## DT Tutor: A Decision-Theoretic, Dynamic Approach for Optimal Selection of Tutorial Actions (2000)

### Cached

### Download Links

- [www.pitt.edu]
- [www.pitt.edu]
- [www.public.asu.edu]
- [www.public.asu.edu]
- [www.cs.ubc.ca]
- DBLP

### Other Repositories/Bibliography

Venue: | Proceedings of Intelligent Tutoring Systems, 5th International Conference, ITS2000 |

Citations: | 23 - 3 self |

### BibTeX

@INPROCEEDINGS{Murray00dttutor:,

author = {R. Charles Murray and Kurt Vanlehn},

title = {DT Tutor: A Decision-Theoretic, Dynamic Approach for Optimal Selection of Tutorial Actions},

booktitle = {Proceedings of Intelligent Tutoring Systems, 5th International Conference, ITS2000},

year = {2000},

pages = {153--162},

publisher = {Springer}

}

### Years of Citing Articles

### OpenURL

### Abstract

DT Tutor uses a decision-theoretic approach to select tutorial actions for coached problem solving that are optimal given the tutor's beliefs and objectives.

### Citations

962 |
Human problem solving
- Newell, Simon
- 1972
(Show Context)
Citation Context ...a bias corresponds to a depth-first traversal of the problem solution graph and is consistent with both activation-based theories of human working memory [1] and observations of human problem solvers =-=[18]-=-. However, a depth-first bias is not absolute. At any given step, there is some probability that a student will not continue depth-first. To model depth-first bias, when a step first becomes ready or ... |

601 | On the Optimality of the Simple Bayesian Classifier under Zero-One Loss
- Domingos, Pazzani
- 1997
(Show Context)
Citation Context ...e often surprisingly insensitive to imprecision in specification of numerical probabilities [8] and may be accurate enough to infer the correct decision even if some of their assumptions are violated =-=[6], -=-so that precise numbers may not always be necessary. This research has shown that a decision-theoretic approach can indeed be used to select tutorial actions that are optimal, given the tutor’s beli... |

311 |
Decisions with Multiple Objectives
- Keeney, Raiffa
- 1976
(Show Context)
Citation Context ...N can be represented while keeping in memory at most two slices at a time [10]. Decision theory extends probability theory to provide a normative theory of how a rational decision-maker should behave =-=[13]-=-. Utilities are used to express preferences among possible future states of the world. To decide among alternative actions, the expected utility of each alternative is calculated by taking the sum of ... |

156 |
Simulation approaches to general probabilistic inference on belief networks
- Shachter, Peot
- 1990
(Show Context)
Citation Context ...1,000 samples were correct on every trial. Table 1. Action selection response times Response time mean (range) Algorithm Problem A a Problem B b Exact: Clustering [9] Approximate: Likelihood Sampling =-=[22]-=- 108 (107-109) 11 (11-12) 1,000 samples 12 (12-13) 8 (7-8) 10,000 samples Heuristic Importance [22] 106 (104-110) 64 (62-66) 1,000 samples 12 (12-13) 8 (7-8) 10,000 samples 104 (101-106) 64 (60-66) No... |

149 | A.: Inference in belief networks: A procedural guide
- Huang, Darwiche
- 1996
(Show Context)
Citation Context ...ed, the approximate algorithms using 1,000 samples were correct on every trial. Table 1. Action selection response times Response time mean (range) Algorithm Problem A a Problem B b Exact: Clustering =-=[9]-=- Approximate: Likelihood Sampling [22] 108 (107-109) 11 (11-12) 1,000 samples 12 (12-13) 8 (7-8) 10,000 samples Heuristic Importance [22] 106 (104-110) 64 (62-66) 1,000 samples 12 (12-13) 8 (7-8) 10,0... |

117 | On-line student modeling for Coached problem solving using Bayesian Networks
- Conati, Gertner, et al.
- 1997
(Show Context)
Citation Context ...r’s beliefs about the student’s problem-related knowledge using a belief network obtained directly from a problem solution graph, a hierarchical dependency network representing solutions to a prob=-=lem [2, 11]-=-. Nodes in the graph represent (1) problem steps, and (2) domain rules licensing each step. Problem steps include the givens and every goal and fact along any path towards the solution. We currently m... |

