Network Business Cases
In this lesson, we’ll see how network science can be applied to business in order to help with analysis in various cases.
Detecting Fraud in Banks with Networks
Banks lose billions of dollars every year due to fraud. To combat this, they seek and implement diverse methods that can help them in the detection of fraud.
Traditional methods of fraud detection focus on individual records. They individually analyze each customer or transaction, looking for anomalies. This is a great approach that detects many cases. However, it can be complemented with a network approach that can detect cases where the traditional approach fails.
Sophisticated fraudsters have developed a variety of ways to elude discovery, both by working together and by leveraging various other means of constructing false identities. In order to improve these systems, we need to look beyond the individual data points, and focus on the connections that link them.
https://www.oreilly.com/ideas/there-are-many-use-cases-for-graph-databases-and-analytics
Viral Churn
In previous lessons, we discussed how to predict churn with supervised learning. We were focusing on analyzing individual records. Thus, we could detect if a customer was about to leave because he changed his behavior. However, by then, it may be too late as the decision is already taken.
One of the reasons that may drive individuals to churn is social influence. If my friends migrate their line and tell me about it, there is a chance that their decision might influence me and that I decide to migrate my line too. This pattern cannot be found when analyzing individual records, but it can be detected when analyzing the links.
Viral Marketing
Network analysis is also extremely important for marketing, especially for viral marketing. Suppose we want to launch a marketing campaign on social media and we want this campaign to become viral. So we aim to select some accounts to whom we will pay to propagate our posts. How do we select the targets to optimize the campaign?
If we just analyze individual records, we may decide to select those with more followers. So if this is a health campaign, and I have a budget for 10 accounts, I will select the ten accounts with higher number of followers that talk about health. However this is not optimum.
To understand this, let’s look at the figure below.
Suppose each node represents a Twitter account and the links are who follows whom. Now imagine that the ten accounts with more followers fall in the blue community. In that case, your posts will quickly propagate through that group but will fail to reach the rest of the network. Hence the campaign will not become viral.
If you want it to be viral, you should select an influential user from each group. So using network science, you would first detect the communities. Once you have separated users into communities, look for the top influencers of each community.
Now if you select influential accounts across all the communities, the posts will affect the whole social network. Hence, by analyzing the data and focusing on the links among users, you’re able to conduct a more efficient campaign than by analyzing each user individually.