Complete project work PDF and DOC download titled Applying Deep Learning Methods For Short Text Analysis In Disease Control
Applying Deep Learning Methods For Short Text Analysis In Disease Control
ABSTRACT
Developing countries have been plagued by recurrent cases of infectious disease outbreaks; coupled with the limitation of traditional disease control strategies, other approaches have been explored for disease control, with social media at the forefront. Data from this source is short, noisy, and informal in representation, thus, conventional natural language processing (NLP) methods are not well adapted for their structure. Hence, deep learning approaches for character-level word vector learning were explored to classify disease-related tweets, and an adaptive prediction model for outbreak monitoring was developed, using the Ebola virus disease as a case study. Our system showed better performance for the described task when compared with existing state-of-the-art architectures; also, our predictive model showed correlation with official reported cases, with early warning of fourteen days prior to official.
Keywords: Deep learning, NLP, disease control, short text analysis, word vector learning
Â
TABLE OF CONTENTS
CERTIFICATION………………………………………………………………………………………….2
ABSTRACT…………………………………………………………………………………………….….5
ACKNOWLEDGEMENTS…………………………………………………………………………….….6
CHAPTER ONE BACKGROUND TO THE STUDY……………………………………………….….7
1.1 Introduction…………………………………………………………………………….…….……….7
1.2 Background of the study…………………………………………………………………………….7
1.2.1 Historical account of infectious disease outbreaks in Africa……………………………….8
1.2.2 Disease control……………………………………………………………………….…………9
1.2.3 Deep learning in short text analysis………………………………………………………….11
1.3 Aim and objectives of the study ………………………………………………………………….12
1.4 Research scope ……………………………………………………………………….…………….12
CHAPTER TWO LITERATURE REVIEW………………………………………………………….….13
2.1 Introduction…………………………………………………………………………………………….13
2.2 Text preprocessing: the rudiments of NLP……………………………………………….………13
2.3 Text analysis for structured and unstructured data……………………………….….….………16
2.4 Word vector learning…………………………………………………………………………………17
2.4.1 One-hot encoding………………………………………………………………………………18
2.4.2 Word embedding……………………………………………………………………………….19
2.5 Disease compartment models in epidemiology……………………………………………………25
2.5.1 The SEIR model…………………………………………………….……………………….…26
2.5.2 SITR: The treatment model………………………………………………………….…….….28
2.6 Related work…………………………………………………………………………………………29
2.6.1 Establishing correlation with CDC reports and spurious data effects for influenza………30
2.6.2 Time series modelling and the temporal diversity of different infectious diseases………30
2.6.3 Integrating computational epidemiology models and social media……………………….30
2.6.4 A hybrid approach involving traditional and big data in disease surveillance……………31
CHAPTER THREE ANALYSIS AND PROPOSED METHODS…………………………………….32
3.1 Introduction……………………………………………………………………………………………..32
3.2 Data collection …………………………………………………………………………………….32
3.2.1 Obtaining historic data of disease outbreak…………………………………………………32
3.2.2 Data labelling……………………………………………………………………………………33
3.3 Text analysis…………………………………………………………………………………………33
8
3.3.1 Text preprocessing………………………………………………………………………………33
3.3.2 One-hot encoding for character embedding………………………………………………….34
3.3.3 Deep learning for text classification……………………………………………………………35
3.4 Model specifications………………………………………………………………………………….36
3.4.1 Existing approach……………………………………………………………………………….36
3.4.2 Proposed approaches……………………………………………………………………………38
3.4.3 Model structure………………………………………………………………………………….40
CHAPTER FOUR IMPLEMENTATION AND SIMULATION………………………………………44
4.1 Introduction…………………………………………………………………………………………….44
4.2 Implementation procedure………………………………………………………………….……….44
4.3 Deployment and use case………………………………………………………………….……….44
4.4 Evaluation of results…………………………………………………………………………………45
4.4.1 Correlation with reported cases……………………………………………………………….46
CHAPTER FIVE SUMMARY AND RECOMMENDATION……………………………………………50
5.1 Summary………………………………………………………………………………………………50
5.2 Recommendation…………………………………………………………………………………….50
References…………………………………………………………………………………………………51
9
CHAPTER ONE
BACKGROUND TO THE STUDY
1.1 Introduction
The explosion of data in recent times has marked a new age for human society. Social media platforms such as Facebook, LinkedIn and Twitter offer a place for people to share information in real time; in 2016, Africa aggregated a total of 120 million Facebook users each month and the statistics for other social media platforms have shown similar growth (Fuseware & World Wide Worx, 2014; Parke, n.d.).
The increasing coverage of social media in Africa cannot be over-estimated; likewise, its potential in worthwhile projects of event monitoring, perception evaluation, information extraction and retrieval; thus, its recommendation in disease control strategies.
