W. Cohen and Y. Singer. Learning to query the web. In AAAI Workshop on Internet-Based Information Systems, 1996.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....viewed as a traditional classification problem. We explore a number of machine learning techniques [8, 18] in the design of other variants of our system (Section 2.4. 2) Several techniques use supervised learning to devise queries that match documents about a specific category of interest [15] [7] constructed topic specific directories on the web by training a classifier with a labeled set of documents and then deriving queries to retrieve additional documents. Flake et al. 11] extracted category specific query modifications from a non linear SVM classifier. Recently, Ghani et al. 14] ....

W. Cohen and Y. Singer. Learning to query the web. In Proceedings of the AAAI Workshop on Internet-Based Information Systems, 1996.

....important long term research direction involves closing information integration loop. We have examined information extraction in isola 200 tion, but our techniques must be integrated with work on resource discovery [Bowman et al. 94, Zaiane Jiawei 95] learning to query information resources [Cohen Singer 96, Doorenbos et al. 97] and learning semantic models of information resources [Perkowitz Etzioni 95, Tejada et al. 96] Third, our work has focused on resources whose content is formatted by html tags. Let us emphasize that our techniques do not depend on html or any other particular formatting ....

Cohen, W. and Singer, W. Learning to Query the Web. In Proc. Workshop Internet-based Information Systems, 13th Nat. Conf. Artificial Intelligence, pages 16--25, 1996.

....Web pages test, although SE 10 achieved best accuracy, it was only slightly ( 0. 5 ) better then the 8 By inspection of these top scoring features it seems that their conjunction could be served as a basis for a sharper query (to search engines) for the concept being learned (cf. Cohen and Singer [7]) 9 results for SE 1. On the other hand, SE 1 is much better from eciency considerations (in speed and memory) Our results indicate that the data sets we used are qualitatively di erent. It seems that the authorship tests versus the Web pages test, represent essentially di erent tasks, while ....

W.W. Cohen and Y. Singer. Learning to query the web. In AAAI Workshop on Internet-Based Information Systems, 1996. Available online: http://www.research.att.com/ singer/publications.html.

....database. Both our techniques rely on basic predicates for links and or pages; boolean combinations of these predicates are used both in the queries and on schema edges. Techniques for automatically classifying Web pages or other semi structured documents are the subject of current research [TS93, CS96] Systems based on knowledge representation (KR) techniques [Guh97] classify pages in hierarchies of catagories and can support complex queries over catagories; this technique augments, but is orthogonal to, the information contained in a Web site s structure. Our techniques could rely on KR ....

William Cohen and Yoram Singer. Learning to query the Web. In Proceedings AAAI Workshop on Internet-Based Information Systems, 1996.

....constructed with little effort. Because the Web presents new opportunities for acquiring user feedback, much of the work has gone toward modeling user interests through direct interaction (e.g. 1] 8] Similarly, details of Web organization can be treated as classifications for learning, as in [2]. The use of ontologies as a means to facilitate automatic processing of Web data has also received attention. SHOE [7] is a proposed extension to HTML with which page designers can annotate their pages, associating them in ontologic structures, thereby enabling indexing systems to respond to ....

Willian W. Cohen and Yoram Singer. Learning to query the Web. In Proceedings of the AAAI-96 Workshop on Internet-Based Information Systems, 1996.

....are represented by lists of keywords or by feature vectors, each feature indicating the occurrence of a particular word within the text. A comparison between different machine learning techniques to be used in this task is presented in [Armstrong et al. 1995] A related approach is followed in [Cohen and Singer, 1996], in which the system learns to query the Web, that is it learns how to use a search tool (which keywords have to be used) to retrieve important information for the user. We do not consider systems whose main goal is an automatic classification of documents. They usually represent documents as a ....

....text of a subject category, constructing personalized information filters. Such a representation is used, however, for special class of textual documents such as e mail, Usenet news and Web pages and machine learning techniques have been proposed to learn rules that classify them (see for example [Cohen, 1996, Bloedorn et al. 1996, Goan et al. 1996] Hunters differ from surfers in building a (virtual) common model of the relevant information space. They act as spiders through the Web gathering information for the user. Many special purpose agents (we call them information brokers) have been ....

