Using Seedlink AI to Measure Candidates’ Cultural Fit to The Company
PostedMon 2, Oct 2017

Background of This Study

One of the world’s renowned industrial gases companies sought to hire better fit candidates based on culture fit for long term success. They cooperated with Seedlink to create predictive index benchmark to increase the quality of future successful hires.

Culture Fit Methodology

1. Academic base

Seedlink’s “Culture Fit Prediction” Model is based on 2 theories:

  • People unconsciously give away something about their behavior performance via their languages (Back, et al., 2010; Li, et al., 2014);
  • In terms of language use, the more similar the candidate is to high performers in certain organization, the better the fit to the corporate culture (Pennebaker & King, 1999; Eirinaki & Vazirgiannis, 2003).

Seedlink AI algorithm can help to identify such pattern, correlate the language with behavior and accordingly predict culture fit and future performance of the candidates.

In order to build a company model, the algorithm need two different data sets:

  • The internal employee language data with behavior label and it requires at least 80 – 100 respondents.

  • The training data that requires at least 5000 respondents. “The training data” includes various language patterns of ordinary people and their behavior labels.

2. The Importance of Two Data Sets

Picture 1a. Language Space
Picture 1b. Language Space in graphic

If we see human language as the universe, the training data as the galaxy, then the data from selected employees are solar system. It is almost impossible to accurately locate solar system on the map of the borderless universe. But if we know where the galaxy is, it will be much easier to find solar system.

Likewise, we must ensure that the employee language data are clustered into the close space as that of the training data (as shown in Picture 1) so that AI technology is able to identify the specific pattern and correlate them with high performing behaviors for certain organization(De Fortuny et al., 2014; Gao, et al. 2013; Verhoeven & Daelemans, 2014) .

In this scenario, we have to use the same questions to collect both labelled language and training data. Most importantly, the training data must include the largest number of behavior labels in line with those selected by the organization.

The company’s Culture Fit Model Behaviors Set:

  • Team Player
  • Adhere to Principles
  • Logical and Analytic
  • Keep Learning
  • Planning and Organization
  • Result Driven
  • Respond to Challenges
  • Cope with Pressure

Seedlink AI recommended to use TOP 3 questions in the table below (Table 1), the training data of which includes all behavior labels selected by the company.


  1. Back, M., Stopfer, J., Vazire, S., Gaddis, S., Schmukle, S., Egloff, B. & Gosling, S. (2010). Facebook Profiles Reflect Actual Personality, Not Self-Idealization. Psychological Science, 21(3).
  2. De Fortuny, E., De Smedt, T., Martens, D. & Daelemans, W. (2014). Evaluating and understanding text-based stock price prediction models. Information Processing and Management, 50(2) ,426-441.
  3. Eirinaki, M. & Vazirgiannis, M. (2003). Web mining for web personalization.
  4. Gao, R., Hao, B., Bai, S., Li, L., Li, A. & Zhu, T. (2013). Improving user profile with personality traits predicted from social media content. Conference on Recommender Systems.
  5. Li, L., Li, A., Hao, B., Guan, Z. & Zhu, T. (2014). Predicting active users’ personality based on micro-blogging behaviors.. PLOS ONE, 9(1) , e84997-e84997.
  6. Pennebaker, J. W., & King, L. A.(1999). Linguistic styles : Language use as an individual difference. Journal of Personality and Social Psychology, 77(6) ,1296-1312.
  7. Verhoeven, B. & Daelemans, W. (2014). CLiPS Stylometry Investigation (CSI) corpus: A Dutch corpus for the detection of age, gender, personality, sentiment and deception in text.
Get in Touch with Seedlink!
What would you like to know?
Please leave your information here
We'll arrange a free demo for you.