Automatic Identification of Character Types from Film Dialogs

Area of Science:

• Computational linguistics
• Artificial intelligence
• Natural language processing

Background:

• Identifying fictional character types from dialogue is challenging.
• Linguistic analysis alone is insufficient for comprehensive character identification.
• Existing methods lack the ability to capture interactive and communication nuances.

Purpose of the Study:

• To develop an integrative approach for fictional character type detection.
• To enhance identification performance by combining linguistic analysis with interactive characteristics.
• To introduce a novel dataset for evaluating character type detection in action movies.

Main Methods:

• Developed an integrative approach combining linguistic analysis with interactive and communication characteristics.
• Utilized semantic graphs weighted by linguistic markers of expressivity and social role to capture interactive traits.
• Introduced a new dataset of action movie character types and their dialogue sequences.

Main Results:

• The integrated approach significantly outperformed baseline methods in character type identification.
• Demonstrated improved performance on the newly introduced action movie dataset.
• Comparative analysis highlighted the effectiveness of incorporating interactive and communication features.

Conclusions:

• The integrative approach offers superior performance for fictional character type detection.
• Combining linguistic, interactive, and communication features is crucial for accurate character analysis.
• Future research should explore limitations and further refine character identification models.