A human resources software company wanted to improve the quality of their natural language processing (NLP) in order to more accurately match their clients’ job postings with available resumes.
Because job candidates are able to describe their skills and experiences in an infinite number of ways, their model had difficulty delineating the skills and normalizing them against the company’s internal terminology. As a result, the model was prone to errors, especially due to the inherent subjectivity of the matching process.
To improve the quality of their matching, the software company tapped Sama to help support two annotation workflows, providing named entity recognition (NER) and named entity normalization (NEN) using a three-annotator consensus approach.
“The tasks for this project were challenging, with a high degree of subjectivity, which can make consensus difficult to reach,” said Annepeace Alwala, Sama Vice President, Global Service Delivery. “Because of that subjectivity and the sheer variety of possible answers, it was important for the annotators on this project to have corporate or HR experience.”
In addition to leveraging a team with domain expertise, Sama:
By proactively addressing the issue of subjectivity, Sama was able to help the software company significantly improve NLP for their Gen AI model.
Sama’s methodological approach ensured higher quality results for the client, including:
Example: Employing a living knowledge base to make decisions more efficiently and reduce subjectivity
Example: Proactively flagging and scrutinizing answers with high variances
Example: Using HITL and multiple QA steps to capture the nuanced ways skills were described and normalizing them against the company’s internal ontology
Sama provides ML Professionals and AI team Leads with an indispensable solution for Computer Vision data labeling.
