返回Google圖書搜尋

Leveraging Textual Information for Social Media News Categorization and Sentiment Analysis

Mahmudul HasanTanver AhmedMd. Palash Uddin

出版	SSRN, 2023
URL	http://books.google.com.hk/books?id=1Yk20AEACAAJ&hl=&source=gbs_api

註釋The recent explosion in the popularity of social media platforms has fundamentally altered how people think about professional and personal connections. Data from social networks are increasingly being used for various reasons, including election prediction, sentimental analysis, marketing, communication, business, and education. Sentiment analysis and news categorization are crucial tasks to understand human feelings and access the news without much effort. In this paper, we first apply supervised and unsupervised Machine Learning (ML) algorithms for news categorization. After this, we propose a blending ensemble algorithm that outperforms the classical ML algorithms. Then, we process both structured and unstructured data for sentiment analysis based on the polarity of the text using TextBlob, which builds upon on natural language toolkit for processing textual data. We investigate Support Vector Machine, k-nearest Neighbors, Decision Tree, AdaBoost, Logistic Regression, Stochastic Gradient Descent (SGD), Ridge Classifier (RC), and Naive Bayes as supervised ML algorithms and K-Means Clustering and Non-negative Matrix Factorization as unsupervised ML algorithms. After evaluating the ML algorithms, we preprocess the text and propose an ensemble blending SGDR classifiers that build upon SGD and RC. The performance of the proposed ensemble outperforms all the algorithms. It shows 98.12% accuracy and besides this, the performance of all the algorithms increased after applying the string preprocessing technique at a significant rate. The result also indicates that linear models are more familiar than the tree-based and nonlinear models for news categorization.