[Paper Review] Simultaneous Identification of Tweet Purpose and Position

Rahul Radhakrishnan Iyer, Yulong Pei+1 more|arXiv (Cornell University)|Dec 24, 2019

Text and Document Classification Technologies24 references4 citations

TL;DR

This paper proposes a multi-label classification framework to simultaneously identify tweet purpose (e.g., information sharing, social interaction) and position (supporting, opposing, neutral) toward a topic, leveraging label correlations for improved accuracy. Using RAkEL with post-processing via summation and weighted summation strategies, the method outperforms individual classifiers, achieving a 29.65% improvement in F1-score over KNN on the Obama care dataset.

ABSTRACT

Tweet classification has attracted considerable attention recently. Most of the existing work on tweet classification focuses on topic classification, which classifies tweets into several predefined categories, and sentiment classification, which classifies tweets into positive, negative and neutral. Since tweets are different from conventional text in that they generally are of limited length and contain informal, irregular or new words, so it is difficult to determine user intention to publish a tweet and user attitude towards certain topic. In this paper, we aim to simultaneously classify tweet purpose, i.e., the intention for user to publish a tweet, and position, i.e., supporting, opposing or being neutral to a given topic. By transforming this problem to a multi-label classification problem, a multi-label classification method with post-processing is proposed. Experiments on real-world data sets demonstrate the effectiveness of this method and the results outperform the individual classification methods.

Motivation & Objective

• To address the inefficiency and missed correlation in separate tweet purpose and position classification.
• To exploit the inherent correlation between tweet purpose and user position toward a topic for improved classification.
• To develop a unified framework that simultaneously predicts both tweet purpose and position.
• To introduce post-processing strategies that refine multi-label predictions and improve classification performance.
• To validate the method on real-world Twitter datasets related to political topics like Obama care and death penalty.

Proposed method

• Transforms the joint tweet purpose and position classification task into a multi-label classification problem to capture label correlations.
• Employs the RAkEL algorithm for multi-label classification using n-gram, POS, and statistical features.
• Applies two post-processing strategies—summation and weighted summation—on KNN-based neighbor predictions to refine label outputs.
• Uses KNN to identify the K most similar training instances for each test tweet to inform post-processing decisions.
• Integrates feature sets including 1-gram, 2-gram, POS tags, and statistical features (e.g., word count, punctuation frequency) for improved representation.
• Optimizes hyperparameters such as K (number of neighbors) and evaluates sensitivity across different values.

Experimental results

Research questions

• RQ1 Can joint prediction of tweet purpose and position improve classification performance compared to individual classification?
• RQ2 How effective are post-processing strategies in refining multi-label predictions for tweet classification?
• RQ3 What is the impact of different feature combinations (e.g., n-grams, POS, statistics) on classification accuracy?
• RQ4 How sensitive is the post-processing performance to the choice of K (number of neighbors)?
• RQ5 Can label correlations between purpose and position be effectively captured and leveraged in a multi-label framework?

Key findings

• The RAkEL multi-label classifier with weighted summation post-processing achieved a 29.65% improvement in F1-score over KNN on the Obama care dataset.
• RAkEL with post-processing outperformed both KNN and SVM classifiers on both the Obama care and death penalty datasets.
• The combination of (1+2)-gram, POS, and statistical features yielded the best performance, improving F1 by 10.04% compared to 1-gram features alone on the Obama care dataset.
• The weighted summation post-processing strategy consistently outperformed the simple summation strategy across both datasets.
• The method showed low sensitivity to the K parameter in post-processing, with performance differences less than 0.01 across K values from 2 to 30.
• The results validate that modeling correlations between tweet purpose and position enhances classification accuracy.