Sequential recommendation is of utmost importance for personalized suggestions across various domains, and its effectiveness hinges critically on accurately capturing the intricate sequential dependencies among items. Two advanced architectures are widely adopted for this purpose: self-attention mechanisms (SAM), which excel at modelling global dependencies in sequences, and linear recurrent units (LRU), which specialize in capturing local sequential patterns. However, SAM inherently lacks the capability to convey sequential order information, while LRU may overemphasize item ordering. Crucially, while the complementary strengths of SAM and LRU architectures are promising, their direct integration risks overfitting in noisy real-world recommendation scenarios, as overly refined representations may predominantly capture training noise rather than generalizable patterns. To tackle this issue while preserving the advantages of both architectures, we design a dual-pronged solution: perturbation injection during recurrence enhances robustness, while Gibbs distribution-based uncertainty scaling reduces noise sensitivity. Integrating these strategies, we propose a self-attention gated linear recurrent units for sequential recommendation (SLSRec) model. Specifically, through a novel self-attention gated unit, SLSRec synergistically integrates SAM’s global modeling capability with LRU’s sequential processing strength. The model is optimized using a comprehensive objective function that combines cross-entropy recommendation loss, perturbation injection loss, and uncertainty-incorporated recommendation loss. Extensive experiments on real-world datasets demonstrate that SLSRec outperforms state-of-the-art methods, showing particular effectiveness in handling noisy interactions and recommending long-tail items.

Abstract Social recommendation aims to improve the recommendation performance by learning user interest and social representations from users’ interaction records and social relations. Intuitively, these learned representations entangle user interest factors with social factors because users’ interaction behaviors and social relations affect each other. A high-quality recommender system should provide items to a user according to his/her interest factors. However, most existing social recommendation models aggregate the two kinds of representations indiscriminately, and this kind of aggregation limits their recommendation performance. In this paper, we develop a model called Disentangled Variational autoencoder for Social Recommendation (DVSR) to disentangle interest and social factors from the two kinds of user representations. Firstly, we perform a preliminary analysis of the entangled information on three popular social recommendation datasets. Then, we present the model architecture of DVSR, which is based on the Variational AutoEncoder (VAE) framework. Besides the traditional method of training VAE, we also use contrastive estimation to penalize the mutual information between interest and social factors. Extensive experiments are conducted on three benchmark datasets to evaluate the effectiveness of our model.

Graph Neural Networks (GNNs) have been successfully used to learn user and item representations for Collaborative Filtering (CF) based recommendations (GNN-CF). Besides the main recommendation task in a GNN-CF model, contrastive learning is taken as an auxiliary task to learn better representations. Both the main task and the auxiliary task face the noise problem and the distilling hard negative problem. However, existing GNN-CF models only focus on one of them and ignore the other. Aiming to solve the two problems in a unified framework, we propose a Multi-Mixing strategy for GNN-based CF (M2GCF). In the main task, M2GCF perturbs embeddings of users, items and negative items with sample-noise by a mixing strategy. In the auxiliary task, M2GCF utilizes a contrastive learning mechanism with a two-step mixing strategy to construct hard negatives. Extensive experiments on three benchmark datasets demonstrate the effectiveness of the proposed model. Further experimental analysis shows that M2GCF is robust against interaction noise and is accurate for long-tail item recommendations.