105 | Numerical Uncertainty Management in User and Student Modeling: An Overview of Systems and Issues. User Modeling and User-Adapted Interaction
- Jameson
- 1995
(Show Context)
Citation Context ... of a DDN to tutoring. Probabilistic reasoning is often used in student and user modeling. In particular, Bayesian networks are used in the student models of Andes [2], HYDRIVE [16] and other systems =-=[12]-=-. However, even with a probabilistic student model, other systems select tutorial actions using heuristics instead of decision-theoretic methods. Reye [19] has suggested the use of a decision-theoreti... |

103 |
Motivational techniques of expert human tutors: Lessons for the design of computer-based tutors
- Lepper, Woolverton, et al.
- 1993
(Show Context)
Citation Context ... the more distant past, so the more recently raised steps remain more relevant. Student Emotional State. Human tutors consider the student’s emotional or motivational state in deciding how to respon=-=d [5, 14]-=-. Concern for student morale is likely to be one reason why tutors tend to give negative feedback subtly, to play up student successes and downplay student failures, etc., while maximizing the student... |

100 |
Rules of the Mind, Lawrence Erlbaum Associates
- Anderson
- 1993
(Show Context)
Citation Context ...k on it before starting work on another. Such a bias corresponds to a depth-first traversal of the problem solution graph and is consistent with both activation-based theories of human working memory =-=[1]-=- and observations of human problem solvers [18]. However, a depth-first bias is not absolute. At any given step, there is some probability that a student will not continue depth-first. To model depth-... |

84 |
Collaborative dialogue patterns in naturalistic one-toone tutoring. Applied cognitive psychology
- Graesser, Person, et al.
- 1995
(Show Context)
Citation Context ...interface, so it has not been evaluated with human students. 1 Introduction Tutoring systems that coach students as they solve problems often emulate the turn taking observed in human tutorial dialog =-=[7, 15]. -=-Student turns usually consist of entering a solution step or asking for help. The tutor’s main task can be seen as simply deciding what action to take on its turn. Tutorial actions include a variety... |

75 |
A method for using belief networks as influence diagrams
- Cooper
- 1988
(Show Context)
Citation Context ...ds required to determine the optimal tutorial action. Mean and range are over 10 trials. All tests were performed on a 200MHz Pentium MMX PC with 64MB of RAM. The algorithms were tested with Cooper’=-=s [3]-=- algorithm for decision network inference using belief network algorithms. a 10-step problem, 185-node TACN. b 5-step problem, 123-node TACN.sIn G. Gauthier, C. Frasson, & K. VanLehn (Ed.), Intelligen... |

67 | Automated symbolic traffic scene analysis using belief networks
- Huang, Koller, et al.
- 1994
(Show Context)
Citation Context ...s (synchronic influences) and on earlier values of the same and other variables (diachronic influences). The evolution of a DBN can be represented while keeping in memory at most two slices at a time =-=[10]-=-. Decision theory extends probability theory to provide a normative theory of how a rational decision-maker should behave [13]. Utilities are used to express preferences among possible future states o... |

66 | The automated mapping of plans for plan recognition
- Huber, Durfee, et al.
- 1994
(Show Context)
Citation Context ...r’s beliefs about the student’s problem-related knowledge using a belief network obtained directly from a problem solution graph, a hierarchical dependency network representing solutions to a prob=-=lem [2, 11]-=-. Nodes in the graph represent (1) problem steps, and (2) domain rules licensing each step. Problem steps include the givens and every goal and fact along any path towards the solution. We currently m... |

57 |
The role of probability-based inference in an intelligent tutoring system. User-Modeling and User Adapted Interaction
- Mislevy, Gitomer
- 1996
(Show Context)
Citation Context ...s the first application of a DDN to tutoring. Probabilistic reasoning is often used in student and user modeling. In particular, Bayesian networks are used in the student models of Andes [2], HYDRIVE =-=[16]-=- and other systems [12]. However, even with a probabilistic student model, other systems select tutorial actions using heuristics instead of decision-theoretic methods. Reye [19] has suggested the use... |

52 | B.: Implementation of motivational tactics in tutoring systems
- Soldato, T
- 1995
(Show Context)
Citation Context ... the more distant past, so the more recently raised steps remain more relevant. Student Emotional State. Human tutors consider the student’s emotional or motivational state in deciding how to respon=-=d [5, 14]-=-. Concern for student morale is likely to be one reason why tutors tend to give negative feedback subtly, to play up student successes and downplay student failures, etc., while maximizing the student... |

