W. Cohen.: Learning Trees and Rules with Set-valued Features. In. Proceedings of the Thirteenth National Conference on Artificial Intelligence. AAAI Press, (1996)  Home/Search   Document Details and Download   Summary   Related Articles   Check

....but that fails to articulate how the words are actually used by the model to generate a prediction. I introduce a new technique that of explaining an otherwise opaque or hardto understand model to address this problem. To do so, I make use of the Ripper inductive rule learning system [Coh95, Coh96] to generate and present to the user a set of rules that are human interpretable and representative of the opaque model. The rules are of the the form: if apple and taste and delicious appear in the information item, then the information is interesting . This technique is similar to ....

....This technique is important as it allows any text based method to use numerical features and can thus be directly applied to existing text classification systems without changing algorithms or framework. So far the few algorithms that can easily deal with text and numbers (e.g. Ripper [Coh95, Coh96] Slipper [CS99] C4.5 [Qui93] and Support Vector Machines (SVMs) Vap95] are much slower than many text classifiers, or do not natively handle free text very well while Ripper and Slipper were designed to handle text natively, C4.5 and SVMs do not handle text as easily. My new technique ....

[Article contains additional citation context not shown here]

William Weston Cohen. Learning trees and rules with set--valued features. In Proceedings of the National Conference on Artificial Intelligence, pages 709--716, 1996.

....short query probes generated from document classifiers and has two steps: a training step and a sampling step. In the training step, we start with a comprehensive, predefined topic hierarchy with an associated training set of preclassified documents. Then, we train a rule based document classifier [5] to produce rules like IF lung AND cancer THEN Health . According to this rule, a document having the words cancer and lung will be classified into category Health . In the sampling step, we transform each of these rules into a query probe (a query containing all the words in the antecedent ....

William W. Cohen. Learning trees and rules with set-valued features. In Proceedings of the 13th National Conference on Artificial Intelligence (AAAI-96) Eighth Conference on Innovative Applications of Artificial Intelligence (IAAI-96), pages 709--716, 1996.

....covering) rule learning algorithms [33] In a post processing phase, Ripper uses a global optimization algorithm that allows it to re evaluate and modify the learned rules in the context of the other rules. Ripper is able to deal with text categorization problems via the use set valued features [18]. Basically, Ripper represents a document as a set of words, i.e. it represents a document by the set of words that occur in the document, disregarding their order and their frequencies. For rule learning algorithms, this representation seems to come more natural than the commonly used ....

William W. Cohen. Learning trees and rules with set-valued features. In Proceedings of the 13th National Conference on Artificial Intelligene (AAAI-96), pages 709--716. AAAI Press, 1996.

....per se (Section 3.3) 3.1 Training a Document Classifier Our technique for classifying databases over a set of categories C1 , Ck starts by training a rule based document classifier over those categories. We use RIPPER, an o# the shelf tool developed at AT T Research Laboratories[2, 3]. Given a set of training, pre classified documents, this tool returns a classifier that might consist of rules like the following: Computers IF mac Computers IF graphics windows Religion IF god christian Hobbies IF baseball The first rule indicates that if a document contains the term mac it ....

William W. Cohen. Learning trees and rules with set-valued features. In Proceedings of the 13th National Conference on Artificial Intelligence and Eighth Innovative Applications of Artificial Intelligence Conference, AAAI 96, IAAI 96, Portland, Oregon, volume 1, pages 709--716. American Association for Artificial Intelligence, AAAI Press / The MIT Press, 1996.

....during testing [9] and lead to false positives in testing. The method described above for the commercial scanner was never intended to detect unknown malicious binaries, but the data mining algorithms that follow were built to detect new malicious executables. The next algorithm we used, RIPPER [3], is an inductive rule learner. This algorithm generated a detection model composed of resource rules that was built to detect future 5 examples of malicious executables. This algorithm used libBFD information as features. RIPPER is a rule based learner that builds a set of rules that identify ....

William Cohen. Learning Trees and Rules with SetValued Features. American Association for Artificial Intelligence (AAAI), 1996.

