Metric Learning for Names

Anonymous

Abstract: Many systems rely on the ability to effectively search through databases of personal names. Despite this, techniques for indexing names have yet to leverage recent advances in neural networks, hindering the throughput of these systems due to inefficient indexing. In this work, we present a method for fine-tuning a neural network via metric learning to embed personal names in a vector space which can be used for retrieval. We not only demonstrate up to a 12% improvement over existing systems, but also show that our model significantly outperforms them even in the absence of an external name transliteration engine, which is often required for existing indexing techniques.

Paper Type: short

Research Area: Information Retrieval and Text Mining