Using of n-grams from morphological tags for fake news classification

Area of Science:

• Natural Language Processing
• Computational Linguistics
• Information Retrieval

Background:

• Effective fake news detection is crucial, with morphological analysis showing potential.
• Previous research suggested n-grams and POS tagging were insufficient, but lacked recent empirical validation.
• A gap exists in experimentally evaluating combined n-gram and POS tag usage for fake news classification.

Discussion:

• The research preprocesses a Covid-19 news dataset using morphological analysis to generate POS tag n-grams.
• Three novel POS tag-based techniques are proposed and compared against the TF-IDF baseline.
• The impact of n-gram size and decision tree depth on classification performance is investigated.

Key Insights:

• Proposed POS tag techniques demonstrate comparable performance to the traditional TF-IDF method.
• Morphological analysis, when integrated with TF-IDF, offers statistically significant improvements.
• Specifically, precision for fake news and recall for real news are notably enhanced.

Outlook:

• Further research can explore advanced morphological features and diverse datasets.
• Investigating ensemble methods combining POS tags with other linguistic features could yield better results.
• The findings support the integration of morphological analysis in developing more robust fake news detection systems.