Engaging Digital News Audiences with an AI-powered News Chatbot
Chatbot has become an emerging paradigm for a new kind of user interface driven by recent advances in artificial intelligence. It has become popular in various domains such as e-commerce, tourism, education, healthcare, and entertainment. Nevertheless, few have applied it in the digital media domain, despite the fact that more and more audiences expect to receive news services via chatbot, according to the 2018 State of Chatbots Report. As claimed by BBC News Labs in 2018, news chatbots are the modern trend of delivering public news, and will make the news more engaging and better suited to social media and messaging apps. It can attract younger audiences given its informal and more conversational communication style.
This project aims to develop a news chatbot for meeting the increasing demand of news organizations to interact with audiences in a more engaging way. It will strengthen the news chatbot’s intelligence and practical values from various aspects, such as dedicated recommendations and explanation specialties, dynamic dialogue management for flexible feedback elicitation, and desired user control. Specifically, we aim to achieve the following three objectives:
- In our chatbot, news recommendations will be a core component to engage readers with personalized assistance. In order to address user preference shifts and improve the diversity and recency of recommendations, we plan to accommodate multiple heterogeneous sources of side information on a knowledge graph, in order to capture readers’ hidden and changing interests as well as her/his propensity towards information diversity. Moreover, the chatbot will explain its recommendation in a natural and easy-to-understand style, for increasing the system’s transparency and trustworthiness.
- One major issue with current chatbot systems is that they are mostly system initiatives, while leaving low control for users to freely pose questions or provide feedback. Therefore, in our system, we will build a mixed-initiative, critique-oriented dialogue management framework to maximize the flexibility of users’ feedback provisions. We will build this framework on the basis of a reinforcement learning network, by strengthening its intent prediction and feedback elicitation capabilities.
- The evaluation of a chatbot system counts on users’ assessment of its naturalness, trustworthiness, and effectiveness. We plan to conduct user experiments on a commercial news portal for validating the system’s practical effects on improving real readers’ engagement with news over time, and their tendency to use such a system in their daily life.
This project is based on a long-term strategic collaboration agreement that HKBU has established with Wisers AI Lab of Wisers Information Limited (Wisers).