Research On Key Technology Of News Reduction For New Media

With the rapid development of new media, online news shows a tendency of spring, people are faced with difficulties in information selection, information overload and other issues. A large amount of redundant information on the one hand increase the time of reading information, on the other hand can also increase the difficulties of finding and understanding hot topics.The popularity of mobile devices is driven to produce a large number of news clients, the characteristic of these mobile devices is small screens, while large amount of information can’t display in a single screen, and split to other screen will cause inconvenience to read. Therefore, automatic news reduction technology emerges naturally, news reduction can gain not only simple and clear, depth and attitude of content resources, and can promote the quality of the network news, realizes the network news value-added reading.This paper launches a series of studies on the key technology of news headline compression and news reduction. In view of the news headlines, by means of word alignment the method based on the combination of rules and statistics is applied to compress. For the news text we propose the framework of sentence compression and sentences selection method to generate news text reduction results. In this framework, the paper firstly applies multi-feature fusion for summary sentence pre-selection, which is produced by extracting the most representative sentences based on sentence weight of multi-feature. Then for the pre-selected summary sentences, this paper puts forward a kind of heuristic sentence compression algorithm based on keywords, sentence compression technology is applied to generate the intermediate layer of source sentence compression variants. Finally the method removes redundancy by integer linear programming algorithm to select the most informative and the final news summary.In terms of evaluation, the paper presents a summarization evaluation method on the basis of sentence compression. The evaluation method through importance of information, grammaticality and compression ratio to score the compressed sentence, then combines the compressed sentence scores with the summary estimating to calculate accuracy, recall and F-value are 79.26%, 76.32% and 77.76% respectively. The experimental results show that the news reduction algorithm proposed in this paper is feasible, which can automatically generate a coherent and symbolic high-density news summary for the given news articles.