D. Lewis. Learning in intelligent information retrieval. In Proceedings of the Eighth International Workshop on Machine Learning, pages 235-239, 1991.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....of some the classified documents are used as examples for updating the query vector as a linear combination of the initial query vector and the examples judged by the user. Especially, when the inner product similarity is used, relevance feedback is just a Perceptton like learning algorithm [10]. It is known [8] that there is an optimal way for updating the query vector if the sets of relevant and irrelevant documents are known. Practically it is impossible to derive the optimal query vector, because the full sets of the relevant and irrelevant documents are not available. There are ....

....similarity based relevance feedback algorithm with the inner product similarity. In this sense our work in this paper extends the lower bound results obtained in [9] for the Perceptton algorithm (or in general the linear additive on line learning algorithm) We refer the readers to the work of [15, 10] for discussions of the Perceptton like learning nature of the similarity based relevance feedback algorithm. It was stated in [10] that the most important future direction for research in information retrieval is likely to be machine learning techniques which can combine empirical learning with ....

[Article contains additional citation context not shown here]

D. Lewis. Learning in intelligent information retrieval. In Proceedings of the Eighth International Workshop on Machine Learning, pages 235-239, 1991.

....of some the classified documents are used as examples for updating the query vector as a linear combination of the initial query vector and the examples judged by the user. Especially, when the inner product similarity is used, relevance feedback is just a Perceptron like learning algorithm [16, 9]. There are many different variants of relevance feedback in information retrieval. The most popular one is Rocchio s similarity based relevance feedback algorithm [13, 11, 21] which works in a step by step adaptive refinement fashion as follows. Starting at an initial query vector 1 , the ....

D. Lewis. Learning in intelligent information retrieval. In Proceedings of the Eighth International Workshop on Machine Learning, pages 235--239, 1991.

....User for a solution. Other collaborative systems which are more closely related to CLML were developed for Natural Language acquisition and Information Retrieval (IR) The goal of IR techniques is to nd, within a large database of documents, those documents which satisfy a User information need [11]. Two IR methods of sampling examples, relevance feedback and uncertainty sampling involve Active Learning. Both are di erent to the CLML approach of selecting the trial which minimises the expected cost of experimentation. Relevance feedback [12] involves Users indicating to an IR system which ....

D.D.Lewis. Learning in intelligent information retrieval. In Eighth International Workshop on Machine Learning, pages 235-239, 1991.

....of some the classified documents are used as examples for updating the query vector as a linear combination of the initial query vector and the examples judged by the user. Especially, when the inner product similarity is used, relevance feedback is just a Perceptron like learning algorithm [16, 9]. There are many different variants of relevance feedback in information retrieval. The most popular one is Rocchio s similarity based relevance feedback algorithm [13, 11, 21] which works in a step by step adaptive refinement fashion as follows. Starting at an initial query vector q 1 , the ....

D. Lewis. Learning in intelligent information retrieval. In Proceedings of the Eighth International Workshop on Machine Learning, pages 235--239, 1991.

....engine can be interpreted as an approximation to the collection of the desired documents. Rocchio s similarity based relevance feedback algorithm, one of the most popular query reformation method in information retrieval [16, 14, 24, 2] is in essence adaptive supervised learning from examples [25, 19]. we showed in [11] that for any of the four typical similarity measurements (inner product, cosine coefficient, dice coefficient, and Jaccard coefficient) listed in [24] Rocchio s similarity based relevance feedback algorithm has a lower bound that is at least linear in the dimensionality of the ....

D. Lewis. Learning in intelligent information retrieval. In Proceedings of the Eighth International Workshop on Machine Learning, pages 235--239, 1991.

....of some the classified documents are used as examples for updating the query vector as a linear combination of the initial query vector and the examples judged by the user. Especially, when the inner product similarity is used, relevance feedback is just a Perceptron like learning algorithm [10]. It is known [8] that there is an optimal way for updating the query vector if the sets of relevant and irrelevant documents are known. Practically it is impossible to derive the optimal query vector, because the full sets of the relevant and irrelevant documents are not available. There are many ....

