Leveraging Big Data and Sentiment Analysis for Actionable Insights: A Review of Data Mining Approaches for Social Media

Abstract

The explosive growth of social media has generated vast troves of user-generated data reflecting opinions, sentiments, and interactions. This "big data" offers immense potential for extracting insights to guide business, government, and societal decisions if appropriate analytics techniques can be applied. Sentiment analysis specifically aims to computationally identify and characterize subjective information like stances, attitudes, or feelings. Combining big data and sentiment analysis for social media thus presents both great opportunities and challenges. This paper reviews the state-of-the-art in data mining and sentiment analysis approaches tailored for big social media data. Key background concepts in big data, social media platforms, and sentiment analysis techniques are first introduced. Characterizing big data by its high volume, velocity, and variety, key properties of major social media sites like Facebook, Twitter, Instagram, and YouTube are highlighted that impact analysis methods. An overview of core sentiment analysis approaches - lexicon-based, machine learning, deep learning - establishes basic techniques subsequently extended. Sampling methodologies, feature engineering processes, and learning algorithms for working with large-scale, streaming social data are then discussed. Both supervised and unsupervised strategies are covered, using benchmarks and evaluation metrics to assess performance. This sets the stage for current sentiment analysis models leveraging this big data mining functionality. Lexicon expansion methods, neural networks, and multimodal architectures that combine text, image and video data are reviewed. State-of-the-art capabilities and limitations are objectively presented for these rapidly evolving techniques. The paper concludes with an analysis of impactful applications in public health, marketing, social prediction, and security. Product recommendation systems and customer experience analytics exemplify business uses, while public health monitoring and event warning systems showcase societal benefits. Remaining challenges and opportunities for innovation in models and applications are summarized to chart promising directions for future research. The comprehensive review of data mining and sentiment analysis techniques for big social media data presented in this paper aims to both take stock of current accomplishments and help guide the next wave of advances in this burgeoning research domain.