As firms move towards data-driven decision-making using algorithmic systems, concerns are raised regarding the lack of transparency of these systems which could have ramifications related to users’ trust and the potential for provoking discriminatory decisions. Although previous research has developed methods to improve algorithmic transparency, little empirical evidence exists regarding the extent of the effectiveness of these approaches. Drawing upon Rest’s theory of ethical decision-making and the literature on algorithmic transparency and bias, we investigate the effectiveness of feature importance (FI), a common transparency-enhancing approach, which illustrates the nature and the weights of the features utilised by an algorithm. Through an online experiment employing a fictitious tool that provided recommendations for selecting employees for a promotion-related training programme, we find that FI is effective when biased recommendations include direct discrimination (i.e. when individuals are treated less favourably on protected grounds such as gender); but is of little assistance when discrimination is indirect (i.e. when a criterion or practice that is apparently neutral, disadvantages a group of individuals who are of a protected class). Additionally, we propose a new transparency approach, using aggregated demographic information, to accompany FI in indirect discrimination circumstances and report the results of testing its effects.

The fake news phenomenon has exposed the vulnerability of individuals and societies to information manipulation in social media. We conducted two studies to understand why people believe in fake news and propose a simple IT intervention method that can help in detecting disinformation. In Study 1, we designed a laboratory experiment using behavioral and neurophysiological tools to test two competing theories in the disinformation literature. Both behavioral and neurophysiological evidence support the classical reasoning account hypotheses and reject the motivated reasoning predictions, suggesting that the lack of actively open-minded thinking (AOT) is linked to the belief in fake news. An intervention method was designed (i.e., performance feedback) that reduces individuals’ overconfidence in their ability to detect fake news and encourages more analytical thinking. In Study 2, we conducted an online survey presenting participants with their performance feedback halfway through the survey. The results show that the intervention increased participants’ performance by 14%. Our study contributes to the research on fake news by providing behavioral and neurophysiological evidence in support of the classical reasoning account. It also offers a simple and practical method that increases users’ ability to detect fake news.

More than 4 billion unique users are now using different social media platforms. This provides firms with a new avenue to connect and converse with their targeted customers in a dialogue fashion and create better customer engagement. Although customer engagement has been studied, no study to date has focused on the concept of customer engagement in social media through the lens of users' participation in (co)creation of content on social media. To that end, this study strives to conceptualize and validate the concept of customer engagement with a social media post, proposing a research model which investigates the effects of popularity, discourse logic and argument frame on customer engagement with a social media post. This study also investigates to what extent those effects vary by the product lifecycle stage and social media platform type, using secondary data extracted from Instagram and Twitter on a group of products (i.e., vehicles) at different lifecycle stages to address its research objective. This study's findings would help firms make better decisions on enhancing their customer engagement on social media platforms.

Although significantly more organizations have recently invested in Data Analytics (DA), most business users do not execute DA recommendations. Conceptualizing the novel concept of DA recommendation quality, shaped by tool, data and analyst quality, this study draws on the Stimulus-Organism-Response framework to investigate its effect on shaping users’ perceptions of concordance, actionability, and risk, ultimately influencing their DA recommendation execution. The theoretical model is empirically validated using a sample of senior managers across North America. Enriching DA literature, this study shows that DA recommendation quality is positively associated with recommendation execution, while actionability is the dominant factor in increasing it.

This paper reviews the literature related to the concept of information quality (IQ) in the context of the online environment to understand how it has been conceptualised in the Information Systems (IS) literature, to identify gaps in the literature, and to make recommendations for future research. This study also extends the Information Processing Model (IPM) and proposes a novel dynamic model of online IQ perceptions and impacts. To understand the online IQ construct, an in-depth review of the literature is conducted. Overall, we reviewed 508 journal articles published between 1995 and 2016. This study found that there is a substantial gap in the IS literature in regards to a full understanding of the online IQ construct and its relation with the IPM. More research is needed to understand the way individuals process and respond to online information. Further, although the importance of context in theory development has been emphasised in the literature, very few empirical articles have explored the impact of contextual factors on the IPM as well as perceived online IQ. The findings provide online vendors with important guidelines that should be considered to motivate individuals to acquire more information from their websites that could lead to more satisfied and loyal users.

Though researchers agree on the role of psychological forces on individuals' decision-making and emphasise the need for developing decision-support systems DSS that make individuals aware of these forces, a framework that can guide us in building such systems is still non-existent. In this article, we attempt to bridge this gap by proposing an agent-based debiasing framework for developing investment DSS. Identifying the primary characteristics of major biases influencing investment decisions through a thorough literature review, we propose a taxonomy to categorise them as cognitive, affective or conative. Cognitive biases are information-processing biases. Affective biases involve general moods and emotions. Conative biases are relatively stable personality traits such as overconfidence and inertia. We then outline debiasing strategies for each of these bias categories and identify decision-support characteristics necessary in software agents to carry out the appropriate debiasing tasks. An agent-based DSS architecture is then proposed, and a detailed trading example triggering a sample bias detection and debiasing algorithm is discussed to demonstrate the feasibility of the proposed system.

Compared to offline shopping, the online shopping experience may be viewed as lacking human warmth and sociability as it is more impersonal, anonymous, automated and generally devoid of face-to-face interactions. Thus, understanding how to create customer loyalty in online environments (e-Loyalty) is a complex process. In this paper a model for e-Loyalty is proposed and used to examine how varied conditions of social presence in a B2C e-Services context influence e-Loyalty and its antecedents of perceived usefulness, trust and enjoyment. This model is examined through an empirical study involving 185 subjects using structural equation modeling techniques. Further analysis is conducted to reveal gender differences concerning hedonic elements in the model on e-Loyalty.

We present a framework for clustering and classifying cheque images according to their payee-line content. The features used in the clustering and classification processes are extracted from the wavelet domain by means of thresholding and counting of wavelet coefficients. The feasibility of this framework is tested on a database of 2620 cheque images. This database consists of cheques from 10 different accounts. Each account is written by a different person. Clustering and classification are performed separately on each account using distance-based techniques. We achieved correct-classification rates of 86% and 81% for the supervised and unsupervised learning cases, respectively. These rates are the average of correct-classification rates obtained from the 10 different accounts.

This paper describes a new method for stroke extraction from grayscale images based on an effective local adaptive kernel technique that assigns a mark for each pixel in the image signifying its relative importance with respect to neighboring pixels. The assigned marks are based on accumulating evidence of the relative importance of each pixel based on multiple comparisons with neighboring pixels under different windows. A global threshold is then applied to the marks produced in the first step to reach a conclusion as to whether a pixel ought to be preserved as part of a stroke or be dropped as part of the background. This technique effectively combines local and global thresholding techniques and shows promising performance especially on images with non-uniform backgrounds or embedded light strokes. The advantage of this scheme is also demonstrated in terms of superior overall amount recognition performance for financial document processing systems utilizing it.

A speech recognition system is developed utilising multi-layer perceptrons (MLPs) as speech-frame predictors. A Markov chain is used to control changes in the MLP's weight parameters. Analytical results and speech recognition experiments indicate that when joint (nonlinear/linear) prediction is performed within the hidden layer of the MLP, the model is better at capturing long-term data correlations which improves speech recognition performance. A discriminative training technique based on the maximum mutual information criterion is presented for training this class of models. The performance of the system on vowel classification tasks when trained with this method is shown to be superior to the same system trained using the maximum likelihood training criterion.>