....similarity based relevance feedback algorithm with the inner product similarity. In this sense our work in this paper extends the lower bound results obtained in [9] for the Perceptron algorithm (or in general the linear additive on line learning algorithm) We refer the readers to the work of [15, 10] for discussions of the Perceptron like learning nature of the similarity based relevance feedback algorithm. It was stated in [10] that the most important future direction for research in information retrieval is likely to be machine learning techniques which can combine empirical learning with ....

[Article contains additional citation context not shown here]

D. Lewis. Learning in intelligent information retrieval. In Proceedings of the Eighth International Workshop on Machine Learning, pages 235--239, 1991.

....identify sets of features which can be used to identify USENET news articles of interest. Articles containing these features are then presented to the user. Other genetic techniques, such as mutation and crossover are applied to these features to explore new and possibly interesting USENET topics. Lewis (1991) has examined issues in applying learning techniques to information retrieval. He identifies two types of learning: short term learning, which occurs within a single session, and long term learning which evolves over many sessions. Short term learning can improve short term goals, such as ....

Lewis, D. (1991). Learning in Intelligent Information Retrieval. In Proceedings of the 8th International Machine Learning Workshop, pp. 235--239.

....systems for such challenging applications have to find ways to automate lexical concept learning as a prerequisite and, at the same time, as a constituent part of the text knowledge acquisition process. Unlike the current mainstream with its focus on statistically based learning methodologies (Lewis, 1991; Resnik, 1992; Sekine et al. 1992) we advocate a symbolically rooted learning approach in order to break the concept acquisition bottleneck, one which is based on expressively rich (terminological) knowledge representation models of the underlying domain (Hahn et al. 1996b; Hastings, 1996) ....

Lewis, D. (1991). Learning in intelligent information retrieval. In L. Birnbaum & G. Collins (Eds.), Machine Learning: Proc. of the 8th Intl. Workshop, pp. 235--239. Chicago, Ill., June 1991. San Mateo, CA: Morgan Kaufmann.

....track with these lexical innovations by hand coding is clearly precluded. On the other hand, the medical domain is more or less lexically stable but the sheer size of its sublanguage (conservative estimates range about 10 6 concepts) also cannot reasonably be coded by humans in advance. Unlike Lewis (1991) we advocate a symbolically rooted learning approach in order to break the lexical acquisition bottleneck (Hahn, Schnattinger, Klenner 1995; Hastings 1995) The methodology we propose is not limited to the information extraction task but can also be applied to any application framework where ....

Lewis, D. 1991. Learning in intelligent information retrieval.

....track with these lexical innovations by hand coding is clearly precluded. On the other hand, the medical domain is more or less lexically stable but the sheer size of its sublanguage (conservative estimates range about 10 6 concepts) also cannot reasonably be coded by humans in advance. Unlike Lewis (1991) we advocate a symbolically rooted learning approach in order to break the lexical acquisition bottleneck (Hahn, Schnattinger, Klenner 1995; Hastings 1995) The methodology we propose is not limited to the information extraction task but can also be applied to any application framework where ....

Lewis, D. 1991. Learning in intelligent information retrieval.

....These relations may be subsumption, associative, transitive, or unconstrained. Thesauruses can be either automatically created for which there exist a wide range of techniques, or they can be hand crafted. Srinivasdan[45] describes a range of these approaches. 2.1. 5 Problems with Keywords Lewis[35] recognizes two types of problem with keywords extraction. These are synonyms (two words may mean the same thing) and polysemy (words two two meanings) Pronouns also cause a problem in cases where the number of references to a keyword are counted. 2.2 Measures of Performance Commonly, retrieval ....

David D. Lewis. Learning in intelligent information retrieval. Machine Learning. Proceedings of the Eighth International Workshop on Machine Learning.

.... period of time, and categorizing documents according to their content (e.g. assigning Dewey Decimal numbers to abstracts of articles) Statistical approaches have been widely applied to these systems because of the poor fit of text to data models based on formal logics (e.g. relational databases) [8]. Rather than requiring that users anticipate exactly the words and combinations of words that will appear in documents of interest, statistical IR approaches let users simply list words that are likely to appear in relevant documents. The system then takes into account the frequency of these ....

David D. Lewis. Learning in intelligent information retrieval. In Eighth International Workshop on Machine Learning, pages 235--239, 1991.

No context found.

LEWIS,D.D. (1991). Learning in Intelligent Information Retrieval. Proceedings of the 8th International Machine Learning Workshop,pp. 235--239.

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