Hybrid search

[fxprunayre]

Experiment to evaluate hybrid search (semantic + lexical) with Elasticsearch. For this, one option is to run a model to compute embeddings at index and search time.

The main benefits of the hybrid search are:

• when a lexical search does not return any results, the semantic search can still find records that are semantically related to the query terms.
• it can improve the ranking of the results by combining the lexical and semantic scores.
• it can support multilingual search by using a model that can compute embeddings for multiple languages.

Example:

• Searching for "rivière en afrique" will return the "Hydrological basins of Africa" record even if the words "rivière" or "afrique" are not present in the record title.

See doc for details.

Checklist

• I have read the contribution guidelines
• Pull request provided for main branch, backports managed with label
• Good housekeeping of code, cleaning up comments, tests, and documentation
• Clean commit history broken into understandable chucks, avoiding big commits with hundreds of files, cautious of reformatting and whitespace changes
• Clean commit messages, longer verbose messages are encouraged
• API Changes are identified in commit messages
• Testing provided for features or enhancements using automatic tests
• User documentation provided for new features or enhancements in manual
• Build documentation provided for development instructions in README.md files
• Library management using pom.xml dependency management. Update build documentation with intended library use and library tutorials or documentation

[sonarqubecloud]

Quality Gate failed

Failed conditions
0.0% Coverage on New Code (required ≥ 80%)

See analysis details on SonarQube Cloud