Donna Harman. Relevance feedback revisited. In Proceedings of the 15th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1--10, Copenhagen, Denmark, June 1992.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....This strategy simply adds a weighted sum of the relevant document vectors and subtracts a weighted sum of the non relevant documents from the query. Relevance feedback using this strategy produces a very large improvement in retrieval performance (from 20 80 or more, depending on the collection) [69, 33, 2]. Adding the full document text directly into the query does have its drawbacks. The number of terms in the query grows rapidly with the addition of evaluated documents, causing searches to take longer and longer, as the performance benefits of the inverted index are lost. A full search must be ....

....query, however it is unlikely to be the best strategy. As the query gets longer, there is a significant danger of overfitting. Many of the terms within a document may not be characteristic of the underlying topic that is relevant. Therefore, some method of term selection may be in order. Harman [33] finds that adding only the twenty most important terms to the query, where importance is judged using term weights, is far more effective than adding all of the terms from the relevant documents. We will address this issue in more detail in the next chapter. 3.3.3 Assumptions of Relevance ....

[Article contains additional citation context not shown here]

Donna Harman. Relevance feedback revisited. In Proc. 15th Int'l Conference on R&D in IR (SIGIR), pages 1--10, 1992.

....of the objects has been the major obstacle to more successful retrieval systems. Relevance feedback and hidden annotation have been shown to be two of the most powerful tools for bridging the gap between low level features and high level semantics. Widely used in text retrieval [2], relevance feedback was first proposed by Rui et al. 3] as an interactive tool in content based image retrieval. Since then it has been proven to be a powerful tool and has become a major focus of research in this area [4] 7] Relevance feedback moves the query point toward the relevant ....

D. Harman, "Relevance feedback revisited," in Proc. 15th Annu. Int. ACM SIGIR Conf. Research and Development in Information Retrieval, 1992, pp. 1--10.

.... 25 retrieved documents as positive exemplars and presuming the lowest 75 ranked out of 1000 were irrelevant, we produced a set of 60 weighted terms for each query that included the original query terms; this is analogous to both query expansion and query term re weighting as described in Harman [11]. It should be pointed out that the sub collections in each language of the CLEF 2001 evaluation are contemporaneous, so this set of expansion terms might be somewhat better than an arbitrary monolingual collection. We did not investigate global methods for query expansion in the source language ....

D. Harman, `Relevance Feedback Revisited.' In the Proceedings of the 15th International ACM SIGIR Conference on Research and Development in Information Retrieval (SIGIR-92), pp. 1-10, 1992.

....the user relevance information. Moreover, the experiments in [Sme 83] did not yield a consistent performance improvement. On the other hand, the direct use of relevance information, by simply extracting terms from relevant documents, is proved to be effective in interactive information retrieval [Har 92, Sat 90] However, this approach does not provide any help for queries without relevance information. In addition to automatic query expansion, semi automatic query expansion has also been studied [Ekm 92, Han 92, Wad 88] In contrast to the fully automated methods, the user is involved in the ....

....we can take into account the domain knowledge contained in the similarity thesaurus to find the most likely intended interpretation for the user s query. When relevance feedback techniques are used, queries are expanded by adding terms from the retrieved relevant documents. The experiments in [Har 92] show that adding as few as 20 top properly ranked terms, rather than all the terms from the retrieved relevant documents, can result in significant performance improvement. This is the reason we also add only those terms that are ranked in the top positions by the Simqt function. Another reason ....

[Article contains additional citation context not shown here]

Harman, D., Relevance feedback revisited, SIGIR'92, 15th Int. ACM/SIGIR Conf on R&D in Information Retrieval, Copenhagen, Denmark, 1-10, June 1992.

.... form: c 1 c2 : where each c j is a disjunction of literals: l 1 l 2 : a term or the number of retrieved relevant documents containing the term (postings) 6] More ellaborated methods use ratios or probabilities of terms occurring in relevant vs ocurring in non relevant documents [7]. Top ranked terms are supposedly the important ones within the set of relevant documents and, thus, they are a better representation for the set of relevant documents. Instead of expanding the query with all the terms from the relevant documents, the query is expanded with the selected terms. ....

