Intrinsic methods for evaluating information completeness of summaries in task of automatic text summarization

The article presents a classification of existing methods for assessing the quality of summaries in the task of automatic text summarization. A formal formulation of the problem of automatic text summarization is presented. The concept of summary quality is considered. The intrinsic methods of evaluating the information completeness of the summary, such as the metrics of the ROUGE group, cosine similarity, Kullback-Leibler distance, Jensen-Shannon distance, are considered in detail. The advantages and disadvantages of the considered intrinsic methods and methods based on expert assessment are presented.