|
[Clustering] [Order] [Other]
Supervised Clustering is a hybrid task combining features of two common grouping tasks: supervised classification and and clustering. In supervised clustering, each training example is a partition of objects.The task is then to learn from a training set, a rule for partitioning unseen object sets. A general method for learning such partitioning rules is useful in any situation where explicit algorithms for deriving partitions are hard to formalize, while individual examples of correct partitions are easy to specify. In the past, clustering techniques have been applied to such problems, despite being essentially unsuited to the task.
Related Publications [ 031 ]
The term order as a sequence of objects that are sorted according to some property, such as size, preference, or price. Lists of ordered objects are widely used as representational forms. Such ordered objects include Web search results or best seller lists. The analysis of orders is useful for sensory surveys, information retrieval, or decision making.
Supervised Ordering task is composed of two stages: learning and sorting. In the learning stage, a sorting function is acquired from given sample orders. In the sorting stage, the sorting function is used for sorting a unordered object set so that are concordant with an attributed central order (ACO).
More specifically, the supervised ordering task is as follows. Sorted objects are the members of the universal object set X*. In the case of the right figure, X*={A,B,C,D,E}. A supervised ordering task can be considered a regression task whose target variable is ordinal. A regression line corresponds to an ACO. Analogous to the case of a regression line, an ACO is estimated so as to be concordant with both given samples S and unseen samples S', to be generated. Note that an ACO consists of all the objects in X*. This task differs from a regression in two ways. First, the target variable is ordinal. Therefore, the discordance is measured by the distance between orders, such as Spearman's or Kendall's distance. Further, an ACO is represented by a sorting function that sorts unordered objects so as to be concordant with the ACO. Second, almost all the samples are incomplete, that is, sample orders consist of subsets of X*.Hence, there may be objects not observed in given samples (ex., D in the figure). Such objects should be ranked under the assumption that the neighboring objects in the attribute space would be nearly ranked.
Related Publications [ 070, 057, 048, 033 ]
A method of using clustering techniques to partition a set of orders. k-o'means algorithm is a modified version of a k-means method, adjusted to handle orders.
Related Publications [ 069, 042 ]
A recommender system suggests the items expected to be preferred by the users. Recommender systems use collaborative filtering to recommend items by summarizing the preferences of people who have tendencies similar to the user preference (see right Figure). Traditional collaborative filtering algorithms adopted the semantic differential method (Figure (a)), in which preferences are measured using an n-point-scale on which extremes are represented by antonyms. We propose some collaborative filtering algorithms adopting the ranking method (Figure (b)). In the ranking method, the preferences are represented by orders, which are sorted item sequences according to the users' preferences.
Related Publications [ 064, 049, 039 ]
An experimental multimodal disputation system, Mr.Bengo, is a knowledge based system with multimodal user interfaces composed of modules for face recognition, face animation, speech recognition, speech synthesis, and text interface. In a virtual court, an attorney agent interact with users and disputes with a prosecutor agent.
 |
"Mr.Bengo" Homepage (in Japanese) Related Publications [ 013, Demo Video ] |
|