Tongfei Chen

Title: Tongfei Chen and Hierarchical Information Retrieval.

Introduction

Tongfei Chen is an inventor based in Seattle, Washington. His patent work addresses information retrieval, focusing on how document collections can be organized and searched more effectively.

Latest Patents

Chen is associated with 1 patent, titled “Information Retrieval System Using a Hierarchical Corpus Encoder.” The invention adapts a dense encoder to organize documents in a corpus as a hierarchy of vector embeddings. It uses negative samples from sibling nodes in the hierarchy, while jointly training the encoder and document tree with a loss function that accounts for document relationships.

The approach is designed for both supervised settings, where query-document relevance judgments are available, and zero-shot settings, where a query dataset is absent. The patent describes performance improvements over several dense-encoder and generative-retrieval baselines, highlighting the potential benefits of learning a document hierarchy alongside the encoder.

Career Highlights

Chen’s listed patent is associated with Microsoft Technology Licensing, LLC. It reflects work on machine-learning methods for information retrieval and the joint modeling of document representations and corpus structure.

Collaborations

Ankita Sharma is identified as Chen’s coworker. The available information does not specify her role in the patent or the nature of their collaboration.

Conclusion

Tongfei Chen’s patent presents a hierarchical approach to information retrieval that combines document organization with encoder training to support both supervised and zero-shot search.