A Cross-Lingual Real-Time E-Commerce Recommendation Method Based on Siamese Graph Convolutional Network and Bilinear Attention.

Feature interactive graph neural network for KG-based recommendation

Feature interactions are essential for achieving high accuracy in recommender systems. Many studies take into account the interaction between every pair of features. However, this is suboptimal because some feature interactions may not be that relevant to the recommendation result and taking them into account may introduce noise and decrease recommendation accuracy. To make the best out of feature interactions, we propose a graph neural network approach to effectively model them, together with a novel technique to automatically detect those feature interactions that are beneficial in terms of recommendation accuracy. The automatic feature interaction detection is achieved via edge prediction with an L0 activation regularization. Our proposed model is proved to be effective through the information bottleneck principle and statistical interaction theory. Experimental results show that our model (i) outperforms existing baselines in terms of accuracy, and (ii) automatically identifies beneficial feature interactions.

Existing drug recommendation algorithms often ignore the role played by neighbourhood information among drugs in recommendation, and it is difficult to model user features. To address the above problems, we propose a research method for drug recommendation based on feature interaction and graph convolution. First, we introduce a knowledge graph (KG) of drugs and perform feature extraction by graph convolution algorithm, which effectively captures the correlation between the neighbourhoods of medicines. Second, we enrich the vector representation of users and drugs by obtaining node information of users and drugs through a heterogeneous propagation layer and employing a knowledge-aware attention mechanism to distinguish the contributions of tail entities in different triples. In addition, we use a two-mechanism approach to dynamically learn the vector representations of users, which improves the recommendation effect of the model. Finally, the user feature vectors and drug feature vectors are fed into the prediction module to obtain the interaction probability between the user and the drug. Compared with other state-of-the-art mainstream models, the accuracy of the proposed model on drugs dataset and Book-Crossing dataset is improved by 1.01% and 0.8%, respectively. The experimental results demonstrate that the proposed model has high recommendation accuracy and superior overall performance.

Taxi pick-up area recommendation based on GPS data can effectively improve efficiency and reduce fuel consumption. Most of the methods use the long-term GPS data, which makes recommendation accuracy low. Therefore, we propose a novel approach of integrating spatio-temporal contexts into the extreme Deep Factorization Machine (xDeepFM) for taxi pick-up area recommendation. In the training process, the urban area is divided into several grids of equal size, we extract pick-up points from the original GPS data. The pick-up points and points-of-interest (POIs) are mapped into the corresponding grids, we distil the spatio-temporal features from these grids to construct spatio-temporal contexts matrix. Then, the spatio-temporal contexts matrix is input into xDeepFM for training, and we get the taxi pick-up area recommendation model. xDeepFM not only can make feature interactions occur at the vector-wise in both implicit and explicit ways, but also learn both low-order and high-order feature interactions. xDeepFM can effectively enhance recommendation accuracy. Finally, the recommendation model is embedded in the system for testing. Evaluate on the public dataset of DiDi, we compare different recommendation methods. The experimental results show that our approach can effectively cope with the data sparseness problem, obtain excellent performance, and is superior to some state-of-the-art methods. The RMSE is only 0.8%, MAE is about 7%, and the explained variance score is over 98%.KeywordsTrajectory miningLocation-based services (LBS)Taxi pick-up area recommendationSpatio-temporal contextsExtreme Deep Factorization Machine (xDeepFM)

This article addresses the task of building personalized educational recommendations based on a heterogeneous knowledge graph that integrates data from university curricula, job vacancies, and online courses. To solve the problem of course recommendations by their relevance to a user’s competencies, a graph neural network (GNN)-based approach is proposed, specifically utilizing and comparing the Heterogeneous Graph Transformer (HGT) architecture, Graph Sample and Aggregate network (GraphSAGE), and Heterogeneous Graph Attention Network (HAN). Experiments were conducted on a heterogeneous graph comprising various node and relation types. The models were evaluated using regression and ranking metrics. The results demonstrated the superiority of the HGT-based recommendation model as a link regression task, especially in terms of ranking metrics, confirming its suitability for generating accurate and interpretable recommendations in educational systems. The proposed approach can be useful for developing adaptive learning recommendations aligned with users’ career goals.

Trust-aware recommendation based on heterogeneous multi-relational graphs fusion

Under the background of information overload, the recommendation system has attracted wide attention as one of the most important means for this problem. Feature interaction considers not only the impact of each feature but also the combination of two or more features, which has become an important research field in recommendation systems. There are two essential problems in current feature interaction research. One is that not all feature interactions can generate positive gains, and some may lead to an increase in noise. The other is that the process of feature interactions is implicit and uninterpretable. In this paper, a Hierarchical Dual-level Graph Feature Interaction (HDGFI) model is proposed to solve these problems in the recommendation system. The model regards features as nodes and edges as interactions between features in the graph structure. Interaction noise is filtered by beneficial interaction selection based on a hierarchical edge selection module. At the same time, the importance of interaction between nodes is modeled in two perspectives in order to learn the representation of feature nodes at a finer granularity. Experimental results show that the proposed HDGFI model has higher accuracy than the existing models.

