Emotion detection in short texts using deep learning and decision threshold optimization with differential evolution: Reproducibility Package

Resource information

Format

zip

Original repository

22148550

DOI

10.5281/zenodo.22148550

Credit basis

Creators in Zenodo record metadata

Access

Open files available from the original Zenodo deposit

Last checked

Nov 5, 2026

License

Not specified

Last record update

Nov 5, 2026

About this resource

Emotion-Classification: Experimental Code and Supporting Materials This repository contains the experimental code and supporting materials used for emotion classification experiments on Spanish tweets. The resource is intended to facilitate the reproducibility of the methodology, data-processing workflow, model evaluation, and statistical analysis described in the associated research article. The repository provides Python-based scripts covering the main stages of the experimental workflow, including tweet corpus preprocessing, feature extraction and vectorization, model training and evaluation, classification analysis, carbon-emission estimation, and statistical significance analysis. Repository contents The repository is organized into the following directories: Corpus_Tweets/ — Scripts and resources associated with corpus preprocessing. Characteristics/ — Scripts for feature extraction and vectorization, including the use of sentiment lexicons and Spanish word embeddings. Test_Models/ — Scripts used for model training and testing, together with the classification reports and carbon-emission reports generated during the experiments. Statistical_Significance/ — Scripts and results associated with the statistical analysis of the best-performing models. Machine learning and deep learning models The experimental framework includes several machine learning, deep learning, and large language model approaches. The machine learning models include Support Vector Machines implemented with LinearSVC , Logistic Regression, CatBoost, and XGBoost. A Convolutional Neural Network (CNN) was implemented using TensorFlow/Keras. The experiments also include transformer-based models using the Hugging Face Transformers framework, specifically: bert-base-multilingual-cased dccuchile/bert-base-spanish-wwm-cased ClassCat/roberta-base-spanish Large language model experiments include Google's Gemini ( gemini-2.0-flash ) through the Google GenAI library and Meta's Llama 3.1 70B model executed locally through Ollama. Differential Evolution optimization is implemented using scipy.optimize.differential_evolution . External resources The experiments rely on several third-party datasets and resources that are not redistributed with this repository . Users must obtain these resources directly from their respective original sources and comply with their applicable licenses and terms of use. The external resources include: SemEval-2018 Task 1, Subtask E-c — Emotion Classification , Spanish dataset, used as the emotion classification corpus. ML-Senticon , used as a Spanish sentiment lexicon. SEL (Spanish Emotion Lexicon) , used as an emotion-related lexical resource. Full-Strength Lexicon , used as a sentiment-related lexical resource. Spanish FastText word vectors ( wiki.es.vec ) , used as pretrained Spanish word embeddings. Pretrained transformer models , obtained through the Hugging Face Transformers ecosystem. Gemini and Llama 3.1 , accessed through their respective software interfaces and services. Syntactic n-gram extraction script , based on the MultiSngram implementation by Grigori Sidorov, used to obtain syntactic n-grams from the text corpus. The repository provides the scripts and configuration necessary to use these resources in the experimental workflow, but does not redistribute third-party models, embeddings, datasets, or lexical resources. Reproducibility The experiments were implemented in Python and are intended to be executed in a Linux-based environme