Abstract not found

This paper presents the theory, design principles, implementation, and performance results of PicHunter, a prototype content-based image retrieval (CBIR) system that has been developed over the past three years. In addition, this document presents the rationale, design, and results of psychophysical experiments that were conducted to address some key issues that arose during PicHunter’s development. The PicHunter project makes four primary contributions to research on content-based image retrieval. First, PicHunter represents a simple instance of a general Bayesian framework we describe for using relevance feedback to direct a search. With an explicit model of what users would do, given what target image they want, PicHunter uses Bayes’s rule to predict what is the target they want, given their actions. This is done via a probability distribution over possible image targets, rather than by refining a query. Second, an entropy-minimizing display algorithm is described that attempts to maximize the information obtained from a user at each iteration of the search. Third, PicHunter makes use of hidden annotation rather than a possibly inaccurate/inconsistent annotation structure that the user must learn and make queries in. Finally, PicHunter introduces two experimental paradigms to quantitatively evaluate the performance of the system, and psychophysical experiments are presented that support the theoretical claims.

The standard strategies for evaluation based on precision and recall are examined and their relative advantages and disadvantages are discussed. In particular, it is suggested that relevance feedback be evaluated from the perspective of the user. A number of different statistical tests are described for determining if differences in performance between retrieval methods are significant. These tests have often been ignored in the past because most are based on an assumption of normality which is not strictly valid for the standard performance measures. However, one can test this assumption using simple diagnostic plots, and if it is a poor approximation, there are a number of non-parametric alternatives.

We use data from the TREC routing experiments to explore how relevance feedback can be applied incrementally --- using a few judged documents each time --- to achieve results that are as good as if the feedback occurred in one pass. We show that relatively few judgments are needed to get highquality results. We also demonstrate methods that reduce the amount of information archived from past judged documents without adversely affecting effectiveness. A novel simulation shows that such techniques are useful for handling long-standing queries with drifting notions of relevance.

Query Refinement is an essential information retrieval tool that interactively recommends new terms related to a particular query. This paper introduces concept recall, an experimental measure of an algorithm's ability to suggest terms humans have judged to be semantically related to an information need. This study uses precision improvement experiments to measure the ability of an algorithm to produce single term query modifications that predict a user's information need as partially encoded by the query. An oracle algorithm produces ideal query modifications, providing a meaningful context for interpreting precision improvement results. This study also introduces RMAP, a fast and practical query refinement algorithm that refines multiple term queries by dynamically combining precomputed suggestions for single term queries. RMAP achieves accuracy comparable to a much slower algorithm, although both RMAP and the slower algorithm lag behind the best possible term suggestions o ered by the oracle. We believe RMAP is fast enough to be integrated into present dayInternet search engines: RMAP computes 100 term suggestions for a 160,000 document collection in 15 ms on a low-end PC.

Push-based data delivery requires knowledge of user interests for making scheduling, bandwidth allocation, and routing decisions. Such information is maintained as user profiles. We propose a new incremental algorithm for constructing user profiles based on monitoring and user feedback. In contrast to earlier approaches, which typically represent profiles as a single weighted interest vector, we represent user profiles as multipleinterest vectors, whose number, size, and elements change adaptively based on user access behavior. This flexible approach allows the profile to more accurately represent complex user interests. Although there has been significant research on user profiles, our approach is unique in that it can be tuned to trade off profile complexity and quality. This feature, together with its incremental nature, makes our method suitable for use in large-scale information filtering applications such as push-based WWW page dissemination. We evaluate the method by experimentally investigating its ability to categorize WWW pages taken from Yahoo! categories. Our results show that the method can provide high filtering effectiveness with modest profile sizes and can effectively adapt to changes in users’ interests. 1.

Traditional information retrieval (IR)... This paper shows that, in systems that allow access to non-matching documents (e.g. hybrid hypertext and information retrieval systems), the strength of the effect of giving relevance feedback varies between matching and non-matching documents. For positive feedback the results shown here are encouraging as they can be justified by an intuitive view of the process. However, for negative feedback the results show behaviour that cannot easily be justified and that varies greatly depending on the model of feedback used.

Abstract not found

We study the problem of organizing the documents returned by an information retrieval system in response to a natural language query. We consider two well-known approaches -- the ranked list and clustering of the results -- and we show how they can be integrated. This new procedure is designed to accept user feedback and direct the user toward the relevant material as effectively as the traditional relevance feedback approach. We show how our technique can be explained to the user by visualizing the process in two or three dimensions, providing him or her with complete control of the procedure. We show that increasing the dimensionality of the visualization generally improves its quality, albeit only a small amount. Additionally we present the result of a small user study designed to investigate how effective our visualization is in supporting the user navigating the retrieved results.

The rst part of the paper brie y introduces what automatic authoring of a hypertext for information retrieval means. The most di cult part of the automatic construction of a hypertext is the creation of links connecting documents or document fragments that are semantically related. Because of this, to many researchers it seemed natural to use IR techniques for this purpose, since IR has always dealt with the construction of relationships between objects mutually relevant. The second part of the paper presents a survey of some of attempts toward the automatic construction of hypertexts for information retrieval. This part will identify and compare scope, advantages and limitations of di erent approaches. The aim of this survey is to point out the main and most successful current lines of research.