Multidocument Summarization

A lot of information appears in social networks every day: news, user opinions about world events, comments from millions of people and organizations. Currently, there is an increasing need for a service that summarizes information without unnecessary noise and presents it in a readable form. In my work, I study the problem of summarizing consecutive events, such as the Australian fires in early 2020, based on tweets. This paper presents a method for creating a list of key news facts, which describe the selected event, with the dates of their occurrence. The construction algorithm consisted of four stages: collecting data from Twitter using keywords and phrases, text processing, clustering tweets or sentences by time intervals and meaning and extracting summarization. Several different methods were tested, some of which showed a high quality of resulting summary — readable text, a full statement of facts and a structured list. As further ways of development, I suggested several directions — improving the automation of the process for developing a software product, considering other parameters for analysis, for example, a list of keywords or splitting into clusters.