A microblog content credibility evaluation model based on collaborative key points.

Area of Science:

• Computer Science
• Information Security
• Natural Language Processing

Background:

• False content on microblogging platforms poses significant information security risks.
• Credibility evaluation is hindered by the complexity of feature selection due to misinformation.
• Existing methods struggle with effectively identifying and filtering deceptive content.

Purpose of the Study:

• To propose a novel collaborative key point-based content credibility evaluation model (CHECKP).
• To address the challenges in feature selection for credibility assessment in microblogging environments.
• To enhance the accuracy and performance of detecting false content.

Main Methods:

• The CHECKP model extracts key points from microblog text at both word and sentence levels.
• It evaluates content credibility based on the semantics derived from these key points.
• A collaboratively constructed rumor lexicon aids in strengthening word semantics and optimizing feature selection for deep learning.

Main Results:

• The CHECKP model demonstrated improved performance compared to the Att-BiLSTM model.
• F1 scores increased by 3.83% for true evaluations and 3.8% for false evaluations.
• The model effectively addresses feature selection challenges in content credibility evaluation.

Conclusions:

• The proposed CHECKP model offers a more effective approach to content credibility evaluation.
• Optimized feature selection through key point extraction and rumor lexicon significantly boosts performance.
• This research contributes to mitigating information security threats posed by false content online.