, a small language model identifies low-confidence word spans, which are then refined and refined by a large language model. To simultaneously ensure recognition accuracy and improve the query efficiency of the LLM, we propose a Batch-Prompt strategy and an Entity Indexing method. These techniques enable the LLM to process multiple test instances simultaneously while maintaining precise correction results. Experimental results demonstrate that our method achieves significant performance gains on benchmark datasets, offering a cost-effective solution for few-shot nested NER tasks.

Artificial intelligence methods offer objectivity and convenience in automatic depression detection, however, current research often neglects the critical role of facial landmarks. This oversight results in insufficient spatial structure information and a lack of detailed local representation, which fails to capture the nuanced semantic information crucial for identifying depression-related clues. To address these issues, we introduce a novel dual-branch network model comprising the Landmark-Image-Landmark Net (LIL Net) and the Global Context Vision Transformer Net (GCVit Net). Through a dual-stream, multiscale, and cross-fusion strategy, LIL Net is designed to extract original facial image features alongside landmark features, prioritizing the detailed semantic information of potential depression clues. LIL Net employs an innovative LIL Attention approach to jointly learn multiscale features from facial landmarks and images, thereby enhancing the model’s ability to capture fine-grained depression-related cues. Furthermore, the Multi-scale Feature Fusion (MSFF) module fuses the obtained multiscale features, augmenting the semantic expression of potential depression clues within facial landmarks via attention mechanisms. Meanwhile, the GCVit Net branch network supplements global information by extracting global facial features. Finally, the features from both branches are concatenated to enhance the accuracy of depression degree predictions. Experimental results demonstrate that our model has superior performance in detecting depression compared to existing methods. We release our code at https://github.com/xlx777/LIL-Net.

Missing data pose significant obstacles in data analysis. Many imputation methods, operating under the assumption that similar instances exhibit similar feature values, often overlook the essential role of the feature containing the missing value. These methods tend to apply a uniform strategy to all missing data within a single instance. This study proposes selecting neighbor instances for imputation based on both instance and missing feature information. To support this approach, a Structural Causal Model (SCM) delineates the relationships among missing data, features, and instances. The model reveals that the feature with the target missing value acts as a confounder, influencing both the selection of neighbor instances and the prediction of the missing value. Building upon these findings, this study introduces a new imputation method called the Causal Imputation strategy (CauImputation). CauImputation utilizes both feature and instance information to select neighbor instances and then applies causal interventions to compute the missing value. This strategy effectively leverages feature information while mitigating confounding effects. Tests on the National Alzheimer’s Coordinating Center (NACC) dataset and the National Center for Biotechnology Information (NCBI) microarray datasets demonstrate that CauImputation outperforms nine established methods, achieving a 3% to 15% increase in imputation accuracy and reducing root mean squared errors by 0.05 to 0.2. By utilizing the Structural Causal Model, this research introduces a novel perspective on data imputation, enhancing the robustness and reliability of biomedical data analyses.

Multimodal sentiment analysis (MSA) aims to integrate multiple modalities of information to better understand human sentiment. The current research mainly focuses on conducting multimodal fusion, which neglects the under-optimized modal representations generated by the imbalance of unimodal performances in joint learning. Moreover, the size of labeled datasets limits the generalization ability of existing supervised models. To address the above issues, this paper proposes a knowledge-enhanced self-supervised balanced representation approach (KEBR). First, a text-based cross-modal fusion method (TCMF) is constructed, which injects the non-verbal information from the videos into the semantic representation of text to enhance the multimodal representation of text. Then, a multimodal cosine constrained loss (MCC) is designed to constrain the fusion of non-verbal information in joint learning to balance the representation. Finally, with the help of sentiment knowledge and non-verbal information, KEBR conducts sentiment word masking and sentiment intensity prediction. Experimental results show that KEBR outperforms the baseline.

