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.

One of the challenging topics in Intelligent Transportation Systems (ITSs) is the metro passenger flow prediction. It has great practical significance for the daily crowd management and vehicle scheduling of metro passenger flow. Recently, Graph Convolutional Networks (GCN) represent a station of metros by aggregating information of stations directly and indirectly connected with the station, and improve the effectiveness of predicting metro passenger flow. Despite its effectiveness, the neighborhood aggregation scheme also brings two limitations in predicting metro passenger flow. First, it limits to predict accurately the peak of passenger flow with large data fluctuation. Second, it enlarges the impact of noisy data and makes results vulnerable to noisy data of metro passenger flow. To solve these problems, we propose regularized spatial-temporal graph convolutional networks for metro passenger flow prediction (PMR-GCN). More specifically, we first propose a novel personalized enhanced GCN (P-GCN), which defines a trainable diagonal matrix to adaptively learn to control the impact of the neighborhood aggregation scheme for predicting the peak of passenger flow. Then, we introduce the multi-head self-attention mechanism to capture richer spatial-temporal features of passenger flow. In addition, we utilize a Fourier Transform module and a dual structure based on KL-divergence regularization to improve the robustness of the proposed model. Extensive experiments on Shanghai, Chongqing, and Hangzhou datasets demonstrate the superiority of our model over the state-of-the-art baseline methods. The code of PMR-GCN is available at https://github.com/cgao-comp/PMC-GCN.

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.

Recently, the rapid diffusion of malicious information in online social networks causes great harm to our society. Therefore, it is of great significance to localize diffusion sources as early as possible to stem the spread of malicious information. This paper proposes a novel sensor-based method, called greedy full-order neighbor localization (denoted as GFNL), to solve this problem under a low infection propagation in line with the real world. More specifically, GFNL includes two main components, i.e., the greedy-based sensor deployment strategy (DS) and direction-path-based source estimation strategy (ES). In more detail, to ensure sensors can observe a propagation information as early as possible, a set of sensors is deployed in a network to minimize the geodesic distance (i.e., the distance of the shortest path) between the candidate set and the sensor set based on DS. Then when a fraction of sensors observe a propagation, ES infers the source based on the idea that the distance of the actual propagation path is proportional to the observed time. Compared with some state-of-the-art methods, comprehensive experiments have proved the superiority and robustness of our proposed GFNL.

Localizing the root cause of network faults is crucial to network operation and maintenance. Operational expenses will be saved if the root cause can be identified accurately. However, due to the complicated wireless environments and network architectures, accurate root cause localization of network falut meets the difficulties including missing data, hybrid fault behaviors, and short of well-labeled data. In this study, global and local features are constructed to make new feature representation for data sample, which can highlight the temporal characteristics and contextual information of the root cause analysis data. A hybrid tree model (HTM) ensembled by CatBoost, XGBoost and LightGBM is proposed to interpret the hybrid fault behaviors from several perspectives and discriminate different root causes. Based on the combination of global and local features, a semi-supervised training strategy is utilized to train the HTM for dealing with short of well-labeled data. The experiments are conducted on the real-world dataset from ICASSP 2022 AIOps Challenge, and the results show that the global and local feature based HTM achieves the best model performance comparing with other models. Meanwhile, our solution achieves third place in the competition leaderboard which shows the model effectiveness.

Medication combination prediction can be applied to the clinical treatment for critical patients with multi-morbidity. The suitable medication combination can help cure patients and keep the treatment medication safe. However, the complexity and uncertainty of clinical circumstances limit the predictive accuracy of medication combination. Thus, this paper proposes a new medication combination prediction model based on the temporal attention mechanism (TAM) and the simple graph convolution (SGC), named as TAMSGC. More specifically, the TAM can capture the temporal sequence information in the medical records, and the SGC is implemented to acquire the medication knowledge from the complicated medication combination. Experiments in a real dataset show that TAMSGC surpasses the baseline models on the predictive accuracy of medication combination.

