A Wikipedia Two-Way Link Vector Model for Measuring Semantic Relatedness
Xinhua Zhu, Qingsong Guo, Bo Zhang · 2018
The measurement of the semantic relatedness between concepts is an important fundamental research topic in natural language processing. This paper proposes a Wikipedia two-way link vector model to extend the existing Wikipedia one-way out-link vector model. This vector model contains the weighted out-links and weighted in-links of concepts in Wikipedia and uses a TFIDF-based bidirectional weight method to uniformly calculate the strength of the mutual association between a given concept with its out-link or in-link concept. The bidirectional weight is equal to the sum of the weight for the link from the given concept to its link concept and the weight for the link from its link concept to the given concept. Moreover, we also propose a disambiguation strategy based on senses' social awareness 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. The experimental results demonstrate that our model surpasses the existing popular ESA and WCVM methods in the current Wikipedia versions and that our two-way link vector model significantly improves the performance of the existing one-way link vector model.