EmbeddingTree: Hierarchical Exploration of Entity Features in Embedding

TL;DR

EmbeddingTree is a hierarchical exploration algorithm that interprets how entity features are encoded in embedding spaces, aiding understanding and manipulation of high-dimensional embeddings.

Contribution

It introduces EmbeddingTree, a novel hierarchical method for interpreting entity features in embeddings, along with an interactive visualization tool.

Findings

Effective in interpreting industry-scale merchant embeddings

Assists in feature denoising and injecting in embedding training

Supports generation of embeddings for unseen entities

Abstract

Embedding learning transforms discrete data entities into continuous numerical representations, encoding features/properties of the entities. Despite the outstanding performance reported from different embedding learning algorithms, few efforts were devoted to structurally interpreting how features are encoded in the learned embedding space. This work proposes EmbeddingTree, a hierarchical embedding exploration algorithm that relates the semantics of entity features with the less-interpretable embedding vectors. An interactive visualization tool is also developed based on EmbeddingTree to explore high-dimensional embeddings. The tool helps users discover nuance features of data entities, perform feature denoising/injecting in embedding training, and generate embeddings for unseen entities. We demonstrate the efficacy of EmbeddingTree and our visualization tool through embeddings generated for industry-scale merchant data and the public 30Music listening/playlists dataset.