Multilingual Retrieval on Yelp Search Engine
Hi there! đź‘‹ In this repository, we develop a Multilingual Information Retrieval model that support 15 different languages, and it will be used on Yelp search engine after further online experiments.
Figure 1: Yelp’s search interface
Background
The current search engine of Yelp is based on NrtSearch. However, inverted index-based lexical matching on Lucene-based search engine such as NrtSearch falls short in several key aspects:
- Lack of understanding of hypernyms, synonyms, and antonyms. For example, “sneaker” might match the intent of the query “running shoes”, but may not be retrieved.
- Fragility of morphological variants (e.g. woman vs. women)
- Sensitivity to spelling errors
- Inability to support multilingual