Search

Feature selection is a prevalent data preprocessing paradigm for various learning tasks. Due to the expensive cost of acquiring supervision information, unsupervised feature selection sparks great interests recently. However, existing unsupervised feature selection algorithms do not have fairness considerations and suffer from a high risk of amplifying discrimination by selecting features that are over associated with protected attributes such as gender, race, and ethnicity. In this paper, we make an initial investigation of the fairness-aware unsupervised feature selection problem and develop a principled framework, which leverages kernel alignment to find a subset of high-quality features that can best preserve the information in the original feature space while being minimally correlated with protected attributes. Specifically, different from the mainstream in-processing debiasing methods, our proposed framework can be regarded as a model-agnostic debiasing strategy that eliminates biases and discrimination before downstream learning algorithms are involved. Experimental results on multiple real-world datasets demonstrate that our framework achieves a good trade-off between utility maximization and fairness promotion.

Recent conversational recommender systems (CRS) provide a promising solution to accurately capture a user's preferences by communicating with users in natural language to interactively guide them while pro-actively eliciting their current interests. Previous research on this mainly focused on either learning a supervised model with semantic features extracted from the user's responses, or training a policy network to control the dialogue state. However, none of them has considered the issue of popularity bias in a CRS. This paper proposes a human-in-the-loop popularity debiasing framework that integrates real-time semantic understanding of open-ended user utterances as well as historical records, while also effectively managing the dialogue with the user. This allows the CRS to balance the recommendation performance as well as the item popularity so as to avoid the well-known "long-tail'' effect. We demonstrate the effectiveness of our approach via experiments on two conversational recommendation datasets, and the results confirm that our proposed approach achieves high-accuracy recommendation while mitigating popularity bias.

Recent years have seen growing interest in modeling visitor engagement in museums with multimodal learning analytics. In parallel, there has also been growing concern about issues of fairness and encoded bias in machine learning models. In this paper, we investigate bias detection and mitigation techniques to address issues of algorithmic fairness in multimodal models of museum visitor visual attention. We employ slicing analysis using the Absolute Between-ROC Area (ABROCA) statistic to detect encoded bias present in multimodal models of visitor visual attention trained with facial expression and posture data from visitor interactions with a game-based museum exhibit about environmental sustainability. We investigate instances of gender bias that arise between different combinations of modalities across several machine learning techniques. We also measure the effectiveness of two different debiasing strategies—learned fair representations and reweighing—when applied to the trained multimodal visitor attention models. Results indicate that patterns of bias can arise across different modality combinations for the different visitor visual attention models, and there is often an inherent tradeoff between predictive accuracy and ABROCA. Analyses suggest that debiasing strategies tend to be more effective on multimodal models of visitor visual attention than their unimodal counterparts

Compositional action recognition is a novel challenge in the computer vision community and focuses on revealing the different combinations of verbs and nouns instead of treating subject-object interactions in videos as individual instances only. Existing methods tackle this challenging task by simply ignoring appearance information or fusing object appearances with dynamic instance tracklets. However, those strategies usually do not perform well for unseen action instances. For that, in this work we propose a novel learning framework called Counterfactual Debiasing Network (CDN) to improve the model generalization ability by removing the interference introduced by visual appearances of objects/subjects. It explicitly learns the appearance information in action representations and later removes the effect of such information in a causal inference manner. Specifically, we use tracklets and video content to model the factual inference by considering both appearance information and structure information. In contrast, only video content with appearance information is leveraged in the counterfactual inference. With the two inferences, we conduct a causal graph which captures and removes the bias introduced by the appearance information by subtracting the result of the counterfactual inference from that of the factual inference. By doing that, our proposed CDN method can better recognize unseen action instances by debiasing the effect of appearances. Extensive experiments on the Something-Else dataset clearly show the effectiveness of our proposed CDN over existing state-of-the-art methods.

