S. E. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of the Third Text REtrieval Conference, Gaithersburg, USA, November 1994.  Home/Search   Document Not in Database   Summary   Related Articles   Check

....problem. The framework unifies several existing retrieval models within one general probabilistic framework, and facilitates the development of new principled approaches to text retrieval. In traditional retrieval models, such as the vector space model [12] and the BM25 retrieval model [11], the retrieval parameters have almost always been introduced heuristically. The lack of a direct modeling of queries and documents makes it hard for these models to incorporate, in a principled way, parameters that adequately address special characteristics of queries and documents. For example, ....

....of a document. As a result, heuristic parameters must be used (see, e.g. the pivot length normalization method [14] Similarly, in the BM25 retrieval formula, there is no direct modeling of queries, making it necessary to introduce heuristic parameters to incorporate query term frequencies [11]. One important advantage of the risk minimization retrieval framework [4] over these traditional models is its capability of modeling both queries and documents directly through statistical language modeling. Although a query and a document are similar in the sense that they are both text, they ....

Robertson, S. E., Walker, S., Jones, S., M.Hancock-Beaulieu, M., and Gatford, M. (1995). Okapi at TREC-3. In Harman, D. K., editor, The Third Text REtrieval Conference (TREC-3).

....### # ### (2) which clearly measures the cosine of the angle between ### and . The terms # # and # appearing in the denominator of (1) have the effect of, at least partially, normalising the distance between documents with respect to document length. The popular BM25 weighting scheme [22] incorporates more sophisticated document length normalisation in its assignment of term weights and distances between documents are measured by the unnormalised inner product: ### ### ######## (3) # # ### (4) In text classification each document in the document collection is ....

....normalised by document length. Using the standard ### # weight longer documents obtain higher weights simply because they contain more terms, document length normalisation combats this. Method BM25: ### # # ####### # ### # ### ### ### # (9) this is the well known method BM25 [22], the values for variables are chosen according to the qualities of the document collection. The term controls the degree of document length normalisation, # ### assumes verbose documents and assumes multitopic documents. The terms determines the amount ### reacts to changes in the ....

S.E. Robertson, S. Walker, S. Jones, and M. M. Hancock-Beaulieu. Okapi at trec-3. In TREC-3, 1994.

....more rigorous evaluations of the tasks. Researchers have studied lexical, phrasal, co occurrence and dependency relations that can be pulled out from spans of text [11, 5] Pseudo relevance feedback methods find words that are related to the query terms to improve the effectiveness of retrieval [24, 19, 14] . Those uses of statistics and others like them are similar to our building of ELMs in that they find related words. We know of no work that viewed the probability distribution of text around a word as a model of its meaning. 3. MODEL QUALITY In the above discussion we said that an ELM is ....

S. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-2. In The Second Text retrieval Conference, pages 21--34, 1994.

....data set. 5.1 System design Our Interactive Tracking system uses the vector space approach where each document is represented by a vector of term weights V . The weight of the ith term in the vocabulary, v i is computed using the Inquery weighting formula, which uses Okapi s tf score [14] and Inquery s normalized idf score: v i = tf tf 0.5 1.5 doclen avgdoclen log( colsize 0.5 docf ) log(colsize 1) where tf is the number of times the term occurs in the document, docf is the number of documents the term occurs in, doclen is the number of terms in the document, ....

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In D. Harman and E. Voorhees, editors, Third Text REtrieval Conference (TREC-3). NIST, 1995.

....is an effective technique commonly used to add more useful words to a query [13, 18] Unfortunately, casual users seldom provide a system with the relevance judgements needed in relevance feedback. For such users, a commonly used approach to query expansion, referred to as adhoc or blind feedback [6, 7, 12, 3], takes the form of pseudo relevance feedback, where actual input from the user is not required. In this method, a small set of (say 20) documents is retrieved using the original user query; these documents are all assumed to be relevant without any intervention by the user. These assumed ....

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In D. K. Harman, editor, Proceedings of the Third Text REtrieval Conference (TREC-3). NIST Special Publication 500-225, 1995.

....Table 1 depicts the summarised statistics for this 100 library test bed. Resource selection is based on the CMU 300 document samples acquired by query based sampling. For both the sample (for resource selection) and the actual collections (for evaluation) we used a BM25 weighting scheme [18]: P(t d) tf (t, d) tf (t, d) 0.5 1.5 dl(d) avgdl log DL df (t) log . Here, tf (t, d) is the term frequency (number of times term t occurs in document d) dl(d) denotes the document length (number of terms in document d) avgdl the average document length, the size of ....

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, A. Gull, and M. Lau. Okapi at TREC. In Text REtrieval Conference, pages 21--30, 1992.

