Naslov
Lexical Sense Labeling and Sentiment Potential Analysis using Corpus-Based Dependency Graph

Autori
Ban Kirigin, Tajana ; Bujačić Babić, Sanda ; Perak, Benedikt

Izvornik
Mathematics (2227-7390) 9 (2021), 12; 1449, 22

Vrsta, podvrsta i kategorija rada
Radovi u časopisima, članak, znanstveni

Ključne riječi
lexical graph analysis ; corpus ; knowledge representation and reasoning ; affective computing ; sentiment analysis

Sažetak
This article describes a graph method for labeling word senses and identifying sentiment potential of lexemes by integrating the corpus- based syntactic-semantic dependency graph layer, lexical semantic resources and sentiment dictionaries. The method, implemented as ConGraCNet application, projects a semantic function to a particular syntactical dependency corpus layer and constructs a seed lexeme graph with high conceptual similarity collocates. Seed lexeme graph is clustered into subgraphs that reveal polysemous semantic nature of a lexeme in a corpus. Using a WordNet, a hypernym graph is created to assert a set of synset labels for a lexical cluster of the seed lexeme, thus providing a generalization of community features. In addition, by integrating sentiment dictionaries, we describe graph propagation methods for sentiment analysis. Original dictionary sentiment values are integrated into ConGraCNet lexical graph in order to calculate sentiment values of node lexemes and lexical clusters, and to ultimately identify sentiment potential of a seed lexeme with respect to a specific corpus. The method can be used for resolving sparsity of sentiment dictionaries and enriching the sentiment evaluation of lexical structures in sentiment dictionaries, revealing relative sentiment potential of polysemous lexemes with respect to a specific corpus. We exemplify the application of the methodology on several lexemes in different languages and corpora and present the evaluation results of two surveys. The proposed approach has the potential to be used as a complementary method to other NLP contemporary resources for the enrichment of various semantic tasks including word disambiguation, domain relatedness, sense structure, synonymy, antonymy and metaphoricity, as well as establish a cross- and intra-cultural discourse variations of prototypical conceptualization patterns and knowledge representations.

Izvorni jezik
Engleski

Znanstvena područja
Matematika, Računarstvo, Filologija

Napomena
This article belongs to the Special Issue New
Trends in Graph and Complexity Based Data Analysis
and Processing

POVEZANOST RADA