Recommender systems (RS) have demonstrated great success in information seeking. Recent years have witnessed a large number of work on inventing recommendation models to better fit user behavior data. However, user behavior data is observational rather than experimental. This makes various biases widely exist in the data, including but not limited to selection bias, position bias, exposure bias. Blindly fitting the data without considering the inherent biases will result in many serious issues, e.g., the discrepancy between offline evaluation and online metrics, hurting user satisfaction and trust on the recommendation service, etc. To transform the large volume of research models into practical improvements, it is highly urgent to explore the impacts of the biases and develop debiasing strategies when necessary. Therefore, bias issues and solutions in recommender systems have drawn great attention from both academic and industry. In this tutorial, we aim to provide an systemic review of existing work on this topic. We will introduce six types of biases in recommender system, along with their definitions and characteristics; review existing debiasing solutions, along with their strengths and weaknesses; and identify some open challenges and future directions. We hope this tutorial could stimulate more ideas on this topic and facilitate the development of debiasing recommender systems.

As it becomes prevalent that user information exists in multiple platforms or services, cross-domain recommendation has been an important task in industry. Although it is well known that users tend to show different preferences in different domains, existing studies seldom model how domain biases affect user preferences. Focused on this issue, we develop a casual-based approach to mitigating the domain biases when transferring the user information cross domains. To be specific, this paper presents a novel debiasing learning based cross-domain recommendation framework with causal embedding. In this framework, we design a novel Inverse-Propensity-Score (IPS) estimator designed for cross-domain scenario, and further propose three kinds of restrictions for propensity score learning. Our framework can be generally applied to various recommendation algorithms for cross-domain recommendation. Extensive experiments on both public and industry datasets have demonstrated the effectiveness of the proposed framework.

Popularity bias is a long-standing challenge in recommender systems: popular items are overly recommended at the expense of less popular items that users may be interested in being under-recommended. Such a bias exerts detrimental impact on both users and item providers, and many efforts have been dedicated to studying and solving such a bias. However, most existing works situate the popularity bias in a static setting, where the bias is analyzed only for a single round of recommendation with logged data. These works fail to take account of the dynamic nature of real-world recommendation process, leaving several important research questions unanswered: how does the popularity bias evolve in a dynamic scenario? what are the impacts of unique factors in a dynamic recommendation process on the bias? and how to debias in this long-term dynamic process? In this work, we investigate the popularity bias in dynamic recommendation and aim to tackle these research gaps. Concretely, we conduct an empirical study by simulation experiments to analyze popularity bias in the dynamic scenario and propose a dynamic debiasing strategy and a novel False Positive Correction method utilizing false positive signals to debias, which show effective performance in extensive experiments.

Word vector embeddings have been shown to contain and amplify biases in data they are extracted from. Consequently, many techniques have been proposed to identify, mitigate, and attenuate these biases in word representations. In this tutorial, we will review a collection of state-of-the-art debiasing techniques. To aid this, we provide an open source web-based visualization tool and offer hands-on experience in exploring the effects of these debiasing techniques on the geometry of high-dimensional word vectors. To help understand how various debiasing techniques change the underlying geometry, we decompose each technique into interpretable sequences of primitive operations, and study their effect on the word vectors using dimensionality reduction and interactive visual exploration.

Recommender systems rely on user behavior data like ratings and clicks to build personalization model. However, the collected data is observational rather than experimental, causing various biases in the data which significantly affect the learned model. Most existing work for recommendation debiasing, such as the inverse propensity scoring and imputation approaches, focuses on one or two specific biases, lacking the universal capacity that can account for mixed or even unknown biases in the data. Towards this research gap, we first analyze the origin of biases from the perspective of \textit{risk discrepancy} that represents the difference between the expectation empirical risk and the true risk. Remarkably, we derive a general learning framework that well summarizes most existing debiasing strategies by specifying some parameters of the general framework. This provides a valuable opportunity to develop a universal solution for debiasing, e.g., by learning the debiasing parameters from data. However, the training data lacks important signal of how the data is biased and what the unbiased data looks like. To move this idea forward, we propose \textit{AotoDebias} that leverages another (small) set of uniform data to optimize the debiasing parameters by solving the bi-level optimization problem with meta-learning. Through theoretical analyses, we derive the generalization bound for AutoDebias and prove its ability to acquire the appropriate debiasing strategy. Extensive experiments on two real datasets and a simulated dataset demonstrated effectiveness of AutoDebias. The code is available at \url{https://github.com/DongHande/AutoDebias}., Comment: Accepted by SIGIR 2021