D. Harman. Relevance feedback revisited. In Proc. SIGIR92, the 15th ACM International Conference on Research and Development in Information Retrieval, pages 1--10, Copenhagen, Denmark, June 1992.

....results and get even better results [10] Also in Harman s work, the query is expanded or modified based on the (explicit) feedback of documents from the user. The number of terms added to the query is experimented to achieve best precision improvement when 20 40 terms are added to the query [3]. From another perspective, two models have been adopted in query expansion for relevance feedback. One is the vector space model initiated by Rocchio in 1971 [8] The other is the probabilistic model proposed by Robertson and Sparck Jones in 1976 [6] The basic module in Rocchio s algorithm is ....

.... weighting in probabilistic model are compared in [7] by Robertson and Walker) Since the probabilistic model did not envision query expansion but only the reweighting of terms based on relevance judgments, several reasonable sets of sorting are tried to select non query terms for query expansion [3]. 3 Concept based Information Search The difficulty to formulate a request and the inherent word ambiguity in natural language can be overcome by concept based relevance feedback. To demonstrate the idea, let us first give the search result of the query TREC conference proceeding as an example ....

D. Harman. Relevance feedback revisited. In Proc. of ACM SIGIR Intl. Conf. on Research and Development in Information Retrieval, pages 1--10, 1992.

....we can say that the use of the convex combination in term nodes will help to improve the results. Also, the more sophisticated techniques (d3 and d4) can be considered better than the simpler ones (d2 and d2 5 ) The best results of these experiments are similar to those obtained by other models [5,10,11,14]. Anyway, these values are not totally comparable, because the experiments have been carried out with different models and under different experimental conditions. The future works will be centred in the development of new relevance feedback methods for the BNR model based on the underlying ....

D.Harman. Relevance feedback revisited. In Proceedings of the 15 ACM--SIGIR Conference on Research and development in information retrieval, pages 1--10, 1992.

.... refers to algorithms for automatic query expansion based on feedback provided by users about the desirability of specific documents (Salton and McGill, 1983) Improvements in query performance attributed to relevance feedback have been reported widely in literature (e.g. Salton and Buckley, 1990; Harman, 1992). Relevance feedback algorithms typically use term frequency statistics to determine which words should be added to the query. Thus high frequency terms that are found in only a few documents are potentially useful for query expansion. See Lee (1995) for a summary of some common term weighting and ....

....if terms from prior, more specific queries were used to focus the new query and to disambiguate the search results. One side effect of this addition was that the number of terms per query tended to increase. In addition to increasing search time, this could also affect query effectiveness. Harman (1992) found that performance decreased when more than 40 or so terms were added to relevance feedback queries. Thus it was necessary to compare retrieval performance 3 of weighting schemes with and without terms from prior queries. Four strategies (Table 6 2) for expanding queries were identified. ....

Harman, D. (1992) Relevance feedback revisited. In Proceedings of SIGIR '92. ACM Press. pp. 1-10.

....results. This suggests that the effects described here are probably real. It may be interesting to test these ideas further by analyzing relevance judgment data from other TREC experiments. 4.5. 2 Query construction How many terms should be included in the queries derived from links Harman [8] observes that after about 20 to 30 terms, marginal gains in precision tend to decrease with added terms, while the cost of computing these queries tends to increase linearly with the number of terms. In our prototype, we have not filtered the terms selected by users (except stop words, which are ....

Harman, D. (1992) Relevance Feedback Revisited. In Proceedings of SIGIR '92 (Copenhagen, Denmark June 1992), ACM Press. pp. 1-10.

