NLP - Embedding Dimensionality Reduction Approaches

Project repository of group 14, nlp course @ university of bonn

Members

• Arwah Jawwad
• Behnam Fanitabasi
• Behzad Shomali
• Mahnaz Mirhaj
• Sven Knauer

Project Outline

• project_outline

Dataset

• We use three different datasets: Aspect Sentiment Classification Dataset - CheckThat Lab 2023 - Subjectivity Detection - FEVER
• The embeddings are calculated using SBERT.

Approach 1

• Variational Autoencoder

Architecture of Variational Autoencoder (source)

Approach 2

• Traditional Dimensionality Reduction techniques such as PCA, LDA, and t-SNE.

Approach 3

• Inspired by: A novel approach for dimension reduction using word embedding: An enhanced text classification approach Link Singh et al., International Journal of Information Management Data Insights Volume 2, Issue 1, April 2022

Results and Report

• Experiments are done with ML classifiers such as SVM, MLP, and Random Forest on the three datasets, Naive Bayes, and KNN. F1-Scores, Time Inference, and Memory Consumption Plots: Result
• Report: Final report