An efficient approach for measuring semantic relatedness using Wikipedia bidirectional links
An efficient approach for measuring semantic relatedness using Wikipedia bidirectional links
复制标题
DOI：
10.1007/s10489-019-01452-1
复制
发表时间：
2019-05
影响因子：
5.3
通讯作者：
Xin-hua Zhu;Qingsong Guo-;Bo Zhang;Fei Li
中科院分区：
文献类型：
--
作者：
Xin-hua Zhu;Qingsong Guo-;Bo Zhang;Fei Li
概念之间语义相关性的度量是自然语言处理中的一个重要的基础研究课题。基于链接的模型是基于维基百科的度量中最有前途的相关性方法，因为它在维基百科中手动定义的链接是精炼的并且接近人类的语义。本文提出一种维基百科双向链接模型来扩展现有维基百科单向出链模型，该模型具有低维度、高效率、易于实现和重复的特点。首先，该模型利用维基百科中概念的出链和入链组合成概念语义解释器的双向链接向量，并使用基于TF*IDF的双向权重方法来统一计算给定概念与其出链或入链概念之间的相互关联强度。其次，我们提出了一种基于语义社会意识的消歧策略，该策略直接按照消歧页面中出现的顺序对消歧页面内的外链进行排序，并采用可调整的阈值来确定将选择多少个语义。此外，我们还提出了基于对数和指数的新向量相似性度量，以提高基于维基百科链接的语义相关性测量的综合性能。在一些公认的数据集上的实验结果表明，我们的模型超越了当前维基百科版本中现有流行的朴素显式语义分析（Naïve-ESA）和基于维基百科外链接向量的测量（WOLM）方法，并且我们的双向链接模型在实际应用中显着提高了现有单向链接模型的性能。
The measurement of the semantic relatedness between concepts is an important fundamental research topic in natural language processing. The link-based model is the most promising relatedness method in Wikipedia-based measures because its manually defined links in Wikipedia are refined and close to the semantics of humans. This paper proposes a Wikipedia two-way link model to extend the existing Wikipedia one-way out-link model, which has a low dimension and a high efficiency, as well as being easy to implement and repeat. First, this model utilizes the out-links and in-links of concepts in Wikipedia to combine into a bidirectional link vector for concept semantic interpreter and uses a TF*IDF-based bidirectional weight method to uniformly calculate the strength of the mutual association between a given concept and its out-link or in-link concept. Second, we propose a disambiguation strategy based on the social awareness of senses that directly sorts the out-links within a disambiguation page in the order in which they occur in the disambiguation page and adopts an adjustable threshold to determine how many senses will be selected. Moreover, we also propose new vector similarity metrics based on logarithm and exponent to improve the comprehensive performance of the semantic relatedness measurements based on Wikipedia links. The experimental results on some well-recognized datasets demonstrate that our model surpasses the existing popular Naïve Explicit Semantic Analysis (Naïve-ESA) and Wikipedia Out-Link vector-based Measure (WOLM) methods in the current Wikipedia versions and that our bidirectional link model significantly improves the performance of the existing one-way link model in practical applications.