ID 原文 译文
56838 例如,给定查询文本"一个人在打篮球"时,现有检索系统将根据整个查询文本和的视频的特征,或者关注于文本与视频中所表现的实体(如"人","篮球")来计算合适的视频片段,而缺乏对于"人打篮球"这类语义关系的考虑.因此,它们将难以辨别语义关系上的不同,从而限制了检索质量的提升. For example, given the query text “a person is playing basketball”,existing retrieval systems may incorrectly return a video moment of “a person holding a basketball” without theconsidering the semantic relationship of “a person playing basketball”.
56839 为了解决这个问题,本文提出跨模态关系对齐的图卷积框架CrossGraphAlign,通过分别构建文本关系图(textural relationship graph)与视觉关系图(visual relationship graph)来建模查询文本与视频片段中的语义关系,再通过跨模态对齐的图卷积网络来评估文本关系与视觉关系的相似度,从而帮助构建更加精准的视频片段检索系统. Therefore, this paper proposes a cross?modal relationship alignment framework, which we refer to as CrossGraphAlign, for cross-modal video momentretrieval. The proposed framework constructs a textual relationship graph and a visual relationship graph tomodel the query semantics in text and video segment relations, and then evaluates the similarity between textrelations and visual relations through cross-modally aligned graph convolutional networks to help construct amore accurate video moment retrieval system.
56840 在公开的跨模态视频片段检索数据集TACoS和ActivityNet Captions上的实验结果表明,本文提出的方法可以有效地利用语义关系来提升跨模态视频片段检索的召回率. Experimental results on the publicly available cross-modal videoretrieval datasets TACoS and ActivityNet Captions demonstrate that the proposed method can effectively utilizethe semantic relationships to improve the recall rate in cross-modal video moment retrieval.
56841 随着万维网的发展,知识图谱数据大量增长,并在面向智能应用的研究中受到广泛关注. The development of the World Wide Web has triggered substantial growth of knowledge graphs (KG).
56842 知识图谱用RDF (resource description framework)三元组描述实体相关的事实. Research into using KGs for intelligent applications has increased significantly. A KG describes facts about entitiesusing RDF triples, and an entity description may contain a large number of triples.
56843 在知识图谱中,关于一个实体的描述可能包含大量三元组,在一些需要直接呈现实体信息的应用中,为了避免用户信息过载,并适应有限的呈现空间,就需要进行实体摘要.实体摘要任务是从实体描述的众多三元组中选出最有代表性的子集作为摘要,以呈现给用户阅读. In applications where entityinformation is presented directly, entity summarization is required to prevent user information overload and tofit the presentation capacity. Here, the task is to select the most representative subset of triples from the richentity description.
56844 本文提出一种新的实体摘要方法 ESSTER以生成具备高可读性和低冗余性的实体摘要. In this paper, we propose an innovative entity summarization method, which we refer to asESSTER, to generate summaries with both high readability and low redundancy.
56845 该方法结合三元组的结构与文本特征,基于结构特性度量知识图谱中三元组的重要性,基于N元语法和文本语料度量三元组的可读性,基于逻辑推理、数值比较和文本相似判断三元组间的冗余关系. The proposed method combinesstructural and textual features. The importance of a triple is measured based on its structural features in the KG. The text readability of a triple is measured based on n-grams in a text corpus, and redundancy in a set of triplesis measured by logical reasoning, numeric comparison, and text similarity.
56846 综合这3种技术要素,将实体摘要问题建模为组合优化问题进行求解. Entity summations is modeled and bycombining these three measures and solved as a combinatorial optimization problem.
56847 本文在两个由人工标注的公开数据集上与6种现有方法进行了对比实验,结果表明本文提出的方法效果达到了当前最佳水平. We conducted experimentsand compared the proposed method to six existing methods on two publicly available datasets of manually labeledsummaries. Experimental results demonstrate that the proposed method achieves state of the art results.