PRIN 2022 CALL. DECREE NO. 104 OF 2 FEBRUARY 2022
PROJECT TITLE: CAMEO - Conversational Agents: Mastering, Evaluating, Optimizing
CODICE CUP: D53D23008920006
Budget: € 77,794.00
P.I. or U.R. Manager: Tommaso Di Noia
Other Research Units or any Sub-Units:
UNIPD - University of Padua
CNR - National Research Council
POLIBA - Polytechnic University of Bari
ROMA1 - Sapienza University of Rome
Brief project description
Conversational agents are now widely used in chatbots, smartphones, and smart home devices such as Google Home, Alexa, and Siri. However, they still struggle to support natural, engaging, and context-aware real-time interactions.
The CAMEO project addressed these limitations by leveraging user context as an internal knowledge source, capturing information from past interactions and enriching it with domain knowledge. This enabled more effective dialogue management, query answering, personalized search, and recommendation.
Purpose
A key contribution of CAMEO is the integration of search and recommendation within a unified framework, along with improved evaluation methodologies to better understand and optimize conversational agent behavior.
Expected results
Definition of a semantic search engine for querying the thesaurus that will make it possible to retrieve headwords and analyse their semantic evolution, highlighting statistics, such as the frequency of a term over time, variations in the meaning of a term, more co-occurring terms. Querying the thesaurus will make it possible to trace and highlight the cultural reflection that took place between the two world wars in France. The thesaurus will be available in open access.
The dissemination activity includes an international conference at which the results of the research will be presented and a subsequent publication of the proceedings in an open access journal. The project also includes extensive public engagement activities, such as lectures in higher education and summer schools.
Achieved results (this section will not be completed for the PRIN 2022 projects subject to progression, as they are still ongoing)
The project successfully established a novel paradigm of joint conversational search and recommendation, integrating catalogue-based recommendation with open-world search and advancing the theoretical foundations of conversational systems. A key outcome is the development and release of CosRec, the first human-annotated dataset specifically designed for joint conversational search and recommendation, providing a unique and reproducible benchmark for future research.
The project also laid the groundwork for the integration of conversational systems with emerging Large Language Models (LLMs), highlighting both their potential as natural language interfaces and current limitations in personalization and reliability. By grounding conversational agents in structured knowledge sources, the project offers a robust foundation for future developments. Overall, the results significantly advance the state of the art and provide essential resources and directions for the continued evolution of conversational search and recommendation systems.
