PACINO - Personality And Conversational INformatiOn Access

Over the past few years, an increasing amount of attention has been directed toward developing methods that enable Conversational Information Retrieval (CIR) and has brought forward the initial development of CIR systems. A CIR system is an IR system that enables the entire information access interaction to be carried out as a conversation with a search agent. Thus, a CIR system should be able to understand the user query, interact with the user to get a better understanding of the user information need, ask clarifying questions, improve the search results in the light of the clarifications provided, and extract and present search results/answers in a simple and clear way.

PACINO aims at advancing the state-of-the-art of CIR systems by incorporating passages of the retrieved information into a knowledge distillation process that will drive a mixed-initiative conversational system and predict when to ask a clarifying question, select/generate the clarifying question, and rank retrieved documents according to users’ responses to the questions. We also plan to enable long-term memory that can span across various conversational sessions by building and attending to user personal knowledge graphs. Finally, we will advance research in the evaluation of CS systems by enriching the training data using synthetic query variations and adding various imperfections.

Furthermore, PACINO will also build a CIR that, while answering the user information need, will be able to understand if the user suffers from some mental health disorder, adapt the conversation to avoid worsening it and steer it toward topics that could help alleviate the condition.

Lastly, we are going to explore all issues related to bias and fairness in CIR in order to avoid unfair and biased treatment of sensitive topics in the conversation, in particular, but not only, towards users with personality disorders.

Thus, this proposal is concerned with advancing the state of the art of CIR systems to enable the development of a Personalized CIR system that will interact with the user in a multi-turn and mixed-initiative conversation manner to enable the user to perform exploratory searches on any topic. The CS system will incorporate an understanding of various aspects of the dialogue with a user, including user's personality recognition, personality disorder identification, adaptation, and dealing with unfair and biased search results. In other words, the system will adapt the modality of interaction and dialogue to the user's behavior and personality enabling a more engaging and effective information-seeking experience.