Abstract:LLMs are frequently used tools for conversational generation. Without additional information LLMs can generate lower quality responses due to lacking relevant content and hallucinations, as well as the perception of poor emotional capability, and an inability to maintain a consistent character. Knowledge graphs are commonly used forms of external knowledge and may provide solutions to these challenges. This paper introduces three proposals, utilizing knowledge graphs to enhance LLM generation. Firstly, dynamic knowledge graph embeddings and recommendation could allow for the integration of new information and the selection of relevant knowledge for response generation. Secondly, storing entities with emotional values as additional features may provide knowledge that is better emotionally aligned with the user input. Thirdly, integrating character information through narrative bubbles would maintain character consistency, as well as introducing a structure that would readily incorporate new information.

What problem does this paper attempt to address?

This paper aims to solve several key problems faced by large - language models (LLMs) when generating conversations: 1. **Poor content**: Responses generated by LLMs without additional information may lack valuable content. 2. **Hallucination phenomenon**: LLMs may generate information that does not conform to the facts, the so - called "hallucination". 3. **Lack of emotional ability**: LLMs perform poorly in processing emotional information and have difficulty understanding or expressing emotions. 4. **Poor role consistency**: LLMs may not be able to maintain role consistency when generating conversations, resulting in incoherent role behaviors. To address these problems, the author proposes three schemes using knowledge graphs (Knowledge Graphs, KGs): 1. **Dynamic knowledge graph embedding and recommendation**: - **Dynamic knowledge graph embedding (Dynamic Knowledge Graph Embedding, DKGE)**: By dynamically updating the knowledge graph embedding, LLMs can obtain new information in real - time, reducing the hallucination phenomenon caused by outdated information. - **Recommendation system**: Use the information in the knowledge graph to recommend relevant knowledge and improve the relevance and quality of the generated responses. 2. **Emotional characteristics of entities**: - **Emotional values as features**: Add emotional values (such as Valence, Arousal, Dominance, VAD scores) to entities in the knowledge graph, so that the generated text is more in line with the emotional tendency of user input. - **Emotional alignment**: By comparing the emotional values stored in the user input and the knowledge graph, generate more emotionally - aligned responses and improve the user's perceived emotional ability. 3. **Role dimension based on novel experience**: - **Narrative bubble structure**: Extract role information from novels and organize it into "narrative bubbles", each bubble contains the character's dialogue, facts and summary, in order to maintain the role's consistency and personal experience. - **Dynamic update**: Allow the dynamic addition of new "narrative bubbles" during the interaction process without retraining the model, thereby better maintaining the role's consistency and richness. Through these schemes, the paper hopes to improve the quality of LLMs - generated conversations, making them more natural, emotionally rich and role - consistent.

Information for Conversation Generation: Proposals Utilising Knowledge Graphs

Knowledge Graphs as Context Sources for LLM-Based Explanations of Learning Recommendations

Can LLMs be Good Graph Judger for Knowledge Graph Construction?

AGENTiGraph: An Interactive Knowledge Graph Platform for LLM-based Chatbots Utilizing Private Data

Knowledge Graph for NLG in the context of conversational agents

EARL: Informative Knowledge-Grounded Conversation Generation with Entity-Agnostic Representation Learning

Response Generation in Longitudinal Dialogues: Which Knowledge Representation Helps?

Exploring In-Context Learning for Knowledge Grounded Dialog Generation

Assessing LLMs Suitability for Knowledge Graph Completion

A Preliminary Roadmap for LLMs as Assistants in Exploring, Analyzing, and Visualizing Knowledge Graphs

Knowledge Graphs, Large Language Models, and Hallucinations: An NLP Perspective

GKA-GPT: Graphical Knowledge Aggregation for Multiturn Dialog Generation

Neural response generation for task completion using conversational knowledge graph

GraphEval: A Knowledge-Graph Based LLM Hallucination Evaluation Framework

Graphologue: Exploring Large Language Model Responses with Interactive Diagrams

Combining LLMs and Knowledge Graphs to Reduce Hallucinations in Question Answering

ChatGPT is not Enough: Enhancing Large Language Models with Knowledge Graphs for Fact-aware Language Modeling

LLM-assisted Knowledge Graph Engineering: Experiments with ChatGPT

Logic Augmented Generation

Generate-on-Graph: Treat LLM as both Agent and KG in Incomplete Knowledge Graph Question Answering

Supporting Student Decisions on Learning Recommendations: An LLM-Based Chatbot with Knowledge Graph Contextualization for Conversational Explainability and Mentoring