Recommender systems have become valuable resources for users seeking intelligent ways to search through the enormous volume of information available to them. One crucial unsolved problem for recommender systems is how best to learn about a new user. In this paper we study six techniques that collaborative filtering recommender systems can use to learn about new users. These techniques select a sequence of items for the collaborative filtering system to present to each new user for rating. The techniques include the use of information theory to select the items that will give the most value to the recommender system, aggregate statistics to select the items the user is most likely to have an opinion about, balanced techniques that seek to maximize the expected number of bits learned per presented item, and personalized techniques that predict which items a user will have an opinion about. We study the techniques thru offline experiments with a large preexisting user data set, and thru a live experiment with over 300 users. We show that the choice of learning technique significantly affects the user experience, in both the user effort and the accuracy of the resulting predictions.

This paper surveys methods for representing and reasoning with imperfect information. It opens with an attempt to classify the different types of imperfection that may pervade data, and a discussion of the sources of such imperfections. The classification is then used as a framework for considering work that explicitly concerns the representation of imperfect information, and related work on how imperfect information may be used as a basis for reasoning. The work that is surveyed is drawn from both the field of databases and the field of artificial intelligence. Both of these areas have long been concerned with the problems caused by imperfect information, and this paper stresses the relationships between the approaches developed in each.

To date, software engineers charged with debugging complex software packages have had few automated reasoning tools to assist them with identifying the sources of error and with prioritizing their effort. We describe methods, based on a synthesis of logical and probabilistic reasoning, that can be employed to identify the likely source and location of problems in complex software. The methods have been applied to diagnosing run-time errors in the Sabre system, the largest timeshared reservation system in the world. The results from our validation suggest that such methods can be of value in directing the attention of software engineers to program execution paths and program instructions that have the highest likelihood of harboring a programming error. Keywords: Software maintenance, decision theory, automated diagnosis, probability, Bayesian reasoning Introduction Software errors abound in the world of computing. Sophisticated computer programs rank high on the list of the most comple...

We present a method for calculation of myopic value of information in influence diagrams (Howard & Matheson, 1981) based on the strong junction tree framework (Jensen et al., 1994) . An influence diagram specifies a certain order of observations and decisions through its structure. This order is reflected in the corresponding junction trees by the order in which the nodes are marginalized. This order of marginalization can be changed by table expansion and use of control structures, and this facilitates for calculating the expected value of information for different information scenarios within the same junction tree. In effect, a strong junction tree with expanded tables may be used for calculating the value of information between several scenarios with different observation-decision order. We compare our method to other methods for calculating the value of information in influence diagrams.

We review the functionality of a modular, component-based tool kit for Bayesian network development and inference. Beyond its operation as a standalone modeling and inference environment, MSBNx facilitates the development of standalone applications by providing a set of run-time components that provide Bayesian reasoning services when integrated into other programs. MSBNx inferential operations provide both inference about states of inference and about the value of information for unobservered evidence. The services and modeling environment supports both diagnostic and troubleshooting mingles observations and repair operations. MSBNx facilitates the development and use of new add-in components. The modeling environment provides a means for assessing distinctions and beliefs, and special interfaces and tools for representing the asymmetric nature of probability distributions. 1.

Sequential decision models are an important element of expert system optimization when the cost or time to collect inputs is significant and inputs are not known until the system operates. Many expert systems in business, engineering, and medicine have benefited from sequential decision technology. In this survey, we unify the disparate literature on sequential decision models to improve comprehensibility and accessibility. We separate formulation of sequential decision models from solution techniques. For model formulation, we classify sequential decision models by objective (cost minimization versus value maximization) knowledge source (rules, data, belief network, etc.), and optimized form (decision tree, path, input order). A wide variety of sequential decision models are discussed in this taxonomy. For solution techniques, we demonstrate how search methods and heuristics are influenced by economic objective, knowledge source, and optimized form. We discuss open research problems to stimulate additional research and development. 1.

. Effective history-taking systems need to dynamically reduce the number of questions to ask. This can be done either categorically or probabilistically, by exploiting previous patient's answers. In this paper, we propose a probabilistic information-based history-taking strategy that combines synergistically two information-content measures for reducing the number of questions asked. We have applied this strategy to an existing history-taking system and some preliminary results seem to confirm our initial intuitions. 1 Introduction History-taking systems collect information from patients about their condition, past-current treatments, habits, and family history [9]. To ask the whole predetermined set of questions available to the history taker, without regard to their relevance to the specific case, can be extremely slow and tedious. Patients become bored when they are asked too many questions, and when a patient is bored and feels the conversation is unnatural, the quality of the inf...

DIAVAL is an expert system for the diagnosis of heart diseases, based on several kinds of data, mainly from echocardiography. The first part of this paper is devoted to the causal probabilistic model which constitutes the knowledge base of the expert system in the form of a Bayesian network, emphasizing the importance of the OR gate. The second part deals with the process of diagnosis, which consists of computing the a posteriori probabilities, selecting the most probable and most relevant diagnoses, and generating a written report. It also describes the results of the evaluation of the program.

AbstractÐThe fatality rate associated with severe bacterial infections is about 30 percent and appropriate antibiotic treatment reduces it by half. Unfortunately, about a third of antibiotic treatments prescribed by physicians are inappropriate. We have built a causal probabilistic network (CPN) for treatment of severe bacterial infections. The net is based on modules, each module representing a site of infection. The general configuration of a module is as follows: Major distribution factors define groups of patients, each of them with a definite prevalence of infection caused by a given pathogen. Minor distribution factors multiply the likelihood of one pathogen, without changing much of the prevalence of infection. Infection caused by a pathogen causes local and generalized signs and symptoms. Antibiotic treatment is appropriate if it matches the susceptibility of the pathogens in vitro and appropriate treatment is associated with a gain in life expectancy. This is balanced against the cost of the drug, side effects, and ecological damage, to reach the most cost effective treatment. The net was constructed in such a way that the data for the conditional probability tables will be available, even if it meant sometimes giving up on fine modeling details. For data, we used large databases collected by us in the last 10 years and data from the literature. The CPN was a convenient way to combine data from databases collected at different locations and times with published information. Although the net is based on detailed and large databases, its calibration to new sites requires data that is available in most modern hospitals. Index TermsÐCausal probabilistic networks, bacterial infections, bacteremia, antibiotic treatment, cost-effectiveness. 1

Abstract not found