If you’ve ever shopped on Amazon or Flipkart, watched a movie on Netflix or YouTube or for that matter read articles on LinkedIn, you must have experienced recommendation systems in action. Online retailers and entertainment providers make real-time recommendations or show advertisements based on what a customer is viewing or listening. Social networking sites such as LinkedIn and Facebook suggest additional connections or friends based on your existing network. You often wonder how these systems can predict what you may like and showcase the product/services/articles which are of interest to you.
Search and Recommendations are two important facets of user engagement. During the search, a user is seeking “information” while with later a relevant “information” reaches out to the user. While Search is more generic in nature, the recommendation is more targeted and hence the likelihood of getting the user attention is much higher.
A lot goes behind building a world class recommendation system. The system analyses historical buying behavior and makes recommendations in real time. Most of this highly intensive computing system uses Lamda Architecture, leveraging “insight” from a batch data as well as real-time inputs for high predicted clicks.
It’s a win-win situation for both customer and business. Customers get personalized, helpful suggestions on additional products or services. Business proactively builds better customer relationships, boost retention and increase sales by an estimated 5 to 20 percent. For example, when Slideshare moved from Solr’s “More Like This” based recommendation to “Collaborative Filtering” based recommendations, it saw a huge jump in user engagement leading to more than 20% increase in pageviews.
The world has matured a lot around building recommendations for various use cases discussed above however the consumer based transportation system remains untapped. Have you ever wondered, if Uber can come back with recommended taxi availability when your search fails? Or based on customer’s booking patterns, can system comes with right suggestions across various product offerings?
At Zoomcar, we have tried to solve that problem for our users. We give variety of recommendations based on
- Location
- Car Group
- Timing
To achieve this, we run a parallel search for adjacent time slots across different car groups and location combination in real time. While the current version is a pure function of search but with time we are planning to include several key attributes of users and inventory to generate personalized recommendations in real time. Stay Tuned!!
