Traffic event description based on Twitter data using Unsupervised Learning Methods for Indian road conditions

TL;DR

This paper proposes an unsupervised learning approach using Twitter data and word embeddings to classify traffic-related tweets, aiming to improve real-time traffic event detection in Indian roads where traditional methods are limited.

Contribution

It introduces a novel unsupervised model leveraging social media data and semantic similarity for traffic event description in Indian road conditions.

Findings

Achieved 94.7% test score in tweet classification

Enhanced traffic event detection using social media data

Addressed data sparsity issues in Indian traffic monitoring

Abstract

Non-recurrent and unpredictable traffic events directly influence road traffic conditions. There is a need for dynamic monitoring and prediction of these unpredictable events to improve road network management. The problem with the existing traditional methods (flow or speed studies) is that the coverage of many Indian roads is very sparse and reproducible methods to identify and describe the events are not available. Addition of some other form of data is essential to help with this problem. This could be real-time speed monitoring data like Google Maps, Waze, etc. or social data like Twitter, Facebook, etc. In this paper, an unsupervised learning model is used to perform effective tweet classification for enhancing Indian traffic data. The model uses word-embeddings to calculate semantic similarity and achieves a test score of 94.7%.