Mittendorf, E. and Schauble, P. (1994). Document and passage retrieval based on hidden markov models. In Proceedings of SIGIR'94, pages 318--327.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....then combining that probability with the probability from the second HMM that a particular passage matched the query. They used the Viterbi algorithm to determine the highest probability path through the concatenated HMMs and from that path identified the best ranked passage for a particular query [Mittendorf and Schauble, 1994]. Mittendorf and Schauble s technique has the advantage that the document collection need not be segmented prior to IR. But, it has the disadvantage that the passages it 135 retrieves begin and end with words from the query. The passages are therefore unlikely to begin and end at sentence ....

Mittendorf, E. and Schauble, P. (1994). Document and passage retrieval based on hidden Markov models. In Proceedings of the ACM-SIGIR Conference on Research and Development in Information Retrieval, pages 318--327, Dublin, Ireland. ACM.

....can be obtained by use of dynamic programming to perform exact and global optimization. This has been used by Ponte and Croft [26, 34] Heinonen [14] Utiyama and Isahara [33] and others. Finally, probabilistically motivated approaches to text segmentation include the use of hidden Markov models [4, 21, 37] and Beeferman s work [1, 2] which utilizes word usage statistics, cue words and several other features to construct a probability distribution on segment boundaries. 2 THE SEGMENTATION ALGORITHM 2.1 Representation Suppose that a text contains T sentences and has a vocabulary of L distinct ....

Mittendorf, E. and Schuble, P. (1996). "Document and passage retrieval based on hidden Markov models". In Proceedings of the 19th Annual International of Association of Computer Machinery - Special Interest Group on Information Retrieval (A CM / SIGIR) Conference on Research and Development in Information Retrieval, pp. 318-327.

....of passage retrieval as a query evaluation mechanism. Many definitions of passage have been proposed in the literature, including passages based on: document markup [8, 10, 25, 26, 32, 33, 39] such as sections, paragraphs, and groups of sentences; boundaries at which topic shifts can be inferred [16, 19]; sequences of paragraphs of similar aggregate length [39] and fixed length sequences of words [7, 16, 17, 29] which can be either disjoint or overlapping. Several of these definitions of passage rely on semantic properties such as sentence boundaries, and intuitively it is reasonable to suppose ....

E. Mittendorf and P. Schauble. Document and passage retrieval based on hidden Markov models. In W.B. Croft and C.J. van Rijsbergen, editors, Proceedings of the Seventeenth Annual International ACM-SIGIR Conference on Research and Development in Information Retrieval, pages 318--327, July 1994.

....weights for each word. An alternative access method, which is the topic of this paper, is to regard each document as a set of passages , where a passage is a contiguous block of text. Instead of computing the similarity of each document to a query, a similarity is computed for each passage [5, 14, 23, 32, 47, 52]. The units retrieved can then be the documents from which the most similar passages are drawn so that passages provide an alternative mechanism for document ranking or can be the passages themselves. Passage level access has several advantages over document level access. First, if passages ....

....weights are calculated using the prior and current concentrations of terms in text. This approach has a more theoretical framework than fixed length arbitrary passages but incorporates many variables and is computationally expensive [22] Instead of defining passages, Mittendorf and Schauble [23] use inferred passage boundaries, by employing a hidden Markov model to determine passages appropriate to each query. This approach is analogous to TextTiling [14] but passage boundaries are determined at query time instead of indexing time. This approach necessitates processing of the full text ....

E. Mittendorf and P. Schauble. Document and passage retrieval based on hidden Markov models.InB.W.CroftandC.J.vanRijsbergen,editors,Proceedings of the 17th Annual International pages 318--327, Dublin, Ireland, July 1994.

....URL: http: www.foyer.de euromedia 2 By a video item we understand any coherent video material; this may be a short report of only a few minutes, a whole documentary or feature film etc. 3 At first site, video retrieval might seem similiar to passage (text) retrieval (Cormack et al. 1999, Mittendorf et al. 1996), but retrieving relevant text passages in a document is only half of the story of video retrieval. The goal here is not to identify and rank documents by taking account of relevant passages included there, but This basically means that recall and precision in their original definition are no ....

