D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.  Home/Search   Document Not in Database   Summary   Related Articles   Check

.... exible and allows to de ne explicitly the structure of the inverted le. As a consequence, we could design and build an inverted le able to e ciently store documents as DNF formulas. A total of 50 TREC topics were used in this experimentation. Topics #151 #200 from TREC 3 adhoc retrieval task [5] were used to generate automatically DNF queries for representing user needs. We used a stoplist of 571 words and terms were stemmed using Porter s algorithm [19] 3.1 Evaluating the PLBR model Two main strategies were applied to de ne logical queries. First, a baseline with at query structure ....

D. Harman. Overview of the third text retrieval conference. In Proc. TREC-3, the 3rd text retrieval conference, 1994.

....using arithmetic character based adaptive coding. Tests were carried out on Linux Red Hat 7.2 operating system, running on a computer with a Pentium III processor at 500 MHz and 128 Mbytes of RAM. For the experiments we selected di erent size collections of WSJ, ZIFF and AP, from TREC 3 [Har95] Several characteristics of the collections are shown in Table 1. We concatenated les so as to obtain approximately similar subcollection sizes from the three collections, so the size in Mbytes is approximate. The structuring of the collections is similar: they have only one level of ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 119, 1995. NIST Special Publication 500-207.

....reaching the highlighted nodes. With leftrank and rightrank we nd that their ranges are [1,9] and [13,14] respectively. Figure 9 shows the second part. We search for the reverse pre xes of ala, namely la and a, in RevT rie. The nodes reached are highlighted. Their ranges are, respectively, [10,10] and [2,5] Finally, Figure 10 shows the last part of the search. We join pre x a with sux la, obtaining a 2 dimensional rank range (2,13) 5,14) and pre x al with sux a, obtaining a 2 dimensional range (10,1) 10,9) Both ranges are searched for in Range, and all the block identi ers found ....

....data structures used one by one, as there are interesting lessons on theory versus practice for each of them. To demonstrate the results in practice, we have chosen two di erent text collections. The rst, ziff, contains 83.37 megabytes (Mb) obtained from the ZIFF 2 disk of the TREC3 collection [10]. The second, dna, contains 51.48 Mb from GenBank (Homo Sapiens DNA, http: www.ncbi.nlm.nih.gov) with lines cut every 60 characters. We use the whole collections as well as incremental subsets of them. Our tests have been run on a Pentium IV processor at 2 GHz, 512 Mb of RAM and 512 kilobytes ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1-19, 1995. NIST Special Publication 500-207.

....600 5 266 with 266 MHz 21164 Alpha processors and 768 Mb of RAM, running Digital Unix 4.0B. The machine was not performing other heavy tasks while the experiments ran. We measure user times (CPU times) We searched 10 megabytes of English text extracted from the Wall Street Journal 1987 [Har95]. Any data point corresponds to an average over 100 di erent search patterns (the same for all the algorithms) The choice of patterns is always problematic when dealing with regular expressions, since there is no clear concept of what a random regular expression is and, as far as we know, there ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1-19, 1995. NIST Special Publication 500-207.

....topics had more than fifty new. Four of these topics had more than twenty new relevant documents found in their last round of feedback during the creation phase. Although our pooling process is radically different, these findings agree with Harman s analysis of the TREC 3 relevance judgements [1], as well as those of Zobel [11] that the largest topics (those with the most relevant documents) tend to yield even more relevant documents upon further searching. We have seen that such topics tend to have a greater number of relevant documents found in the last round of judging. In retrospect ....

D K Harman. Overview of the Third Text REtrieval Conference (TREC--3). In D K Harman, editor, Proceedings of the Third Text REtrieval Conference (TREC--3), pages 1--20. Gaithersburg, MD: NIST, 1994. NIST Special Publication 500-225.

....investigated more extensively in the next section. 4 Experiments and Results In order to determine the usefulness of the techniques described above, we test them on a variety of tasks. We use the TREC collections in our experiments. Our methods are evaluated on the adhoc tasks for TRECs 3 6 [8, 9, 20, 21]. The query sets and document collections used in these tasks are shown in Table 1. Since we are interested in studying short queries, we use only the Description field for queries 151 300. For the TREC 6 queries (numbered 301 350) we use the Title field in addition a. Our experiments use ....

D. K. Harman. Overview of the Third Text REtrieval Conference (TREC-3). In D. K. Harman, editor, Proceedings of the Third Text REtrieval Conference (TREC-3). NIST Special Publication 500-226, 1995.

....an inadvertently deleted mail message could have disastrous consequences. Thus it is more important not only that the agent have a model of the user but also that the user has a model of the agent, in order to build up trust. aAs characterized by the ad hoc queries in the TREC conferences [Harman, 1995]. Social or collaborative filtering systems Social filtering systems share our goal described in section 5 of capitalizing on the shared interests of users. The key process here is to match up similar users, and make use of evaluations or annotations supplied by others. Examples of such systems ....

....ranked by the user as relevant actually outperformed profiles hand crafted by users. More recently experiments in the TREC 3 conference showed similar results, with automatically generated profiles for the routing task outperforming even skilled human searchers using an interactive IR system [Harman, 1995]. The absence of queries in our utomated browsing paradigm can be beneficial to users in a number of ways: they no longer have to grapple with different interfaces or query languages, they do not need to have any understanding of the vocabulary or other properties of the space of documents ....