Notwithstanding the success of social media approaches in politics and business, and its prospects in epidemiology as being timely, collaborative and populace-centric; extensive analysis is imperative, as the tendency of information to be misconstrued is common when machines process natural language. Hence, the need for deep learning methods in social media analytics.
1.2 Background of the study
Africa has been plagued with epidemic disease outbreaks, preceding the 15th century; these occurrences tend to retard both the growth of the human population in the region and the development expectations (Spinage & House, 2012). Study of these cases shows a trend of recurrence of disease outbreaks in previously affected nations and a migration to neighbouring countries, which may be attributed to the ecological changes in the region (Kebede, Duales, Yokouide, & Alemu, 2010; Spinage & House, 2012).
12
The recurrence frequency and the associated mortality rate raises questions about the level of preparedness, surveillance efficacy and control efforts in place; thus, over the years, different approaches have been explored and fused with existing methods to mitigate disease propagation.
1.2.i Historical account of infectious disease outbreaks in Africa
Records show a number of viral diseases prevalent in Africa, such as cholera, meningitis, influenza, yellow fever, rickettsia, smallpox, HIV/AIDS, Lassa fever and Ebola. The total mortality score ranges in the millions, with Ebola and HIV/AIDS accounting for over 3 million recorded deaths (Spinage & House, 2012).
Table 1 shows a selected number of disease outbreak cases in Africa, with the estimated casualty scores.
Table 1: History of disease outbreaks in Africa
13
1.2.ii Disease control
Walter R. Dowdle defined disease control as ‘the reduction of disease incidence, prevalence, morbidity or mortality to a locally acceptable level as a result of deliberate efforts; continued interventions are required to maintain the reduction’ (Dowdle, 1998).
Efforts in disease control interventions are targeted to reduce the contact rate of transmission, keep the infectious population low, shorten the infection span of the prevalent disease and obtain a disease-free equilibrium (DFE) in the population (Brauer & Castillo-Chavez, 2014).
Disease control involves:
• Prevention activities for disease event surveillance, preparedness, and rapid response.
• Eradication activities for isolation, treatment, and rehabilitation of infectious people.
A good disease control strategy involves both prevention and eradication, though they tend to overlap and can be carried out in varying orders during the intervention cycle.
A number of organizations in conjunction with the World Health Organization (WHO), Centre for Disease Control (CDC) and the health ministries of different countries are in affiliation for disease control purposes. These bodies have been active in epidemic preparedness and response (EPR) activities and integrated disease surveillance response (IDSR) strategies. Adapting traditional methods for data collection, disease identification, outbreak events and predictions, casualty estimation, and all the other metrics of disease outbreak.
1.2.ii.1 Social media in disease control
Due to the decision-pipeline involved in traditional methods, though unavoidable because of the sensitivity of health matters; information collection, validation and dissemination tend to be gradual. Judging from the fatality and transmission rate of the recent outbreaks of the Ebola virus disease
14
(EVD) and Lassa fever in 2014 and 2017, one can only conclude that a swift response would have averted the disaster and reduced the mortality rate.
Social media in disease control has been explored in recent years to get timely information for disease event surveillance, disease prevalence and detection in spatial locations, thus causing its potential in disease prediction and prevention to increase (Choi, Cho, Shim, & Woo, 2016).
Unlike documents which are over 1,000 characters, formal and follow the syntax of the target language, social media data is characterized by texts less than 500 characters and do not follow syntax hence, deep learning methods are favoured for processing their structure.
1.2.iii Deep learning in short text analysis
For short text analysis, word representation by word embedding has been adjudged most suitable for word similarity and sentence classification tasks (Komninos, 2016), which include: sentiment analysis, machine translation, question type classification, topic categorization; and word similarity for web queries and search processing.
Ground-breaking works by Bengio, Ducharme, Vincent, and Janvin (2003), Mikolov, Sutskever, Chen, Corrado, and Dean (2013), and Mikolov, Corrado, Chen, and Dean (2013) which applied multilayer convolutional neural networks to capture word semantics and syntactic properties paved the way for further advancement in the field of natural language processing (NLP).
In the event of disease outbreak, timely intervention is necessary to control the mortality scores; this can be achieved by effective disease monitoring and control. Information shared over social media and text messages could portend disease prevalence only if its curation is timely and accurate and the parameters estimated from the text data can be integrated into statistical models to forecast the disease dynamics. Though the possibilities with social media data are encouraging, it is also characterized by a high noise level and language processing barriers because of its informality.
Do you the complete work from chapters one to five titled Applying Deep Learning Methods For Short Text Analysis In Disease Control, click below
Do you need help? Talk to us right now: (+234) 08060082010, 08107932631 (Call/WhatsApp). Email: [email protected].IF YOU CAN'T FIND YOUR TOPIC, CLICK HERE TO HIRE A WRITER»
