Learning LexRank——Graph-based Centrality as Salience in Text Summarization(一)
(1)What is Sentence Centrality and Centroid-based Summarization ?
Extractive summarization works by choosing a subset of the sentences in the original documents. This process can be viewed as identifying the most central sentences in a (multi-document) cluster that give the necessary and sufficient amount of information related to the main theme of the cluster.
The centroid of a cluster is a pseudo-document which consists of words that have tf×idf scores above a predefined threshold, where tf is the frequency of a word in the cluster, and idf values are typically computed over a much larger and similar genre data set.
In centroid-based summarization (Radev, Jing, & Budzikowska, 2000), the sentences that contain more words from the centroid of the cluster are considered as central. This is a measure of how close the sentence is to the centroid of the cluster.
(2)Centrality-based Sentence Salience:
All of our approaches are based on the concept of prestige in social networks. A social network is a mapping of relationships between interacting entities (e.g. people, organizations, computers). Social networks are represented as graphs, where the nodes represent the entities and the links represent the relations between the nodes.
A cluster of documents can be viewed as a network of sentences that are related to each other. We hypothesize that the sentences that are similar to many of the other sentences in a cluster are more central (or salient) to the topic.
There are two points to clarify in this definition of centrality:
1.How to define similarity between two sentences.
2.How to compute the overall centrality of a sentence given its similarity to other sentences.
To define similarity, we use the bag-of-words model to represent each sentence as an N-dimensional vector, where N is the number of all possible words in the target language. For each word that occurs in a sentence, the value of the corresponding dimension in the vector representation of the sentence is the number of occurrences of the word in the sentence times the idf of the word. The similarity between two sentences is then defined by the cosine between two corresponding vectors:
A cluster of documents may be represented by a cosine similarity matrix where each entry in the matrix is the similarity between the corresponding sentence pair.
Figure 1 shows a subset of a cluster used in DUC 2004, and the corresponding cosine similarity matrix. Sentence ID dXsY indicates the Y th sentence in the Xth document.
Figure 1: Intra-sentence cosine similarities in a subset of cluster d1003t from DUC 2004.
This matrix can also be represented as a weighted graph where each edge shows the cosine similarity between a pair of sentence (Figure 2).
Figure 2: Weighted cosine similarity graph for the cluster in Figure 1.
(3)Degree Centrality:
Since we are interested in significant similarities, we can eliminate some low values in this matrix by defining a threshold so that the cluster can be viewed as an (undirected) graph.
Figure 3 shows the graphs that correspond to the adjacency matrices derived by assuming the pair of sentences that have a similarity above 0.1, 0.2, and 0.3, respectively, in Figure 1 are similar to each other. Note that there should also be self links for all of the nodes in the graphs since every sentence is trivially similar to itself. Although we omit the self links for readability, the arguments in the following sections assume that they exist.
-----------------------------------------------------------------------
-----------------------------------------------------------------------
Figure 3: Similarity graphs that correspond to thresholds 0.1, 0.2, and 0.3, respectively, for the cluster in Figure 1.
A simple way of assessing sentence centrality by looking at the graphs in Figure 3 is to count the number of similar sentences for each sentence. We define degree centrality of a sentence as the degree of the corresponding node in the similarity graph. As seen in Table 1, the choice of cosine threshold dramatically influences the interpretation of centrality. Too low thresholds may mistakenly take weak similarities into consideration while too high thresholds may lose many of the similarity relations in a cluster.
Table 1: Degree centrality scores for the graphs in Figure 3. Sentence d4s1 is the most central sentence for thresholds 0.1 and 0.2.
Learning LexRank——Graph-based Centrality as Salience in Text Summarization(一)的更多相关文章
- Deep Learning of Graph Matching 阅读笔记
Deep Learning of Graph Matching 阅读笔记 CVPR2018的一篇文章,主要提出了一种利用深度神经网络实现端到端图匹配(Graph Matching)的方法. 该篇文章理 ...
