Comparing and Benchmarking Semantic Measures Using SMComp

• Type: Article in International Conference Proceedings Book
• Date: 2016

Authors

Costa, T (Author) Other; José Paulo Leal (Author) FCUP

Conference proceedings International

Title: SLATE

Pages: 4:1-4:13

5th Symposium on Languages, Applications and Technologies, SLATE 2016

20 June 2016 through 21 June 2016

Indexing

Scopus - 0 Citations

Other information

Authenticus ID: P-00K-K0F

DOI: 10.4230/OASIcs.SLATE.2016.4

Abstract (EN): The goal of the semantic measures is to compare pairs of concepts, words, sentences or named entities. Their categorization depends on what they measure. If a measure only considers taxonomy relationships is a similarity measure; if it considers all type of relationships it is a relatedness measure. The evaluation process of these measures usually relies on semantic gold standards. These datasets, with several pairs of words with a rating assigned by persons, are used to assess how well a semantic measure performs. There are a few frameworks that provide tools to compute and analyze several well-known measures. This paper presents a novel tool - SMComp - a testbed designed for path-based semantic measures. At its current state, it is a domain-specific tool using three different versions of WordNet. SMComp has two views: one to compute semantic measures of a pair of words and another to assess a semantic measure using a dataset. On the first view, it offers several measures described in the literature as well as the possibility of creating a new measure, by introducing Java code snippets on the GUI. The other view offers a large set of semantic benchmarks to use in the assessment process. It also offers the possibility of uploading a custom dataset to be used in the assessment. © Teresa Costa and José Paulo Leal;licensed under Creative Commons License CC-BY.

Language: English

Type (Professor's evaluation): Scientific