Automatic Analysis of Sentiment Changes in News Texts with Emergence of New Information

Internet allows modern people to get quick answers to almost any questions. However, in some cases people still have to analyze large volumes of information to get a general idea of the subject. One such case is analysis of information available on upcoming products, such as digital devices, which a person might consider buying. At the same time, people are often interested in comments of other readers. All this leads to the task of automatic monitoring of news sources to extract relevant information about such products, and analysis of people's opinion about them. This work presents an approach for building such a system. We use topic modeling to extract new information about the products from news posts. To extract the mentioned products and their attributes we use a set of rules with a dictionary and a taxonomy of product attributes. Articles with new information are then grouped based on their headings' similarity. From the comments of the readers for each group, we extract attributes of the mentioned products, after which sentiment analysis is performed to get the opinion of people about the product. We then build a visualization of all the collected data. The resulting system allows people to get the general idea of the available pieces of information about the product and see the reaction of other readers to the appearance of each piece of information.