Design of a personal agent integrated with a Linked Data Indexing System

Alberto Mardomingo Mardomingo. (2015). Design of a personal agent integrated with a Linked Data Indexing System. Trabajo Fin de Titulación (PFC). ETSI Telecomunicación, Universidad Politécnica de Madrid.

Abstract:
This project has focused on the design and implementation of a Personal Agent integrated with a Linked Data System, allowing users to interact with it using natural language. In order to achieve our goal, we have studied current technologies that would allow us to analyse natural language, retrieve and index semantic information, and present it to the user. We have therefore proposed an architecture for our system that would allow us to perform these tasks, integrating a question answering system, a conversational agent with natural language processing capabilities, as well as an information retrieval system, along with a linked data indexing system. Based on this architecture, we developed several prototypes that would allow us to test the architecture. The first one of them focus on a simple example of e-learning in Spanish, providing students with a platform to solve their questions about the Java programming language, and offering new topics so they can delve into their study. A different prototype has analysed the implementation of a similar system for a hetero- geneous collection of documents in English, testing the student ability to design a modular system that can be easily extended. Finally, we have studied our first prototype in a real scenario, gathering information about its efficiency, the success rate while tested with a corpus, as well as proposing an experiment with users. The experiment compared the system with traditional question answering systems, analysing the improvement in the results, the user experience, as well as the patterns in user behaviour when faced with a system like ours for the first time.