Thesis (M.A.)--Bowling Green State University, 2009. / Document formatted into pages; contains xi, 122 p. Includes bibliographical references.

Thesis (Ph. D.)--University of Washington, 2000. / Vita. Includes bibliographical references (leaves 142-148).

Master-Arbeit Univ. St. Gallen, 2004.

Master-Arbeit Univ. St. Gallen, 2007.

Universiẗat, Diss., 2004--Münster (Westfalen).

Proefschrift Universiteit Maastricht. / Met lit. opg. - Met samenvatting in het Nederlands.

abstract: Exabytes of data are created online every day. This deluge of data is no more apparent than it is on social media. Naturally, finding ways to leverage this unprecedented source of human information is an active area of research. Social media platforms have become laboratories for conducting experiments about people at scales thought unimaginable only a few years ago. Researchers and practitioners use social media to extract actionable patterns such as where aid should be distributed in a crisis. However, the validity of these patterns relies on having a representative dataset. As this dissertation shows, the data collected from social media is seldom representative of the activity of the site itself, and less so of human activity. This means that the results of many studies are limited by the quality of data they collect. The finding that social media data is biased inspires the main challenge addressed by this thesis. I introduce three sets of methodologies to correct for bias. First, I design methods to deal with data collection bias. I offer a methodology which can find bias within a social media dataset. This methodology works by comparing the collected data with other sources to find bias in a stream. The dissertation also outlines a data collection strategy which minimizes the amount of bias that will appear in a given dataset. It introduces a crawling strategy which mitigates the amount of bias in the resulting dataset. Second, I introduce a methodology to identify bots and shills within a social media dataset. This directly addresses the concern that the users of a social media site are not representative. Applying these methodologies allows the population under study on a social media site to better match that of the real world. Finally, the dissertation discusses perceptual biases, explains how they affect analysis, and introduces computational approaches to mitigate them. The results of the dissertation allow for the discovery and removal of different levels of bias within a social media dataset. This has important implications for social media mining, namely that the behavioral patterns and insights extracted from social media will be more representative of the populations under study. / Dissertation/Thesis / Doctoral Dissertation Computer Science 2017

The generation of predictions shapes our experience of the world around us. By making inferences about what is likely to happen within a given scenario, we can conserve cognitive resources and enhance our prospects of survival. Predictive coding accounts of perception indicate that this is achieved by minimally processing information that is consistent with our expectations, and prioritising the processing of unexpected or meaningful information. Predictions are also beneficial in situations where accurate perception is difficult, and clues like contextual information allow expectations to ‘fill in the blanks’ when sensory information is noisy or ambiguous. This comes at a cost, however, and a reliance upon expectations can lead to perceptual biases, and in certain cases misperceptions. According to Assimilation Contrast Theory (ACT) and the Affective Expectation Model (AEM), when we attempt to judge affectively ambiguous stimuli, our judgements are biased by expectations in a similar manner. If stimuli are within an acceptable range of an existing expectation, minor discrepancies will be ignored and judgements of those stimuli will fall in line with expectations (assimilation). Alternatively, if the affective discrepancy between expectation and stimulus is so large that it is acknowledged, the extent of that discrepancy will be exaggerated instead (contrast). This thesis aimed to investigate the boundaries and time-course of these effects. A series of behavioural experiments were conducted to investigate: (i) whether predictive cues promoted a state of affective readiness, where judgements across a range of stimuli were biased based upon the assumption that they were broadly part of a positive or negative category (chapters 3 and 4); (ii) whether affective biases (assimilation effects) persisted over time (chapters 5 and 6); and (iii) whether the boundaries of affective and perceptual assimilation effects remained consistent over time (chapter 6 and 7). Psychophysical measures of affective bias indicated that predictive cues influenced participants to judge the same stimuli differently, according to whether they expected those stimuli to be positive or negative. Furthermore, after expectations were learned, judgements of the same stimuli continued to be biased toward expectations after a period of one week. When stimuli from affectively or perceptually distinct categories were manipulated slowly over time, to the point where they became identical, judgements of those stimuli continued to be influenced by the expectation that they should remain distinct. These findings indicate that the boundaries of perceptual and affective assimilation effects may not be static, and if deviations from expectation are small enough to go generally unnoticed, people may update their internal representations of items over time, and the boundaries of acceptance which surround those representations.

