Researchers from the University of Twente, a method for reading comprehension of text by computers in the control room of the future’. By this Twitter monitoring, emergency services faster behind that something is going on. In the next few months tests.
The use of the software is part of the project TEC4SE. In this initiative to put government organizations, businesses, and educational institutions in the region Twente their knowledge to people, hardware, information and services connect to each other via a network. Goal is to make the flow of information for the emergency services to improve.
The University of Twente puts this knowledge in to computers better comprehension to read. “We apply the software in the control room of the fire brigade and the police in the region Twente”, explains Maurice van Keulen, associate professor Data Management Technology at the University of Twente. The police is particularly interested in the uses for Twitter as a source of information, ” he tells Tweakers: “In large-scale events such as football matches, want the emergency services to provide as soon as possible if something happens and what is going on.” Of Cologne, mentions as an example a defined as: “On a channel like Twitter, you can see what people there say about it.
The software of the UT may use the information that institutions want to have automated from tweets, not only on the basis of the content, but also on the basis of hashtags and the author. In the first instance will need to manually a selection on the basis of, for example, hashtags need to be made; the analysis is on the flow of messages released, which, for example, also the location on the basis of coordinates are taken.
“In Enschede, you have, for example, a brasserie that The Hangover is called. If there what happens it is useful if the emergency services know that it is in tweets to that cafe and not a cat, and that they are directly over the location. “We are working to improve the system,” says the professor, “such as the ability to analyze all tweets in a certain region.” In the coming months to start testing in which the software applied in practice.
The existing techniques for reading comprehension make use of a superficial analysis of words, where a lot of text needed to learn. The method of the UT can also work efficiently if there is only little text is present, such as in tweets, and in addition, she works not only for English texts, but for all languages.
The technique makes use of the recognition of so-called ‘named entities‘. This can be the names of such persons, places or organizations are. The meaning depends on the context in which the name is used. The method of UT, let the computer not only to recognize which part of the text is a named entity, but analyze the context to determine what the entity is meant.
The operation proceeds in a few stages. The first thing the software on the basis of the piece of text all of the possible candidates for named entities, places Of Cologne. “The difference with other methods is that we in the first instance, not to the best candidates, but to possible candidates.” That collection is used to create a large database with entities to consult, where matches be enriched with more candidates. “The enormous amount of information generated is machine learning released to see which possibilities fit together,” says Cologne. The system can discern when an ambiguous word such as ‘you’ as a named entity should be considered, and when it is not.
Cologne gives as examples the named entity ‘Paris Hilton’ :”Is this a piece of text a hotel in Paris, the celebrity or the perfume? Another example is the ‘rijksmuseum’. “It depends on the context in which the rijksmuseum is meant. This may have to do with the author, the subject of discussion, what before or after it is said and sometimes even with the location or the time. Attends the sender in Enschede, he’s referring most likely to the rijksmuseum in Enschede. But it can also be about one of the many other national museums in the Netherlands.”
Comments
(73)