Abstract:In recent years,deep learning has been a hot research topic in traffic flow prediction.Graph convolutional networks outperform traditional convolutional neural networks in spatial feature modeling,in view of their powerful capabilities in processing non-Euclidean data such as topological map,distance map and flow similarity map.Therefore,graph convolutional network and its variants have become a research hotspot in traffic flow prediction,and many attractive research results have been obtained.This article classifies and summarizes traffic flow prediction models based on graph convolutional networks in recent years.First,the graph convolution is elaborated by combining the definitions of spatial convolution and spectral convolution.Second,in view of the network structure of the prediction model,the graph convolutional network based traffic flow prediction models are divided into two major categories of combined type and improved type,each of which are analyzed and discussed in detail with representative model structures.In addition,typical datasets commonly used in traffic flow prediction for model performance comparison are reviewed,and a simulation test is conducted using one real dataset to demonstrate the prediction performance of four traffic flow prediction models based on graph convolutional networks.Finally,the future research hotspots and challenges in traffic flow prediction based on graph convolutional networks are prospected.

Abstract:In view of the problems that aspect level sentiment analysis tasks cannot give full consideration to syntactic comprehensiveness and semantic relevance,and the graph volume used in most studies only considers the top-down dissemination of information and ignores the bottom-up aggregation of information,this paper proposes a sentiment analysis model based on attention and dual channel network.While expanding the dependency representation,the model uses self attention to obtain the information matrix with semantic relevance,and uses a dual channel network to combine comprehensive syntactic and semantic relevance information.The dual channel network focuses on the semantic features of top-down propagation and the structural features of bottom-up aggregation respectively.The graph convolution output in the channel will interact with the information matrix,pay attention to complement the residual,and then complete the tasks in the channel through average pooling.Finally,the final sentiment classification features are obtained by the fusion of semantic based and structure based decision-makings.The experimental results show that the accuracy and F1 value of the model are improved on three public data sets.

Abstract:Peoples awareness about their nutrition habits is increasing.Keeping track of what we eat will be helpful for us to follow a healthier diet.Currently,nutrient recognition of food images is mainly focused on food categories recognition,or is tackled as multi-label task recognition.These two approaches,however,are not very discriminative owing to their neglect of potential relationship between ingredients.In this paper,we introduce the relationship between ingredients to identify food nutrients based on previous work.The recognition approach includes two modules,namely the image feature extraction module and the ingredients relationship learning module.The low-dimensional image feature vectors are extracted by convolutional neural network (CNN),and the relationship between ingredients is learned through a graph convolutional network (GCN).Specifically,GCN uses graph data where nodes represent food ingredients as word embedding and edges represent the correlation between nodes.Then the GCN directly map the graph data into a set of interdependent classifiers.Finally,the low-dimensional image feature vectors are fused to make detailed classification.We conducted experiments on food data sets of Food-101 and VireoFood-172.Compared with state of the art food recognition methods,our GCN-based multi-label food image classification method offers very promising results and can effectively improve the recognition performance.