Post-click conversion, as a strong signal indicating the user preference, is salutary for building recommender systems. However, accurately estimating the post-click conversion rate (CVR) is challenging due to the selection bias, i.e., the observed clicked events usually happen on users' preferred items. Currently, most existing methods utilize counterfactual learning to debias recommender systems. Among them, the doubly robust (DR) estimator has achieved competitive performance by combining the error imputation based (EIB) estimator and the inverse propensity score (IPS) estimator in a doubly robust way. However, inaccurate error imputation may result in its higher variance than the IPS estimator. Worse still, existing methods typically use simple model-agnostic methods to estimate the imputation error, which are not sufficient to approximate the dynamically changing model-correlated target (i.e., the gradient direction of the prediction model). To solve these problems, we first derive the bias and variance of the DR estimator. Based on it, a more robust doubly robust (MRDR) estimator has been proposed to further reduce its variance while retaining its double robustness. Moreover, we propose a novel double learning approach for the MRDR estimator, which can convert the error imputation into the general CVR estimation. Besides, we empirically verify that the proposed learning scheme can further eliminate the high variance problem of the imputation learning. To evaluate its effectiveness, extensive experiments are conducted on a semi-synthetic dataset and two real-world datasets. The results demonstrate the superiority of the proposed approach over the state-of-the-art methods. The code is available at https://github.com/guosyjlu/MRDR-DL., 10 pages, 3 figures, accepted by SIGIR 2021

Conference Name: EASE 2021: Evaluation and Assessment in Software Engineering Date of Conference: June 2021 Reviewer selection in modern code review is crucial for effective code reviews. Several techniques exist for recommending reviewers appropriate for a given pull request (PR). Most code reviewer recommendation techniques in the literature build and evaluate their models based on datasets collected from real projects using open-source or industrial practices. The techniques invariably presume that these datasets reliably represent the “ground truth.” In the context of a classification problem, ground truth refers to the objectively correct labels of a class used to build models from a dataset or evaluate a model’s performance. In a project dataset used to build a code reviewer recommendation system, the recommended code reviewer picked for a PR is usually assumed to be the best code reviewer for that PR. However, in practice, the recommended code reviewer may not be the best possible code reviewer, or even a qualified one. Recent code reviewer recommendation studies suggest that the datasets used tend to suffer from systematic labeling bias, making the ground truth unreliable. Therefore, models and recommendation systems built on such datasets may perform poorly in real practice. In this study, we introduce a novel approach to automatically detect and eliminate systematic labeling bias in code reviewer recommendation systems. The bias that we remove results from selecting reviewers that do not ensure a permanently successful fix for a bug-related PR. To demonstrate the effectiveness of our approach, we evaluated it on two open-source project datasets —HIVE and QT Creator— and with five code reviewer recommendation techniques —Profile-Based, RSTrace, Naive Bayes, k-NN, and Decision Tree. Our debiasing approach appears promising since it improved the Mean Reciprocal Rank (MRR) of the evaluated techniques up to 26% in the datasets used.

A growing proportion of human interactions are digitized on social media platforms and subjected to algorithmic decision-making, and it has become increasingly important to ensure fair treatment from these algorithms. In this work, we investigate gender bias in collaborative-filtering recommender systems trained on social media data. We develop neural fair collaborative filtering (NFCF), a practical framework for mitigating gender bias in recommending career-related sensitive items (e.g. jobs, academic concentrations, or courses of study) using a pre-training and fine-tuning approach to neural collaborative filtering, augmented with bias correction techniques. We show the utility of our methods for gender de-biased career and college major recommendations on the MovieLens dataset and a Facebook dataset, respectively, and achieve better performance and fairer behavior than several state-of-the-art models.

