Prediction of box office revenue of movies

In the ever-evolving field of prediction of box office revenue of movies, the integration of state-of-the-art neural networks, especially BERT with traditional FNN offers promising avenues for research. This paper investigates the effectiveness of BERT-based models combined with FNNs in predicting m...

Full description

Saved in:
Bibliographic Details
Main Author: Er, Erica Ming Chee
Other Authors: Li Boyang
Format: Final Year Project
Language:English
Published: Nanyang Technological University 2023
Subjects:
Online Access:https://hdl.handle.net/10356/171916
Tags: Add Tag
No Tags, Be the first to tag this record!
Institution: Nanyang Technological University
Language: English
Description
Summary:In the ever-evolving field of prediction of box office revenue of movies, the integration of state-of-the-art neural networks, especially BERT with traditional FNN offers promising avenues for research. This paper investigates the effectiveness of BERT-based models combined with FNNs in predicting movie box office revenues. Leveraging a comprehensive dataset comprising 36,108 entries from TMDB and enriched with metrics from IMDb, the study presents a two-pronged approach: analyzing pre-release and all-available data to simulate varying real-world scenarios. Three distinct models were proposed and assessed: a pre-trained BERT embedding model, a fine-tuned BERT variant, and an integrated hybrid model encompassing both textual and numerical data. Comparative evaluations based on loss curves, predicted vs. actual values, and overall performance metrics unveiled the superior efficacy of the integrated hybrid model, particularly when fed with comprehensive data from the all-available dataset. The results underscore the importance of a cohesive architecture that effectively processes both textual and numerical data, emphasizing the value of comprehensive data and thoughtful model selection in maximizing predictive accuracy in the realm of box office revenue forecasting.