EUVIMEDEUVIMED entdecken
Open Access · CC BY-NC 4.0

Sentiment Analysis of Social Media Data Using Natural Language Processing Techniques

Hemlata, Diksha, Urvinder, Suman

International Journal of Research and Review in Applied Science, Humanities, and Technology · 28.12.2024 · 243-251

DOI: 10.71143/m2avs350 ↗

Kurz gefasst

Fachartikel in der Zeitschrift International Journal of Research and Review in Applied Science, Humanities, and Technology, online erschienen am 28.12.2024. Verfasst von Hemlata und 3 weiteren Autor:innen. Der Volltext ist frei zugänglich (Open Access, CC BY-NC 4.0).

AbstractBibliografische AngabenZitierenVerwandte Studien

Abstract (Originalsprache)

Social media platforms carry millions of short, informal messages every day, and identifying the sentiment they express is valuable for brand monitoring, public-opinion research and customer service. Tweets are difficult to analyse because of slang, hashtags, user mentions, elongated words, emoticons and negation. This paper presents a lightweight natural language processing (NLP) pipeline for three-class (negative, neutral, positive) sentiment analysis of tweets. The pipeline normalises social-media text, extracts word n-gram and character n-gram TF-IDF features, adds sentiment-lexicon scores from VADER and TextBlob, and classifies the fused feature vector with class-balanced logistic regression. The approach is evaluated on the public TweetEval sentiment benchmark derived from SemEval-2017 Task 4 (45,615 training and 12,284 test tweets) and compared with two lexicon-based methods and six machine learning models. The proposed model achieves an accuracy of 64.21%, a macro recall (the official metric) of 65.01% and a macro F1-score of 63.89%, outperforming all baselines; the best single machine learning model, logistic regression on word features, reaches 62.08% macro recall and VADER reaches 57.01%. The model also exceeds the published SVM, FastText and LSTM results on the same benchmark, while remaining below large, pre-trained transformers. It classifies a tweet in about 1.7 ms on a single CPU core. Ablation, learning-curve and error analyses show that lexicon features give the largest gain, that the model is data-efficient, and that most remaining errors involve the neutral class.

Zitieren

APA

Hemlata, Diksha, Urvinder, & Suman (2024). Sentiment Analysis of Social Media Data Using Natural Language Processing Techniques. International Journal of Research and Review in Applied Science, Humanities, and Technology, 243-251. https://doi.org/10.71143/m2avs350

Vancouver

Hemlata, Diksha, Urvinder, Suman. Sentiment Analysis of Social Media Data Using Natural Language Processing Techniques. International Journal of Research and Review in Applied Science, Humanities, and Technology. 2024:243-251. doi:10.71143/m2avs350

BibTeX
@article{Hemlata2024,
  author = {Hemlata and Diksha and Urvinder and Suman},
  title = {Sentiment Analysis of Social Media Data Using Natural Language Processing Techniques},
  journal = {International Journal of Research and Review in Applied Science, Humanities, and Technology},
  year = {2024},
  pages = {243--251},
  doi = {10.71143/m2avs350}
}
Kostenlos starten →