Conference papers

Stock Market Prediction Without Sentiment Analysis: Using a Web-Traffic Based Classifier and User-Level Analysis

Pierpaolo Dondio — Tue, 26 Mar 2013 02:46:15 PDT

This paper provides further evidence on the predictive power of online community traffic with regard to stock prices. Using the largest dataset to date, spanning 8 years and almost the complete set of SP500 stocks, we train a classifier using a set of features entirely extracted from web-traffic data of financial online communities. The classifier is shown to outperform the predictive power of a baseline classifier solely based on price time-series, and to have similar performances as the classifier built considering price and traffic features together. The best predictive performances are achieved when information about stock capitalization is coupled with long-term and midterm web traffic levels. In the second part of the paper we show how there exists a group of users whose traffic patterns constantly outperform the other users in predictive capacity. The findings set interesting future works in the definition of novel market indicators for market analysis

Domain Independent Sentiment Classification with Many Lexicons

Bruno Ohana et al. — Mon, 07 Jan 2013 01:31:44 PST

Sentiment lexicons are language resources widely used in opinion mining and important tools in unsupervised sentiment classification. We present a comparative study of sentiment classification of reviews on six different domains using sentiment lexicons from different sources. Our results highlight the tendency of a lexicon’s performance to be imbalanced towards one class, and indicate lexicon accuracy varies with the target domain. We propose an approach that combines information from different lexicons to make classification decisions and achieve more robust results that consistently improve our baseline across all domains tested. These are further refined by a domain independent score adjustment that mitigates the effect of the recall imbalance seen on some of the results

A Case-Based Approach to Cross Domain Sentiment Classification

Bruno Ohana et al. — Mon, 07 Jan 2013 01:27:11 PST

This paper considers the task of sentiment classification of subjective text across many domains, in particular on scenarios where no in-domain data is available. Motivated by the more general applicability of such methods, we propose an extensible approach to sentiment classification that leverages sentiment lexicons and out-of-domain data to build a case-based system where solutions to past cases are reused to predict the sentiment of new documents from an unknown domain. In our approach the case representation uses a set of features based on document statistics, while the case solution stores sentiment lexicons employed on past predictions allowing for later retrieval and reuse on similar documents. The case-based nature of our approach also allows for future improvements since new lexicons and classification methods can be added to the case base as they become available. On a cross domain experiment our method has shown robust results when compared to a baseline single-lexicon classifier where the lexicon has to be pre-selected for the domain in question.

Retention in Computer Science: a Level 6 Experience in an Irish Third Level Institute.

Jane Ferris — Thu, 25 Oct 2012 01:56:08 PDT

The current World economic recession has brought significant changes to the lives of many Irish citizens and brought many back into the education system. In 2010 in response to the need to provide retraining opportunities for the long term unemployed some third level institutes introduced re-training programmes. This paper describes the retention experience of a Computer Science department of an Irish third level institute managing level 6 part-time computing students. The paper reports the barriers for students in continuing on the programmes from exit interviews and identifies potential strategies to engage students in their chosen programmes.

2010 School of Computing Undergraduate Student Survey.

Jane Ferris — Thu, 25 Oct 2012 01:41:34 PDT

This paper summaries the findings from a voluntary online survey distributed to all Computer Science undergraduates (part time and full time) of a large Irish third level institute in 2010. The objective of the survey was to investigate, identify and report the cohorts’ personal circumstances, computing interests, preferred learning methods and reasons for selecting Computer Science as a profession. The survey had a good response rate from most student groups. Identified from the two cohorts of part and full time Computer Science undergraduates are typical students, their personal computing interests, learning motivation and the methods of learning they indicate as most significant.

Context Cues For Classification Of Competitive And Collaborative Overlaps

Catharine Oertel et al. — Tue, 09 Oct 2012 01:46:03 PDT

Being able to respond appropriately to users’ overlaps should be seen as one of the core competencies of incremental dialogue systems. At the same time identifying whether an interlocutor wants to support or grab the turn is a task which comes naturally to humans, but has not yet been implemented in such systems. Motivated by this we first investigate whether prosodic characteristics of speech in the vicinity of overlaps are significantly different from prosodic characteristics in the vicinity of non-overlapping speech. We then test the suitability of different context sizes, both preceding and following but excluding features of the overlap, for the automatic classification of collaborative and competitive overlaps. We also test whether the fusion of preceding and succeeding contexts improves the classification. Preliminary results indicate that the optimal context for classification of overlap lies at 0.2 seconds preceding the overlap and up to 0.3 seconds following it. We demonstrate that we are able to classify collaborative and competitive overlap with a median accuracy of 63%.

