Improving Sentiment Analysis Through Word Embeddings and Recurrent Neural Networks

This research proposes a sentiment classification framework integrating Word2Vec embeddings with a multilayer LSTM-based recurrent neural network to effectively capture contextual and sequential dependencies in text. The model was trained using a batch size of 30 and optimized with the Adam optimizer at a learning rate of 0.001, ensuring stable and efficient parameter updates. Word2Vec embeddings of dimension 32 were generated from a dataset of 2142 sentences and provided as input features to the LSTM layers. Training accuracy reached 99.6%, with testing accuracy peaking at 87.8%, demonstrating robust generalization. The cross-entropy loss showed consistent convergence, reducing from 0.56 in the first epoch to 0.03 in the final epoch. Compared to traditional approaches, the proposed model achieved higher precision, recall, and <tex xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">$\mathbf{F 1}$</tex>-scores, confirming its effectiveness in sentiment classification tasks and highlighting its potential for broader natural language processing applications.

Ҳали таржима қилинмаган