BACKGROUND: Understanding the working mechanism of the brain is one of the grandest challenges for modern science. Toward this end, the BigNeuron project was launched to gather a worldwide community to establish a big data resource and a set of the state-of-the-art of single neuron reconstruction algorithms. Many groups contributed their own algorithms for the project, including our mean shift and minimum spanning tree (M-MST). Although M-MST is intuitive and easy to implement, the MST just considers spatial information of single neuron and ignores the shape information, which might lead to less precise connections between some neuron segments. In this paper, we propose an improved algorithm, namely M-AMST, in which a rotating sphere model based on coordinate transformation is used to improve the weight calculation method in M-MST. RESULTS: Two experiments are designed to illustrate the effect of adapted minimum spanning tree algorithm and the adoptability of M-AMST in reconstructing variety of neuron image datasets respectively. In the experiment 1, taking the reconstruction of APP2 as reference, we produce the four difference scores (entire structure average (ESA), different structure average (DSA), percentage of different structure (PDS) and max distance of neurons' nodes (MDNN)) by comparing the neuron reconstruction of the APP2 and the other 5 competing algorithm. The result shows that M-AMST gets lower difference scores than M-MST in ESA, PDS and MDNN. Meanwhile, M-AMST is better than N-MST in ESA and MDNN. It indicates that utilizing the adapted minimum spanning tree algorithm which took the shape information of neuron into account can achieve better neuron reconstructions. In the experiment 2, 7 neuron image datasets are reconstructed and the four difference scores are calculated by comparing the gold standard reconstruction and the reconstructions produced by 6 competing algorithms. Comparing the four difference scores of M-AMST and the other 5 algorithm, we can conclude that M-AMST is able to achieve the best difference score in 3 datasets and get the second-best difference score in the other 2 datasets. CONCLUSIONS: We develop a pathway extraction method using a rotating sphere model based on coordinate transformation to improve the weight calculation approach in MST. The experimental results show that M-AMST utilizes the adapted minimum spanning tree algorithm which takes the shape information of neuron into account can achieve better neuron reconstructions. Moreover, M-AMST is able to get good neuron reconstruction in variety of image datasets.

Distance measure is quite important for pattern recognition. Utilizing invariance in image data, tangent distance is very powerful in classifying handwritten digits. For this measure a set of invariant transformations must be known a priori. But in many practical problems, it is very difficult to know these transformations. In this paper, an algorithm is proposed to approximate the invariant tangent distance exclusively from the data. By virtue of ideas arising from manifold learning, the algorithm needs no prior transformations and can be applied to more classification problems. k-nearest neighbor rule based on the new distance are implemented for classification problems. Experimental results on synthetic and real datasets illustrate its validity.

Peculiarity-oriented mining (POM) is a new data mining method consisting of peculiar data identification and peculiar data analysis. Peculiarity factor (PF) and local peculiarity factor (LPF) are important concepts employed to describe the peculiarity of points in the identification step. One can study the notions at both attribute and record levels. In this paper, a new record LPF called distance based record LPF (D-record LPF) is proposed, which is defined as the sum of distances between a point and its nearest neighbors. It is proved mathematically that D-record LPF can characterize accurately the probability density function of a continuous m-dimensional distribution. This provides a theoretical basis for some existing distance based anomaly detection techniques. More important, it also provides an effective method for describing the class conditional probabilities in the Bayesian classifier. The result enables us to apply peculiarity analysis for classification problems. A novel algorithm called LPF-Bayes classifier and its kernelized implementation are presented, which have some connection to the Bayesian classifier. Experimental results on several benchmark data sets demonstrate that the proposed classifiers are effective.

Peculiarity oriented mining (POM), aiming to discover peculiarity rules hidden in a dataset, is a new data mining method. In the past few years, many results and applications on POM have been reported. However, there is still a lack of theoretical analysis. In this paper, we prove that the peculiarity factor (PF), one of the most important concepts in POM, can accurately characterize the peculiarity of data with respect to the probability density function of a normal distribution, but is unsuitable for more general distributions. Thus, we propose the concept of local peculiarity factor (LPF). It is proved that the LPF has the same ability as the PF for a normal distribution and is the so-called µ-sensitive peculiarity description for general distributions. To demonstrate the effectiveness of the LPF, we apply it to outlier detection problems and give a new outlier detection algorithm called LPF-Outlier. Experimental results show that LPF-Outlier is an effective outlier detection algorithm.

Distance metric learning and nonlinear dimensionality reduction are two interesting and active topics in recent years. However, the connection between them is not thoroughly studied yet. In this paper, a transductive framework of distance metric learning is proposed and its close connection with many nonlinear spectral dimensionality reduction methods is elaborated. Furthermore, we prove a representer theorem for our framework, linking it with function estimation in an RKHS, and making it possible for generalization to unseen test samples. In our framework, it suffices to solve a sparse eigenvalue problem, thus datasets with 105 samples can be handled. Finally, experiment results on synthetic data, several UCI databases and the MNIST handwritten digit database are shown.