Learning heterogeneous graph embedding for Chinese legal document similarity

Mixed training at high and low accuracy levels leads to perceptual learning without feedback

Existing recommendation methods suffer from the data sparsity problem which means that most of users have rated only a very small number of items, often resulting in low recommendation accuracy. In addition, for cold start users evaluating only few items, rating predictions with the methods also produce low accuracy. To address these problems, we propose a novel data imputation method that effectively substitutes missing ratings with probable values (i.e., imputed values). Our method successfully improves accuracy of recommendation methods from the following three aspects: (1) exploiting a trust network, (2) imputing only a part of missing ratings, and (3) applying them to any recommendation methods. Our method employs a bidirectional connection structure within a distance level for finding reliable users in exploiting a trust network as useful information. In addition, our method imputes only some missing ratings, called fillable ratings, whose imputed values are expected to be accurate with a sufficient level of confidence. Moreover, our imputation method is independent of, thus applicable to, any recommendation methods that may include application-specific ones and the most accurate one in each domain. We conduct experiments on three real-life datasets which arise from Epinions and Ciao. Our experimental results demonstrate that our method has recommendation accuracy better than existing recommendation methods equipped with imputation methods or trust networks, especially for cold start users.

Digital libraries suffer from the overload problem, which makes the researchers have to spend much time to find relevant papers. Fortunately, recommender system can help to find some relevant papers for researchers automatically according to their browsed papers. Previous paper recommendation methods are either citation-based or content-based. In this paper, we propose a novel recommendation method with a heterogeneous graph in which both citation and content knowledge are included. In detail, a heterogeneous graph is constructed to represent both citation and content information within papers. Then, we apply a graph-based similarity learning algorithm to perform our paper recommendation task. Finally, we evaluate our proposed approach on the ACL Anthology Network data set and conduct an extensive comparison with other recommender approaches. The experimental results demonstrate that our approach outperforms traditional methods.

Recently, Graph Neural Networks (GNNs) have been widely used for fraud detection. GNNs first generate node embedding by aggregating neighboring information under different relations, and then use the final node embedding to detect the node’s suspiciousness. However, traditional GNNs employing only a single type of aggregator fail to capture neighbor information from multiple perspectives and treating different relations equally inevitably weakens the semantic information of heterogeneous graphs. Meanwhile, expressive ability of GNNs is limited by using conventional concatenating or averaging operations to update the center node. Also, camouflaged entities could damage GNN-based models. To handle these problems, a novel heterogeneous GNN model called <italic xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">Multiple Aggregators and Feature Interactions Network</i> (MAFI) is proposed in this paper to conduct fraud detection tasks. Concretely, multiple types of aggregators are applied on different relations to aggregate neighbor information and aggregator-level attention is utilized to learn the importance of different aggregators. Also, relation-level attention is leveraged to learn the importance of each relation. Besides, conventional update operations are replaced with vector-wise implicit and explicit feature interactions. Moreover, a trainable neighbor sampler is employed to filter camouflaged fraudsters. Comprehensive experiments on two real-world fraud datasets indicate that the proposed MAFI outperforms existing GNN-based fraud detectors.

Alleviating data sparsity problem in time-aware recommender systems using a reliable rating profile enrichment approach

With the diversification of human activity and travel demand in urban space, recommending ROIs (region-of-interest) to users is important for both satisfying commercial demands and better understanding user urban lifestyles. Current researches mainly resort to the traditional POI-level (point-of-interest) or neural network-based recommendation methods for ROI recommendation, in disregard of the rich heterogeneous graph information, such as user-region-user, region-category-region, just to name a few. In this work, we employ the heterogeneous graph to address this issue, considering heterogeneous graph contains more comprehensive information and rich semantics. We propose a novel meta-path based graph attention network for ROI recommendation, called MRec. MRec is a newly devised heterogeneous graph neural network, which is equipped with both node-level and semantic-level attentions. Specially, the node-level attention aims to learn the importance between a node and its meta-path based neighbors, while the semantic-level attention is to learn the importance of different meta-paths. This mechanism contributes to effectively embedding users and ROIs in a hierarchical manner of fully considering both node and semantic-level component information. An extensive experiment on two real-world datasets demonstrates the effectiveness of the proposed framework.

Attraction recommendation plays an important role in tourism, such as solving information overload problems and recommending proper attractions to users. Currently, most recommendation methods are dedicated to improving the accuracy of recommendations. However, recommendation methods only focusing on accuracy tend to recommend popular items that are often purchased by users, which results in a lack of diversity and low visibility of non-popular items. Hence, many studies have suggested the importance of recommendation diversity and proposed improved methods, but there is room for improvement. First, the definition of diversity for different items requires consideration for domain characteristics. Second, the existing algorithms for improving diversity sacrifice the accuracy of recommendations. Therefore, the article utilises the topic ‘features of attractions’ to define the calculation method of recommendation diversity. We developed a two-stage optimisation model to enhance recommendation diversity while maintaining the accuracy of recommendations. In the first stage, an optimisation model considering topic diversity is proposed to increase recommendation diversity and generate candidate attractions. In the second stage, we propose a minimisation misclassification cost optimisation model to balance recommendation diversity and accuracy. To assess the performance of the proposed method, experiments are conducted with real-world travel data. The results indicate that the proposed two-stage optimisation model can significantly improve the diversity and accuracy of recommendations.