Have you ever heard the phrase, “the best restaurant is the one that knows my name”? If a restaurant knows what you like and serves you according to your preferences, it can surely become your favorite place to eat. This logic is not necessarily limited to restaurants. Individualized service, just like the restaurant knowing your taste, learns about each customer’s preference and then based on this, provides the right services and products.
Two companies known globally for their individualized services are Amazon and Netflix.
Amazon has been providing individualized service for over a decade, and it all began with their book recommendation service. This service is considered one of the key factors that led Amazon to success.
Individualized services make up a large portion (over 35%) of Amazon’s overall sales. Many other companies began to show more interest in individualized service after Amazon’s success.
Another case is Netflix, America’s biggest online streaming service corporation. Netflix started with a postal DVD rental service, and increased its annual sales up to $4 billion by jumping into the online streaming service market in 2009.
The key factor of such tremendous growth was the individualized content recommendation service for movies and TV shows. Netflix created 76,800 categories to classify their contents and used these categories to offer individualized recommendations which have an accuracy rate of over 80%.
Other than these well-known success stories, there are countless other individualized services. You may have experienced some of them without realizing what they were.
What businesses need to provide high quality individualized services are analysis and metadata.
Individualized services don’t blatantly ask what their consumers like. Instead, they analyze consumer records that show which products and services each consumer has bought, used, or shown interest in, so the company can figure out personal preferences and make recommendations accordingly.
Analysis, therefore, is a critical part of individualized service. This is why Amazon and Netflix continue to invest large amounts of money in order to improve the quality of their analysis.
Another critical factor is metadata, which defines the products’ (or services’) characteristics and properties that can be used for analysis. There can’t be any quality individualized services without well-defined metadata, no matter how great the analytical algorithms are.
Netflix attaches a 36-page-long document to each movie summarizing its characteristics in order to create detailed metadata. This task can’t be automated, and is therefore done by their staff manually.
As we see, metadata requires accuracy, detail, and continuous quality control. Companies will soon understand the importance of metadata, and start making continuous investments and efforts to improve it.
What methods are used in analysis for individualized service, then? There are a number of them, and I’d like to introduce collaborative filtering, which became popular since Amazon first used it for their service.
Collaborative filtering is a method that predicts users’ interests by using collected items or their preferences on products. This method is commonly used for books, movies, VOD, and music recommendations.
Preference on products can be calculated based on the information collected from a user’s purchase and search history as well as reviews. The preference model should differ based on the industry domain and the company’s business characteristics.
Collaborative filtering can be divided into user-based and item-based filtering.
As you can see in the image, user-based collaborative filtering finds other users with similar preferences, and classifies them into the same group. The commonly preferred items in the user’s group are then recommended by the system.
On the other hand, item-based collaborative filtering recommends different items with high correlations to preferred items based on the user’s purchase and search history. Amazon also uses item-based collaborative filtering for their recommendation system.
Users with similar preferences or the items which are highly related to a previously purchased item can be found through various distance calculation methods (such as Cosine and Euclidean) based on the users’ preference information on each item. This is called ‘similarity’.
In short, user-based collaborative filtering recommends items based on similarities between users, while item-based collaborative filtering makes recommendations based on similarities of different items.
Now that we have covered well-known cases of individualized service and its critical factors as well as the algorithms commonly used for this service, let’s have a look at some of its prospects.
The recommendation model is advancing in order to improve the accuracy and diversity of the recommendations. The problem here is that high accuracy tends to lower the diversity, and diversity tends to lessen the accuracy.
This means recommending only the popular items will keep accuracy high but ends up being less diverse, while recommending long tail products for more diversity lowers the overall accuracy.
Academia and companies continue to study in order to achieve both accuracy and diversity by improving the individualized recommendation model. The final goal of this recommendation model with both high accuracy and diversity is called ‘serendipity’.
At the point of serendipity, the company can recommend new products that their consumers had not previously thought of but will still be satisfied with. Such improvement can present unforeseen joy to consumers, while increasing a company’s sales.
Big data is advancing continuously with more specified analysis. Have you thought about products and services that you would love, but never even imagined to have available? This is the idea big data analytics is going for.
Written by Hwanmuk Lee, Advisor in LG CNS Big Data Business Group
Big Data Analytics to Prevent IT Infrastructure Failure -Big Data Analytics (1)-
Integrated Security Logs and Finding Risk Patterns -Big Data Analytics (2)-
FDS for Safer Payments with Big Data -Big Data Analytics (3)-
Hybrid DW, the Future of DW -Big Data Analytics (4)-