An Investigation into Feature Selection of Oncological Survival Prediction

Dmitry Strunkin et al. — Tue, 09 Oct 2012 01:32:17 PDT

In machine learning based clinical decision support (CDS) systems the features used to train prediction models are of paramount importance. Strong features will lead to accurate models, whereas as weak features will have the opposite effect. Feature sets can either be designed by domain experts, or automatically extracted for unstructured data that happens to be available from some process other than a CDS system. This paper compares the usefulness of structured expert-designed features to features extracted from unstructured data sources in an oncological survival prediction application scenario.

Feeling the Ambiance: Uusing Smart Ambiance to Increase Contextual Awareness in Game Agents

Colm Sloan et al. — Tue, 09 Oct 2012 01:32:16 PDT

The behaviour of non-player character game agents can be made more interesting and believable through the use of increased contextual awareness. In this paper, we present smart ambiance which allows information about the am- biance of an environment (determined by the environment itself, objects in the environment and recent events) to be used in agent plan generation. We demonstrate how this leads to contextually in uenced action selection and, in turn, more interesting and believable character behaviour.

Feasibility Study of Utility-Directed Behaviour for Computer Game Agents

Colm Sloan et al. — Tue, 09 Oct 2012 01:32:14 PDT

Utility-based control (UBC) hasn’t been widely adopted for commercial game AI. Some of the reasons for this are that UBC is perceived to be: (1) resource intensive, (2) difficult to design complex behaviours with, and (3) difficult to scale for use in complex environments. This paper investigates these perceptions to see if UBC is suitable for controlling the behaviour of non-player characters in commercial games. The investigation compares agents using a UBC system against two control systems that are more frequently used in commercial games: finite state machines (FSMs), considered a simple control system, and goal-oriented action planning (GOAP), considered a complex control system. We present a case study which suggests that: (1) UBC is more resource intensive than FSMs and less than GOAP; (2) it was reasonably simple to create complex behaviours with UBC; (3) UBC didn’t scale as well as FSMs or GOAP for use in complex environments.

Exploring the Frontier of Uncertainty Space

Rong Hu et al. — Mon, 21 May 2012 04:42:45 PDT

We aim to investigate methods balancing exploitation with exploration in active learning to improve the performance of uncertainty sampling. Two exploration guided sampling methods are compared to uncertainty sampling on various real-life datasets from the 2010 Active Learning Challenge. Our initial experiments seems to indicate that combining exploration with uncertainty sampling improves performance on certain datasets but not all.

SVM Based Active Learning with Exploration

Patrick Lindstrom et al. — Mon, 21 May 2012 04:42:43 PDT

EGAL: Exploration Guided Active Learning for TCBR

Rong Hu et al. — Mon, 21 May 2012 04:42:42 PDT

The task of building labelled case bases can be approached using active learning (AL), a process which facilitates the labelling of large collections of examples with minimal manual labelling effort. The main challenge in designing AL systems is the development of a selection strategy to choose the most informative examples to manually label. Typical selection strategies use exploitation techniques which attempt to refine uncertain areas of the decision space based on the output of a classifier. Other approaches tend to balance exploitation with exploration, selecting examples from dense and interesting regions of the domain space. In this paper we present a simple but effective exploration only selection strategy for AL in the textual domain. Our approach is inherently case-based, using only nearest-neighbour-based density and diversity measures. We show how its performance is comparable to the more computationally expensive exploitation-based approaches and that it offers the opportunity to be classifier independent.

Off to a Good Start: Using Clustering to Select the Initial Training Set in Active Learning

Rong Hu et al. — Mon, 21 May 2012 04:42:41 PDT

Active learning (AL) is used in textual classification to alleviate the cost of labelling documents for training. An important issue in AL is the selection of a representative sample of documents to label for the initial training set that seeds the process, and clustering techniques have been successfully used in this regard. However, the clustering techniques used are nondeterministic which causes inconsistent behaviour in the AL process. In this paper we first illustrate the problems associated with using non-deterministic clustering for initial training set selection in AL. We then examine the performance of three deterministic clustering techniques for this task and show that performance comparable to the non-deterministic approaches can be achieved without variations in behaviour.

Drift Detection Using Uncertainty Distribution Divergence

Patrick Lindstrom et al. — Wed, 18 Jan 2012 02:11:08 PST

Concept drift is believed to be prevalent inmost data gathered from naturally occurring processes andthus warrants research by the machine learning community.There are a myriad of approaches to concept drift handlingwhich have been shown to handle concept drift with varyingdegrees of success.

However, most approaches make the keyassumption that the labelled data will be available at nolabelling cost shortly after classification, an assumption whichis often violated. The high labelling cost in many domainsprovides a strong motivation to reduce the number of labelledinstances required to handle concept drift. Explicit detectionapproaches that do not require labelled instances to detectconcept drift show great promise for achieving this.

Ourapproach Confidence Distribution Batch Detection (CDBD)provides a signal correlated to changes in concept without usinglabelled data. We also show how this signal combined with atrigger and a rebuild policy can maintain classifier accuracywhile using a limited amount of labelled data.

