Skip to main content
main-content

Über dieses Buch

This SpringerBrief presents a typical life-cycle of mobile data mining applications, including:

data capturing and processing which determines what data to collect, how to collect these data, and how to reduce the noise in the data based on smartphone sensors

feature engineering which extracts and selects features to serve as the input of algorithms based on the collected and processed data

model and algorithm designIn particular, this brief concentrates on the model and algorithm design aspect, and explains three challenging requirements of mobile data mining applications: energy-saving, personalization, and real-time

Energy saving is a fundamental requirement of mobile applications, due to the limited battery capacity of smartphones. The authors explore the existing practices in the methodology level (e.g. by designing hierarchical models) for saving energy. Another fundamental requirement of mobile applications is personalization. Most of the existing methods tend to train generic models for all users, but the authors provide existing personalized treatments for mobile applications, as the behaviors may differ greatly from one user to another in many mobile applications. The third requirement is real-time. That is, the mobile application should return responses in a real-time manner, meanwhile balancing effectiveness and efficiency.

This SpringerBrief targets data mining and machine learning researchers and practitioners working in these related fields. Advanced level students studying computer science and electrical engineering will also find this brief useful as a study guide.

Inhaltsverzeichnis

Frontmatter

Chapter 1. Introduction

Smartphones are usually equipped with various sensors by which the personal data of the users can be collected. To make full use of the smartphone data, mobile data mining aims to discover useful knowledge from the collected data in order to provide better services for the users. In this chapter, we introduce some background information about mobile data mining, including what data can be collected by smartphones, what applications can be built upon the collected data, what are the key steps for a typical mobile data mining task, and what are the key characteristics and challenges of mobile data mining.
Yuan Yao, Xing Su, Hanghang Tong

Chapter 2. Data Capturing and Processing

The first step towards mobile data mining is collecting smartphone sensors’ readings, and processing the raw data in order to remove various noises. The key questions that need to be answered during data capturing and processing include: what data should be collected and how to collect such data, and what are the characteristic patterns behind these data and how to clean the data so that these patterns can be easily identified. In this chapter, we first introduce the smartphone sensors and their readings. Next, to facilitate discussions, we present the details of data collection and data denoising with an example of travel mode detection.
Yuan Yao, Xing Su, Hanghang Tong

Chapter 3. Feature Engineering

So far, the collected data are time series data of different sensors’ readings. To make use of these time series in the following learning models, we usually need to first slice the time series data into data segments, and then extract features from these segments. Meanwhile, the data segmentation and feature extraction also affect the aspects like energy efficiency, model accuracy, and response time. In this chapter, we first discuss the data segmentation method, and then introduce the feature extraction which extracts features from a segment with the principle that the extracted features should be informative and discriminative. To further save time, we also discuss the feature selection method which selects a subset of the features for the current task.
Yuan Yao, Xing Su, Hanghang Tong

Chapter 4. Hierarchical Model

After the features are ready to use, we start to present the learning models for mobile data mining applications. The learning models mainly need to consider two challenges: energy-saving and personalization. In this chapter, we present a hierarchical learning framework for mobile data mining tasks with the goal of energy-saving. Specially, we illustrate the idea with the travel mode detection task. We classify the six modes into wheeled modes and unwheeled modes, where the wheeled modes include outdoor modes (biking) and indoor modes (taking a subway, driving a car, and taking a bus), and the unwheeled modes include walking and jogging. Corresponding to the classification, the hierarchical model consists of three layers. It is based on the results of group feature analysis in the previous chapter. That is, not all sensor data are required for a certain task. For example, we find that only wheeled modes require the full sensor data while the majority of the sensors (except for accelerometer and gyroscope) are turned off in other cases.
Yuan Yao, Xing Su, Hanghang Tong

Chapter 5. Personalized Model

Another challenge for mobile data mining tasks is personalization. On one hand, we usually do not have plenty labels for a certain user, and thus we need to borrow the labeled data from other users. One the other hand, different users tend to behavior differently and thus have different patterns of sensor readings, making data borrowing non-trivial. In this chapter, we introduce a personalized treatment for mobile data mining tasks. The basic idea can be divided into two stages. In the first stage, for a given target user, we select and borrow some data from the users whose labeled data are already collected in the database. In the second stage, we reweight the borrowed data and use them as the training data for the target user. The proposed method is able to estimate the sample distributions, and then reweight the samples based on the estimated distributions so as to minimize the model loss with respect to the target user’s data.
Yuan Yao, Xing Su, Hanghang Tong

Chapter 6. Online Model

Model updating is important for mobile data mining tasks. When more labeled data are available for a given target user, the mobile data mining algorithm should incorporate such data in order to provide more accurate services for the user. However, directly re-training the model with both the old and the new data would be resource consuming especially for mobile applications. A more desired way is to incrementally update the model in an online fashion. In this chapter, we introduce an online model for the mobile data mining tasks. The online model is orthogonal with the hierarchical model and personalized model. The basic idea is to adopt the stochastic sub-gradient descent method and updates the learning models with a small portion of new data.
Yuan Yao, Xing Su, Hanghang Tong

Chapter 7. Conclusions

In this chapter, we first present a brief summary of the content in this book, and then discuss the interesting future directions.
Yuan Yao, Xing Su, Hanghang Tong

Backmatter

Weitere Informationen

Premium Partner

BranchenIndex Online

Die B2B-Firmensuche für Industrie und Wirtschaft: Kostenfrei in Firmenprofilen nach Lieferanten, Herstellern, Dienstleistern und Händlern recherchieren.

Whitepaper

- ANZEIGE -

Best Practices für die Mitarbeiter-Partizipation in der Produktentwicklung

Unternehmen haben das Innovationspotenzial der eigenen Mitarbeiter auch außerhalb der F&E-Abteilung erkannt. Viele Initiativen zur Partizipation scheitern in der Praxis jedoch häufig. Lesen Sie hier  - basierend auf einer qualitativ-explorativen Expertenstudie - mehr über die wesentlichen Problemfelder der mitarbeiterzentrierten Produktentwicklung und profitieren Sie von konkreten Handlungsempfehlungen aus der Praxis.
Jetzt gratis downloaden!

Bildnachweise