Search

Aspect-level sentiment classification has been widely used by researchers as a fine-grained sentiment classification task to predict the sentiment polarity of specific aspect words in a given sentence. Previous studies have shown relatively good experimental results using graph convolutional networks, so more and more approaches are beginning to exploit sentence structure information for this task. However, these methods do not link aspect word and context well. To address this problem, we propose a method that utilizes a hierarchical multi-head attention mechanism and a graph convolutional network (MHAGCN). It fully considers syntactic dependencies and combines semantic information to achieve interaction between aspect words and context. To fully validate the effectiveness of the method proposed in this paper, we conduct extensive experiments on three benchmark datasets, which, according to the experimental results, show that the method outperforms current methods. [ABSTRACT FROM AUTHOR]

Recently, social networks have been regarded as auxiliary information to mitigate the data sparsity issue in recommender systems. However, most existing social recommendation methods fail to effectively capture the relations between multiple behaviors, resulting in the correlated behaviors being unable to make semantic complements to the target behavior and sparse behavior data features. To alleviate the above problems, we propose a novel method based on graph neural network, namely Multiple Behavior Interactive Enhanced Social-aware Recommendation (MISR), which can dynamically acquire more fine-grained relations and differences between different behaviors and combine features of temporal sequences to capture potential interactions. In addition, we develop a global enhanced module to fully learn the enhanced user representation, empowering MISR to capture jointly the heterogeneous strengths of global social context and social relations. Extensive experiments on three real-world recommendation datasets validate the rationality and effectiveness of the proposed method. [ABSTRACT FROM AUTHOR]

Sequential Recommendation systems play a crucial role in predicting users’ preferences based on their behavioral history. However, the existing methods ignore the extrapolation nature of sequences and do not make deep use of item provider information. This oversight limits the model’s ability to fully utilize relevant external properties. To alleviate these limitations, we design a recommendation model that incorporates Position encoding and external reinforcement (Item -Provider), named DPSRec. Specifically, we design an Embed Encoding layer, in order to distinguish the Position Embedding of previous sequence models, we combine the time variability with the position encoding with extrapolation property, and encode the item and item provider sequences accordingly. Meanwhile, to avoid the noise that the initial item embeddings might cause with Position Encoding, we calculate the position Encoding separately from the item embedding. In addition, we design a Cross Propagation layer to capture implicit higher-order dependencies between item sequences. Extensive experiments on three real-world datasets demonstrate that the proposed model generally outperforms the baselines by about 1–12.5%. Our source code will be published after the paper is published. [ABSTRACT FROM AUTHOR]