Back to top

Meta Model based Natural Language Generation for Automatic Abstractive Text Summarization (A-SUM)

Last modified Feb 13, 2020

More than 5000 years after its invention, written language is still the most important medium to document and communicate knowledge. While the production of texts is simplified and accelerated by word processing software, template systems and other technologies, the consumption of texts is still a comparably little by technology supported process.

The current state-of-the-art in automatic text summarization are mostly so called extractive methods, which extract the n most important sentences of a text by the means of metrics like TF/IDF. These summaries consist of mostly incoherent sentences.

The goal of the BMBF founded Softwarecampus Project "Meta Model based Natural Language Generation for Automatic Abstractive Text Summarization" (A-SUM) is the creation of coherent, personalized automatic abstractive summaries. Instead of just extracting sentences, A-SUM aims to extract facts from texts and store them in an intermediate, meta model based, structured format. Based on user preferences and contextual information, this information can be personalized.

By applying Natural Language Generation techniques, we aim to transform the structured representation of information back to a coherent text, which gives the recipient a quick and tailored overview of the content of the original text.