signature based

8/2/2019 Signature Based

1/12

A Signature-Based Indexing Methodfor Efficient Content-Based Retrieval

of Relative Temporal Patterns


2/12

AbstractA number of algorithms have been proposed for thediscovery of temporal patterns. However, since the number ofgenerated patterns can be large, selecting which patterns toanalyze can be nontrivial. There is thus a need for algorithms

and tools that can assist in the selection of discovered patternsso that subsequent analysis can be performed in an efficientand, ideally, interactive manner. In this project, we propose asignature-based indexing method to optimize the storage andretrieval of a large collection of relative temporal patterns.


3/12

Introduction MANY rule discovery algorithms in data mining generate a large number of

patterns/rules, sometimes even exceeding the size of the underlying database,with only a small fraction being of interest to the user It is generally understoodthat interpreting the discovered patterns/rules to gain insight into the domainis an important phase in the knowledge discovery process. However, when

there are a large number of generated rules, identifying and analyzing thosethat are interesting becomes difficult. For example, providing the user with alist of association rules ranked by their confidence and support might not be agood way of organizing the set of rules as this method would overwhelm theuser and not all rules with high confidence and support are necessarilyinteresting for a variety of reasons.

Therefore, to be useful, a data mining system must manage thegenerated rules by offering flexible tools for rule selection. In the caseof association rule mining, several approaches for the post processingof discovered association rules have been discussed. One approach is togroup similar rules which works well for a moderate number of rules.However, for a larger number of rules it produces too many clusters.


4/12

A more flexible approach is to allow the identification of rules that areof special importance to the user through templates or data miningqueries.This approach can complement the rule grouping approach

and has been used to specify interesting and uninteresting classes ofrules (for both association and episodic rules).The importance of datamining queries has been highlighted by the introduction of theinductive database concept, which allows the user to both query thedata and query patterns, rules, and models extracted from these data.


5/12

Existing System

Mine-Rule DMQL and OLE DB . These languages are designedto generate the rules from the data rather than allow queriesover the discovered rules.

Post processing of association rules, an area in which littleresearch to date has been conducted.


6/12

De-Merits of Existing System In previous number of algorithm proposed for discover the temporal

pattern, in that generated patterns are large .

Mine-Rule, DMQL to generate the rules from the data rather thanallow queries over the discovered rules.


7/12

Proposed SystemEfficiently retrieving subsets of a large collection of previouslydiscovered temporal patterns.

Focuses on supporting content-based queries of temporalpatterns, as opposed to point- or range-based queries .


8/12

Advantages of Proposed System

we address the problem of efficiently retrieving subsets of a large

collection of previously discovered temporal patterns.

improves the performance of temporal pattern retrieval.

This retrieval system is currently being combined with

visualization techniques for monitoring the behavior of a single

pattern or a group of patterns over time.


9/12

Dataflow DiagramLarge Database

Finding temporalpattern similarity

ConstructingSignature

Filesfor temporal patterns

AnsweringContent-Based

Queries Using theSignature File

Effectively retrievedthe data


10/12

Modules and Description Finding temporal pattern similarity

1. Temporal patterns are variable-length objects thatcannot be represented in a k-dimensional metricspace.

2. Each pattern contains a list of states.3. Each pattern contains a set of state relationships.

Constructing Signature Files for temporal patterns.

Given that T and Q denote the target set and query set,respectively, three commonly used set-valued queries are

1. Subset query T Q. The target set is a superset ofthe query set.

2. Superset query T Q. The target set is a subset ofthe query set.

3. Equality query T Q. The target set is equal to the

query set.


11/12

Answering Content-Based Queries Using the SignatureFile.

1.Subpattern Queries

2.Superpattern Queries

3.Equality Queries


12/12

Hardware:

PROCESSOR : PENTIUM IV 2.6 GHzRAM : 512 MB DD RAMMONITOR : 15 COLORHARD DISK : 20 GB

FLOPPY DRIVE : 1.44 MBCDDRIVE : LG 52XKEYBOARD : STANDARD 102 KEYSMOUSE : 3 BUTTONS

Software:

Front End : Java.Back End : MS SqlOperating System : WindowsXP

System Requirements

signature based

Documents