Using Universal Sentence Encoder for Semantic Search of Employee Data

Abstract

This paper describes an application that performs a semantic search on an employee database. It helps Human Resources employees to target relevant people for their events and trainings. Syntactic or lexical searching involves keyword matching but does not match synonyms and other contextually related data. By using regular keyword search, a document either contains the given word or not, and there is no middle ground. Semantic Search allows the matching of data contextually linked with the search term. High dimensional vectors, also known as embeddings, are generated for a complete sentence and are then used for searching. Under the hood, Google’s Universal Sentence Encoder. The Universal Sentence Encoder encodes text into high dimensional vectors that can be used for text classification, semantic similarity, clustering, and other natural language tasks to provide better performance of the model as compared to a custom trained Convolution Neural Network which also requires more training data.