Adaptive Recommendations based on Spatiotemporal Data Streams

No Thumbnail Available

URL

Journal Title

Journal ISSN

Volume Title

Perustieteiden korkeakoulu | Master's thesis

Date

2018-06-18

Department

Major/Subject

Innovation and Entrepreneurship

Mcode

SCI3081

Degree programme

Master's Programme in ICT Innovation

Language

en

Pages

78 + 6

Series

Abstract

In recent times mobile devices, i.e.\ smart-phones, IoT sensors, GPS equipment and wearables are getting sophisticated in hardware. Hence, developers are allow to implement complex software that helps to produce more than position location of devices such as GPS positions, WiFi fingerprinting signals, or Bluetooth beacons. Intelligent mobile devices are often connected through Internet or any other network to Location-Based Services which stores the Geo-positioning location of the entity together with useful information about the state of the object, for instance, this can be distance of a trip, walking steps, etc. In parallel, streaming analytics is being more popular with Big Data adopters that believe stream analytics is the next step in the revolution of real-time processing and Big Data. MapReduce of Apache Hadoop was the first approach to tackle the problematic of processing large sets of data. This process is called batch processing but it is not able to process data streams. However, Apache Hadoop has built a big ecosystem that takes MapReduce architecture as the core of any other implementation which runs on top of its architecture. Then, Stream frameworks like Apache Spark, Apache Flink and Apache Storm took Hadoop architecture to solve Big Data issues such as Velocity and Volume. Those frameworks offer libraries and extensions which allow to use machine learning algorithms and data mining techniques within data streams. Although, stream processing is able to process large and fast data streams, there has to be a method to reduce the dimension of data without losing information of this set of samples. Therefore, feature selection is an important technique that helps to reduce high dimensional datasets. This thesis will discuss one of those techniques in order to reduce dimensionality of datasets and to identify the most suitable dimensions which helps to formulate models for the construction of adaptable recommendations. This work introduces the design and implementation of a system together with guidelines for the development of adaptable recommendations based on spatiotemporal data streams. For this reason, it is fundamental to be capable of managing heterogeneous data sources and unstructured information. Another main goal of this thesis is to produce meaningful knowledge for improving real-time Adaptive Systems in the field of Location-Based Services.

Description

Supervisor

Heljanko, Keijo

Thesis advisor

Heljanko, Keijo

Keywords

adaptable recommender systems, location-based services, real-time analysis, spatiotemporal data, streaming analysis

Other note

Citation