Evaluation of Text Summaries Based on Linear Optimization of Content Metrics

In the context of Automatic Text Summarization, the Evaluation of Text Summaries (ETS) represents a challenging task because it requires certain “intelligence” in the process. For human judgments are expensive and time consuming, the automatic evaluation has been adopted to examine the content of text summaries.
To solve the problem that existing automated methods do not strongly correlate enough concerning human judgments, this book addresses the before-mentioned issues, providing a comprehensive discussion and new insights about linear optimization of content metrics to improve the correlation between automatic and human evaluations. It represents four years of arduous research by the authors.
The readers are expected to learn about, through reading this book, the background and fundamentals of the ETS, the state-of-the-art evaluation methods, experimental evidence of how linear optimization has improved the performance of Natural Language Processing tasks, the stages of a new methodology based on linear optimization of metrics for the ETS and the experiments and results of how linear optimization improves the evaluation of single-and multi-document summaries.