A Semi-Supervised Learning Model for Predicting User Attributes Based on Ladder Network

Abstract

In recent years, marketing has often relied on using attribute information associated with the accounts registered in online services. However, most people use such services without a personal account, and it is impossible to obtain attribute information for these unregistered users. To deal with this situation, semi-supervised learning is an effective way to increase the number of users with attribute information. Attributes can be predicted from the historical data of unregistered users, having the historical data of registered users who have attribute information. One such semi-supervised learning method is the ladder network, which is a neural-network-based model that adds and removes noise. This model provided highly accurate predictions for image data and is also considered to be useful for predicting user attributes from historical data, where the feature vector is high-dimensional. However, this method does not account for cases where the label takes an ordered value, such as the user’s age category. In this study, we propose an extended model based on a ladder network that incorporates a mechanism that can appropriately predict a user’s attribute information, including ordinal-scale variables. We also conducted an evaluation experiment using browsing history data to demonstrate the effectiveness of the proposed method.