본문 바로가기
  • Home

Minimizing the Loss Values in the CBOW Models

  • Journal of The Korea Society of Computer and Information
  • Abbr : JKSCI
  • 2025, 30(1), pp.65-72
  • Publisher : The Korean Society Of Computer And Information
  • Research Area : Engineering > Computer Science
  • Received : December 6, 2024
  • Accepted : December 30, 2024
  • Published : January 31, 2025

Gukbeom Yoon 1 Rhee,Jung-Soo 1

1부산외국어대학교

Accredited

ABSTRACT

The Continuous Bag of Words (CBOW) model is a popular technique in natural language processing (NLP) used to generate word Embeddings. It predicts a target word given its surrounding context words. The model consists of an input layer, a hidden layer, and an output layer. The PTB, which is commonly used as a benchmark for evaluating the performance of the CBOW model, is a medium-sized corpus that plays an important role in natural language processing and computational linguistics. It consists of 2,499 articles extracted from the Wall Street Journal in 1989, containing approximately 1 million words and 49,208 sentences. This paper aims to improve the average loss value of the loss function by applying batch normalization to the CBOW model and training it on the PTB dataset. To achieve the objective of this paper, experiments were conducted in a notebook environment equipped with CuPy, comparing the original CBOW model with the batch normalized CBOW model. The results shows that average loss value decreased from 1.25 to 0.65. Therefore, this paper demonstrates the effectiveness of batch normalization in improving the performance of the CBOW model and is expected to contribute to refining distributional representation of words for transfer learning.

Citation status

* References for papers published after 2023 are currently being built.

This paper was written with support from the National Research Foundation of Korea.