Depression is a global problem, causing disability and economic burden. Many people currently do not obtain treatment. Development of more accessible, cost-effective treatments is essential. An identified mechanism by which depression treatments work is through modifying underlying negative cognitive biases, which mediate changes in mood. A specific negative information-processing bias in depression is a tendency to interpret ambiguous facial expressions as sad rather than happy. The emotion recognition task is a treatment paradigm developed as a cognitive bias modification intervention to target this emotion recognition bias. Previous studies showed promising signs that this novel intervention could modify biases in people with low mood outside of laboratory conditions and potential to increase positive affect within laboratory conditions. The current study built on these developments, aiming to investigate, using a randomised controlled trial with follow-up at 2 and 6 weeks, whether a web-based version of the emotion recognition task could reduce depressive symptoms in addition to modifying emotion recognition biases. An analogue sample of 124 participants with low mood was recruited. Evidence was found that the intervention modified participants’ biases, compared to the control group but there was no evidence of improvement in mood. Study limitations included a high rate of attrition and non-adherence to the intervention. Future recommendations include modifying the intervention to increase acceptability, investigating generalizability of increased positive bias to different stimuli, and identifying consistent reductions in symptoms of depression before examining its efficacy with a clinical population.

It has been suggested that depressed individuals have difficulties decoding emotional facial expressions in others contributing to a negative cycle of interpersonal difficulties. Some studies have demonstrated global deficits in the processing of emotional facial expressions compared to non-depressed participants, whereas others have noted differences for specific emotions. Methodological issues, including the operationalization of accuracy and bias and the examination of a limited range of emotion and intensity, can partially explain the mixed findings. The aim of the current study was to examine differences in accuracy in the detection of emotional facial expressions in participants with MDD (currently depressed, partially remitted, and those with a lifetime history of MDD) and non-depressed comparisons. Methodological limitations of previous studies were addressed by: (a) using the unbiased hit rate (Wagner, 1993), which is a more precise measure of accuracy for specific emotion, (b) using a more precise measure of judgment bias, taking into account the overuse or underuse of specific emotion categories, (c) including the six basic emotions, and (d) incorporating expressions ranging from 20%-100% intensity. Of secondary interest was to determine whether transient mood state is predictive of accuracy scores regardless of diagnostic status. Thirty-seven depressed and 34 non-depressed participants recruited from the ROHCG Mood Disorders program and the University of Ottawa took part in this study. Clinical status was assessed using the Structured Clinical Interview for the DSM-IV (SCID-IV) and the Beck Depression Inventory-II (BDI-II). Participants also completed the Profile of Mood States-Bipolar (POMS-BI) form to assess mood state at the time of testing. The facial recognition task consisted of happiness, sadness, anger, fear, disgust, and surprise at 20%-100% intensity, presented for 500 ms. Participants pressed a computer key to identify the emotion that was presented. Results indicated that both groups of depressed participants were more accurate than non-depressed participants in detecting anger at 20% intensity. Depressed participants also showed a bias away from surprise. Group differences at high intensity were non-significant, however, participants with current depression and partial remission showed a bias towards anger at 50% intensity. Regression analyses were performed using the POMS-Agreeable Hostile (POMS-AH) and POMS-Elated Depressed (POMS-ED) scales to determine whether mood state was predictive of accuracy in the detection of anger and sadness. Regression models predicting accuracy were non-significant. Results of this study are considered in the context of cognitive and cognitive-interpersonal theories of MDD.