ACM (2025) — LinkedIn profile characteristics & professional success

Summary:

This research paper looks at whether parts of a LinkedIn profile are linked to career success. The authors define success in three ways: promotions, follower count, and how fast someone seems to move forward in their career.

This research can serve as a proxie to understanding how founder growth and public persona can help with founder-led sales.

It was written by researchers at Texas Tech University and published as an ACM conference paper. The main goal was to see which profile features seem to matter most, and whether machine learning models can predict these success measures from profile data.

The study used a Kaggle dataset with 62,706 anonymized LinkedIn profiles.

The authors cleaned the data, filled in missing values, encoded categories, scaled numeric fields, removed extreme outliers, and built a custom career progression rate metric.

They then trained several machine learning models, including Random Forest, Gradient Boosting, XGBoost, and CatBoost, and compared how well those models predicted promotions and follower counts.

This is an observational modelling study based on existing profile data, not an experiment.

The biggest finding is that promotions were very easy for the models to predict, with very high R2 scores above 0.97 across all models. Career progression rate was the strongest predictor of promotions, followed by current position length and total experience.

Follower count was much harder to predict. Beauty score was the most important feature for follower prediction across models. Current position length, age, career progression rate, and total experience also mattered for followers, but less consistently.

The paper says follower growth seems more complex and less structured than promotions, which may mean important drivers were missing from the dataset.

Some notes:
The source data came from Kaggle rather than directly from LinkedIn, so data quality depends on that third-party dataset. The paper does not clearly explain how the original Kaggle profiles were sampled, so representativeness is uncertain.