TCS Framework: Generating Text Summaries of Specified Style Using Classifier and Summarizer

There have been a lot of excellent researches in the field of natural language processing for text summarization tasks. However, there is almost no research on automatic text summarization with a specified style. Usually, people only need a piece of text in the style of content we want in many cases. Therefore, we propose an extended task of the traditional text summarization task: the task of generating text summaries with a specified style. We designed the TCS framework to accomplish this task. The TCS framework mainly includes classifiers, summarizers and data annotation processing rules. We trained the classifier by fine-tuning ALBERT. Fine-tune the T5 model to train the summarizer. By comparing the performance of T5 and other summary models, the PEGASUS model is also used as one of our classifiers. We randomly used a small number of samples in the XSUM dataset to verify the TCS framework. The results show that the TCS framework can generate text summaries with specified styles.