.... used linear retrieval function can be used [16, 18] cj Gq condition weight indexing weight Here, the query q consists of conditions c (for text retrieval, a condition simply is a term) with weight Pr(q cj) Furthermore, Pr(cj d) denotes the indexing weight (e.g. a normalised BM25 weight [14]) E(rellq , D) Pr(rellq, d ) f(Pr(q d) In practice, we do not have the probabilities Pr(q d) or Pr(rellq, d) respectively) for each document d. But, let us assume that Pr(q d) is the score of the library search engine, and that we know the distribution of the scores. Then, ....

....bed. To reduce the evaluation time, we only used the AP 1988 1990 collection and split it into 24 collections, according to the CMU 100 collection partition [4] We removed all TREC stop words, and applied the Porter stemruer. As indexing weights, we used a BM 25 variant (for original BM 25, see [14]) if(t, d) tfw(l,d) if( d) 0.5 1.5. dyd) avgdl log af(t) idf(l) log ]D Lcoection ] tfw(t, if(t) Here, tf(t, d) denotes the number of times term t appeared in document d (term kequency) dl(d) the number of tokens in d (document length) avgdl the average document length ....

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, M. Gatford, and A. Payne. Olmpi at TREC-4. In Text REtrieval Conference 6, pages 73-96, 1996.

.... 0.5 Nw 0.5 (3) evaluates the relevance of term w inside the collection. The model implies two parameters k 1 and b to be empirically estimated over a development sample. As in previous work, the setting k 1 = 1.5 and b = 0.4 were used. An explanation of the involved terms can be found in [7] and other papers referred in it. 3.2 Language Model According to this model, the match between a query random variable Q and a document random variable D is expressed through the following conditional probability distribution: P r(D Q) P r(Q D)P r(D) P r(Q) 4) where P r(Q D) ....

Robertson, S. E., S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford, 1994. Okapi at TREC-3. In Proc. of 3rd TREC . Gaithersburg, MD.

....document length ( l) global vocabulary size (V ) and mean document vocabulary size ( Vd ) Terms Stop l V Vd text no 225 160K 134 base forms no 225 126K 129 stems no 225 101K 126 base forms yes 103 125K 80 stems yes 103 100K 77 3 Information Retrieval Models 3. 1 Okapi Model Okapi [9] is the name of a retrieval system project that developed a family of weighting functions in order to evaluate the relevance of a document d versus a query q. In this work, the following Okapi weighting function was applied: s(d) w#q#d f q (w) c d (w) idf(w) 1) where: c d (w) f d (w) k ....

....of w in d, and the inverted document frequency: idf(w) log N Nw 0.5 Nw 0.5 (3) evaluates the relevance of w inside the collection. The model implies two parameters k 1 and b to be empirically estimated over a development sample. An explanation of the involved terms can be found in [9] and other papers referred in it. 3.2 Statistical Model A statistical retrieval model was developed based on previous work on statistical language modeling [2] The match between a query q and a document d can be expressed through the following conditional probability distribution: P (d q) ....

Robertson, S. E., S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford, 1994. Okapi at TREC-3. In Proceedings of the 3rd Text REtrieval Conference. Gaithersburg, MD.

....likely to include content that is a relevant response to the query. Nq is the total number of queries and qf(t) is the number of queries in which t occurs. Those terms which occur more frequently in queries are more likely to This equation is BM11, which corresponds to BM25 in the case of b = 1 [11]. be stop words such as documents and thing. We decrease the scores of stop words by using . K category and K location are extended numerical terms that are introduced to improve precision of results. K category uses the category information of the document found in newspapers, such as the ....

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at trec-3. In TREC-3, 1994.

....) 8 ) 7 ) i j i i j i i j i i i j i j i i j i i i CE q q q k d d q q q q k d n N d i n N d w 4.3. Extending to the 2 Poisson Model Our method can be extended to incorporate the term dependence into the state of the art 2 Poisson model [13] in particular Okapi BM25 [12], using the Chow Expansion. O P O PG O P The weight of a term t in a 2 Poisson model is represented as in the following Eq. 10) 13] 1 ( 1 ( 1 ( 1 ( p e p e q q q e q e p p w tf = # # # # # where tf is the frequency of term ....

....zero. Then we can safely remove the components, and e will be small, so the approximation will be: 1 ( 1 ( q p Since we cannot estimate (10) directly, the alternative is w in Eq. 9) From the above results, Eq. 10) can be transformed into a simple formulation, such as BM25 [12], as given below. 1 ) 1 ( tf avdl dl b b k tf qtf w = where qtf is query term frequency, tf is the frequency of term t, dl is document length, avdl is average length of documents, and k 1 and b are constant parameters. We de f i ne MS BM25 Chow1(N) a query document scoring function which ....

S.E. Robertson, S. Walker, S. Jones, M.M. Hancock-Beaulieu, M. Gatford. Okapi at TREC-3. In Overview of the Third Text Retrieval Conference(TREC-3). 109-126, 1995.

