Abstract:Driven by advancements in Artificial Intelligence (AI) techniques, significant progress has been made in the improvement of Natural Language Processing (NLP). One area of particular interest within NLP is the extraction of events from textual data. While studies on the event extraction has traditionally concentrated on the sentence level, the increasing real-world implementation of event extraction has significantly raised the demand for document-level event extraction. Two common challenges in the research of document-level event extraction are event arguments scattering across different sentences and more than one event existing in one input document. Additionally, we recognized that the class imbalance problem exists in the benchmark dataset. A novel deep-learning approach for document-level Chinese financial event extraction by using axial attention and adaptive focal loss is proposed in this paper. Previous studies showed that relation information among entities is critical to solving the argument scattering issue and multi-event issue since both of the issues are due to a complex semantic understanding of input text. Our method leverages up-to-date axial attention mechanisms to capture relations among entities and found that the benchmark dataset does not include enough argument-scattering data to cover all of the five event types. To solve the problem of class imbalance, we leverage an adaptive focal loss function that adjusts the important weights of samples during training dynamically, enabling the model to focus more on minority class instances and improve overall performance. In conclusion, our methodology leverages axial attention and adaptive focal loss and solves the problems of multi-event occurrences, cross-sentence relationships among event arguments, and class imbalance. We conducted extensive experiments on a large-scale Chinese financial event extraction dataset, demonstrating that our proposed approach outperforms baseline methods.

A Deep Learning Approach for Document-level Chinese Financial Event Extraction

DCFEE: A Document-level Chinese Financial Event Extraction System based on Automatically Labeled Training Data

Doc2EDAG: An End-to-End Document-level Framework for Chinese Financial Event Extraction

Knowledge Graph Enhanced Event Extraction in Financial Documents

Advancing document-level event extraction: Integration across texts and reciprocal feedback

DocEE-zh: A Fine-grained Benchmark for Chinese Document-level Event Extraction

CFERE: Multi-type Chinese financial event relation extraction

A Practical Approach to Chinese Emergency Event Extraction Using BiLSTM-CRF

Chinese Event Extraction Using DeepNeural Network with Word Embedding

A Survey on Deep Learning Event Extraction: Approaches and Applications

New Word Extraction from Chinese Financial Documents.

DocEE: A Large-Scale and Fine-grained Benchmark for Document-level Event Extraction

DEERE: Document-Level Event Extraction as Relation Extraction

A Convolution BiLSTM Neural Network Model for Chinese Event Extraction

Enabling and Analyzing How to Efficiently Extract Information from Hybrid Long Documents with LLMs

A deep learning approach of financial distress recognition combining text

The advance of event relation extraction based on deep learning

Event Extraction Via Dynamic Multi-Pooling Convolutional Neural Networks.

Document-Level Event Role Filler Extraction using Multi-Granularity Contextualized Encoding

A Novel Joint Framework for Multiple Chinese Events Extraction

Beyond Single-Event Extraction: Towards Efficient Document-Level Multi-Event Argument Extraction