Depression is a prevalent and severe mental illness that significantly impacts patients’ physical health and daily life. Recent studies have focused on multimodal depression assessment, aiming to objectively and conveniently evaluate depression using multimodal data. However, existing methods based on audio–visual modalities struggle to capture the dynamic variations in depression clues and cannot fully explore multimodal data over a long time. In addition, they rely heavily on insufficient single-stage multimodal fusion, which limits the accuracy of depression assessment. To address these limitations, we propose a novel parallel multiscale bridge fusion network (PMBFN) for audio–visual depression assessment. PMBFN comprehensively captures subtle multilevel dynamic changes in depression expression through parallel multiscale dynamic convolutions and long short-term memories (LSTMs) and effectively solves the problem of long-term audio–visual sequence information loss by using spatiotemporal attention pooling modules. Furthermore, the multimodal bridge fusion module is proposed in PMBFN to achieve multistage interactive recursive multimodal fusion, enhancing the expressive capacity of multimodal depression-related features to improve the accuracy of assessment. Extensive experiments on the DAIC-WOZ and E-DAIC datasets demonstrate that our method outperforms current state-of-the-art methods and clearly shows our method's effectiveness eventually.

Studies have shown that life stories can help caregivers better understand older adults, leading to better care. The original life stories are often redundant and disordered, hindering the discovery of valuable information. There has been little research on organizing older adults’ life stories automatically. This article proposes the ALBERT Based Text Extraction Network (ABTE-NET) to generate valuable event timelines to address this problem. To evaluate the proposed method, we created an Older Adults’ Life Story dataset with 80 older adults’ life stories. In experiments, we verify that the timelines generated by ABTE-NET have good readability and summarization for life stories. A survey of 33 caregivers from two nursing homes shows that timelines can help caregivers understand older adults and build positive relationships, just like life stories. More importantly, timelines are better organized and more concise than original life stories, reducing the cognitive load and helping caregivers form a preliminary understanding of the older adult quickly.

与线性因果相比, 非线性因果具有更复杂的特点和内涵. 本文主要讨论非线性因果中的若干个问题, 并着重强调因果域的概念. 本文基于广泛应用的计算模型和方法, 围绕非线性因果分析与计算以及因果域的识别问题提出相应观点和建议, 并通过几个具体案例揭示非线性因果在处理复杂因果推断问题中的重要性和现实意义.

BACKGROUND: Visualizing data by dimensionality reduction is an important strategy in Bioinformatics, which could help to discover hidden data properties and detect data quality issues, e.g. data noise, inappropriately labeled data, etc. As crowdsourcing-based synthetic biology databases face similar data quality issues, we propose to visualize biobricks to tackle them. However, existing dimensionality reduction methods could not be directly applied on biobricks datasets. Hereby, we use normalized edit distance to enhance dimensionality reduction methods, including Isomap and Laplacian Eigenmaps. RESULTS: By extracting biobricks from synthetic biology database Registry of Standard Biological Parts, six combinations of various types of biobricks are tested. The visualization graphs illustrate discriminated biobricks and inappropriately labeled biobricks. Clustering algorithm K-means is adopted to quantify the reduction results. The average clustering accuracy for Isomap and Laplacian Eigenmaps are 0.857 and 0.844, respectively. Besides, Laplacian Eigenmaps is 5 times faster than Isomap, and its visualization graph is more concentrated to discriminate biobricks. CONCLUSIONS: By combining normalized edit distance with Isomap and Laplacian Eigenmaps, synthetic biology biobircks are successfully visualized in two dimensional space. Various types of biobricks could be discriminated and inappropriately labeled biobricks could be determined, which could help to assess crowdsourcing-based synthetic biology databases' quality, and make biobricks selection.