Recommender systems have been widely used in our life in recent years to facilitate our life. And it is very important and meaningful to improve recommendation performance. Generally, recommendation methods use users’ historical ratings on items to predict ratings on their unrated items to make recommendations. However, with the increase of the number of users and items, the degree of data sparsity increases, and the quality of recommendations decreases sharply. In order to solve the sparsity problem, other auxiliary information is combined to mine users’ preferences for higher recommendation quality. Similar to rating data, review data also contain rich information about users’ preferences on items. This paper proposes a novel recommendation model, which harnesses an adversarial learning among auto-encoders to improve recommendation quality by minimizing the gap of the rating and review relation between a user and an item. The empirical studies on real-world datasets show that the proposed method improves the recommendation performance.

We propose a novel pairwise unified recommendation model (short for pairwise URM). The pairwise URM combines two pairwise ranking-oriented collaborative filtering approaches, namely Collaborative Less-is-More Filtering (CLiMF) and Bayesian Personal Ranking (BPR). By sharing common latent features of users and items in BPR and CLiMF, the pairwise URM can benefit from the two methods to improve recommendation qualities. The experimental evaluation is conducted on two real-world datasets with different scales and demonstrates the positive effect of the performance of the pairwise URM.

This research is aimed to analysis the resting EEG features in depression and the application in clinic. Sixteen patients with depression and sixteen healthy controls were involved in this study. Both features from the alpha and beta frequency bands were selected to analysis in this study. First the features' sensitivity to the group-difference and the correlation to the clinical HAMD scale score were analyzed, and then the classification method was used to further test the role of the resting EEG features in depression. The results showed that the difference between depression and healthy controls in the absolute power of beta band in the left prefrontal lobe was significant. And the alpha left-right asymmetry in the prefrontal cortex had a correlation with HAMD scale score. In addition, the classification based on the features showed that there was a relative higher accuracy rate to identify the depressions than to identify the healthy controls. Specifically, the classification based on alpha asymmetry was higher than that based on beta asymmetry, and the absolute power in beta band was higher than that in alpha band. Alpha asymmetry is a traditional sensitive resting EEG features for depression, this study provide new evidence to support the view. The findings here further suggest that absolute power in beta band would be important biomarker in depression.

Recommendation methods usually use users' historical ratings on items to predict ratings on their unrated items to make recommendations. However, the sparse rating data limit the recommendation quality. In order to solve the sparsity problem, other auxiliary information is combined to mine users' preferences for higher recommendation quality. This paper proposes a novel recommendation model, which harnesses an adversarial learning among auto-encoders to improve recommendation quality by minimizing the gap of rating and review relation of users and items. The empirical studies on real-world datasets prove that the proposed method improves recommendation performance.

Recommender systems aim to provide users with preferred items to address the information overload problem in the Web era. Social relations, item connections, and user-generated item reviews and ratings play important roles in recommender systems as they contain abundant potential information. Many methods have been proposed to predict users’ ratings by learning latent topic factors from their reviews and ratings of corresponding items. However, these methods ignore the relationships among items and cannot make full use of the complicated relations between reviews and ratings. Motivated by this observation, we integrate ratings, reviews, user connections and item relations to improve recommendations by combining matrix factorization with the Latent Dirichlet Allocation (LDA) model. Experimental results on two real-world datasets prove that item–item relations contain useful information for recommendations, and our model effectively improves recommendation quality.

Top-N recommendation tasks aim to solve the information overload problem for users in the information age. As a user's decision may be affected by correlations among items, we incorporate such correlations with the user and item latent factors to propose a Poisson-regression-based method for top-N recommendation tasks. By placing priori knowledge and using a sparse structure assumption, this method learns the latent factors and the structure of the item-item correlation matrix through the alternating direction method of multipliers (ADMM). The preliminary experimental results on two real-world datasets show the improved performance of our approach.

