Home Products & Services

Heavy information extraction significant progress, MIT use intensive learning from the external netw www.av7788.com

Heavy | information extraction using reinforcement learning MIT significant progress, pull data from external network technology – Sohu selected from MIT Author: Larry Hardesty machine heart compiler participation: empirical methods, Natural Language Processing conference hosted by Du Xiade Jiang Siyuan in the last week to the Association for Computational Linguistics (EMNLP), computer science and artificial intelligence research laboratory are from MIT by virtue of a subversion of the traditional machine learning information extraction method won the best paper award. Click here to download this article. There are a lot of valuable information on the Internet is open, mostly in the form of text. The data needed to answer countless problems — including, the association between industrial use of the specific chemical substances and disease events, association between – or news reports and voting results — maybe all over the internet. However, it is very time-consuming to extract and organize the data from the plain text and analyze it. The extraction of information, or the automatic classification of data items, is a major topic in the study of artificial intelligence (YISHION). The empirical method of Natural Language Processing last week to host the Association for Computational Linguistics (EMNLP), computer science and artificial intelligence research laboratory from MIT who by virtue of a subversion of the traditional machine learning information extraction method won the best paper award. Most machine learning systems rely on the combination of training samples and the corresponding classification provided by human annotations. For example, humans may make a set of part of speech in the text of the label, the machine learning system will try to solve the ambiguity recognition mode — for example, when "her" is a direct object and when "her" is an adjective. In general, computer scientists will try to train the machine learning system with as much data as possible. This is usually more likely to get a system that can handle a difficult problem. In contrast, the MIT researchers trained the system in the absence of data because they had all the data available in the case they were investigating. They found that the problem of limited information is easy to solve. "In information extraction, usually in Natural Language Processing, you have an article, you need to do anything to extract the correct content from the things on this article," another author Regina Barzilay of the paper said. "It’s different from what you or I would do. When you read an article that you don’t understand, you will find a piece of information that you can understand. A machine learning system will probably assign a confidence score for each classifier, which is a statistical measure of the likelihood that the classification is correct. The researchers used the new system, if the confidence score is too low, the system automatically generates a search query, and then extract the relevant data from a text of these new texts, then reconcile the results with the initial extraction of content. If the confidence level is still low, it will move to the next search character n相关的主题文章: