To classify reviews into topics that matter to managers and researchers, the Culture 500 team built a custom dictionary from the ground up by combining machine learning tools with human expertise. This approach dramatically improved the accuracy of their models.

Custom Culture 500 Dictionary Combining Machine Learning and Human Expertise

The varied language in Glassdoor reviews paints a nuanced picture of corporate culture but defies simplistic textual analysis. When the Culture 500 team initially applied standard natural language processing tools to the Glassdoor data, the results were disappointing.

Google

When describing culture, Glassdoor reviewers use rich and varied language, including slang, idioms, acronyms, buzzwords, and colorful phrases. For a topic such as integrity, employees used many synonyms, antonyms, idioms, and slang terms.

Linguistic Variety in Glassdoor Culture Reviews

MIT Sloan Management Review project page on measuring culture in leading companies, including the Culture 500 dataset. https://sloanreview.mit.edu/projects/measuring-culture-in-leading-companies/

MIT Sloan Management Review

There can be a large gap between how many reviewers discuss a cultural topic and how many use the topic's specific label word. In the Culture 500 data, nearly 80,000 employees discussed integrity in their reviews, but fewer than 5,000 used the specific word.

Learn Before

Related

Learn After