WE model: A Machine Learning Model Based on Data-Driven Movie Derivatives Market Prediction

Yaoyao Ding, Chenghao Wu, Xinyu Liu, Yuntao Zou, Peng Zhou

Abstract

The mature development and the extension of the industry chain make the income structure of the film industry. The income of the traditional film industry depends on the box office and also includes movie merchandising, advertisement, home entertainment, book sales etc. Movie merchandising can even become more profitable than the box office. Therefore, market analysis and forecasting methods for multi-feature merchandising of multi-type films are particularly important. Traditional market research is time-consuming and labour-intensive, and its practical value is restricted. Due to the limited research method, more effective predictive analysis technology needs to be formed. With the rapid development of machine learning and big data, a large number of machine learning algorithms for predictive regression and classification recognition have been proposed and widely used in product design and industry analysis. This paper proposes a high-precision movie merchandising prediction model based on machine learning technology: WE model. This model integrates three machine learning algorithms to accurately predict the movie merchandising market. The WE model learns the relationship between the movie merchandising market and movie features by analyzing the main feature information of movies. After testing, the accuracy rate of prediction and evaluation in the merchandising market reaches 72.5%, and it has achieved a strong market control effect.

Introduction

In the modern cinematic landscape, the "long tail" of revenue—comprising toys, apparel, and other derivatives—frequently eclipses the initial box office take. Despite the economic weight of this sector, predicting which films will successfully translate into merchandising gold has historically remained a qualitative endeavor. Industry veterans often rely on intuition and labor-intensive market research, which are prone to subjectivity and slow turnaround times.

While machine learning has been aggressively applied to box office forecasting, the derivatives market has lacked a robust, quantitative framework. This paper addresses this gap by introducing the WE model (Weighted Ensemble), a machine learning framework designed to predict the market value of movie merchandising with high precision.

Methodology

The researchers developed a four-stage pipeline to transition from raw cinematic data to actionable market predictions.

1. Data Collection and Feature Engineering

The study utilized a custom-labeled dataset of 441 influential films released between 1970 and 2020. The features included 18 distinct variables, such as genre, MPAA rating, box office performance, production companies, and the number of entries in a series (sequel count).

2. The Delphi Method for Labeling

Because no standardized "merchandising score" exists in public databases, the researchers employed the Delphi method to establish a ground truth. Twenty industry experts—including directors and producers—iteratively scored films on a scale of 0–25 across categories like stationery, toys, and apparel. This iterative process continued until a consensus was reached (indicated by low variance), providing a reliable target variable for supervised learning.

3. Model Training and the Weighted Ensemble

The team evaluated four primary algorithms: Linear Regression, XGBoost, LightGBM, and LASSO. Recognizing that different algorithms captured different nuances of the market, they developed the WE model. This ensemble integrates the top three performers (LightGBM, LASSO, and XGBoost) using optimized weighting coefficients ($\alpha_1, \alpha_2, \alpha_3$) to maximize overall predictive accuracy.

Key Insights

The "Franchise Factor"

The analysis revealed that being part of an established series is the most significant predictor of merchandising success. Films belonging to franchises with five or more entries almost universally demonstrated higher merchandising potential, suggesting that brand longevity is the primary driver of consumer derivative interest.

The Rating Paradox

Interestingly, the study found a negligible correlation between IMDB ratings and merchandising value. High critical acclaim does not necessarily translate to "toy-ability." Conversely, films with mediocre critical reception often achieved high merchandising scores, highlighting the disconnect between artistic merit and commercial derivative potential.

Addressing Data Sparsity

Approximately 50% of the films in the dataset had a merchandising score of zero. This sparsity created a challenge for traditional gradient backpropagation, which often struggled with "gradient disappearance." The ensemble approach was specifically chosen to mitigate this issue, providing more stable predictions across a sparse matrix.

Results

The WE model significantly outperformed standalone algorithms. While traditional Linear Regression achieved only 43.75% accuracy, the individual machine learning models performed better:

  • XGBoost: 57.50%
  • LASSO: 58.75%
  • LightGBM: 62.50%
  • WE Model (Ensemble): 72.50%

By weighting the strengths of the three best models, the WE framework achieved a 10% improvement over the best individual performer (LightGBM), demonstrating the power of ensemble learning in complex, high-dimensional market environments.

Limitations and Future Work

The current model is limited by a relatively small sample size (441 movies), a constraint imposed by the scarcity of truly "influential" films with accessible data. Furthermore, the high frequency of zero-value scores remains a challenge for model bias. Future research aims to expand the dataset to include animated television series and streaming-exclusive content, while also incorporating granular data such as social media sentiment and specific character design metrics to further refine the WE model’s accuracy.

Summary updated

Abstract

The mature development and the extension of the industry chain make the income structure of the film industry. The income of the traditional film industry depends on the box office and also includes movie merchandising, advertisement, home entertainment, book sales etc. Movie merchandising can even become more profitable than the box office. Therefore, market analysis and forecasting methods for multi-feature merchandising of multi-type films are particularly important. Traditional market research is time-consuming and labour-intensive, and its practical value is restricted. Due to the limited research method, more effective predictive analysis technology needs to be formed. With the rapid development of machine learning and big data, a large number of machine learning algorithms for predictive regression and classification recognition have been proposed and widely used in product design and industry analysis. This paper proposes a high-precision movie merchandising prediction model based on machine learning technology: WE model. This model integrates three machine learning algorithms to accurately predict the movie merchandising market. The WE model learns the relationship between the movie merchandising market and movie features by analyzing the main feature information of movies. After testing, the accuracy rate of prediction and evaluation in the merchandising market reaches 72.5%, and it has achieved a strong market control effect.
BibTeX
@misc{ding2022we,
  author = {Yaoyao Ding and Chenghao Wu and Xinyu Liu and Yuntao Zou and Peng Zhou},
  title = {{WE model: A Machine Learning Model Based on Data-Driven Movie Derivatives Market Prediction}},
  year = {2022},
  eprint = {2207.02423},
  archivePrefix = {arXiv},
  url = {https://arxiv.org/abs/2207.02423},
}

Loading related content...

Figure