Capturing Preferences of New Users in Generative Tasks with Minimal Interactions Collaborative Filtering Using Siamese Networks and Soft Clustering

Abstract

Prediction of user preferences is a challenge, in particular when the objective is to learn them without requiring the user to provide a profile or a significant number of interactions. Many collaborative filtering algorithms exist but all of them require the availability of huge datasets of user information and expensive computations. In this paper, a novel architecture is introduced which aims to predict a new user’s interests in the context of previous users’ interactions with minimal feedback interactions. Here, a Siamese Network is used to generate an embedding space for data from existing users. This information is then used in a Gaussian Mixture Model to generate multiple soft clusters. Based on the embedding space, system responses to the user are generated using a Conditional Generative Adversarial Network which uses a vector drawn from the Gaussian Mixture in embedding space from the Siamese Network as the conditional input. The predictive model then interacts with the new user and based on their feedback adjusts the Gaussian Mixture to find the distribution with the highest probability of generating the user’s preferred data. The approach is applied in the context of an image generation task where the goal is to learn to generate images that match the preferences of the user using only a minimal number of direct user interactions. Testing in this domain has shown promising results that exemplify the ability of the approach to capture the user’s preferences while presenting only a minimal number of image examples.