使用LRCN模型進行情感分析
No Thumbnail Available
Date
2023
Authors
Journal Title
Journal ISSN
Volume Title
Publisher
Abstract
情感分析是近年來自然語言處理領域中的一個熱門研究方向。然而傳統的深度學 習分類模型往往難以捕捉到文本內在的複雜特徵，導致分類效果不佳。因此，本研究 使用深度學習的長期循環卷積網路(long term recurrent convolutional networks, LRCN)模型來進行文本的情感分析。同時我們也討論 Word2Vec 和GloVe 兩個預訓練模型所建構的詞向量空間對文本分析的影響，並以Reddit 和 Twitter 兩個資料集來探討這兩種模型的分類表現。同時我們也將LRCN 模型與傳統的深度學習模型：Convolutional Neural Network( 卷積神經網路 、Recurrent Neural Network ( 循環神經網路 、Long short term memory( 長短記憶神經網路 做比較，我們發現LRCN 能夠更有效的去捕捉字與字之間的空間特徵，通過循環網路層提取文字序列間的關係，有較佳的模型表現。
Sentiment analysis has emerged as a popular research direction in the field of natural language processing in recent years. However, traditional NLP classification models often struggle to capture the intricate features within texts, leading to subpar cla ssification results.Therefore, this study employs the deep learning model of Long term Recurrent Convolutional Networks (LRCN) for sentiment analysis of texts. I n addition, we also discuss the impacts of word vector spaces constructed by two pre trained m odels, Word2Vec and GloVe, on text analysis. We evaluate the classification performances of these two models using datasets from Reddit and Twitter. Furthermore , we compare the LRCN model with traditional deep learningmodels, including Convolutional Neura l Networks (CNN), Recurrent Neural Networks (RNN), and Long Short term Memory Networks (LSTM). Our findings suggest that the LRCN model can more effectively capture spatial features between words and extract relationships among text sequences through recur rent layers, leading to superior modelperformance.
Sentiment analysis has emerged as a popular research direction in the field of natural language processing in recent years. However, traditional NLP classification models often struggle to capture the intricate features within texts, leading to subpar cla ssification results.Therefore, this study employs the deep learning model of Long term Recurrent Convolutional Networks (LRCN) for sentiment analysis of texts. I n addition, we also discuss the impacts of word vector spaces constructed by two pre trained m odels, Word2Vec and GloVe, on text analysis. We evaluate the classification performances of these two models using datasets from Reddit and Twitter. Furthermore , we compare the LRCN model with traditional deep learningmodels, including Convolutional Neura l Networks (CNN), Recurrent Neural Networks (RNN), and Long Short term Memory Networks (LSTM). Our findings suggest that the LRCN model can more effectively capture spatial features between words and extract relationships among text sequences through recur rent layers, leading to superior modelperformance.
Description
Keywords
深度學習, 情感分析, 自然語言, Deep learning, Sentiment Analysis, NLP