Benchmarking Twitter sentiment models

In this project, a comparison was made between different sentiment models for the automatic categorization of Flemish Tweets.

Objective

The aim of this project is to compare different sentiment models for the production of a new statistic based on Flemish Tweets.

Data

For this project, a random sample of 47,000 Dutch tweets from Belgian Twitter accounts was collected over a period of 5 years.

Methods

In this project, different sentiment models were compared:

lexicon-based sentiment models
traditional machine learning models
standard deep learning models
attention-based deep learning models

Results

This research is currently being evaluated for publication in an academic journal. Once this research is published, a link to the publication will appear here.