23 |
Tutoring: Guided learning by doing
- Merrill, Reiser, et al.
- 1995
(Show Context)
Citation Context ...interface, so it has not been evaluated with human students. 1 Introduction Tutoring systems that coach students as they solve problems often emulate the turn taking observed in human tutorial dialog =-=[7, 15]. -=-Student turns usually consist of entering a solution step or asking for help. The tutor’s main task can be seen as simply deciding what action to take on its turn. Tutorial actions include a variety... |

16 |
A tutorial introduction to stochastic simulation algorithms for belief networks
- Cousins, Chen, et al.
- 1993
(Show Context)
Citation Context ...rk inference is still NP-hard in the worst case. However, many stochastic sampling algorithms have an anytime property that allows an approximate result to be obtained at any point in the computation =-=[4]. DT-=- Tutor represents the tutor’s beliefs about the student’s problem-related knowledge using a belief network obtained directly from a problem solution graph, a hierarchical dependency network repres... |

16 |
The reification of goal structures in a calculus tutor: Effects on problem solving performance. Interactive Learning Environments
- Singley
- 1990
(Show Context)
Citation Context ...rates problems for two reasons. First, the number of steps per problem is non-trivial without being too large, so results obtained should be generalizable to other real world domains. Second, Singley =-=[23] d-=-eveloped an interface for this domain with the purpose of reifying goal structures. We assume an extension to Singley’s interface that makes all problem solving actions observable. This makes it eas... |

13 |
Two-phases updating of student models based on dynamic belief networks
- Reye
- 1998
(Show Context)
Citation Context ...nstead of decision-theoretic methods. Reye [19] has suggested the use of a decision-theoretic architecture for tutoring systems and the use of dynamic belief networks to model the student’s knowledg=-=e [20, 21]. -=-2 Detailed Solution This section describes TACNs in more detail, along with their implementation to form DT Tutor’s action selection engine. 2.1 Major TACN Components and Their Interrelationships Fi... |

11 |
A belief net backbone for student modeling
- Reye
- 1996
(Show Context)
Citation Context ...nstead of decision-theoretic methods. Reye [19] has suggested the use of a decision-theoretic architecture for tutoring systems and the use of dynamic belief networks to model the student’s knowledg=-=e [20, 21]. -=-2 Detailed Solution This section describes TACNs in more detail, along with their implementation to form DT Tutor’s action selection engine. 2.1 Major TACN Components and Their Interrelationships Fi... |

10 |
W.B.: Human tutoring : why do only some events cause learning
- Lehn, Siler, et al.
- 2003
(Show Context)
Citation Context ...qual, DT Tutor preferred to address rules rather than problem-specific steps. Effective human tutoring is correlated with teaching generalizations that go beyond the immediate problem-solving context =-=[24]. • DT-=- Tutor considered the effects of its actions on the student’s emotional state as well as the student’s knowledge state. Human tutors consider both as well [14]. • DT Tutor prioritized its action... |

7 |
A goal-centred architecture for intelligent tutoring systems
- Reye
- 1995
(Show Context)
Citation Context ...of Andes [2], HYDRIVE [16] and other systems [12]. However, even with a probabilistic student model, other systems select tutorial actions using heuristics instead of decision-theoretic methods. Reye =-=[19] h-=-as suggested the use of a decision-theoretic architecture for tutoring systems and the use of dynamic belief networks to model the student’s knowledge [20, 21]. 2 Detailed Solution This section desc... |

5 |
Why is diagnosis in belief networks insensitive to imprecision in probabilities
- Henrion, Pradhan, et al.
- 1996
(Show Context)
Citation Context ... would be beneficial as well. However, an encouraging result from prior research is that Bayesian systems are often surprisingly insensitive to imprecision in specification of numerical probabilities =-=[8]-=- and may be accurate enough to infer the correct decision even if some of their assumptions are violated [6], so that precise numbers may not always be necessary. This research has shown that a decisi... |

2 |
A dynamic, decision-theoretic model of tutorial action selection
- Murray
- 1999
(Show Context)
Citation Context ...he exact algorithm on the larger problem did not return responses quickly enough, and in any case, faster response times are desirable. Fortunately, a number of speedups are feasible, as discussed in =-=[17]-=-. In addition, the anytime property of the approximate algorithms could be used to continually improve results until a response is required. For many applications, including this one, it is sufficient... |