[Paper Review] Emora: An Inquisitive Social Chatbot Who Cares For You

Sarah E. Finch, James D. Finch+10 more|arXiv (Cornell University)|Sep 10, 2020

Topic Modeling15 references7 citations

TL;DR

Emora is a personality-driven, inquisitive social chatbot developed at Emory University that prioritizes personal experience and opinion-sharing over information retrieval, using a hybrid pipeline of rule-based and neural components. It achieved a 3.61 average user rating in the Alexa Prize competition, with personal-experience-focused conversations scoring significantly higher (3.85 vs. 3.79) and current-event awareness further boosting user satisfaction.

ABSTRACT

Inspired by studies on the overwhelming presence of experience-sharing in human-human conversations, Emora, the social chatbot developed by Emory University, aims to bring such experience-focused interaction to the current field of conversational AI. The traditional approach of information-sharing topic handlers is balanced with a focus on opinion-oriented exchanges that Emora delivers, and new conversational abilities are developed that support dialogues that consist of a collaborative understanding and learning process of the partner's life experiences. We present a curated dialogue system that leverages highly expressive natural language templates, powerful intent classification, and ontology resources to provide an engaging and interesting conversational experience to every user.

Motivation & Objective

• To develop a social chatbot that emulates human-like, experience-focused conversation rather than relying on information retrieval.
• To enhance user engagement by embedding personality, opinions, and active curiosity about the user’s life experiences.
• To create a self-consistent, personalized dialogue system using a hybrid architecture combining rule-based and data-driven components.
• To investigate whether personal-experience-centered dialogue improves user satisfaction compared to fact- or opinion-based topics.
• To explore the impact of current events and holidays on user perception and interaction quality.

Proposed method

• Emora uses a pipelined architecture with two-stage NLP processing for intent classification and entity extraction, leveraging curated templates and ontologies.
• The dialogue manager employs a state machine with flexible topic transitions, enabling dynamic shifts between topics based on user input and emotional cues.
• A combination of handcrafted templates and neural intent classification ensures expressive, natural, and contextually appropriate responses.
• The system integrates user-specific data and conversation history to maintain personality consistency and personalization across sessions.
• Current-event awareness is implemented via dynamic 'openings' that adapt to holidays and global events like COVID-19, enhancing relevance and emotional resonance.
• The system uses Amazon’s CoBot framework for scalable deployment on AWS, with DynamoDB for state and user data persistence.

Experimental results

Research questions

• RQ1 Does prioritizing personal experience and opinion-sharing in dialogue lead to higher user satisfaction compared to traditional information-focused chatbots?
• RQ2 How does the inclusion of a personalized, opinionated persona affect user engagement and perceived authenticity?
• RQ3 To what extent do current events and seasonal holidays influence user ratings and perception of the chatbot?
• RQ4 Can a hybrid, rule-based and neural approach achieve better consistency and control than purely data-driven models in open-domain social chat?
• RQ5 How do topic transitions and emotional continuity contribute to the perception of a coherent, caring personality?

Key findings

• Emora achieved an average user rating of 3.61 over the last seven days of the Alexa Prize semifinal, indicating strong overall performance.
• Conversations focused on personal experiences scored 3.85 on average, significantly higher than the 3.79 average for 'info and opinions' topics (p < 0.10), showing a preference for experiential dialogue.
• The use of holiday-specific openings—such as for Valentine’s Day, Easter, and April Fools’ Day—resulted in a positive trend in user ratings, suggesting enhanced user satisfaction during event-aware interactions.
• The COVID-19-themed opening had a sustained positive impact on ratings, indicating that long-term, globally relevant events can improve user perception over time.
• The system demonstrated that combining curated templates, intent classification, and ontology resources enables expressive, consistent, and emotionally engaging dialogue without relying solely on large-scale neural training.
• The results support the hypothesis that experience- and opinion-oriented dialogue strategies significantly enhance user engagement and perceived social connection in chatbots.