Abstract:Mind-map generation aims to process a document into a hierarchical structure to show its central idea and branches. Such a manner is more conducive to understanding the logic and semantics of the document than plain text. Recently, a state-of-the-art method encodes the sentences of a document sequentially and converts them to a relation graph via sequence-to-graph. Though this method is efficient to generate mind-maps in parallel, its mechanism focuses more on sequential features while hardly capturing structural information. Moreover, it's difficult to model long-range semantic relations. In this work, we propose a coreference-guided mind-map generation network (CMGN) to incorporate external structure knowledge. Specifically, we construct a coreference graph based on the coreference semantic relationship to introduce the graph structure information. Then we employ a coreference graph encoder to mine the potential governing relations between sentences. In order to exclude noise and better utilize the information of the coreference graph, we adopt a graph enhancement module in a contrastive learning manner. Experimental results demonstrate that our model outperforms all the existing methods. The case study further proves that our model can more accurately and concisely reveal the structure and semantics of a document. Code and data are available at <a class="link-external link-https" href="https://github.com/Cyno2232/CMGN" rel="external noopener nofollow">this https URL</a>.

What problem does this paper attempt to address?

The problems that this paper attempts to solve are the two main problems existing in the existing methods when generating mind maps: 1. **Insufficient capture of structural information**: Although the existing sequence - based methods can efficiently generate mind maps in parallel, their mechanisms focus more on sequence features and it is difficult to capture the structural information of documents. This leads to the generated mind maps may not accurately reflect the logical and semantic relationships of the document. 2. **Difficulty in modeling long - distance semantic relationships**: As the length of the document increases, the existing methods become more and more difficult in modeling long - distance semantic relationships. This makes the generated mind maps less effective when dealing with long documents. To solve these problems, the author proposes a Coreference - guided Mind Map Generation Network (CMGN). It introduces explicit graph - structure information by constructing a coreference graph and uses a graph - enhancement module to improve the quality of graph representation. Specifically: - **Coreference graph construction**: By identifying the coreference relationships in the document, a coreference graph is constructed to explicitly introduce graph - structure information. - **Coreference graph encoder**: A Graph Convolutional Network (GCN) encoder is used to further explore the potential semantic relationships between sentences, especially long - distance sentence pairs. - **Graph - enhancement module**: A graph - enhancement module based on Graph Contrastive Learning (GCL) is used to eliminate noise and make better use of the information of the coreference graph. Through these methods, CMGN can more accurately capture the structural and semantic information of the document when generating mind maps, especially when dealing with long documents. The experimental results show that CMGN is superior to the existing methods in multiple evaluation indicators.

Coreference Graph Guidance for Mind-Map Generation

Enhancing Structure Preservation in Coreference Resolution by Constrained Graph Encoding

MindMap: Knowledge Graph Prompting Sparks Graph of Thoughts in Large Language Models

Scene Graph Generation Via Multi-Relation Classification and Cross-Modal Attention Coordinator.

Knowledge Enhanced Coreference Resolution Via Gated Attention.

CoreGen: Contextualized Code Representation Learning for Commit Message Generation

Scene Graph Generation With Hierarchical Context

Enhancing Scene Graph Generation with Hierarchical Relationships and Commonsense Knowledge

Graph-based Multi-hop Reasoning for Long Text Generation.

Learning Hierarchical Document Graphs From Multilevel Sentence Relations

Building the Directed Semantic Graph for Coherent Long Text Generation

Local-to-global GCN with knowledge-aware representation for distantly supervised relation extraction

Text Pared into Scene Graph for Diverse Image Generation.

Interconnected Question Generation with Coreference Alignment and Conversation Flow Modeling

Capturing High-Level Semantic Correlations via Graph for Multimodal Sentiment Analysis

Enhancing Structure Modeling for Relation Extraction with Fine-Grained Gating and Co-Attention.

Multi-Modal Coreference Resolution with the Correlation between Space Structures

Improving Coreference Resolution by Leveraging Entity-Centric Features with Graph Neural Networks and Second-order Inference

Rethinking Relation Classification with Graph Meaning Representations

CorefQA: Coreference Resolution As Query-based Span Prediction.

A Graph Sequence Neural Architecture for Code Completion with Semantic Structure Features