Matrix factorization (MF) models and their extensions are widely used in modern recommender systems. MF models decompose the observed user-item interaction matrix into user and item latent factors. In this paper, we propose mixture models which combine the technology of MF and the embedding. We show that some of these models significantly improve the performance over the state-of-the-art models on two real-world datasets, and explain how the mixture models improve the quality of recommendations.

Recommender systems aim to provide users with preferred items to tackle the information overload problem in the Web era. Social relations, item connections, and user-generated reviews on items contain abundant potential information. By combining matrix factorization with latent Dirichlet allocation, we integrate ratings, reviews, user similarity and item similarity in recommender systems. The experimental result on a real-world dataset proves that both item connection and user connection contain useful sources for recommendation, and our model can effectively improve recommendation quality.

With the increasing popularity of location-based services, location recommendation is one of important applications. In this paper, according to a user’s preference, we recommend locations to a user by extending an information retrieval model, namely, belief network model. We regard the user’s preference as a query, a location as a document, categories as index terms. And then, we use the belief network model to recommend locations to a user by adding expert information. Experimental results on a real world data set show that recommendation effectiveness can be improved.

One of the fundamental tasks of targeted marketing is to elicit associations between customers and products. Based on the results from information retrieval and utility theory, this article proposes a unified framework of targeted marketing. The customer judgments of products are formally described by preference relations and the connections of customers and products are quantitatively measured by market value functions. Two marketing strategies, known as the customer‐oriented and product‐oriented marketing strategies, are investigated. Four marketing models are introduced and examined. They represent, respectively, the relationships between a group of customers and a group of products, between a group of customers and a single product, between a single customer and a group of products, and between a single customer and a single product. Linear and bilinear market value functions are suggested and studied. The required parameters of a market value function can be estimated by exploring three types of information, namely, customer profiles, product profiles, and transaction data. Experiments on a real‐world data set are performed to demonstrate the effectiveness of the proposed framework.

Web software systems provide information and service for end users through the Web interface. The interactive process between the user and the software is the process of software systems to perceive the environment and user, to adjust own configuration and provide appropriate services, monitor and eliminate untrustworthy factors, complete own evolvement and achieve trusted result finally. This paper aims to establishing the online trust evolution model and then building the corresponding trustworthy software framework during the human-Web interactive process. Our research will consider the human and software factors together and combine related methodologies and tools for developing trust model in a user-centric way. Our study can provide theoretical and technical support for developing trustworthy Web software systems.

Peculiarity rules are a new type of interesting rules which can be discovered by searching the relevance among peculiar data. A main task of mining peculiarity rules is the identification of peculiarity. Traditional methods of finding peculiar data are attribute-based approaches. This paper extends peculiarity oriented mining to relational peculiarity oriented mining. Peculiar data are identified on record level, and peculiar rules are mined and explained in a relational mining framework. The results from preliminary experiments show that relational peculiarity oriented mining is very effective.

TMS (Targeted Marketing System) is an integrated system and toolkit for profit-driven and cost-effective marketing. The system consists of three components: a Web-based user interface, a market value inference engine, and a presentation and evaluation module. It supports marketing decision making for a company or an organization by combining results from information retrieval, data mining, information theory, and utility theory.

The linear model of market value functions is a new method for direct marketing. Just like other methods in direct marketing, attribute reduction is very important to deal with large databases. We apply the algorithm of attribute reduction, which is based on the combination of rough set theory with the boosting algorithm, to the linear model of market value functions. Experimental results compared with the ELSA/ANN model show that the proposed algorithms can be used effectively in the linear model of market value functions.

Targeted marketing typically involves the identification of customers or products having potential market values. We propose a linear model for solving this problem by drawing and extending results from information retrieval. It is assumed that each object is represented by values of a finite set of attributes. A market value function, which is a linear combination of utility functions on attribute values, is used to rank objects. Several methods are examined for mining market value functions. The main advantage of the model is that one can rank objects of interest according to their market values, instead of classifying the objects. Both the theoretical and experimental results are reported in this paper. It establishes a basis on which further studies and experimental evaluation can be carried out.