Using transfer learning to quantify user color bias in movie poster selection for neuromarketing applications
Date:
Original Presentation can be found here ; award certificate can be found here
Transfer learning has emerged as a powerful technique in the field of neuromarketing, particularly in understanding user color bias in movie poster selection. By leveraging pre-trained deep learning models, researchers can quantify the extent to which individuals exhibit preferences for certain colors when choosing movie posters. The process involves training the model on a large dataset of movie posters and their associated user ratings, capturing the complex relationships between colors and subjective preferences. By transferring this knowledge to a new dataset of user choices, researchers can analyze the extent to which color bias influences the selection process.
The application of transfer learning in quantifying user color bias holds immense potential for neuromarketing. By understanding the underlying psychological factors that influence individuals’ color preferences in movie poster selection, marketers can optimize their strategies to cater to specific target audiences. For example, if the analysis reveals a strong bias towards vibrant colors among a particular demographic, marketers can design movie posters that incorporate these hues to enhance the appeal and engagement of their campaigns. This approach allows for a more data-driven and personalized marketing approach, enabling companies to effectively target their desired consumer segments and maximize the impact of their promotional efforts.
The present article and the ensuing research serve to bridge that gap. In our article, we elucidate the salient points of the research. We conducted using movie posters as a means of estimating and studying the effects of visual bias. We built a Web application to survey and collect user ratings for movie posters belonging to different genres and thus having different visual effects. Based on the user’s input, a bias mapping was done, and the result of which was the genre that the user was most visually partial to. Contingent on that, neural style transformation was applied to movie posters, and the augmented results were presented to the user to rate. By comparing the initial genre-wise ratings to the transformed genre ratings, the extent of visual bias was detected and thus analyzed. Our study not only confirmed the empirical existence of such bias, but the detection methodology outlined here may serve as a visual bias manipulation tool that can utilize real-time machine learning to optimize AdSense and the likes.
The results of the study were presented at the IEMIS 2022: 3rd International Conference on Emerging Technologies in Data Mining and Information Security at the Technical Session 1.5 on Advanced Computing where the article won the Best Paper Award.