- Learning Context Graph for Person Search
Learning Context Graph for Person Search 2019-06-24 09:14:03 Paper:http://openaccess.thecvf.com/cont ...
- Learning Conditioned Graph Structures for Interpretable Visual Question Answering
Learning Conditioned Graph Structures for Interpretable Visual Question Answering 2019-05-29 00:29:4 ...
- 《Deep Learning of Graph Matching》论文阅读
1. 论文概述 论文首次将深度学习同图匹配(Graph matching)结合,设计了end-to-end网络去学习图匹配过程. 1.1 网络学习的目标(输出) 是两个图(Graph)之间的相似度矩阵 ...
- DAG-GNN: DAG Structure Learning with Graph Neural Networks
目录 概 主要内容 代码 Yu Y., Chen J., Gao T. and Yu M. DAG-GNN: DAG structure learning with graph neural netw ...
- Graph Based SLAM 基本原理
作者 | Alex 01 引言 SLAM 基本框架大致分为两大类:基于概率的方法如 EKF, UKF, particle filters 和基于图的方法 .基于图的方法本质上是种优化方法,一个以最小化 ...
- 论文解读( N2N)《Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximization》
论文信息 论文标题:Node Representation Learning in Graph via Node-to-Neighbourhood Mutual Information Maximiz ...
- 论文解读(GMT)《Accurate Learning of Graph Representations with Graph Multiset Pooling》
论文信息 论文标题:Accurate Learning of Graph Representations with Graph Multiset Pooling论文作者:Jinheon Baek, M ...
- Learning Latent Graph Representations for Relational VQA
The key mechanism of transformer-based models is cross-attentions, which implicitly form graphs over ...
随机推荐
- 模板引擎 Velocity
模板引擎 Velocity 一.Velocity简介: Velocity是一个基于java的模板引擎(template engine).它允许任何人仅仅简单的使用模板语言(template langu ...
- 【剑指Offer学习】【面试题40:数组中仅仅出现一次的数字】
题目:一个整型数组里除了两个数字之外.其它的数字都出现了两次,请敲代码找出这两个仅仅出现一次的数字. 要求时间复杂度是O(n),空间复杂度是O(1). 举例说明 比如输入数组{2, 4, 3, 6, ...
- JVM经常使用的调优參数
-Xms512m -Xmx1024m -XX:MaxPermSize=1204m -XX:+UseConcMarkSweepGC -XX:+CMSClassUnloadingEnabled -XX:+ ...
- chrome 全屏无法退出
起因:chrome 按F11 全屏后,再按F11 死活无法退出全屏,关掉后,再打开chrome,自动全屏,F11无法退出全屏 系统:centos 6.3 解决:找到 chrome的用户数据存储目录,( ...
- ZooKeeper监控
http://jm-blog.aliapp.com/?p=1450 在公司内部,有不少应用已经强依赖zookeeper,zookeeper的工作状态直接影响它们的正常工作.目前开源世界中暂没有一个比较 ...
- [D3] 2. Basics of SVG
1. svg should use 'fill' prop instead 'background-color'2. svg width & height no need 'px'3. att ...
- onInterceptTouchEvent和onTouchEvent调用时序
onInterceptTouchEvent()是ViewGroup的一个方法,目的是在系统向该ViewGroup及其各个childView触发onTouchEvent()之前对相关事件进行一次拦截,A ...
- 使用like时left outer join和inner join的区别
--select top 10000 * into #s from search set statistics time on set statistics io on select userId,c ...
- android开发之重写Application类
在android应用开发中,重写Application也算是比较常见的,以前开发的一些程序太过于简单,都不要重写这个类,但是在真正的商业开发中,重写Application类几乎是必做的. 为什么要重写 ...
- win8 需要管理员权限才能删除此应用程序
在win8系统里面 安装了 dotfuscator,安装在c盘,结果发现运行是提示需要有管理员权限,想重新删除后安装也不行,删除的时候提示也需要管理员权限,网上查了一下资料,windows自xp开始, ...