Dynamic graphs such as the user-item interactions graphs and financial transaction networks are ubiquitous nowadays. While numerous representation learning methods for static graphs have been proposed, the study of dynamic graphs is still in its infancy. A main challenge of modeling dynamic graphs is how to effectively encode temporal and structural information into nonlinear and compact dynamic embeddings. To achieve this, we propose a principled graph-neural-based approach to learn continuous-time dynamic embeddings. We first define a temporal dependency interaction graph() that is induced from sequences of interaction data. Based on the topology of this , we develop a dynamic message passing neural network named TDIG-MPNN, which can capture the fine-grained global and local information on. In addition, to enhance the quality of continuous-time dynamic embeddings, a novel selection mechanism comprised of two successive steps, i.e., co-attention and gating, is applied before the above TDIG-MPNN layer to adjust the importance of the nodes by considering high-order correlation between interactive nodes'-depth neighbors on. Finally, we cast our learning problem in the framework of temporal point processes (TPPs) where we use TDIG-MPNN to design a neural intensity function for the dynamic interaction processes. Our model achieves superior performance over alternatives on temporal interaction prediction (including tranductive and inductive tasks) on multiple datasets.
A novel sports genre categorization algorithm based on representative shot extraction and geometry visual phrase(GVP) is presented in this paper. Performance of sports classification can be observably improved by generating reduced image set containing representative information and encoding spatial information into bag-of-words (BOW) model. Firstly, Shots containing significant information of videos are chosen by key-frame clustering. Secondly, GVP are searched by the co-occurrence of visual words in a spatial layout based on scale invariant feature transform (SIFT). Then visual words and GVP are concatenated to form enhanced histograms before SVM based classifying procedure. Compared with most existing methods, our algorithm is domain knowledge free as well as fully automatic and thus provides better extensibility. Experiments on a database of 10 sport genres with over 10257 minutes of videos from different sources achieved an average accuracy of 87.3%, which validates the robustness of our proposed algorithm over large-scale database.
In this paper, a novel non-supervised macro segmentation algorithm is presented by detecting duplicate sequences of large-scale TV videos. Motivated by the fact that "Inter-Programs" are repeatedly inserted into the TV videos, the macro structure of the videos can be effectively and automatically generated by identifying the special sequences. There are four sections in the algorithm, namely, keyframe extraction, discrete cosine transformbased feature generation(a fixed-size 64D signature), Locality-Sensitive Hashing (LSH)-based frame retrieval and macro segmentation through the duplicated sequence detection and the dynamic programming. The main contributions are: (1) supply one effective and efficient algorithm for the macro segmentation in the large-scale TV videos, (2) LSH can quickly query the similar frames, and (3) the non-supervised learned duplicate sequence models are used to find the lost duplicate sequences by the dynamic programming. The algorithm has been tested in 15-day different-type TV streams. The F -measure of the system is greater than 96%. The experiments show that it is efficient and effective for the macro segmentation.
scite is a Brooklyn-based organization that helps researchers better discover and understand research articles through Smart Citations–citations that display the context of the citation and describe whether the article provides supporting or contrasting evidence. scite is used by students and researchers from around the world and is funded in part by the National Science Foundation and the National Institute on Drug Abuse of the National Institutes of Health.
hi@scite.ai
10624 S. Eastern Ave., Ste. A-614
Henderson, NV 89052, USA
Copyright © 2024 scite LLC. All rights reserved.
Made with 💙 for researchers
Part of the Research Solutions Family.