Ernesto William De Luca




| ITC-IRST | EURAC | OVGU-Magdeburg | TU-Berlin | FHP | GEI | All |

Berlin Institute of Technology


  • Project: SPIGA ( 2009-2012 )



    The goal of the project SPIGA (Language-independent personalized information provision with global orientation) is the development an information service that automatically creates document collections for semantically described topics. Such a service can be used in products such as personalized newsletters or welcome pages, or in press reviews. The idea of this project is to exploit both the semantics and multilinguality of the news articles. Working within the news domain has the advantage that often the same news event is covered in different languages, such that there are multiple text variants of the same information. The core approach focuses on the extraction of knowledge - or rather semantic concepts - from news, in order to link the news documents. Each news event is thus represented by its concepts, and can hence be managed in a language-independent fashion. The research focus lies on the development and evaluation of methods for the Disambiguation of Named Entities. The representation of news documents as a set of concepts and concept relations then allows to search and group documents on the basis of these concepts, e.g. to create a press review.
  • Project: SERUM ( 2009-2012 )



    SERUM (Semantic Recommendations based on Large Unstructured Datasets) establishes the basis for a semantic recommender system that calculates high quality recommendations based on a semantic analysis of user behavior and news articles. The aim of the project is to develop a recommender system that computes recommendations independently from a specific use case or domain. The recommendations are personalized and adapted to the specific needs of a user based on personal interests and preferences. Within the SERUM research project, the goal is to recommend news articles based on the previous reading behavior of a user. This reading behavior is analyzed to create a personalized news digest for each user. The recommendation system is connected to a semantic knowledge base, which is modeled and managed as an ontology. The semantic knowledge is being linked with information from current news articles. Based on this semantic network, new algorithms were developed that analyze the semantic information and the user behavior to compute high quality recommendations.
  • Project: KMulE ( 2009-2012 )



    Context-aware recommender systems are becoming a popular topic, yet there are still many untouched aspects. In the project KMulE (Context-based Multimedia Recommender System), we study context identification and the concepts involved in hybrid and context-aware systems. The goal of the project is to implement a conceptual architecture for a context-aware recommender system for movies and TV shows. The system consists of a number of modules for context identification and recommendation. Key contextual features are identified and used for the creation of several sets of recommendations, based on the predicted context. The resulting prototype system will be evaluated and incorporated into the recommendation engine of movie and TV recommendation website Moviepilot.
  • Project: PIA ( 2009-2012 )



    The goal of the Personal Information Assistant (PIA) project is to provide a comprehensive agent-based solution for the personalized and device-independent supply of information. The user receives information that is relevant to his personal needs and interests. This includes daily news, background knowledge on work issues, or information on leisure time plans and activities. Besides a typical web search engine interface, the PIA system allows users to define and save searches which are then continually monitored by search agents for any new developments. The architecture of the PIA system is designed to allow information sources to be flexibly integrated into the system. Information is analyzed and filtered using advanced filtering methods, e.g. content-based or collaborative filtering techniques. The use of multiple filtering techniques, which are guided by integrating user feedback from a learning and user modelling component, guarantees a high accuracy of search results.
  • Project: SPREE ( 2009-2012 )



    The main focus of the "SPREE - A Community-Based Information Exchange Network" project is the implementation of an online portal for an efficient knowledge transfer between its users. Therefore the platform has to be capable of identifying the best qualified users for answering a given query in real time, and the quality of the algorithm doing the matching between query and experts will be of central importance. Moreover the portal will provide means - such as chat - allowing users and experts to directly communicate with each other. The SPREE project provides an online portal offering the functionality of automatically identifying the best qualified users (experts) for a given query. Furthermore the system will offer means such as chat for a direct communication between user and expert in order to maximize the result quality.