Employee Ranking Based on Intercompany Relations

You may have heard about the PageRank algorithm which made Google so successful in it's early days. This algorithm ranks websites based on their relation to other websites. This algorithm could be used to apply the same logic on employees and their relationships to other employees in order to determine their importance.

The PageRank Algorithm

The PageRank algorithm checks which website links to which other website and infers a score based on the amount and quality of relations. A link from a high ranking website is more valuable than a link from a low ranking website. Since the rank of a page changes over time (based on the incoming links) the algorithm needs to "re-calculate" the ranks many times to come to a conclusion.

The algorithm itself shall not be a topic of this post and can be checked out at https://en.wikipedia.org/wiki/PageRank.

Possible Conclusions and Analysis

While this is most certainly a controversial topic it is an interesting though experiment. You could:

• Learn a lot about who the key people in the company are
• Find critical positions which need a backup
• Possible communication issues (which people/departments have a low rank and are possibly not involved enough)
• Which are the social groups in a company (this could be paired with a cluster analysis). Changes to single positions/people may effect the whole group/cluster (e.g. employee leaves the company, becomes sick, becomes demotivated etc.)

As with most analysis you need to understand the source data and how the result is calculated in order to make valid conclusions. There are various scenarios where the wrong conclusions can be made if you don't understand the underlying data and business structure.

If you only use your own emails received/send you should exclude yourself from the ranking since your own score will be skewed. Additionally, people you are working with a lot, usually have higher ratings and will be skewed

Data basis

The most simple data basis which most companies have are emails. Let's say we could use all the email history within the company to track who is communicating with whom and how often. In terms of the PageRank algorithm the people would be the "websites" and the emails sent to other employees are the "links" between websites. Contrary to the normal PageRank algorithm it may make sense to consider the amount of emails from one person to another person. In a simple approach we could weight the relationship between two individuals simply by considering every single email as its own connection to the other person instead of reducing all emails from one person to another person to a single connection.

Implementation

class PageRank:
    def __init__(self, relations, isUnique, damping = 0.85):
        self.pageRanks = {}
        self.relations = {}
        self.outgoing = {}
        self.damping = damping

        for key in relations:
            self.outgoing[key] = len(relations[key])

            if key not in self.relations:
                self.relations[key] = []

            for linkTo in relations[key]:
                if linkTo not in self.relations:
                    self.relations[linkTo] = []

                if linkTo not in self.outgoing:
                    self.outgoing[linkTo] = 0

                if isUnique or key not in self.relations[linkTo]:
                    self.relations[linkTo].append(key)

    def calculateRanks(self, iterations = 20, startRank = None):
        if startRank is not None:
            self.pageRanks = startRank
        else:
            for key in self.relations:
                self.pageRanks[key] = 0.0

        i = 0
        while i < iterations:
            for key in self.relations:
                PR = 0.0

                for linkFrom in self.relations[key]:
                    PR += self.pageRanks[linkFrom] / self.outgoing[linkFrom]

                self.pageRanks[key] = 1 - self.damping + self.damping * PR


            i += 1

        return self.pageRanks