Mark E. Rorvig

Images of similarity: a visual exploration of optimal similarity metrics and scaling properties of TREC topic-document sets

Multiple similarity measures for five TREC topic-document sets from the LDC TREC Collection Disk 1 are derived from the full text of documents. Each measure on each set is scaled using SAS MDS under ordinal, interval, and MLE assumptions. The resulting 75 permutations are ploted. It is suggested that cosine-vector and overlap measures for similarity appear to recover optimal data relationships among the documents of the five sets. MLE assumptions appear to be required to model the data adequately.

The Simple Scalability of Documents.

The relationship between scaling practice and scaling theory remains a controversial problem in Information Retrieval research and experimentation. This article reports a test of a general theory of scaling, i.e., Simple Scalability, applied to the stimulus domain of documents represented as abstracts. The significance of Simple Scalability is that it implies three important properties of scales: transitivity, substitutibility, and independence. The test results indicate that, with some reservations, this theory of scaling is applicable to documents. This finding is further applied to the construction of test collections for Information Retrieval research that could more sensitively measure retrieval system alterations through the use of documents scaled not merely by relevance, but rather, by preference.

The NASA image collection visual thesaurus

This article is an unpublished manuscript of a paper presented at the 17th Annual Mid-Year Meeting of the American Society for Information Science at Ann Arbor, Michigan, May 15–18, 1988. It is the first recorded proposal for visual support of indexing. The system itself was built and deployed at NASA in the Johnson Space Center Image Archives from 1989 to 1992.

A method for automatically abstracting visual documents

Visual documents—motion sequences on film, videotape, and digital recordings—constitute a major source of information for the Space Agency, as well as all other government and private sector entities. This article describes a method for automatically selecting key frames from visual documents. These frames may in turn be used to represent the total image sequence of visual documents in visual libraries, hypermedia systems, and training guides. The performance of the abstracting algorithm reduces 51 minutes of video sequences to 134 frames; a reduction of information in the range of 700:1.

Visualization and scaling of TREC topic document sets

Abstract Although numerous new document visualization tools are emerging throughout academia and industry, reliable test data for such tools has not yet been established. This paper explores the applicability of the TREC Information Retrieval Test Collection for this purpose using commonly available data and statistical methods.

A visual exploration of the orderliness of TREC relevance judgments

TREC topic specification statements 1-50 are converted to a similarity matrix, scaled, and plotted. Two close topics and two distant topics are selected from within the topic visual field. Subsequent scaling and visualization of documents associated with the close topics reveals a strong mixing of documents from both topic sets. Scaling and visualization of documents associated with the distant topics reveals a bifurcated distribution of documents from both topic sets. Relevant documents in both cases present near the center of both visualizations. Scaling and visualization of documents by multidimensional scaling using a maximum likelihood estimation method is shown to accurately model token similarity relationships among topic specification statements. The implications of these findings for prior critical arguments regarding IR test collections generally, and TREC specifically, by other scholars is examined.

A new machine classification method applied to human peripheral blood leukocytes

Human beings judge images by complex mental processes, whereas computing machines extract features. By reducing scaled human judgments and machine extracted features to a common metric space and fitting them by regression, the judgments of human experts rendered on a sample of images may be imposed on an image population to provide automatic classification.

Scaled structure in visualized TREC data and query feedback

Abstract A specific hypothesis arising from an earlier exploratory study (Rorvig & Fitzpatrick, 1998) was tested in this experiment. The data confirm that in the use of the query feedback technique in retrieval, query surrogate documents chosen from within dense inter-document structures appearing in scaled and visualized TREC Topic-document datasets raise the number of relevant documents subsequently retrieved by 26–46%, depending on the method of calculation. Use of documents as query surrogates which were chosen from the periphery of such dense inter-document structures did not differ significantly from the results retrieved by use of the topic specification document itself, submitted as a query. Some anomalies appear in one of the datasets from which a specific pattern is suggested for further exploration. Retrieval performance of IR systems on datasets for which relevant documents are located primarily outside dense inter-document structures was predicted to be poor in Rorvig and Fitzpatrick. This prediction was confirmed. It is suggested that some aspects of these findings may be incorporated into commercial IR systems.

Foundations of advanced information visualization for information retrieval systems

An adjusting tool for performing precision adjustment of rotary elements disposed in inaccessible areas including an elongated barrel. A driver is rotatably mounted on one end of the barrel and has a coupling configuration for mating with the rotary element to be adjusted. An elongated shaft is journalled within the barrel and a transmission connects the same to the driver. A knob is disposed at the opposite end of the barrel and is connected to the shaft to rotate the same. A scale is carried by the actuator for movement therewith and an index device is disposed in proximity to the scale and is mounted on the tool for movement relative to the scale. A brake is employed for selectively holding the index device against movement on the tool.

Modeling question-response patterns by scaling and visualization (poster session)

The evaluation of question difficulty is usually considered the domain of Latent Trait Theory. However, these methods require standardized question sets normalized by large populations, rendering them inefficient for use in the numerous areas where questions must be evaluated. A new technique is illustrated that models the question-response cycle well, but without the procedural difficulty of the traditional methods.