HouseCritic is a meta-learning based and semi-supervised deep neural network in order to estimate a specific user’s satisfying degree for a given housing estate.
Concretely, it first captures the user’s preference and the house’s representation from a collection of extracted features. Then, the user preference is used as the meta-knowledge to derive the parameter weights of the house representation such that we can explicitly model the selection causality (the decision-making process for users to choose a house according to their preferences) and accordingly provide a satisfying degree of the given house.
Figure 1 shows the structure of the HouseCritic, which consists of three components:
- A user module, which captures the user preferences and generates weights of the house embedding based on the user preferences.
- A house module, which embeds features of the housing estate.
- A selection module, which obtains the houses' estimated satisfying degree of a user. The component is a Meta-FCN, which uses the house embedding as input and the user preference as meta-knowledge (weights). As a result, the satisfying degree can be estimated by modeling the selection causality between the user and the housing estate.
Figure 1: Structure of HouseCritic