.... unconfirmed) How does a modem work What should I consider when purchasing a PC for under 2,000 What is top3 and where can i learn more about it Who was Cleopatra Why do dogs have wet noses Where is the Taj Mahal TREe systems (including at least one based on the Okapi BM25 formula [10], used in the present study) were found to be highly effective in answering queries of this type, even in comparison to commercial search engines [6] The AltaVista examples indicate that different user needs exist. Where is the CNN home page is clearly different from What is CNN . However, ....

S. E. FI.obertson, S. Walker, M.M. Hancock-Beaulieu, and M. Gatford. Okapi at TFI.EC-3. In Proceedings of TREC-3, pages 109 126, Gaithersburg MD, November 1994. NIST special publication 500-225.

....frequency, which measures the rarity of a unigram. Normalized F actor is defined as follows: 1 if Sentence Length 40 Sentence Length 40 if Sentence Length 40 Another sentence ranking function is designed based on modification of the Okapi ranking function used for document ranking [18]. It s defined as follows: Score(S) T#Q tf idf 0.5 1.5 Sentence Length Sentence Lengthavg tf (4) where Sentence Length is the length of a sentence in words and Sentence Lengthavg is the average sentence length in the top 20 documents returned from a search. tf, idf is the ....

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-4. In D. K. Harman, editor, Proceedings of the Fourth Text Retrieval Conference, pages 73--97. NIST Special Publication 500-236, 1996.

....This means that shorter documents will have more aggressive discounting, while longer stories will not assign a lot of significance to a single occurrence of a term. This form of the tf component is generally referred to as Okapi tf since it was first introduced as part of the Okapi system. [2] The idf comp component is the logarithm of the inverse probability of the term in the collection, normalized to be between 0 and 1. N denotes the total number of documents in the collection, while df shows in how many of those documents the term occurs. This particular idf formulation arises ....

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In D. K. Harman, editor, The Third Text REtrieval Conference (TREC-3). NIST, 1995.

....out a comparative investigation into retrieval of electronic text, spoken documents and document images. In this investigation we explored the behaviour of term weighting and pseudo relevance feedback (PRF) for each collection. These experiments used our summary based PRF method described in [7] based on a modification of the Robertson expansion term selection methods [6] Experiments were carried out to optimise the number of assumed relevant documents and the number of expansion terms to be added. Results from this investigation showed that as we might expect, PRF gave improvement for ....

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In D. K. Harman, editor, In Proceedings of the Third Text REtrieval Conference (TREC-3), pages 109-126. NIST, 1995.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of the Third Text REtrieval Conference, Gaithersburg, USA, November 1994.

No context found.

S. E. Robertson, S. Walker, and M. Hancock-Beaulieu. Okapi at TREC-7. In Proceedings of the Seventh Text REtrieval Conference, Gaithersburg, USA, November 1998.

No context found.

S. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of the Third Text REtrieval Conference, November 1994.

No context found.

S. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In TREC-3, 1994.

No context found.

S. E. Robertson, S. Walker, M.M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of TREC-3, pages 109--126, November 1994. NIST special publication 500-225. Alastair G. Smith. Does metadata count? a webometric investigation. In Proceedings of the International Conference on Dublin Core and Metadata for e-communities, pages 133--138. Firenze University Press, 2002.

No context found.

S.E. Robertson, S. Walker, S. Jones, M.M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In D.K. Harman, editor, Proc. Text Retrieval Conference (TREC), pages 109--126, Gaithersburg, Maryland, 1994. NIST Special Publication 500225.

No context found.

Robertson, S. E., Walker, S., Jones, S., M.Hancock-Beaulieu, M., and Gatford, M. (1995). Okapi at TREC-3. In Harman, D. K., editor, The Third Text REtrieval Conference (TREC-3).

No context found.

ROBERTSON, S. E., WALKER, S., JONES, S., M.HANCOCK-BEAULIEU, M., AND GATFORD, M. 1995. Okapi at TREC-3. In The Third Text REtrieval Conference (TREC-3), D. K. Harman, Ed. 109--126.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, A. Gull, and M. Lau. Okapi at TREC. In Text REtrieval Conference, 1992.

No context found.

Robertson, S.E., Walker S., Jones S., Hancock-Beaulieu, M.M. and Gatford, M. 1995. Okapi at TREC-3. In Proceedings of the Third Text REtrieval Conference(TREC-3), NIST Special Publication 500-225, Washington D.C., 109-126.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at trec-3. In D. K. Harman, editor, Overview of the Third Text REtrieval Conference (TREC-3), pages 109--126. Department of Commerce, National Institute of Standards and Technology, 1994.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. M.Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In D. K. Harman, editor, The Third Text REtrieval Conference (TREC-3), pages 109--126, 1995.

