Reply is the place to meet an incredible variety of enthusiastic, passionate, ideas-driven people, who want to make a difference and an impact.Would you like to know more?
Recommender engines are systems that help users discover items they may like through the use of machine learning technologies. Product Recommenders are becoming one of the most important means of communication between retail companies and their customers. One-to-one marketing helps both to increase Customer Engagement whilst boosting sales on E-commerce platforms. On the other hand automated analytics helps improve the efficiency and the reactivity of the actions. For example, Netflix and Amazon are gaining a large share of their revenues thanks to their recommender engines. In fact, 35% of what consumers purchase on Amazon and 75% of what they watch on Netflix come from product recommendations based on modern algorithms.
Data Reply, who specialise in data management with Big Data & Advanced Analytics methodologies, have developed a tailor made version of a Recommender Engine which can take advantage of various types of interactions of the customer with the product (purchases, ratings, web page views, social sharing actions…) and is capable of automatically adapting over time to specific customer needs by listening to the feedback actions collected. A book recommender engine has been customised, starting from this general framework, for a client in the Publishing Industry.
Different sources producing structured and unstructured data, including transactional data, web clickstream, social rating and reviews are directed into a central Data Lake. This approach enables data to be stored in a single place in a raw format which historically has been confined in silos and never correlated with each other.
Data is then crunched by several algorithms with the capability of identifying customer preferences and tastes with a one-to-one approach, instead of the classical segmentation in groups, even when the information level available is very different among customers.
Recommending books is a multi-objective problem: you want to optimize at the same time relevance, engagement, diversity, novelty, business requirements, and more.
This can be achieved by creating an ensemble of algorithms, each one of which is designed to achieve one specific objective. One last algorithm is responsible for finding the right mix for each customer: for example, a customer can be interested in niche products rather than best sellers.
The system learns in two ways: firstly, by obtaining new data (such as the purchase of a new product) it can improve the knowledge of a customers preferences and as a consequence provide more relevant recommendations; secondly, by using the interactions of the customer with the system (the clicks on the recommendations displayed) with a methodology called “reinforcement learning”, very similar to the try-and-adapt approach the robots use when learning to walk.
The whole system works both with daily re-training steps and real-time adjustments to the recommendations provided. It can be easily integrated with the most diverse systems (mobile, web, intelligent totems…) thanks to a simple API interface.