Feature-based News Summarization System exploiting Real-time Search Keywords and Paragraph Vector

Abstract

As the dissemination of news through the Internet becomes more active, the number of users who access news articles through online news, rather than newspapers, is rapidly increasing. Behind the growth of online news, there is a debate on how to effectively provide news articles that contain important information while attracting public attention due to the vast amount of news articles available on the Internet. In this paper, we use real-time search keyword provided by the portal site to identify topic that the public is interested in. The real-time search keyword chart is a ranking chart of the topics that are currently attracting interest based on the number of searches per portal site. According to search engine share statistics, the keywords appearing on the real-time search keyword charts of the top 2 portal sites that are heavily used can be seen as the main topic of public interest. Using this, The score of each sentence is calculated considering the TextRank algorithm and the general features of the news articles in the set of news articles searched for each keywords, and the final score is calculated by taking each weight into account. We arrange the scores in descending order and generate the top k sentences as summary and present them to users. Finally, we evaluated the performance of the proposed system through ROUGE value comparison experiments in all three experiments and compared it with the results of the existing system.