Tomonobu Ozaki

Discovery of Quantitative Sequential Patterns from Event Sequences

In this paper, we consider the problem of frequent pattern mining in databases of temporal events with intervals. Since quantitative temporal information might play important roles in many application domains, it is critical to discover patterns to which numerical attributes are associated. To this end, we consider two kinds of temporal patterns with quantitative information on the durations and time differences of events, and propose corresponding algorithms by incorporating numerical clustering techniques into existing temporal pattern miners. The effectiveness of the proposed algorithms was assessed by using real world datasets.

Respiration Reflecting Musical Expression: Analysis of Respiration during Musical Performance by Inductive Logic Programming

In this paper, we describe the analysis of respiration during musical performance by Inductive Logic Programming (ILP). For effective musical performance, it is one of the most important factors to control one’s respiration in response to the aspects of the music performed. It is, however, often difficult even for experts to explain how to do so clearly. We measured respiration during cello performance by using a respiration sensor, and tried to extract rules of respiration from the data together with musical/performance background knowledge such as harmonic progression and bowing direction. As a result it was found that there was repeatability and regularity in performers’ respiration pattern during musical performance, and then consistency in respiration with regard to musical structure was confirmed. It was also discovered that players tend to exhale at the beginning of new large musical structures, and inhale immediately before the change of keys.

Frequent Pattern Discovery from a Single Graph with Quantitative Itemsets

In this paper, we focus on a single graph whose vertices contain a set of quantitative attributes. Several networks can be naturally represented in this complex graph. An example is a social network whose vertex corresponds to a person with some quantitative items such as age, salary and so on. Although it can be expected that this kind of data will increase rapidly, most of current graph mining algorithms do not handle these complex graphs directly. Motivated by the above background, by effectively combining techniques of graph mining and quantitative itemset mining, we developed an algorithm named FAG-gSpan for finding frequent patterns from a graph with quantitative itemsets.

Mining correlated subgraphs in graph databases

In this paper, we bring the concept of hyperclique pattern in transaction databases into the graph mining and consider the discovery of sets of highly-correlated subgraphs in graph-structured databases. To discover frequent hyperclique patterns in graph databases efficiently, a novel algorithm named HSG is proposed. By considering the generality ordering of subgraphs, HSG employs the depth-first/breadth-first search strategy with powerful pruning techniques based on the upper bound of h-confidence measure. The effectiveness of HSG is assessed through the experiments with real world datasets.

Interaction site prediction by structural similarity to neighboring clusters in protein-protein interaction networks

BackgroundRecently, revealing the function of proteins with protein-protein interaction (PPI) networks is regarded as one of important issues in bioinformatics. With the development of experimental methods such as the yeast two-hybrid method, the data of protein interaction have been increasing extremely. Many databases dealing with these data comprehensively have been constructed and applied to analyzing PPI networks. However, few research on prediction interaction sites using both PPI networks and the 3D protein structures complementarily has explored.ResultsWe propose a method of predicting interaction sites in proteins with unknown function by using both of PPI networks and protein structures. For a protein with unknown function as a target, several clusters are extracted from the neighboring proteins based on their structural similarity. Then, interaction sites are predicted by extracting similar sites from the group of a protein cluster and the target protein. Moreover, the proposed method can improve the prediction accuracy by introducing repetitive prediction process.ConclusionsThe proposed method has been applied to small scale dataset, then the effectiveness of the method has been confirmed. The challenge will now be to apply the method to large-scale datasets.

Weighted Frequent Subgraph Mining in Weighted Graph Databases

We focus on the problem of pattern discovery from externally and internally weighted labeled graphs because the target data can be modeled more naturally and in detail by using weighted graphs. For example, while external weight can be used for representing a degree of importance and reliability of a graph itself, internal weight reflects utility and significance of each component in a graph. Therefore, we can expect to realize more precise knowledge discovery by employing weighted graphs. From these backgrounds, in this paper, we discuss two pattern mining problems with external and internal weighted frequencies, and propose two algorithms to solve them efficiently.

Decision Tree Construction from Multidimensional Structured Data

Since most structured data mining techniques specialize in mining from single structured data, it cannot handle more realistic data which consist of different and plural kinds of structured data. To cope with this problem, we propose an algorithm for constructing decision trees from multidimensional structured data by introducing the techniques for mining correlated and closed patterns with effective pruning capabilities into the traditional TDIDT approach. The results of the experiments with real world data show the effectiveness of the proposed algorithm

Correlation and Contrast Link Formation Patterns in a Time Evolving Graph

A time evolving graph is becoming increasingly abundant in a wide variety of application domains. While several classes of advanced frequent patterns in time evolving graphs are proposed, in this paper, correlation and contrast patterns on link formations are developed, which can be regarded as nontrivial upgrades of corresponding patterns in item set mining into the graph domain. More concretely, hyper clique patterns and conditional contrast patterns are adopted to develop new correlation and contrast link formation patterns, respectively. In addition, another novel correlation pattern is derived by the analogy of conditional contrast patterns. Discovery of sets of link formation patterns having opposite characteristics, i.e. correlation and contrast, can expect to obtain deep understanding of target dataset which in turn brings new findings. To discover correlation and contrast patterns efficiently, a series of algorithms is developed by using conventional methods in closed sub graph discovery, item set mining and pseudo clique enumeration. Experiments using real world datasets confirm the effectiveness of the proposed framework.

Extraction of Binding Sites in Proteins by Searching for Similar Local Molecular Surfaces

There is much research on the automatic extraction of new binding sites in proteins by searching for common sites in proteins with identical functions. While many binding sites consist of concave structures, it is difficult to compare such concaves directly due to the various sizes of concaves. To cope with this difficulty and to realize detailed and precise comparisons between concaves, we propose a method of searching for and comparing concaves by gradually changing the size. By experiments with enzyme proteins, we confirmed that extraction accuracy for the binding sites is improved.

Efficient Mining of Closed Induced Ordered Subtrees in Tree-structured Databases

In this paper, two new closed ordered subtree miners are given which are based on the breadth-first and depth-first/breadth-first enumeration strategy, respectively. Through the experiments with synthesized and real world data, we discuss the effects of the difference of the search strategies in mining closed induced ordered subtrees