Combining Machine Learning and Knowledge-Based Systems for Summarizing Interviews

摘要

Achieving optimal results of an automatic summarization process is frequently conditioned by the knowledge of the domain. The performance of general methods is always lower than what can be achieved by introducing custom modifications taking into account the context. Nevertheless, these type of custom adjustments represents a hard work by experts and developers, which is not always possible to achieve due to the high costs. In this work we aim to leverage the features of the documents in order to classify them by using machine learning methods. Once the typology is identified, the application of improvements is done by a knowledge-based system that allows users to easily customize both the summarization process, and the presentation to the final user. The proposed method has been applied with promising results to interviews in a real environment of a major Spanish media group.