Be it E-Commerce OR online streaming services OR news/article platforms OR banking/insurance services OR any other industry, content on websites can directly/indirectly decide whether a visitor will become our customer or not.
Ultimately if we do not show relevant content to website visitors, they will bounce soon. Hence, it becomes not only sufficient but necessary that we provide personalized/relevant content to visitors.
This objective is achieved in 2 ways:
1. One can simply put a generic content (not specific to any particular type of viewer), surely this will not offend any one but this will not also motivate visitors to spend more time on websites.
2. One can analyze behavior of different type of users, their browsing pattern or profile of uses and accordingly recommend content to users. This can be done through a development of recommendation engine.
Types of Recommendation Engine:
In the era of big data, it is very much possible to identify what a customer’s choice will be. Even if we do not use advanced deep learning, we can define preferences by exploratory data analysis and business knowledge.
If we explore recommendation systems, we see there are many types of recommendation systems are available:
1. Market Basket Analysis (Association Rule Mining): Here we use Apriori algorithm to identify association between products which were bought together
2. Collaborative Filtering (User Based /Item Based): We apply similarity matrix using items selected (OR videos viewed) and users to rank preference of items for users.
3. Content Based Filtering: It helps in identifying relevant content to users. We can apply vector space model or latent semantic indexing.
4. Look Alike Models: Many companies cluster customers (Using unsupervised ML methods) and explore products with respect to individual clusters. By virtue of this method, we can conclude that this type of segments normally buy these types of products OR they normally watch these types of videos.
5. Domain based filters: It is the traditional, most basic but most insightful application. One simply recommends items based upon country/age/language. No hard-core data science technique is used.
Application of recommendation systems on websites:
1. Video analytics: Streaming services such as Netflix/ YouTube use this functionality a lot . Now a days many companies are transforming their web content into videos. They simply host videos on websites so users don’t need to read content instead they can watch videos also. Normally collaborative filtering and market basket analysis are adopted to develop recommendation engine although many analytics team develop simply rule based engine to recommend videos.
2. Content Analytics: Static websites which host a huge number of contents/articles/product information such as banks/insurance companies use various data science methods to personalized contents. We can optimize content using most influencing keywords and removing least useful texts. We will also be able to see where to place 'click' buttons on website to reduce bounce rates.
3. Product analytics: Companies widely use market basket analysis and user-user based collaborative filtering but many organizations who do not have a good infra or big data setup simply apply rule based algorithms such as recommend education products to 21-26 year population etc. We can also apply look alike model approach to recommend relevant products to customers.
About the Author : Ankit has 7 years of industry experience in analytics and machine learning. He is, currently, a part of the marketing analytics team at leading indian banking firm where he is responsible for Data Science programs. His previous stint was in telecom, IT sector and manufacturing sectors.