We want to deliver the search results users want, even when queries contain mistyped words or abbreviations.” - We believe this is an issue that many services face, and LINE MUSIC is one of them. In this session, we will talk about our efforts to solve this problem using machine learning-based search techniques. We plan to cover the two approaches we have implemented - vector search using a word embedding model and tie-up abbreviation extraction using generative AI - and provide technical explanations of the architecture and implementation ideas, as well as the process leading up to production deployment. We hope this session will be useful in your product development.



DAY 1
14:20-14:50 JST
Main Room C
JaEnKo
Streaming
Challenges in Ambiguous Search Using Hybrid Search and Abbreviation Extraction in "LINE MUSIC"
Speaker

I joined LINE Corporation as a new graduate in 2020. I'm responsible for server-side development for LINE MUSIC, mainly worked on developing features related to search and data processing.