Masayuki Okabe

Semisupervised Query Expansion with Minimal Feedback

Query expansion is an information retrieval technique in which new query terms are selected to improve search performance. Although useful terms can be extracted from documents whose relevance is already known, it is difficult to get enough of such feedback from a user in actual use. We propose a query expansion method that performs well even if a user makes practically minimum effort, that is, chooses only a single relevant document. To improve searches in these conditions, we made two refinements to a well-known query expansion method. One uses transductive learning to obtain pseudorelevant documents, thereby increasing the total number of source documents from which expansion terms can be extracted. The other is a modified parameter estimation method that aggregates the predictions of multiple learning trials to sort candidate terms for expansion by importance. Experimental results show that our method outperforms traditional methods and is comparable to a state-of-the-art method.

Learning filtering rulesets for ranking refinement in relevance feedback

In this paper we propose an approach for refining a document ranking by learning filtering rulesets through relevance feedback. This approach includes two important procedures. One is a filtering method, which can be incorporated into any kinds of information retrieval systems. The other is a learning algorithm to make a set of filtering rules, each of which specifies a condition to identify relevant documents using combinations of characteristic words. Our approach is useful not only to overcome the limitation of the vector space model, but also to utilize tags of semi-structured documents like Web pages. Through experiments we show our approach improves the performance of relevance feedback in two types of IR systems adopting the vector space model and a Web search engine, respectively.

Clustering by Learning Constraints Priorities

A method for creating a constrained clustering ensemble by learning the priorities of pair wise constraints is proposed in this paper. This method integrates multiple clusters produced by using a simple constrained K-means algorithm that we modify to utilize the constraints priorities. The cluster ensemble is executed according to a boosting framework, which adaptively learns the constraints priorities and provides them for the modified constrained K-means to create diverse clusters that finally improve the clustering performance. The experimental results show that our proposed method outperforms the original constrained K-means and is comparable to several state-of-the-art constrained clustering methods.

Query Expansion with the Minimum User Feedback by Transductive Learning

Query expansion techniques generally select new query terms from a set of top ranked documents. Although a users manual judgment of those documents would much help to select good expansion terms, it is difficult to get enough feedback from users in practical situations. In this paper we propose a query expansion technique which performs well even if a user notifies just a relevant document and a non-relevant document. In order to tackle this specific condition, we introduce two refinements to a well-known query expansion technique. One is application of a transductive learning technique in order to increase relevant documents. The other is a modified parameter estimation method which laps the predictions by multiple learning trials and try to differentiate the importance of candidate terms for expansion in relevant documents. Experimental results show that our technique outperforms some traditional query expansion methods in several evaluation measures.

Interactive Spam Filtering with Active Learning and Feature Selection

This paper proposes an interactive spam filtering method that utilizes active learning and feature selection. Identifying effective features are very important in spam filtering because spam mails include so many meaningless words that are slightly different from each other. Thus identifying effective and ineffective features is promising approach.Although traditional feature selection methods have been done based on some amount of labeled training data, this assumption does not hold in interactive spam filtering. We propose a method to identify effective features through active learning in spam filtering using naive Bayes approach. Experimental results show that our method outperforms traditional methods that operate with no feature selection.

A theory of communication for user interface design

This chapter investigates the phenomenon of communication, and proposes a theory for the design and analysis of user interfaces of distributed information systems. It is argued that the explanation of communication offered by the classical theories is not adequate to fully account for computer-mediated communication. Therefore, a new model of communication is proposed to compensate for the theoretical shortcomings found. The process of communication is represented by a partial sequence of semiosis processes defined recurrently, as emerging from the interaction of at least one psychic system with one or more social systems. The psychic and social systems are characterized as self-organizing systems and their dynamics are described. We have conducted a pilot study on the basis of which we draw conclusions.

Communication of Social Agents and the Digital City - A Semiotic Perspective

This paper investigates the concept of digital city. First, a functional analysis of a digital city is made in the light of the modern study of urbanism; similarities between the virtual and urban constructions are pointed out. Next, a semiotic perspective on the subject matter is elaborated, and a terminological basis is introduced to treat a digital city as a self-organizing meaning-producing system intended to support social or spatial navigation. An explicit definition of a digital city is formulated. Finally, the proposed approach is discussed, conclusions are given, and future work is outlined.

Clustering with Constrained Similarity Learning

This paper proposes a method of learning a similarity matrix from pairwise constraints for interactive clustering. The similarity matrix can be learned by solving an optimization problem as semi-definite programming where we give additional constraints about neighbors of constrained pairwise data besides original constraints. For interactive clustering, since we can get only a few pairwise constraints from a user, we need to extend such constraints to richer ones. Thus this proposed method to extend the pairwise constraints to space-level ones is effective to interactive clustering. First we formalize clustering with constrained similarity learning, and then introduce the extended constraints as linear constraints. We verify the effectiveness of our proposed method by applying it on a simple clustering task. The results of the experiments shows that our method is promising.

Tele-operation between USA and Japan using humanoid robot hand/arm

This paper presents a tele-control system constructed by a robot hand/arm and operator. The angle of the robot hand is controlled by the angle of the operators finger, and the operator feels the environmental force, as detected by the touch sensor of the robot hand, constituting so-called bilateral master/slave control. The position of the robot arm is controlled by the position of the operators arm. So far, there has been little research using a multi-fingered humanoid robot hand in a network environment where communication delay exists. The purpose of our study is to achieve tele-operation between an operators hand/arm and a multi-fingered humanoid robot hand/arm with delayed time. Therefore, this study constructs a system that can grasp and manipulate objects stably, despite the communication delay. In the experiments, the operator operates humanoid robot hand/arm of Toyohashi University of Technology from USA. The operator grasp and moves an object in Japan by operating the humanoid robot hand/arm. The staff in Japan gives the operation instruction to the operator in USA during experiment. The operator operates robot according to the voice. The experimental results show that this system can grasp and manipulate objects stably, despite the communication delay. In addition, the operator grasped the object using a multi-fingered humanoid robot hand/arm by master/slave control feeling fingertip force by tele-operation between USA and Japan.

Learning Similarity Matrix from Constraints of Relational Neighbors

This paper describes a method of learning similarity matrix from pairwise constraints assumed used under the situation such as interactive clustering, where we can expect little user feedback. With the small number of pairwise constraints used, our method attempts to use additional constraints induced by the affinity relationship between constrained data and their neighbors. The similarity matrix is learned by solving an optimization problem formalized as semidefinite programming. Additional constraints are used as complementary in the optimization problem. Results of experiments confirmed the effectiveness of our proposed method in several clustering tasks and that our method is a promising approach.