Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Automatic Design of Semantic Similarity Ensembles Using Grammatical Evolution

Jorge Martinez-Gil

TL;DR

This paper addresses the challenge of reliably measuring semantic similarity by automatically designing ensembles of similarity measures using grammatical evolution (GE). By encoding aggregation strategies as syntactically valid programs through a domain-specific Backus–Naur Form grammar, GE evolves ensemble formulas that maximize correlation with human judgments across multiple benchmarks, including MC30, GeReSiD50, and WS353. Empirical results show GE often outperforming traditional GP-based ensembles and simple baselines in SRCC, while offering interpretability and transferability through a grammar-guided, code-ready representation. The work provides a practical, scalable approach to adaptive similarity modeling and suggests pathways for future comparisons with deep learning models and extensions to other NLP tasks.

Abstract

Semantic similarity measures are a key component in natural language processing tasks such as document analysis, requirement matching, and user input interpretation. However, the performance of individual measures varies considerably across datasets. To address this, ensemble approaches that combine multiple measures are often employed. This paper presents an automated strategy based on grammatical evolution for constructing semantic similarity ensembles. The method evolves aggregation functions that maximize correlation with human-labeled similarity scores. Experiments on standard benchmark datasets demonstrate that the proposed approach outperforms existing ensemble techniques in terms of accuracy. The results confirm the effectiveness of grammatical evolution in designing adaptive and accurate similarity models. The source code that illustrates our approach can be downloaded from https://github.com/jorge-martinez-gil/sesige.

Automatic Design of Semantic Similarity Ensembles Using Grammatical Evolution

TL;DR

This paper addresses the challenge of reliably measuring semantic similarity by automatically designing ensembles of similarity measures using grammatical evolution (GE). By encoding aggregation strategies as syntactically valid programs through a domain-specific Backus–Naur Form grammar, GE evolves ensemble formulas that maximize correlation with human judgments across multiple benchmarks, including MC30, GeReSiD50, and WS353. Empirical results show GE often outperforming traditional GP-based ensembles and simple baselines in SRCC, while offering interpretability and transferability through a grammar-guided, code-ready representation. The work provides a practical, scalable approach to adaptive similarity modeling and suggests pathways for future comparisons with deep learning models and extensions to other NLP tasks.

Abstract

Semantic similarity measures are a key component in natural language processing tasks such as document analysis, requirement matching, and user input interpretation. However, the performance of individual measures varies considerably across datasets. To address this, ensemble approaches that combine multiple measures are often employed. This paper presents an automated strategy based on grammatical evolution for constructing semantic similarity ensembles. The method evolves aggregation functions that maximize correlation with human-labeled similarity scores. Experiments on standard benchmark datasets demonstrate that the proposed approach outperforms existing ensemble techniques in terms of accuracy. The results confirm the effectiveness of grammatical evolution in designing adaptive and accurate similarity models. The source code that illustrates our approach can be downloaded from https://github.com/jorge-martinez-gil/sesige.
Paper Structure (34 sections, 12 equations, 9 figures, 6 tables, 1 algorithm)

This paper contains 34 sections, 12 equations, 9 figures, 6 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. State-of-the-art
  3. Semantic Similarity
  4. Grammatical evolution
  5. Differences between Genetic Programming and Genetic Algorithm
  6. Contribution over the state-of-the-art
  7. Problem Statement
  8. Methods
  9. Mathematical Foundation
  10. Genotype
  11. Grammar
  12. Mapping Function
  13. Output
  14. Summary
  15. Fitness Function
  16. ...and 19 more sections

Figures (9)

  • Figure 1: Results for the a) PCC and b) SRCC over the MC30 benchmark dataset
  • Figure 2: Evolution of different variables during the ensemble learning process for PCC over the MC30 benchmark dataset
  • Figure 3: Evolution of different variables during the ensemble learning process for SRCC over the MC30 benchmark dataset
  • Figure 4: Results for the a) PCC and b) SRCC over the GeReSiD50 benchmark dataset
  • Figure 5: Evolution of key variables during the ensemble learning process for PCC over the GeReSiD50 dataset
  • ...and 4 more figures