Donna Harman. Overview of the third Text REtrieval Conference (TREC-3). In Proceedings of the 3 ra Text REtrieval Conference, Gaithersburg, MD, November 1995.

....empirically analyze our model and evaluate its performance. Tests were carried out on Linux Red Hat 7.2 operating system, running on a computer with a Pentium 4 processor at 1.4 GHz and 128 Mbytes of RAM. For the experiments we selected di erent size collections of WSJ, ZIFF and AP, from TREC 3 [Har95] The average speed to compress all collections is around 128 Kbytes sec. In this value we include the time needed to model, merge dictionaries and compress. The time for merging dictionaries is included in this gure, and it ranges from 4.37 seconds for 1 Mb to 40.27 seconds for 100 Mb. The ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 119, 1995. NIST Special Publication 500-207.

....file. The exceptions file is separated because it has to be accessed in parallel with the classical file. Our implementation, in C under Linux, is careful with the space usage. We have applied the algorithms to 100 Kbytes (i.e. 800 Kbits) of articles extracted from The Wall Street Journal 1989 [6]. Figure 2 shows the behavior of the classical algorithm as k grows. On top, we show the sizes of both parts of the compressed file (the antidictionary representation and the compressed text) On the bottom we show the amount of main memory needed to compress the file. As it can be seen, reaching ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

....will see, this extension makes grep very powerful and closer to our goal of a smooth degradation in efficiency as the pattern gets more complex. The experiments were carried out over 100 Mb of English text extracted from Wall Street Journal articles of 1987, which are part of the trec collection [11]. Two different machines were used: sun is a Sun UltraSparc 1 of 167 MHz with 64 Mb RAM running Solaris 2.6, and intel is an i686 of 550 MHz with 64 Mb RAM running Linux Red Hat 6.2 (kernel 2.2.14 5.0) Both are 32 bit machines (w = 32) To illustrate the complexity of the code, we show the sizes ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

....Huffman code is faster than decompression of binary Huffman code. All techniques for efficient encoding and decoding mentioned in [20] can easily be extended to this case. 3 4. Compression and Decompression Performance For the experimental results we used literary texts from the trec collection [11]. We have chosen the following texts: ap Newswire (1989) doe Short abstracts from doe publications, fr Federal Register (1989) wsj Wall Street Journal (1987, 1988, 1989) and ziff articles from Computer Selected disks (ZiffDavis Publishing) Table 1 presents some statistics about the five ....

D. K. Harman. Overview of the third text retrieval conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, Gaithersburg, Maryland, 1995. National Institute of Standards and Technology Special Publication.

....signal processing and computational biology [23] The particular case of a dictionary is of interest in spelling applications. The second metric space is that of documents under the cosine similarity measure [8] We took the 25,960 documents of the fr (Federal Register) collection of trec 3 [18]. We took the vocabulary of each document (considering letters and digits and mapping them to lower case) and created for each document a vector where each vocabulary word is a coordinate. If the vocabulary word t i appears f ij times in document d j and it appears in n i documents out of a total ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1-19, 1995. NIST Special Publication 500-207.

....Therefore, the scheme does also get worse as the dimension grows. However, it worsens much slower than the exact algorithm. 5 Real life Examples We show the performance of our method on two real applications. The rst one is a database of text lines from the Federal Register collection of TREC 3 [11]. We used edit distance: the minimum number of character insertions, deletions and substitutions to make the two strings equal. This model is commonly used in text retrieval, signal processing and computational biology applications. Figure 6 (left) shows the results for di erent k values. As can ....

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. TREC3, pages 1-19, 1995. NIST Special Publication 500-207.

....by the user [SM83] Relevance [Fro94, SEN90, Sch94, Sar75] is one of the most debated and central concepts in information science. For an exhausting survey on relevance see [Miz96] Classical IR research defines relevance mostly as topicality (e.g. the text retrieval conference competition [TREC, Har94] Other researchers also take task oriented utility (also called informativeness [Boy82] into account [Boy82, Soe94, Har92, GL91, Coo73a, Coo73b, Coo78, Bar67] Because of its subjective nature, task oriented utility is harder to grasp and to express in a query. While topicality is often ....

D.K. Harman. Overview of the third Text REtrieval Conference (TREC-3). In Proceedings of the Third Text Retrieval Conference (TREC-3), pages 119, D.K. Harman (Ed.), National Institute of Standards and Technology (NIST) Special Publication 500-225, Gaithersburg, Maryland, 1995. Online at http://potomac.ncsl.nist.gov/TREC. 177

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1-19, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 119, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. TREC-3, pages 119, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 119, 1995. NIST Special Publication 500-207.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1-19, 1995. NIST Special Publication 500-207.

No context found.

Donna K. Harman. Overview of the third text retrieval conference. In The Third Text REtrieval Conference), pages 1--20, Gaithersburg, Maryland, USA, November 2-4 1994. National Institute of Standards and Technology.

No context found.

D. Harman. Overview of the Third Text REtrieval Conference. In Proc. Third Text REtrieval Conference (TREC-3), pages 1--19, 1995. NIST Special Publication 500-207.

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