Discovery of Sequential Patterns Based on Sequential Interestingness and Constraint Conditions

Discovery of Sequential Patterns Based on Sequential Interestingness and Constraint Conditions

Shigeaki Sakurai (Toshiba Corporation, Japan)
Copyright: © 2015 |Pages: 11
DOI: 10.4018/978-1-4666-5888-2.ch169
OnDemand PDF Download:
$30.00
List Price: $37.50

Chapter Preview

Top

Introduction

We can easily collect a large amount of data due to the progress of the computer and network environments. It is anticipated that the data includes knowledge leading to a smart society. The discovery task of the knowledge from the data is put on a significant position in the information science. Many researchers try to discover the knowledge. Recently, the amount of data expands more and more, leading to the creation of a buzzword such as BigData in the information communication technology field. Even if BigData is not always well-defined, it is characterized by three 'V': Volume, Variety, and Velocity. It means that large and complex data is speedily brought up from our world. Many companies and institutions aggressively try to activate BigData.

This article focuses on sequential data because the sequential data is explosively expanding according to the progress of Twitter and YouTube, the high interest for smart grid and smart community, and so on. The sequential data is important parts of BigData and is a set of sequences. Each sequence is a row of item sets. In the case of retail field, an item is goods, an item set is a receipt, and a sequence is receipts which are gathered and sequentially arranged per a customer. Also, this article focuses on the discovery task of characteristic sequential patterns. The patterns are characteristic subsequences extracted from given sequential data. The task evaluates whether a subsequence is characteristic or not based on given evaluation criteria of sequential patterns. The discovered sequential patterns are activated for floor design, order planning, goods recommendation, and so on. Recently, techniques developed for the task are applied into various application fields. For example, they are applied to the analysis of word sequences and healthcare data. In the case of the word sequences, a word corresponds to an item. In the case of the healthcare data, a discretized test value or its change does so. Initial researches for the task have regarded frequent sequential patterns as characteristic ones. That is, the characteristic sequential patterns are sequential patterns with high frequency in given sequential data. However, the analysts are not always interested in the patterns because they still know the patterns. Therefore, many researchers have developed techniques discovering characteristic sequential patterns that are not simply frequent.

This article introduces an evaluation criterion, sequential interestingness (Sakurai, Kitahara, & Orihara, 2008). The evaluation criterion can simultaneously evaluate the frequency and conditional probability of a sequential pattern. Also, this article introduces the activation method of background knowledge. The knowledge can represent constraints related to the time between items (Sakurai, Ueno, & Orihara, 2008). The criterion and the knowledge can acquire other types of characteristic sequential patterns. On the other hand, this article introduces an application task based on these techniques in order to clarify the necessity of the discovery task in real world. The task is an analysis task of periodical medical examination (Sakurai, Kitahara, Orihara, Iwata, Honda, & Hayashi, 2008). This task discovers change of health situation as characteristic sequential patterns. The patterns are activated for the healthcare guidance by an industrial doctor and the improvement of healthcare situation. Lastly, this article introduces future research directions in this field.

Key Terms in this Chapter

Apriori Property: It expresses monotonic decrease of an evaluation criterion accompanying with the progress of a sequential pattern. It is activated in order to efficiently discover all frequent sequential patterns.

Item: It is a minimum unit composing of sequential data or a sequential pattern. In a retail field, it corresponds to goods included in a receipt.

Sequential Data: It is a row of item sets. Each item set does not include multiple same items. Each item in the same item set has the same time stamp.

Time Constraint: It is a constraint related to time between items. It represents the background knowledge of analysts. It is used in order to discover characteristic sequential pattern coinciding with the interests of analysts.

Confidence: It is one of evaluation criteria for sequential patterns. It corresponds to conditional probability of a sequential pattern in the case that one of its sequential subpatterns is given. It does not satisfy the Apriori property.

Sequential Interestingness: It is one of evaluation criteria for sequential patterns. It simultaneously evaluates their frequencies and conditional probabilities. It also satisfies the Apriori property.

Frequent Sequential Pattern: It is a sequential pattern whose support is larger than or equal to the minimum support. The minimum support is given by an analyst. The discovery method of frequent sequential patterns aims at efficiently discovering all of them.

Support: It is one of evaluation criteria for sequential patterns. It evaluates their relative frequencies. It satisfies the Apriori property.

Complete Chapter List

Search this Book:
Reset