A FoReLab research paper was awarded with the Best Paper Award Honorable Mention at SIGIR 2023
A new method to evaluate and predict the effectiveness of Artificial Intelligence systems, such as voice assistants, in carrying on a conversation that meets user expectations.
This is the focus of the article awarded with the "ACM SIGIR 2023 Best Paper Award Honorable Mention" at the recent SIGIR conference (Conference on Research and Development in Information Retrieval” https://sigir.org/sigir2023/), held in Taiwan, which annually brings together the leading academic and industrial experts of information retrieval, the science behind search engines.
The research wa carried on by the Department of Information Engineering of the University of Pisa in collaboration with the University of Padua and CNR.
"Search engines - explains Nicola Tonellotto, professor of computer engineering at the Department of Information Engineering and co-author of the article - are evolving from the simple "question-multiple answers" scenario (the famous 10 blue links) to a " conversational”, where the interaction is composed of multiple exchanges of questions and answers. A typical example is the interaction between humans and digital assistants, whose implementation is often based on "deep" neural networks. The article proposes a new tool to predict the success of a digital assistant in answering the next question in a conversation, or to modify the behavior of a digital assistant on the fly in retrieving information to improve the quality of the elaborated answers. mathematical model estimates the "semantic distance" between a question and an answer, and therefore evaluates the system that best responds to user requests".
The work on the interaction between human beings and artificial intelligence is part of the activities of the Forelab, the laboratory of the Department carrying on cutting-edge research on technologies for the new paradigm of society and industry 5.0