Mittendorf, E. & Schuble, P. (1996): Document and Passage Retrieval Based on Hidden Markov Models. Proceedings of SIGIR96, p. 318 - 327.

....searches) without affecting the existing applications. b) The IR system may incorporate more sophisticated query evaluation techniques (e.g. similarity thesauri, automatic query expansion, etc. and useful document inspection tools like highlighting words of the query or passage retrieval [MS94] which are difficult to build into a DB server. c) The throughput of query evaluations and updates within a single IR server is optimal because the IR server s work is reduced to a minimum: indexing documents and IR queries is performed outside the IR server by the synchronizer (if necessary ....

Mittendorf, E., and Schauble, P. (1994). Document and Passage Retrieval Based on Hidden Markov Models. In ACM SIGIR Conference on R&D in Information Retrieval, pp. 318--327.

....storing, retrieving, and managing the data in multimedia systems. Recent papers related to multimedia database systems can be categorized in the following application domains: speech recognition, word recognition, signal processing, handwriting recognition, and document passage retrieval [1] 6] [7] [8] 9] However, the focus of the above researches is on the lowlevel feature recognition of multimedia data; while our approach addresses the need for a mechanism at the database management point of view. Toward this end, we have proposed a unified model that 1 allows us to query different ....

E. Mittendorf and P. Schauble, "Document and passage retrieval based on Hidden Markov Models, " In ACM-SIGIR Conference on Research and Development in Information Retrieval, pp. 318-327, 1994.

....attenuated, we shall refer to the HMM as a four parameter model, allowing us to concentrate on the state transitions and character emissions. HMM s have been used widely in speech recognition [22] and recently also in the prediction of protein structure [1, 15] and information retrieval research [20]. In our application the model has two states, one for being in a cluster and another one for being between clusters. In Figure 1 we have an example of an HMM, taken from TLF: it corresponds to the word extravagant, which appears in 392 of the 39000 documents. The probability of starting in state ....

Mittendorf E., Schauble P., Document and passage retrieval based on Hidden Markov models, Proc. 17-th ACM-SIGIR Conf., Dublin (1994) 318--327.

....The TPI model [6] is a combination of the binary independence model and the 2 Poisson model. A previous approach which extended the binary independence model to cover multiple term occurrences is described in [11] and [12] Another approach in which text is generated by a stochastic process is [4], which employs a hidden Markov model. 6 Conclusions and future work The multinomial model gives a simple account of how tf can come into probabilistic models, through modeling the generation of text by a stochastic process. It also takes document length into account in a new and well justified ....

Mittendorf, Elke and Peter Sch¨auble. Document and passage retrieval based on hidden Markov models. Proc. ACM SIGIR Conference on R & D in Information Retrieval pp. 318-327, 1994

....sentences have very many, or indeed any, words in common. In that case, measuring similarity of within topic sentences does not provide enough information. As mentioned earlier, much of the work on passage retrieval has used passages of fixed length. An exception to this is Mittendorf and Sh auble [5] in which Hidden Markov Models (HMMs) were used to retrieve relevant passages of variable length. This is an interesting approach and is somewhat related to our work in that in both cases the text is broken up using a sequential (Markov) decision process. The difference is that in their approach, ....

Mittendorf E. and P. Sh¨auble, "Document and Passage Retrieval Based on Hidden Markov Models", In Proceedings of the Seventeenth Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, Dublin, Ireland, July, 1994 (pp. 318-327).

....segment of each document; in other words, that their discussion must overlap, rather than be distributed throughout the document. It is important to realize that this strategy is different than the best segment or best passage strategies that have been explored recently in the literature ([14, 31, 25, 24, 4]) in that documents are not assigned a rank based on how well the best segment matches the query; rather, the algorithm simply eliminates from initial consideration documents that do not have any promising subtopic units at all. 2.4 Use in Interactive Information Access This kind of query ....

Elke Mittendorf and Peter Schauble. Document and passage retrieval based on hidden markov models. In Proceedings of the 17th Annual International ACM/SIGIR Conference, pages 318--327, Dublin, Ireland, 1994.

....searches) without affecting the existing applications. b) The IR system may incorporate more sophisticated query evaluation techniques (e.g. similarity thesauri, automatic query expansion, etc. and useful document inspection tools like highlighting words of the query or passage retrieval [Mittendorf and Schauble, 1994] which are difficult to build into a DB server. c) The throughput of query evaluations and updates within a single IR server is optimal because the IR server s work is reduced to a minimum: indexing documents and IR queries is performed outside the IR server by the synchronizer (if necessary ....

Mittendorf, E., and Schauble, P. (1994). Document and Passage Retrieval Based on Hidden Markov Models. In ACM SIGIR Conference on R&D in Information Retrieval, pp. 318--327.

....respect to r, the estimation formula (1) is not yet realistic because the constant c does not have the same effect for low and high values of r. The influence of c on the retrieval effectiveness is subject to further investigations. 3 Passage Retrieval In previous experiments [Knaus et al. 1994, Mittendorf Schauble, 1994] we have shown that Hidden Markov Models (HMM) provide a natural and effective method for retrieving relevant passages from documents. In this section we summarize the ideas of HMM based passage retrieval and we describe the refinements and changes we have made in the passage retrieval method ....

Mittendorf, E., & Schauble, P. (1994). Document and Passage Retrieval Based on Hidden Markov Models. In ACM SIGIR Conference on R&D in Information Retrieval, pp. 318--327.

....method is called the basic method B. It consists of a well known weighting scheme which is sometimes referred to as lnc:ltc (Knaus Schauble, 1993) Passage retrieval is used as a second source of evidence of relevance. We used our passage retrieval method P which is based on Hidden Markov Models (Mittendorf Schauble, 1994). This approach has the advantage that the passages are not restricted to a fixed length. Furthermore, the parameters can be estimated automatically by means of the Baum Welch reestimation formula. A so called link method L is used as a third source of evidence of relevance. The link method was ....

....(HMM) and the Viterbi algorithm provide a natural method for retrieving passages without knowing anything about the structure of documents and without assuming anything about the format and the size of the passages. Document and passage retrieval based on Hidden Markov Models are described in (Mittendorf Schauble, 1994). For our purposes we roughly assume that with respect to a query each document can be segmented into three passages: an irrelevant passage, followed by a relevant passage, again followed by an irrelevant passage. The HMM which models these assumptions is visualized in Figure 1. An HMM can be ....

Mittendorf, E., & Schauble, P. (1994). Document and Passage Retrieval Based on Hidden Markov Models.

No context found.

Mittendorf, E. and Schauble, P. (1994). Document and passage retrieval based on hidden markov models. In Proceedings of SIGIR'94, pages 318--327.

No context found.

E. Mittendorf, P. Schauble. Document and passage retrieval based on hidden Markov models. In the Proceedings of the 17 Annual International ACM SIGIR Conference on Research and Development in IR,

No context found.

E. Mittendorf and P. Schauble. Document and passage retrieval based on hidden markov models. In Proceedings of the Seventeenth Annual International ACM SIGIR Conference on Research and Development in Information Retrieval, Dublin, Ireland, 1994.

No context found.

Mittendorf E. and Schauble P., Document and Passage Retrieval Based on Hidden Markov Models. In Proceedings of the 17th Annual International Conference on Research and Development in Information Retrieval, pages 318-327, July 3-6, Dublin, Ireland, 1994.

No context found.

MITTENDORF, E. And SCHAUBLE, P. Document and passage Retrieval Based on Hidden Markov Models. In: Proceedings of SIGIR'94, Dublin, Springer-Verlag, 318-327, 1994.

No context found.

E. Mittendorf and P. Sch¨auble. Document and passage retrieval based on hidden markov models. In Proc. ACM-SIGIR International Conference on Research and Development in Information Retrieval, pages 318-- 327, Dublin, Ireland, 1994.

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