....meanings of the objects has been the major obstacle to more successful retrieval performance. Relevance feedback and hidden annotation have been shown to be two of the most powerful tools for bridging the gap between low level features and high level semantics. Widely used in text retrieval [2][3] relevance feedback was first proposed by Rui et al. as an interactive tool in content based image retrieval [4] Since then it has been proven to be a powerful tool and has become a major focus of research in this area [5] 6] 7] 8] 9] 10] In MindReader, Ishikawa et al. formulated a ....

Donna Harman, "Relevance Feedback Revisited", Proceedings of the Fifteenth Annual International ACM SIGIR conference on Research and development in information retrieval, pp. 1-10, 1992.

....that have high relevance weights are likely to improve retrieval effectiveness if they are added to the user s query. Both these components have been shown to be powerful techniques for improving retrieval effectiveness, 7] with query expansion often being the more successful of the two, e.g. [8, 16]. Query expansion may be automatic, in which case the system selects which terms to add to the user s query and how many terms to add. An alternative is interactive query expansion, where the user selects the new expansion terms, based on a set chosen by the system. Current experimental evidence ....

D. Harman. Relevance feedback revisited. Proceedings of the Fifteenth Annual International ACM SIGIR Conference on Research and Development in Information Retrieval. pp 1-10. Copenhagen. 1992.

....to term i , we want to arrive at a new query expression: Q 0 0 = q 0 1 ; q 0 2 ; q 0 t ) such that the weights are adjusted so that new terms can be introduced into the vector representation, while other terms can effectively be removed by reducing their respective weights to 0. Harman (1992) describes the operational procedure underlying this process as the merging of document and query vectors. More specifically, this means that query terms not in the original query but appearing in the relevant documents are added to the initial query expression. The expansion occurs using both ....

Harman, D. (1992). Relevance feedback revisited. In Proceedings of ACM SIGIR-92.

No context found.

Donna Harman. Relevance feedback revisited. In Proceedings of the 15th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, pages 1--10, Copenhagen, Denmark, June 1992.

No context found.

Donna Harman. Relevance feedback revisited. In Proceedings of the 15th ACM SIGIR, pages 1--10, 1992.

No context found.

D. Harman. Relevance feedback revisited. In Proceedings of SIGIR 1998.

No context found.

Harman, Donna (1992a): Relevance Feedback Revisited. In: 15th Proc. Annual Intl. ACM SIGIR Conf on Research and Development in Information Retrieval. New York, USA, pp. 1-10.

No context found.

Harman, D.K. Relevance feedback revisited. In > Proceedings of the Eleventh ACM SIGIR Conference. New York: ACM Press, 1992, pp. 321--331.

No context found.

Harman, D. Relevance feedback revisited. Proceedings of the 15th Annual International ACM SIGIR Conference on Research and Development in Information Retrieval. 1-10. Copenhagen. 1992.

No context found.

D. Harman. Relevance feedback revisited. In Proceedings of the Fifth International SIGIR Conference on Research and Development in Information Retrieval, pages 1--10, 1992.

No context found.

D. Harman. Relevance feedback revisited. In Proc. SIGIR-92, the 15th ACM International Conference on Research and Development in Information Retrieval, pages 110, Copenhagen, Denmark, June 1992.

No context found.

Donna Harman, "Relevance Feedback Revisited", Proceedings of the Fifteenth Annual International ACM SIGIR conference on Research and development in information retrieval, pp. 1-10, 1992.

No context found.

Donna Harman, "Relevance Feedback Revisited", Proceedings of the Fifteenth Annual International ACM SIGIR conference on Research and development in information retrieval, pp. 1-10, 1992.

No context found.

Donna Harman, "Relevance Feedback Revisited", Proceedings of the Fifteenth Annual International ACM SIGIR conference on Research and development in information retrieval, pp. 1-10, 1992.

No context found.

D. Harman. Relevance feedback revisited. In Proc. of the fifteenth annual international ACM SIGIR ConferenceonResearch and Development in Information Retrieval. June 1992.

No context found.

Donna Harman, "Relevance feedback revisited," in 15th Ann Int'l SIGIR '9/Denmark-6/9, 1992.

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