Gabrilovich and Markovitch IJCAI 2007

From Cohen Courses
Jump to navigationJump to search

Citation

Evgeniy Gabrilovich and Shaul Markovitch. 2007. Computing Semantic Relatedness using Wikipedia-based Explicit Semantic Analysis. IJCAI'07 Proceedings of the 20th international joint conference on Artifical intelligence, 1606-1611.

Online version

AAAI

Summary

The paper presents a system for computing Semantic Relatedness.

The dataset used for evaluation is WordSimilarity-353 collection, and a collection 50 documents from the Australian Broadcasting Corporation's news mail service. WordSimilarity-353 collection has 353 pairs of words, and the other collection has 1,225 pairs of documents. Both have human judgments as gold standards.

They propose a method, called Explicit Semantic Analysis, which represents the meaning of any text in terms of natural concepts defined on large-scale knowledge repository such as Wikipedia [1] and Open Directory Project (ODP) [2] . After building a semantic interpreter that maps fragments of natural language text into a weighted sequence of Wikipedia concepts ordered by their relevance to the input. Then, input texts are represented as weighted vectors of concepts. Then, the semantic relatedness is calculated comparing their vectors, for example, using the cosine metric.The method is described in the picture below.

SemanticRelatedness Wikipedia.png

The methodology introduced shows substantial improvements over other ones as below.

SemanticRelatedness result.png


Their system can compute relatedness between words, sentences, and texts with the same method while some other researches on Semantic Relatedness limit the target as one of three.

Key Contribution

This system is the first statistical model on FrameNet solving the semantic role labeling problem, and future systems use the features introduced in this paper as a baseline. This paper is also very worth to read in that it describes the whole process of semantic role labeling in detail. In addition, they did many various experiments to find out which features, algorithms, and techniques affect the performance of the system.