Learn2Sum: A New Approach to Unsupervised Text Summarization Based on Topic Modeling

Abstract

Due to the enormous volume of data on the web, it is hard for the user to retrieve effective and useful information within the right time. Thus, it has become a need to generate a brief summary from a large amount of textual data according to the user profile. In this context, text summarization is used to identify important information within text documents. It aims to generate shorter versions of the source text, by including only the relevant and salient information. In recent years, the research on summarization techniques based on topic modeling techniques has become a hot topic among researchers thanks to their ability to classify, understand a large text corpora and extract important topics on the text. However, existing studies do not provide the support of personalization when generating summaries because they need to know not only which documents are most helpful to the users, but also which topics and keywords are more or less related to the user' interests. Thus, existing studies lack of the support of adaptive user modeling for user applications in the emerging areas of automatic summarization, topic modeling and visualization. In this context, we propose a new approach of automated text summarization based on topic modeling techniques and taking into account the user's profile which helps to semantically extract relevant topics of textual documents, summarizing information according to the user' topics interests and finally visualize them through a hyper-graph Experiments have been conducted to measure the effectiveness of our solution compared to existing summarizing approaches based on text content. The results show the superiority of our approach.