Skip to Main content Skip to Navigation
Theses

Sequential Pattern Generalization for Mining Multi-source Data

Julie Bu Daher 1
1 KIWI - Knowledge Information and Web Intelligence
LORIA - AIS - Department of Complex Systems, Artificial Intelligence & Robotics
Abstract : Huge amounts of digital data have been created across years due to the increasing digitization in our everyday life. As a consequence, fast data collection and storage tools have been developed and data can be collected in huge volumes for various research and business purposes. The collected data can come from multiple data sources and can be of heterogeneous kinds thus forming heterogeneous multi-source datasets, and they can be analyzed to extract valuable information. Data mining is an important task in discovering interesting information from datasets. Different approaches in this domain have been proposed, among which pattern mining is the most important one. Pattern mining, including sequential pattern mining, discovers statistically relevant patterns (or sequential patterns) among data. The challenges of this domain include discovering important patterns with a limited complexity and by avoiding redundancy among the resulting patterns. Multi-source data could represent descriptive and sequential data, making the mining process complex. There could be problems of data similarity on one source level which leads to a limited number of extracted patterns. The aim of the thesis is to mine multi-source data to obtain valuable information and compensate the loss of patterns due to the problem of similarity with a limited complexity and by avoiding pattern redundancy. Many approaches have been proposed to mine multi-source data. These approaches either integrate multi-source data and perform a single mining process which increases the complexity and generates a redundant set of sequential patterns, or they mine sources separately and integrate the results which could lead to a loss of patterns. We propose G_SPM, a general sequential pattern mining algorithm that takes advantage of multi-source data to mine general patterns which compensates the loss of patterns caused by the problem of data similarity. These rich patterns contain various kinds of information and have higher data coverage than traditional patterns. G_SPM adopts a selective mining strategy of data sources where a main source is first mined, and other sources are mined only when similarity among patterns is detected, which limits the complexity and avoids pattern redundancy. The experimental results confirm that G_SPM succeeds in mining general patterns with a limited complexity. In addition, it outperforms traditional approaches in terms of runtime and pattern redundancy.
Document type :
Theses
Complete list of metadata

https://hal.univ-lorraine.fr/tel-03184696
Contributor : Thèses Ul <>
Submitted on : Monday, March 29, 2021 - 5:14:11 PM
Last modification on : Wednesday, March 31, 2021 - 3:32:14 AM

Identifiers

  • HAL Id : tel-03184696, version 1

Citation

Julie Bu Daher. Sequential Pattern Generalization for Mining Multi-source Data. Computer Science [cs]. Université de Lorraine, 2020. English. ⟨NNT : 2020LORR0204⟩. ⟨tel-03184696⟩

Share

Metrics

Record views

47

Files downloads

15