11-08, 11:40–12:20 (US/Eastern), Central Park East
The goal of this project is to predict the box office outcome of an upcoming movie based on a wide variety of feature set like poster (images), revenue, language, title, genre, and popularity score. This project aims to help a content creator to gauge how their trailers are going to be received by audiences, based on the historic data, so that can gauge their work and improve as needed before publishing their work. Using a simple convolutional neural network (CNN) with one convolutional layer and three fully connected layers to extract visual features from the posters, the model was trained using MSE loss function with Adam Optimizer. Early stopping was implemented to prevent overfitting. The model was able to predict the outcome of a movie with an accuracy of 89% based on features extracted from posters alone. In addition to still image posters, trailers convey the idea of the movie. The tonality of trailers depend on the visuals, narrative, editing style, and audio. The tonality factors will be scored out of 10 so that the creator can choose to tweak their work based on any parameter. This project explores various attributes of videos to study the impact of multimedia promotions on the viewership indirectly by using revenue. The addition of video tonality to this prediction model should improve the prediction accuracy of the box office success of movies.
Description:
Objective: To build a predictive model to estimate the box office performance of movies, by analyzing multimedia features such as the visual aesthetics of posters, the auditory impact of music, and the emotional tonality of movie posters and trailers.
Methods:
Data Sources: A dataset of 5000+ movies was curated using TMDb and YouTube.
Technology: Convolutional neural networks (CNNs) were employed for image analysis of posters. Transformers were utilized for assessing trailer tonality.
1. Poster Analysis: CNNs were used to extract visual features from movie posters
2. Video Tonality Analysis: Transformers will be used to analyze the following features of the video trailers to determine the overall tonality of the video. Trailer scripts will be analyzed using BERT or similar models for sentiment detection, and to evaluate how the following tonalities correlate with box office performance.
o Emotion: Analysis of the emotional impact conveyed through trailer dialogues and visuals.
o Narrative: Assessment of the storyline and pacing as presented in the trailer.
o Audio: Evaluation of background music, sound effects, and their contribution to the overall tone.
o Visuals: Analysis of visual elements such as lighting, color schemes, color grading, camera techniques and editing style.
3. Model Development: A multi-input model architecture, using concatenation layers will be used to merge features from posters, music, and trailers into a unified neural network model. Deep learning techniques will be applied for prediction of box office success of movies based on the above-mentioned features extracted from previously released movies.
Previous knowledge expected
I consider myself fortunate to have been born in the 90s, the era which connects “everything but tech” with “everything tech”. To be sure my brother and I did not miss the dotcom wave, my parents invested in a personal computer at the start of the millennium. Little did I know that this would shape my career, two decades later.
As a teen, my tech career had officially begun when I saw my grandparents recording their monthly expenses in a notebook and I developed an expensing app called “Cash +” to address it. I had it on a floppy disk (the most accessible tech at the time) to give their user experience (UX) flexibility when they travel. It was easier to carry the disk than the physical notebook.
Once I realized I had the acumen for tech, I decided to pursue engineering and then an MBA in highly ranked universities. My challenging MBA program took me to 3 culturally diverse cities each semester - Singapore, Sydney & Dubai where I had to learn to be frugal, fluid, and adaptable and all these qualities helped my career post MBA.
At graduation, I set a goal for myself - to explore different management functions to select the path in tech that I would thrive in. Looking back at my career experimentation, I’m glad that I was able to build & and launch products at Jio, take products & and services to the market through strategic B2B partnerships at Redington, and increase brand awareness through online marketing at Zero&One.
Last year, I decided to segue into a more technical path in Data Science and Machine Learning. I’m currently building my network in the US and proactively connecting with tech leaders here in the Data Science and Machine Learning fields.