Logo Delphi


The global spread of information technology and the popularity of digital content puts media and digital content into the hands of users in any context, including e-commerce.
This new opportunity poses new problems for managing and organising information and multimedia assets, and drives the search for new solutions to improve user experience in the field. 
 Recommendation systems are of huge importance in a scenario where interactivity, recommendations and reviews are determining factors. The value of "recommendation" systems is that they provide users with a personalised item review tool before proceeding to the purchase stage.


Cuma is a recommendation and review service designed for e-commerce platforms.
Recommendation systems provide users with personalised review tools based on algorithms related to their own interests, and therefore to potential items they may be interested in buying. Recommendations are formulated by combining information obtained from item purchase and shipping habits.
Cuma provides advice and recommendations for generic articles to both profiled and non-profiled users. The platform chooses items from time to time that may meet users' needs.

Basic features

The platform comprises a series of services designed and developed in a service-oriented architectural logic. It offers a range of services and an orchestrator (recommendation strategy advisor) so that the system remains independent of the application domain.
The system is also streamlined to operate in any condition for database population. In particular, Cuma integrates and revisits even fast computing processes where there are large amounts of data that are frequently updated.

Some of the services/functionality that Cuma provides:

    • knowledge based recommendation services | Services based on inferential methods and user profiling. These services allow users to view recommendations tailored to the specific attributes of each application domain. Specifically, data relating to the application domain is cross-referenced with user needs and preferences, allowing recommendations for relevant desired items;
    • utility based recommendation services | Services based on evaluation mechanisms for the usefulness of an item in relation to specific user needs. These mechanisms do not build broad user stereotypes but develop single specific preferences based on matching item usefulness with user need. The concept of usefulness is based on calculations derived from user profile data;
    • profit based recommendation services |Services that factor in the profitability of a specific item in relation to the provider's profit requirements. The recommendation must take into account the reliability of the recommendation system in the eyes of the user, in spite of the profitability aspect; 
    • collaborative filtering (CF) servicesMechanisms for generating and evaluating  recommendations that take into account customer product browsing history and historical data on interest raised by individual products. CF services match reviews and recommendations for a specific article and identify correlations with users based on established ratings. Recommendations in this case arise from comparisons and similarities between various users. This type of filter is based on the assumption that two users that generate similar reviews have similar tastes.
      CF services are divided into two main subcategories:
      - Memory-based, which work on user history to produce forecasts, using the principle of "similar users";
      - Model-based, which work on models that use learning tools, including neural and Bayesian networks and semantic indexing techniques.
      The CF identifies similar users by suggesting articles that have been positively reviewed by other users. These algorithms are not effective where there is limited data to produce forecasts. Cuma overcomes this cold start scenario by using correction tools operated by the Advisor element

Advanced Features

In addition to the services described above, Cuma offers a range of services for managing the relationship between supply and demand, which is scalable and adaptable to the application domain. These services are based on the so-called "knowledge cards" constituted by ontologies that enable the creation of scalable and adaptable recommendations in relation to specific types of users and products.

In Cuma there is also a form of semantic matchmaking between user requests and offers using ontologies. In this case too, the algorithm used has the advantage of being of general and therefore completely independent of any specific application domain. More specifically, the service that we have designed performs semantic matchmaking between supply and demand scalably, both B2B and B2C, with the addition of domain knowledge cards and standard metadating.

The Cuma platform is enhanced with features that can be used in various fields such as community platforms (social) or multi-channel information portals, where the user's search target is not a product for purchase but information content to inform consultation.


The Framework Recommender consists of a set of services and an orchestrator. The system interacts with the external e-commerce platform through web-service. The framework also provides a data synchronisation component to keep the database continuously aligned with the e-commerce database. 

депутат лобановскийникас кернесхарьков nikasстроительство бани под ключ