How Rovo solves search challenges through entity linking

Summary

The article discusses how Atlassian addresses search challenges through advanced entity linking, transforming unstructured text into actionable knowledge. It highlights the importance of accurately associating textual mentions with unique entries in a Knowledge Graph to enhance the relevance of tools like Rovo Chat and Search. The piece elaborates on the architecture of the entity linking system, which includes a detection stage using a transformer-based model, a recall stage with multiple paths for candidate generation, and a ranking stage that employs both coarse and fine-grained disambiguation techniques. The challenges of linking team names, such as non-standard formats and high operational noise, are also discussed, along with the need for a solution that respects tenant isolation and privacy.

Key Learnings

• 1Entity linking is critical for improving search relevance in enterprise AI applications.
• 2The architecture consists of a multi-stage pipeline that separates detection, recall, and ranking to balance precision and latency.
• 3Challenges in entity linking include handling ambiguous names, non-standard formats, and the need for real-time updates due to high churn rates of entities.
• 4Personalization and context are essential for disambiguating entities effectively.
• 5The system must be adaptable to different corporate cultures and naming conventions while maintaining privacy.

Who Should Read This

Senior Machine Learning Engineers focusing on natural language processing and entity resolution challenges in enterprise applications.

Test Your Knowledge

Topics