Tunepal: the Traditional Musician's Toolbox

Bryan Duggan — Mon, 12 Dec 2011 02:38:03 PST

In this paper we present Tunepal, a search engine and music retrieval tool for traditional musicians that runs on an iPhone/iPod Touch (2nd generation)/iPad. Tunepal connects musicians the scores and metadata of 13,290 traditional Irish, Welsh, Scottish and Breton dance tunes. These tunes are drawn from community sources, such as the website thesession.org and “standard” references including O’Neills Dance Music of Ireland and Brendan Breathneach’s Ceol Rince Na hÉireann series. Tunes can be retrieved by typing in a title or by playing a twelve second extract from the tune on a traditional instrument. Tunepal can be used in sitiu in traditional music sessions, classes and concerts. This paper presents background information on the sources of music contained in the Tunepal corpus and describes the functionality, operation, development and usage of the app.

TSB technique: Increasing a User's Sense of immersion with Intelligent Virtual Agents

Mark T. Dunne et al. — Fri, 09 Dec 2011 08:06:35 PST

The field of Intelligent Virtual Agent (IVA) research depends heavily on immersive techniques when presenting virtual agents to end users. This sense of immersion relies on the user believing the agent to be real and present in their en- vironment, creating the Illusion of Life [21]. This poster describes the on-going research into using a combination of three rendering techniques, Twisting, Stretching and Boxing (TSB), to create a fully immersive 3D illusion for an end-user from any viewpoint, as they move freely in front of displays distributed across large populated environments. The novel approach outlined in this poster, uses head tracking, face de- tection and the TSB technique to increase the user's sense of immersion and subsequently their sense of the agent's pres- ence within the environment. Using only web-cameras, our approach is a hardware-light solution which does not require the end-user to wear any additional apparatus, such as the LED headset used in [10]. The poster primarily discusses our Approach, the Preliminary Experiments & Results, Pro- posed Evaluation Process and Future Work.

The Scenario-Oriented Method for Recording and Playing-Back Healthcare Information

Yi Ding et al. — Wed, 16 Nov 2011 06:49:00 PST

This paper proposes a new method, called the scenario-oriented method, to support the idea of recording and replaying the healthcare information such that the reporting and decision-support capabilities can be enhanced. In order to play back the changing history of certain information units, the scenario- oriented method attempts to organize related information and knowledge elements as a context so that the history of real medical activity can be recorded, and then be queried as a continuous, on-the-fly, understandable and playing-back information scenario through replay operations.

Measuring Design Metrics In Websites

Emilio Navarro et al. — Thu, 20 Oct 2011 04:06:22 PDT

The current state of the World Wide Web demands website designs that engage consumers in order to allow them to consume services or generate leads to maximize revenue. This paper describes a software quality factor to measure the success of websites by analyzing web design structure and not relying only on websites traffic data. It is also documents the requirements and architecture to build a software tool that measures criteria for determining Engagibility. A new set of social criteria to be measured for current website philosophy is also proposed.

Accessible Rich Internet Applications: the Search Engine Challenge

Angela Kielthy — Mon, 26 Sep 2011 06:08:57 PDT

The perception that Rich Internet Applications (RIAs) and Accessible Rich Internet Applications (ARIAs) are inaccessible to search engines is perhaps one of the main factors that hinder their wider adoption by the web development community. Recent announcements that RIAs and ARIAs are becoming more search engine friendly is provoking web developers to look for further information and evidence that will support or refute these announcements.

This paper outlines research undertaken and tests performed to establish if RIAs and ARIAs developed using Adobe Flex are crawlable and indexable by the Google search engine by default.

The conclusion drawn from testing is that RIAs and ARIAs are not yet fully supported by the Google search engine. They can however be made search engine friendly by employing third party software and some imaginative coding techniques.

This conclusion contradicts various published statements from search engine providers such as Google, RIA software providers such as Adobe and numerous field experts.

Locating Mineral Exploration Targets using a Geographical Information System

Finnian O'Connor — Fri, 23 Sep 2011 01:13:22 PDT

This paper outlines the research and development of a complete open source geographic information system (GIS) that offers real-time geoprocessing over the Internet. The premise of the geoprocessing is to locate mineral exploration targets that have high potential for success based on parameters chosen by the end-user of the system.

Components integrated in the system include a spatial database PostGIS, a GIS processing engine GRASS, a GIS server GeoServer, a web server Apache, and front-end technologies OpenLayers and GeoExt. Appropriate data was sourced from the Geological Survey of Ireland to be used for the geoprocessing.

With all the components of the GIS integrated, an individual not specialised in the use of a GIS can interact with and interrogate the data through a web browser. The GIS then provides a vital role as a decision support system for locating mineral exploration targets.