This article was accepted into the Industry Track at SIGIR 2024.
Virtual assistants (VAs) are important information retrieval platforms that help users perform various tasks using spoken commands. The speech recognition system (speech to text) uses previous queries, trained solely on text, to distinguish between phonetically confusing alternatives. Therefore, generating synthetic queries that are similar to existing VA usage can greatly improve VA capabilities, especially for use cases that do not (yet) occur in paired audio/text data.
In this article, we provide a preliminary exploration of using large language models (LLMs) to generate synthetic queries that are complementary to template-based methods. We investigate whether the methods (a) generate queries similar to queries from users of a popular VA and (b) whether the generated queries are specific. We found that LLMs generate more detailed queries, compared to template-based methods, and reference specific aspects of the entity. The generated queries are similar to VA user queries and are specific enough to retrieve the relevant entity. We conclude that queries generated by LLM and templates are complementary.