....E(A) The attribute predicate with the highest in formation gain is used to partition the dataset. Other measures for building decision trees may also be used [8] Recently some decision tree algorithms that take into account a large number of features describing objects have been proposed [3]. The algorithm [3] can be modified to analyze spa tim descriptions of the data. For every classified object a set of generalized predicates that are satisfied by this object is stored. An example of such description is presented in Table 3. For every predicate P from the table we have to find p, ....

....with the highest in formation gain is used to partition the dataset. Other measures for building decision trees may also be used [8] Recently some decision tree algorithms that take into account a large number of features describing objects have been proposed [3] The algorithm [3] can be modified to analyze spa tim descriptions of the data. For every classified object a set of generalized predicates that are satisfied by this object is stored. An example of such description is presented in Table 3. For every predicate P from the table we have to find p, e.g. the number ....

W. W. Cohen. Learning Trees and Rules with Set-valued Features. Proc. of the 13th National Conference on Artificial Intelligence (AAAI), Portland, OR, 1996.

No context found.

William W. Cohen. Learning trees and rules with set-valued features. In William J. Clancey and Dan Weld, editors, Proceedings of the Thirteenth National Conference on Artificial Intelligence, pages 709--716. AAAI Press, Menlo Park, California, 1996.

No context found.

W. Cohen.: Learning Trees and Rules with Set-valued Features. In. Proceedings of the Thirteenth National Conference on Artificial Intelligence. AAAI Press, (1996)

No context found.

W. W. Cohen. Learning trees and rules with set-valued features. In Proceedings of AAAI'96, 1996.

No context found.

Cohen, WW, 1996, Learning trees and rules with set-valued features. In Proceedings of the 13th National Conference on Artificial Intelligence, vol.1,pp. 709--716.

No context found.

(MILCOM Oct. 2000. [12] W. W. Cohen. Fast e#ective rule induction. Proc. of the 12th Intl. Conf. on Machine Learning, 1995. [13] W. W. Cohen. Learning trees and rules with set-valued features. AAAI/IAAI, 1:709--716, 1996. [14] L. Degioanni, F. Risso, and P. Viano. Windump: tcpdump for Windows, Jan. 2002. .

No context found.

Cohen W., "Learning Trees and Rules with Set-Valued Features", American Association for Artificial Intelligence (AAAI), 1996.

No context found.

William Cohen. Learning Trees and Rules with Set-Valued Features. American Association for Artificial Intelligence (AAAI), 1996.

No context found.

Cohen, W. (1996). Learning trees and rules with setvalued features. Proceedings of the Thirteenth National Conference on Artificial Intelligence (pp. 709-- 716). Menlo Park, CA: AAAI Press.

No context found.

COHEN, W. 1996. Learning trees and rules with set-valued features. In Proceedings of the 14th National Conference on Artificial Intelligence. AAAI Press, 709--716.

No context found.

Cohen W (1996) Learning Trees and Rules with Set-valued Features. Proceedings of the Thirteenth National Conference on Artificial Intelligence, pp 709-716, AAAI/MIT Press.

No context found.

Cohen, W. (1996). Learning trees and rules with set-valued features. In Proc. 14th AAAI.

No context found.

Cohen W., Learning Trees and Rules with Set-Valued Features, American Association for Artificial Intelligence (AAAI), 1996.

No context found.

Cohen W., Learning Trees and Rules with Set-Valued Features, American Association for Artificial Intelligence (AAAI), 1996.

No context found.

William Cohen. 1996. Learning trees and rules with setvalued features. In Proc. 14th AAAI.

No context found.

Cohen W (1996) Learning Trees and Rules with Set-valued Features. Proceedings of the Thirteenth National Conference on Artificial Intelligence, pp 709-716, AAAI/MIT Press.

No context found.

W.W. Cohen, Learning trees and rules with set-valued features, in: Proceedings AAAI-96, Portland, OR, 1996, pp. 709--716.

No context found.

W. Cohen, `Learning trees and rules with set-valued features', in Proceedings of the 13th AAAI and 8th IAAI, (1996).

No context found.

Cohen William, Learning Trees and Rules with Set-Valued Features, Proceedings of the AAAI/IAAI Conference, vol. 1, pages 709-716, 1996.

No context found.

William W. Cohen. Learning trees and rules with set-valued features. In Proceedings of the Thirteenth

First 50 documents  Next 50

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