prudsys RECOMMENDATION ENGINE - Technology

Architecture

Conventional recommendation engines generally function on the classic principle of data mining: they evaluate historical customer data, especially master data like transactions (clicks, purchases, etc.) and use the data to calculate models, often in the form of rules such as "the purchase of product A leads to the purchase of products B and C", which is then used, for example, in the web shop to display recommendations.

The prudsys RE, on the other hand, learns not only from historical data but is constantly learning through interaction with the customers. This allows it to continuously adapt its recommendations to changes in purchasing patterns. Equally important is a second advantage: it strategically tests out new recommendations, thus constantly expanding its scope.

Another feature unique to the prudsys RE is that it not simply recommends the closest best product in each step but that it performs an optimisation via the chain of all possible interactions of a user in order to maximise total sales. This means that in some cases a product that is less profitable at the time may be recommended, if it is more likely to lead to more profitable products subsequently. Thus, the prudsys RE is geared towards the long term and is target oriented.

As a mathematical framework, the prudsys RE uses a method from the field of reinforcement learning (RL) in combination with data mining to learn about transition probabilities.

The recommendation engine can be operated in two modes: in offline mode it evaluates historical data and generates the recommendations from them. In online mode it learns from interacting with customers. A combination of the offline and online modes is best: in offline mode the engine initialises the recommendations by using historical transaction data; in online mode this data can be continually adapted and improved.

The prudsys RE comprises several layers: the tried and true business intelligence library XELOPES makes up the core, providing the preprocessing and the meta data management in addition to the analysis algorithms. On top of that is the layer of the actual prudsys RE, which realises the application logic. Sophisticated functionalities like price optimisation and reporting are supported by special modules. Connectors to external systems are also integrated. All of the interfaces for the configuration and communication of the prudsys RE are together in one unified Java API. This enables the prudsys RE itself to be incorporated as a library into third party applications, which is especially advantageous for OEM applications. Finally, an analogue service layer based on the Java API is implemented - especially for web services. This allows the seamless integration of the prudsys RE into service-oriented architectures (SOA).

For more information on the prudsys RECOMMENDATION ENGINE
call +49 371 27093-21 or