Enhanced sentiment extraction architecture for social media content analysis using capsule networks.

Area of Science:

• Artificial Intelligence
• Natural Language Processing
• Machine Learning

Background:

• Deep learning algorithms offer efficient text-based analytics applicable to social media.
• Convolutional Neural Networks (CNNs) and Recurrent Neural Networks (RNNs) show promise but have limitations in social media content analysis.
• Capsule Networks (CapsNets) present a potentially superior alternative for language modeling tasks.

Purpose of the Study:

• To propose and evaluate a novel approach for social media content analysis using capsule networks.
• To demonstrate the effectiveness of capsule networks in processing dynamic social media data, specifically on Twitter.
• To assess the performance of the proposed method without reliance on external linguistic resources.

Main Methods:

• Implementation of a capsule network architecture for text processing.
• Empirical evaluation on benchmark datasets: Twitter Sentiment Gold and CrowdFlower US Airline.
• Comparative analysis of performance against existing deep learning methods.

Main Results:

• The proposed capsule network approach achieved 86.87% accuracy on the Twitter Sentiment Gold dataset.
• An accuracy of 82.04% was obtained on the CrowdFlower US Airline dataset.
• The method demonstrated state-of-the-art performance, highlighting its optimality without linguistic resources.

Conclusions:

• Capsule networks are a viable and effective technique for social media content analysis, particularly for sentiment analysis.
• The proposed architecture offers significant accuracy enhancements in text processing for dynamic social media data.
• This research validates the applicability of capsule networks beyond image classification into natural language processing tasks.