BACKGROUND: As the next-generation sequencing (NGS) technologies producing hundreds of millions of reads every day, a tremendous computational challenge is to map NGS reads to a given reference genome efficiently. However, existing methods of all-mappers, which aim at finding all mapping locations of each read, are very time consuming. The majority of existing all-mappers consist of 2 main parts, filtration and verification. This work significantly reduces verification time, which is the dominant part of the running time. RESULTS: An efficient all-mapper, BitMapper, is developed based on a new vectorized bit-vector algorithm, which simultaneously calculates the edit distance of one read to multiple locations in a given reference genome. Experimental results on both simulated and real data sets show that BitMapper is from several times to an order of magnitude faster than the current state-of-the-art all-mappers, while achieving higher sensitivity, i.e., better quality solutions. CONCLUSIONS: We present BitMapper, which is designed to return all mapping locations of raw reads containing indels as well as mismatches. BitMapper is implemented in C under a GPL license. Binaries are freely available at http://home.ustc.edu.cn/%7Echhy.

The multiple longest common subsequence (MLCS) problem, related to the identification of sequence similarity, is an important problem in many fields. As an NP-hard problem, its exact algorithms have difficulty in handling large-scale data and time- and space-efficient algorithms are required in real-world applications. To deal with time constraints, anytime algorithms have been proposed to generate good solutions with a reasonable time. However, there exists little work on space-efficient MLCS algorithms. In this paper, we formulate the MLCS problem into a graph search problem and present two space-efficient anytime MLCS algorithms, SA-MLCS and SLA-MLCS. SA-MLCS uses an iterative beam widening search strategy to reduce space usage during the iterative process of finding better solutions. Based on SA-MLCS, SLA-MLCS, a space-bounded algorithm, is developed to avoid space usage from exceeding available memory. SLA-MLCS uses a replacing strategy when SA-MLCS reaches a given space bound. Experimental results show SA-MLCS and SLA-MLCS use an order of magnitude less space and time than the state-of-the-art approximate algorithm MLCS-APP while finding better solutions. Compared to the state-of-the-art anytime algorithm Pro-MLCS, SA-MLCS and SLA-MLCS can solve an order of magnitude larger size instances. Furthermore, SLA-MLCS can find much better solutions than SA-MLCS on large size instances.

An enormous amount of sequence data has been generated with the development of new DNA sequencing technologies, which presents great challenges for computational biology problems such as haplotype phasing. Although arduous efforts have been made to address this problem, the current methods still cannot efficiently deal with the incoming flood of large-scale data. In this paper, we propose a flow network model to tackle haplotype phasing problem, and explain some classical haplotype phasing rules based on this model. By incorporating the heuristic knowledge obtained from these classical rules, we design an algorithm FNphasing based on the flow network model. Theoretically, the time complexity of our algorithm is (O(n(2)m+m(2)), which is better than that of 2SNP, one of the most efficient algorithms currently. After testing the performance of FNphasing with several simulated data sets, the experimental results show that when applied on large-scale data sets, our algorithm is significantly faster than the state-of-the-art Beagle algorithm. FNphasing also achieves an equal or superior accuracy compared with other approaches.

The multiple longest common subsequence (MLCS) problem, which is related to the measurement of sequence similarity, is one of the fundamental problems in many fields. As an NP-hard problem, finding a good approximate solution within a reasonable time is important for solving large-size problems in practice. In this paper, we present a new progressive algorithm, Pro-MLCS, based on the dominant point approach. Pro-MLCS can find an approximate solution quickly and then progressively generate better solutions until obtaining the optimal one. Pro-MLCS employs three new techniques: 1) a new heuristic function for prioritizing candidate points; 2) a novel d-index-tree data structure for efficient computation of dominant points; and 3) a new pruning method using an upper bound function and approximate solutions. Experimental results show that Pro-MLCS can obtain the first approximate solution almost instantly and needs only a very small fraction, e.g., 3 percent, of the entire running time to get the optimal solution. Compared to existing state-of-the-art algorithms, Pro-MLCS can find better solutions in much shorter time, one to two orders of magnitude faster. In addition, two parallel versions of Pro-MLCS are developed: DPro-MLCS for distributed memory architecture and DSDPro-MLCS for hierarchical distributed shared memory architecture. Both parallel algorithms can efficiently utilize parallel computing resources and achieve nearly linear speedups. They also have a desirable progressiveness property-finding better solutions in shorter time when given more hardware resources.