Cohen, W. W. and Singer, Y. (1996). Learning to query the web. In AAAI Workshop on Internet-Based Information Systems.

....Wiener et al. 1995; Cohen, 1995b] In this paper, we will investigate the performance of two recently implemented machine learning algorithms on a number of large text categorization problems. The two algorithms considered are set valued Ripper, a recent rule learning algorithm [Cohen, 1995a; Cohen, 1996b] and sleeping experts, a new on line learning method [Freund et al. 1997] These algorithms share several features that make them attractive for large text categorization problems. First, both algorithms are efficient on large, noisy corpora, running in linear or nearly linear time. Second, ....

....some properties that make them useful in certain contexts. If a ruleset is compact, it is relatively easy for people to understand; this may make it easier for users to accept a learned classifier as being reasonable. Rulesets can also be easily converted to queries for a boolean search engine [Cohen and Singer, 1996b] There are a number of subtleties involved in learning rulesets. In particular, relatively straightforward greedy algorithms often give rulesets with unnecessarily high error rates; furthermore, many algorithms that do find accurate rulesets tend to be relatively inefficient for large noisy ....

[Article contains additional citation context not shown here]

William W. Cohen and Yoram Singer. Learning to query the Web. In Proceedings of AAAI-96 Workshop on Internet-Based Information Systems, 1996.

....Wiener et al. 1995; Cohen, 1995b] In this paper, we will investigate the performance of two recently implemented machine learning algorithms on a number of large text categorization problems. The two algorithms considered are set valued Ripper, a recent rule learning algorithm [Cohen, 1995a; Cohen, 1996], and sleeping experts, a new on line learning method. These algorithms share several features that make them attractive for large text categorization problems. First, both algorithms are efficient on large, noisy corpora, running in linear or nearly linear time. Second, both algorithms use what ....

....some properties that make them useful in certain contexts. If a ruleset is compact, it is relatively easy for people to understand; this may make it easier for users to accept a learned classifier as being reasonable. Rulesets can also be easily converted to queries for a boolean search engine [Cohen and Singer, 1996]. Ripper builds a ruleset by repeatedly adding rules to an empty ruleset until all positive examples are covered. Rules are formed by first splitting the training data into two sets, a growing set and a pruning set , and then greedily adding conditions to the antecedent of a rule with an empty ....

[Article contains additional citation context not shown here]

William W. Cohen and Yoram Singer. Learning to query the Web. In Proceedings of AAAI-96 Workshop on Internet-Based Information Systems, 1996.

....Wiener et al. 1995; Cohen, 1995b] In this paper, we will investigate the performance of two recently implemented machine learning algorithms on a number of large text categorization problems. The two algorithms considered are set valued Ripper, a recent rule learning algorithm [Cohen, 1995a; Cohen, 1996], and a new on line learning method, modified for text categorization, sleeping experts. These algorithms share several features that make them attractive for large text categorization problems. First, both algorithms are efficient on large, noisy corpora, running in linear or nearly linear time. ....

....that make them useful in certain contexts. If a ruleset is compact, it is relatively easy for people to understand; this may make it easier for users to accept a learned classifier as being reasonable. Rulesets can also be easily converted to a set of queries for a simple boolean search engine [Cohen and Singer, 1996]. Ripper builds a ruleset by repeatedly adding rules to an empty ruleset until all positive examples are covered. Rules are formed by first, splitting the training data into two sets, a growing set and a pruning set , and then greedily adding conditions to the antecedent of a rule with an ....

[Article contains additional citation context not shown here]

William W. Cohen and Yoram Singer. Learning to query the Web. Submitted to ML-96, 1996.

No context found.

W. Cohen and Y. Singer. Learning to query the web. In AAAI Workshop on Internet-Based Information Systems, 1996.

No context found.

W. Cohen and Y. Singer. Learning to query the web. In AAAI Workshop on Internet-Based Information Systems, 1996.

No context found.

W. Cohen and Y. Singer. Learning to query the web. In AAAI Workshop on Internet-Based Information Systems, 1996.

No context found.

CS96. W. W. Cohen and Y. Singer. Learning to query the web. In AAAI Workshop on Internet-Based Information System, 1996.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC