Ross Wilkinson

Effective retrieval of structured documents

Information systems usually retrieve whole documents as answers to queries. However, it may in some circumstances be more appropriate to retrieve parts of documents. We consider formulas for retrieving whole documents and parts of documents horn a large structured document collection. We consider what information is needed to retrieve effectively and show that knowledge of the structure of documents can lead to improved retrieval performance.

Using the cosine measure in a neural network for document retrieval

An air separation process using pressure swing adsorption techniques, for providing high purity oxygen. Two sections are employed one comprising beds of molecular sieve carbon and the other comprising beds of zeolite molecular sieve. Air is fed to a first of the sections which provides an oxygen-rich gas stream as feedstock for the next section where further enrichment takes place. The zeolite sieve section serves to effect a separation as between oxygen and nitrogen while the carbon sieve section serves to effect a separation as between oxygen and argon and the processes performed at each section are integrated in such a manner as to minimize power consumption and make use of gas recycled from the second section to the first in addition to the flow of gas from the first section to the second.

Efficient retrieval of partial documents

Abstract Management and retrieval of large volumes of text can be expensive in both space and time. Moreover, the range of document sizes in a large collection such as TREC presents difficulties for both the retrieval mechanism and the user. We consider division of documents into parts as a solution to the problem of the range of document sizes, and show that, for databases with long documents, use of document parts can improve the quality of the information presented to the user. We also describe the compressed text database system we use to manage the TREC collection; the compressed inverted files with which it is indexed; and the techniques we use to evaluate the TREC queries, both on whole documents and on document parts.

Automatic link generation

In order to access any kind of stored information, one may store it at a specific location, and in the case of electronic information this could be a file name or a Web address. If the location is not known or the amount of information to be accessed is greater than the number of locations that can be remembered, then it is necessary to find the information based on its attributes, its content, or its relationships to other pieces of information whose location is known. In the first two cases, we search, as in information retrieval, while in the latter we navigate, as in hypertext and thus these two areas of hypertext and information retrieval are tightly related [Agosti 1996].

Preserving digital information forever

Well within our lifetime we can expect to see most information being created, stored and used digitally. Despite the growing importance of digital data, the wider community pays almost no attention to the problems of preserving this digital information for the future. Even within the archival and library communities most work on digital preservation has been theoretical, not practical, and highlights the problems rather than giving solutions. Physical libraries have to preserve information for long periods and this is no less true of their digital equivalents. This paper describes the preservation approach adopted in the Victorian Electronic Record Strategy (VERS) which is currently being trialed within the Victorian government, one of the states of Australia. We review the various preservation approaches that have been suggested and describe in detail encapsulation, the approach which underlies the VERS format. A key difference between the VERS project and previous digital preservation projects is the focus within VERS on the construction of actual systems to test and implement the proposed technology. VERS is not a theoretical study in preservation.

Using clustering and classification approaches in interactive retrieval

Satisfying non-trivial information needs involves collecting information from multiple resources, and synthesizing an answer that organizes that information. Traditional recall/precision-oriented information retrieval focuses on just one phase of that process: how to efficiently and effectively identify documents likely to be relevant to a specific, focused query. The TREC Interactive Track has as its goal the location of documents that pertain to different instances of a query topic, with no reward for duplicated coverage of topic instances. This task is similar to the task of organizing answer components into a complete answer. Clustering and classification are two mechanisms for organizing documents into groups. In this paper, we present an ongoing series of experiments that test the feasibility and effectiveness of using clustering and classification as an aid to instance retrieval and, ultimately, answer construction. Our results show that users prefer such structured presentations of candidate result set to a list-based approach. Assessment of the structured organizations based on the subjective judgement of the experiment subjects suggests that the structured organization can be more effective; however, assessment based on objective judgements shows mixed results. These results indicate that a full determination of the success of the approach depends on assessing the quality of the final answers generated by users, rather than on performance during the intermediate stages of answer construction.

Structured answers for a large structured document collection

There is a simple method for integrating information retrieval and hypertext. This consists of treating nodes as isolated documents and retrieving them in order of similarity. If the nodes are structured, in particular, if sets of nodes collectively constitute documents, we can do better. This paper shows how the formation of the hypertext, the retrieval of nodes in response to content based queries, and the presentation of the nodes can be achieved in a way that exploits the knowledge encoded as the structure of the documents. The ideas are then exemplified in an SGML based hypertext information retrieval system.

Understanding the relationship of information need specificity to search query length

When searching, peoples information needs flowthrough to expressing an information retrieval request posed to asearch engine. We hypothesise that the degree of specificity of anIR request might correspond to the length of a search query. Ourresults show a strong correlation between decreasing query lengthand increasing broadness or generality of the IR request. We foundan average cross-over point of specificity from broad to narrow of 3words in the query. These results have implications for searchengines in responding to queries of differing lengths.

Experiments in spoken document retrieval using phoneme n -grams

Abstract In spoken document retrieval (SDR), speech recognition is applied to a collection to obtain either words or subword units, such as phonemes, that can be matched against queries. We have explored retrieval based on phoneme n -grams. The use of phonemes addresses the out-of-vocabulary (OOV) problem, while use of n -grams allows approximate matching on inaccurate phoneme transcriptions. Our experiments explored the utility of word boundary information, stopword elimination, query expansion, varying the length of phoneme sequences to be matched and various combinations of n -grams of different lengths. Given word-based recognition (WBR), we can match queries to speech using a phoneme representation of the words, permitting us to test whether it was the recognition or the matching process that was most crucial to retrieval performance. Our experiments show that there is some deterioration in effectiveness, but the particular form of matching is less vital if the sequence of phonemes was correct. When phone sequences are recognised directly, with higher error rates than for words, it was more important to select a good matching approach. Varying gram length trades precision against recall; combination of n -grams of different lengths, in particular 3-grams and 4-grams, can improve retrieval. Overall, phoneme-based retrieval is not as effective as word-based retrieval, but is sufficient for situations in which word-based retrieval is either impractical or undesirable.

Information Environments for Supporting Consistent Registrar Medical Handover

This study was two-fold in nature. Initially, it examined the information environment and the use of customary information tools to support medical handovers in a large metropolitan teaching hospital on four weekends (i.e. Friday night to Monday morning). Weekend medical handovers were found to involve sequences of handovers where patients were discussed at the discretion of the doctor handing over; no reliable discussion of all patients of concern occurred at any one handover, with few information tools being used; and after a set of weekend handovers, there was no complete picture on a Monday morning without an analysis of all patient progress notes. In a subsequent case study, three information tools specifically designed as intervention that attempted to enrich the information environment were evaluated. Results indicate that these tools did support greater continuity in who was discussed but not in what was discussed at handover. After the intervention, if a doctor discussed a patient at handover, that patient was more likely to be discussed at subsequent handovers. However, the picture at Monday morning remained fragmentary. The results are discussed in terms of the complexities inherent in the handover process