Multimodal sentiment analysis aims to accurately identify emotional tendencies and expressions from three modalities: text, audio and video, which is useful for opinion guidance and mental health analysis. Current studies have focused on fusing different sources of information from certain practical problems, focusing on the textual modality and ignoring the interaction and complementary information between modalities. In light of the aforementioned limitations, we propose a general model of hypergraph neural network. This model employs a tensor structure to fully obtain the interaction and complementary information during fusion, a low-rank decomposition of the tensor to achieve downscaling and a more accurate feature representation, and a hypergraph convolution and an attention mechanism to facilitate the capture of complex relationships in emotional expression. The experimental results on two benchmark datasets demonstrate the effectiveness of our model. Our approach outperforms existing methods by improving performance while reducing computation time, exhibiting strong generalization ability and high competitiveness.

Multi-label ensemble can achieve superior performance on multi-label learning problems by integrating a number of base classifiers. In existing multi-label ensemble methods, the base classifiers are usually trained with the same original features; it is difficult for each base classifier to capture label-relevant or label subset-relevant information. Meanwhile, the manually designed integrating strategies cannot automatically distinguish the importance of the base classifiers, which also lack flexibility and scalability. In order to resolve these problems, this paper proposes a new multi-label ensemble framework, named Attention-aware Multi-label Ensemble with Label Subset-specIfic FeaTures (AME-LSIFT). It utilizes$c$-means clustering to produce Label Subset-specIfic FeaTures (LSIFT), constructs a neural network based model for each label subset, and integrates the base models with a dynamic and automatic attention-aware mechanism. Moreover, an objective function that considers both the label subset accuracy and ensemble accuracy is developed for training the proposed AME-LSIFT. Experiments conducted on ten benchmark datasets demonstrate the superior performance of the proposed method compared with state-of-the-art approaches.

Spectral clustering is widely used in modern data analysis. Spectral clustering methods speed up the computation and keep useful information by reducing dimensionality. Recently, graph signal filtering (GSF) has been introduced to further speed up the dimensionality reduction process by avoiding solving eigenvectors.

Spectral clustering (SC) is an important data mining method in bioinformatics due to the collection of big datasets. Calculating the number of clusters and feature vectors are two main steps of SC, both of which are related to eigenproblem. We propose a novel SC method using an adaptive filter which accelerates the SC and completely sidestepping the eigensolver. Results: We are inspired by two theories of different disciplines. The first theory is Density of State (DOS), a concept of solid-state physics. DOS roughly represents the distribution of eigenvalues. We propose a more accurate DOS and integrate it to obtain the number of clusters. The second theory is Graph Signal Processing (GSP) in irregular graph. Based on existing dimension reduction method, we use GSP to compute the feature vectors. We use Chebyshev polynomials to compute DOS and GSP and let them share the polynomials. Both the computational complexity and storage complexity are low.