Jeremy Ellman

Simple Approaches of Sentiment Analysis via Ensemble Learning

Twitter has become a popular microblogging tool where users are increasing every minute. It allows its users to post messages of up to 140 characters each time; known as ‘Tweets’. Tweets have become extremely attractive to the marketing sector, since the user can either indicate customer success or presage public relations disasters far more quickly than web pages or traditional media. Moreover, the content of Tweets has become a current active research topic on sentiment polarity as positive or negative. Our experiment of sentiment analysis of contexts of tweets show that the accuracy performance can improve and be better achieved using ensemble learning, which is formed by the majority voting of the Support Vector Machine, Naive Bayes, SentiStrength and Stacking.

A Cost Modelling System for Cloud Computing

An advance in technology unlocks new opportunities for organizations to increase their productivity, efficiency and process automation while reducing the cost of doing business as well. The emergence of cloud computing addresses these prospects through the provision of agile systems that are scalable, flexible and reliable as well as cost effective. Cloud computing has made hosting and deployment of computing resources cheaper and easier with no up-front charges but pay per-use flexible payment methods. However, there is lack of tools to aid decision makers in evaluating the much promised benefits of cloud computing particularly its cost benefit. To fill this gap in tools for evaluating the cost benefit of cloud services as an alternative to on-premise computing, a cost modelling system for cloud computing (CCMS) is proposed. A prototype model was developed to simulate the cost incurred on maintaining an on-premise IT infrastructure under various usage patterns with the purpose of determining the cost benefit of cloud alternatives. CCMS assists decision makers with insights on cost savings of adopting cloud alternatives, plan computing budgets and also demonstrate how the utilization capacity and cost of acquisition of an infrastructure can influence the cost savings from cloud alternatives.

Everyday usage of home telecare services in England

Deployment of telecare is being driven by policy in different European contexts, however little is known about the utilization of telecare following installation. This study provides an analysis of contact data within one telecare call centre in North East England. Characteristics of service users that influence use is presented. 50.6% of contact episodes were due to activation of devices and sensors; 17.7% to service users contacting the call centre for support, referral to other services; and 31.3% to managing the system.

UNN-WePS: Web Person Search using co-Present Names and Lexical Chains

We describe a system, UNN-WePS for identifying individuals from web pages using data from Semeval Task 13. Our system is based on using co-presence of person names to form seed clusters. These are then extended with pages that are deemed conceptually similar based on a lexical chaining analysis computed using Rogets thesaurus. Finally, a single link hierarchical agglomerative clustering algorithm merges the enhanced clusters for individual entity recognition. UNN-WePS achieved an average purity of 0.6, and inverse purity of 0.73.

The role of HTML5 IndexedDB, the past, present and future

Over the past 20 years Web browsers have changed considerably from being a simple text display to now supporting complex multimedia applications. The client can now enjoy chatting, playing games and Internet banking. All these applications have something in common, they can be run on multiple platforms and in some cases they will run offline. With the introduction of HTML5 this evolution will continue, with browsers offering greater levels of functionality. This paper outlines the background study and the importance of new technologies, such as HTML5s new browser based storage called IndexedDB. We will show how the technology of storing data on the client side has changed over the time and how the technologies for storing data on the client will be used in future when considering known security issues. Further, we propose a solution to IndexedDBs known security issues in form of a security model, which will extend the current model.

User Activity Pattern Analysis in Telecare Data

Telecare is the use of devices installed in homes to deliver health and social care to the elderly and infirm. The aim of this paper is to identify patterns of use for different devices and associations between them. The data were provided by a telecare call center in the North East of England. Using statistical analysis and machine learning, we analyzed the relationships between users’ characteristics and device activations. We applied association rules and decision trees for the event analysis, and our targeted projection pursuit technique was used for the user-event modeling. This study reveals that there is a strong association between users’ ages and activations, i.e., different age group users exhibit different activation patterns. In addition, a focused analysis on the users with mental health issues reveals that the older users with memory problems who live alone are likely to make more mistakes in using the devices than others. The patterns in the data can enable the telecare call center to gain insight into their operations and improve their effectiveness in several ways. This study also contributes to automatic analysis and support for decision making in the telecare industry.

TJP: Identifying the Polarity of Tweets from Contexts

The TJP system is presented, which participated in SemEval 2014 Task 9, Part A: Contextual Polarity Disambiguation. Our system is ‘constrained’, using only data provided by the organizers. The goal of this task is to identify whether marking contexts are positive, negative or neutral. Our system uses a support vector machine, with extensive pre-processing and achieved an overall F-score of 81.96%.

Combined clustering models for the analysis of gene expression

Clustering has become one of the fundamental tools for analyzing gene expression and producing gene classifications. Clustering models enable finding patterns of similarity in order to understand gene function, gene regulation, cellular processes and sub-types of cells. The clustering results however have to be combined with sequence data or knowledge about gene functionality in order to make biologically meaningful conclusions. In this work, we explore a new model that integrates gene expression with sequence or text information.