tailieunhanh - Báo cáo khoa học: "A Graphical Environment for Graph-Based NLP"

This interactive presentation describes LexNet, a graphical environment for graph-based NLP developed at the University of Michigan. LexNet includes LexRank (for text summarization), biased LexRank (for passage retrieval), and TUMBL (for binary classification). All tools in the collection are based on random walks on lexical graphs, that is graphs where different NLP objects (., sentences or phrases) are represented as nodes linked by edges proportional to the lexical similarity between the two nodes. We will demonstrate these tools on a variety of NLP tasks including summarization, question answering, and prepositional phrase attachment. costly. . | LexNet A Graphical Environment for Graph-Based NLP Dragomir R. Radev1 2 Giines Erkan1 Anthony Fader3 Patrick Jordan Siwei Shen and James P Sweeney2 Department of Electrical Engineering and Computer Science School of Information Department of Mathematics University of Michigan Ann Arbor MI 48109 radev gerkan afader prjordan shens jpsweeney @ Abstract This interactive presentation describes LexNet a graphical environment for graph-based NLP developed at the University of Michigan. LexNet includes LexRank for text summarization biased LexRank for passage retrieval and TUMBL for binary classification . All tools in the collection are based on random walks on lexical graphs that is graphs where different NLP objects . sentences or phrases are represented as nodes linked by edges proportional to the lexical similarity between the two nodes. We will demonstrate these tools on a variety of NLP tasks including summarization question answering and prepositional phrase attachment. 1 Introduction We will present a series of graph-based tools for a variety of NLP tasks such as text summarization passage retrieval prepositional phrase attachment and binary classification in general. Recently proposed graph-based methods Szummer and Jaakkola 2001 Zhu and Ghahra-mani 2002b Zhu and Ghahramani 2002a Toutanova et al. 2004 are particularly well suited for transductive learning Vapnik 1998 Joachims 1999 . Transductive learning is based on the idea Vapnik 1998 that instead of splitting a learning problem into two possibly harder problems namely induction and deduction one can build a model that covers both labeled and unlabeled data. Unlabeled data are abundant as well as significantly cheaper than labeled data in a variety of natural language applications. Parsing and machine translation both offer examples of this relationship with unparsed text from the Web and untranslated texts being computationally less costly. These can then be used to supplement manually translated .