Research on Generative Text Summarization Fusing Multidimensional Semantic Information
FengXia Jiang, X. Y. Zhou · 2023
Automatic text summarization is a method to streamline text content into essential information to promote quick comprehension of crucial information. Aiming at the problems of the seq2seq model in current generative summarization methods that cannot fully understand the context and the generated summary is inconsistent with the original topic, this paper proposes an automatic text summarization method incorporating multidimensional semantic information. The model is based on the pointer generation network, extracting the global semantic information with the BiLSTM encoder and extracting local semantic information with the local convolution extractor to enhance the fine-grained features of the original text. Meanwhile, to enhance the topic semantic features of the original text, the BERTopic model is utilized to collect topic words. Then the attention mechanism is used to integrate the topic information into the pointer generation network. In the training stage, we design a hybrid learning goal for optimization to avoid the exposure bias problem using the cross-entropy loss function and the reinforcement learning loss function. The experimental results show that the method significantly improves ROUGE evaluation metrics on the publicly available dataset LCSTS.