No context found.

S. E. Robertson, S. Walker, S. Jones, M.M. Hancock-Beaulieu, M. Gatford. Okapi at TREC-3. Proceedings of TREC-3, 1995.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, A. Gull, and M. Lau. Okapi at TREC. In Proc. 1st Text REtrieval Conf. (TREC 2001), pages 21--30. NIST, 1992.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of the 3rd Text REtrieval Conference (TREC3).

No context found.

Robertson, S. E., S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford (1995). Okapi at TREC--3. In Proc. Third Text Retrieval Conference (TREC-3), pp. 109--126.

No context found.

Robertson, S.E.; Walker, S.; Jones, S.; Hancock-Beaulieu, M.M.; and Gatford, M. Okapi at TREC-4. In D.K. Harman (ed.), Proceedings of the Fourth Text Retrieval Conference. Gaithersburg, MD: National Institute of Standards and Technology, 1996, pp. 73--97.

No context found.

S.E. Robertson, S. Walker, S. Jones, M.M. Hancock-Beaulieu, M. Gatford, Okapi at TREC-4, in: D.K. Harman (Ed.), Proceedings of the Fourth Text Retrieval Conference, NIST Special Publication, vol. 500-236, 1996, pp. 73 -- 97.

No context found.

ROBERTSON, S. E., WALKER, S., JONES, S., M.HANCOCK-BEAULIEU, M., AND GATFORD, M. 1995. Okapi at TREC-3. In The Third Text REtrieval Conference (TREC-3), D. K. Harman, Ed. 109--126.

No context found.

Robertson, S. E., Walker, S., Jones, S., M.Hancock-Beaulieu, M., and Gatford, M. (1995). Okapi at TREC-3. In Harman, D. K., editor, The Third Text REtrieval Conference (TREC-3).

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, A. Gull, and M. Lau. Okapi at TREC. In Text REtrieval Conference, pages 21-30, 1992.

No context found.

Robertson, S. E., Walker, S., Jones, S., Hancock-Beaulieu, M. M., and Gatford, M. Okapi at TREC-3. In Overview of the Third Text REtrieval Conference (TREC-3), pages 109-126, Gaithersburg, MD, 1994. National Institute of Standards and Technology.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of TREC-3, November 1994.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at trec-3. In Text REtrieval Conference (TREC-2), pages 109--126, 1994.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, A. Gull, and M. Lau. Okapi at TREC4. In Text REtrieval Conference, pages 21--30, 1992.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proc. of the 3rd Text REtrieval Conference, pages 109--126, Gaithersburg, MD, 1994.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In NIST Special Publication 500-225: The Third Text REtrieval Conference (TREC-3), pp. 109--126, 1994.

No context found.

S. E. Robertson, S. Walker, S. Jones, M. Hancock-Beaulieu, and M. Gatford. Okapi at TREC-3. In Proceedings of the 3rd Text REtrieval Conference (TREC-3), 1994.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, M. Gatford, and A. Payne. Okapi at TREC-4. In Proceedings of the 4th Text REtrieval Conference (TREC-4), 1995.

No context found.

S. E. Robertson, S. Walker, M. Hancock-Beaulieu, A. Gull, and M. Lau. Okapi at TREC. In Text REtrieval Conference, pages 21--30, 1992.

No context found.

Robertson, S.E., Walker, S., Spark-Jones, K., Hancock-Beaulieu, M.M., and Gatford, M. (1995) OKAPI at TREC-3. In Proceedings of the 3 rd Text Retrieval Conference (TREC-3).

No context found.

S.E. Robertson, S. Walker, S. Jones, M.M. Hancock-Beaulieu, M. Gatford, Okapi at TREC-4, in: D.K. Harman (Ed.), Proceedings of the Fourth Text Retrieval Conference, NIST Special Publication, vol. 500-236, 1996, pp. 73 -- 97.

No context found.

Robertson, S.E., Walker, S., Spark-Jones, K., Hancock-Beaulieu, M.M., and Gatford, M. (1996) OKAPI at TREC-3. In Proceedings of the 3 rd Text Retrieval Conference (TREC-3).

No context found.

S. E. Robertson, S. Walker, S. Jones, M. M. Hancock-Beaulieu, M. Gatford. OKAPI at TREC-3, In The Third Text REtrieval Conference (TREC 3), 1994.

No context found.

Robertson, S.E., S. Walker, S. Jones, M.M. Hancock-Beaulieu, M. Gatford. 1995. Okapi at TREC-3. In Harman, Donna K. (editor). Overview of the Third Text REtrieval Conference 195 (TREC-3). NIST Special Publication 500-225, National Institute of Standards and Technology, Gaithersburg, MD, (http://trec.nist.gov/pubs.html), pp.109-126.

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