Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Bridging Emotions and Architecture: Sentiment Analysis in Modern Distributed Systems

Mahak Shah, Akaash Vishal Hazarika, Meetu Malhotra, Sachin C. Patil, Joshit Mohanty

TL;DR

This paper investigates bridging sentiment analysis with distributed systems to handle large-scale text data efficiently. It compares a single-node pipeline against a four-node distributed Spark-based architecture that uses BERT embeddings and logistic regression on the Sentiment140 dataset. Results show a 75% reduction in processing time and a modest accuracy gain (about 3.5 percentage points) with distributed processing, while lowering per-node resource requirements. The work demonstrates the practicality of scalable, real-time capable NLP pipelines for big data contexts and outlines concrete future directions, including federated learning and hardware acceleration.

Abstract

Sentiment analysis is a field within NLP that has gained importance because it is applied in various areas such as; social media surveillance, customer feedback evaluation and market research. At the same time, distributed systems allow for effective processing of large amounts of data. Therefore, this paper examines how sentiment analysis converges with distributed systems by concentrating on different approaches, challenges and future investigations. Furthermore, we do an extensive experiment where we train sentiment analysis models using both single node configuration and distributed architecture to bring out the benefits and shortcomings of each method in terms of performance and accuracy.

Bridging Emotions and Architecture: Sentiment Analysis in Modern Distributed Systems

TL;DR

This paper investigates bridging sentiment analysis with distributed systems to handle large-scale text data efficiently. It compares a single-node pipeline against a four-node distributed Spark-based architecture that uses BERT embeddings and logistic regression on the Sentiment140 dataset. Results show a 75% reduction in processing time and a modest accuracy gain (about 3.5 percentage points) with distributed processing, while lowering per-node resource requirements. The work demonstrates the practicality of scalable, real-time capable NLP pipelines for big data contexts and outlines concrete future directions, including federated learning and hardware acceleration.

Abstract

Sentiment analysis is a field within NLP that has gained importance because it is applied in various areas such as; social media surveillance, customer feedback evaluation and market research. At the same time, distributed systems allow for effective processing of large amounts of data. Therefore, this paper examines how sentiment analysis converges with distributed systems by concentrating on different approaches, challenges and future investigations. Furthermore, we do an extensive experiment where we train sentiment analysis models using both single node configuration and distributed architecture to bring out the benefits and shortcomings of each method in terms of performance and accuracy.

Paper Structure

This paper contains 33 sections, 6 equations, 2 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Sentiment Analysis Techniques
  4. Distributed Computing
  5. Distributed Sentiment Analysis
  6. Methodology
  7. Experimental Setup
  8. Tools and Technologies Used
  9. Model Training
  10. Training on a Single Node
  11. Training in Distributed Architecture
  12. Complexity Analysis
  13. Training Phase Complexity
  14. Testing Phase Complexity
  15. Communication Overhead
  16. ...and 18 more sections

Figures (2)

  • Figure 1: Sentiment Analysis in Distributed and Single-Node Environment
  • Figure 2: Network Topology of the Distributed System