Anime popularity prediction before huge investments: a multimodal approach using deep learning

Area of Science:

• Artificial Intelligence
• Data Science
• Entertainment Industry Analysis

Background:

• Predicting entertainment product success, particularly anime, is critical for the Japanese industry.
• Existing methods often require data only available after significant investment, limiting early-stage decision-making.
• There is a need for a comprehensive dataset and robust model utilizing pre-investment, publicly available data.

Purpose of the Study:

• To introduce a novel, comprehensive dataset for predicting anime popularity using only pre-investment features.
• To propose and evaluate a deep neural network (DNN) architecture for anime popularity prediction.
• To assess the scope and impact of early-stage features on predicting anime success.

Main Methods:

• Development of a comprehensive dataset using freely available internet data, adhering to real-world standards.
• Implementation of a deep neural network architecture combining GPT-2 (for language understanding) and ResNet-50 (for image analysis).
• Evaluation of the model using Mean Squared Error (MSE), R-square (R2), Pearson, and Spearman correlation coefficients, compared against traditional benchmarks.

Main Results:

• The proposed DNN model achieved a Mean Squared Error (MSE) of 0.012, a significant improvement over the benchmark's 0.415.
• The model's R-square (R2) score was 0.142, substantially outperforming the benchmark's -37.591.
• Pearson (0.382) and Spearman (0.362) correlation coefficients indicated relevance of pre-investment features, but not decisiveness.

Conclusions:

• Pre-investment features, while relevant, are insufficient on their own to definitively predict anime popularity.
• The proposed multimodal deep learning approach offers a promising tool for the entertainment industry to mitigate financial risks.
• Further research should explore the interaction of these features with post-investment data for more accurate predictions.