The Prediggo Difference: Semantic + Transactional

    Today's commercial recommendation systems are based on Amazon's Collaborative Filtering technology, commonly known as "Customers Also Bought." This approach predicts purchase behavior using transactional data to compute the "distance" or links between items. Recommendations are made by finding the closest items to the one a user is viewing.

    The limitations of Collaborative Filtering include scenarios typical to eShop marketing such as featuring New Products, Long Tail Products, Items Never Purchased Together and more. To solve these issues and deliver better personalization, Prediggo's founders conceived a patented new technology called Ontology Filtering. Ontology Filtering does not only use past purchase behavior but more importantly it also uses the semantic information of products to formulate recommendations.

    Prod_Detail

    To build more complete product relationship links, Prediggo analyzes the semantic data assigned to each product along with relevant purchase history.

    For example, the semantic descriptors of a bottle of wine - name, grapes, rating, region, country - are critical links in formulating recommendations.

    Optimized Data Structure: Ontologies

    One of the key innovations behind Prediggo's patented technology, as well as its name, come from the advanced data structures we use.

    Ontologies are tree structures where the leaf nodes contain the items to be recommended and edges between the nodes model the relationship between the items.

    Ontologies allow Prediggo to organize the semantic and transactional data in RAM for fast and concurrent access.

    		ontology_structure

    Most importantly, ontologies are easy to update without impacting the performance of the site. While most vendors will refresh their data structure on a weekly basis, Prediggo automatically updates its ontologies daily, without any human intervention. In fact, our News customer's ontologies are updated every hour. Ontology Filtering insures that we always accurately reflect the product catalog and user behavior for each site. 

    Many eShops must deal with multiple markets, their languages and different user behaviors. Ontology filtering deals with this very easily by creating separate ontologies for all your countries and languages, allowing you to offer your customer a truly unqiue experience.

    Real Time, Behavior Targeted Recommendations

    When a user browses the website, each of their actions helps build a unique browsing profile and this profile is automatically injected into the ontology. To make recommendations, we use our patented algorithm to infer the preference of the unseen node from the nodes that contain the user's browsing profile. As the ontology is stored in RAM, this inference process is done in real time (on average, in less than 2 milli seconds) for every single recommendation request.

    Award Winning

    After the patent submission, The core elements of Ontology Filtering technology have been published and peered reviewed in Artificial Intelligence most prestigous worldwide conferences (such as KDD, AAAI and IJCAI). In 2007, it was also awarded the "Breakthrough Technology Award"  from the Dimitris N. Chorafs Foundation.