A well-known challenge in leveraging implicit user feedback like clicks to improve real-world search services and recommender systems is its inherent bias. Most existing click models are based on the examination hypothesis in user behaviors and differ in how to model such an examination bias. However, they are constrained by assuming a simple position-based bias or enforcing a sequential order in user examination behaviors. These assumptions are insufficient to capture complex real-world user behaviors and hardly generalize to modern user interfaces (UI) in web applications (e.g., results shown in a grid view). In this work, we propose a fully data-driven neural model for the examination bias, Cross-Positional Attention (XPA), which is more flexible in fitting complex user behaviors. Our model leverages the attention mechanism to effectively capture cross-positional interactions among displayed items and is applicable to arbitrary UIs. We employ XPA in a novel neural click model that can both predict clicks and estimate relevance. Our experiments on offline synthetic data sets show that XPA is robust among different click generation processes. We further apply XPA to a large-scale real-world recommender system, showing significantly better results than baselines in online A/B experiments that involve millions of users. This validates the necessity to model more complex user behaviors than those proposed in the literature.

This paper connects equal opportunity to popularity bias in implicit recommenders to introduce the problem of popularity-opportunity bias. That is, conditioned on user preferences that a user likes both items, the more popular item is more likely to be recommended (or ranked higher) to the user than the less popular one. This type of bias is harmful, exerting negative effects on the engagement of both users and item providers. Thus, we conduct a three-part study: (i) By a comprehensive empirical study, we identify the existence of the popularity-opportunity bias in fundamental matrix factorization models on four datasets; (ii) coupled with this empirical study, our theoretical study shows that matrix factorization models inherently produce the bias; and (iii) we demonstrate the potential of alleviating this bias by both in-processing and post-processing algorithms. Extensive experiments on four datasets show the effective debiasing performance of these proposed methods compared with baselines designed for conventional popularity bias.

Graph neural networks (GNNs) have shown great power in modeling graph structured data. However, similar to other machine learning models, GNNs may make predictions biased on protected sensitive attributes, e.g., skin color and gender. Because machine learning algorithms including GNNs are trained to reflect the distribution of the training data which often contains historical bias towards sensitive attributes. In addition, the discrimination in GNNs can be magnified by graph structures and the message-passing mechanism. As a result, the applications of GNNs in sensitive domains such as crime rate prediction would be largely limited. Though extensive studies of fair classification have been conducted on i.i.d data, methods to address the problem of discrimination on non-i.i.d data are rather limited. Furthermore, the practical scenario of sparse annotations in sensitive attributes is rarely considered in existing works. Therefore, we study the novel and important problem of learning fair GNNs with limited sensitive attribute information. FairGNN is proposed to eliminate the bias of GNNs whilst maintaining high node classification accuracy by leveraging graph structures and limited sensitive information. Our theoretical analysis shows that FairGNN can ensure the fairness of GNNs under mild conditions given limited nodes with known sensitive attributes. Extensive experiments on real-world datasets also demonstrate the effectiveness of FairGNN in debiasing and keeping high accuracy.

Automated systems for detecting harmful social media content are afflicted by a variety of biases, some of which originate in their training datasets. In particular, some systems have been shown to propagate racial dialect bias: they systematically classify content aligned with the African American English (AAE) dialect as harmful at a higher rate than content aligned with White English (WE). This perpetuates prejudice by silencing the Black community. Towards this problem we adapt and apply two existing bias mitigation approaches: preferential sampling pre-processing and adversarial debiasing in-processing. We analyse the impact of our interventions on model performance and propagated bias. We find that when bias mitigation is employed, a high degree of predictive accuracy is maintained relative to baseline, and in many cases bias against AAE in harmful tweet predictions is reduced. However, the specific effects of these interventions on bias and performance vary widely between dataset contexts. This variation suggests the unpredictability of autonomous harmful content detection outside of its development context. We argue that this, and the low performance of these systems at baseline, raise questions about the reliability and role of such systems in high-impact, real-world settings.

Item-to-item recommendation (e.g., “People who like this also like...”) is a ubiquitous and important type of recommendation in real-world systems. Observational data from historical interaction logs abound in these settings. However, since virtually all observational data exhibit biases, such as time-in-inventory or interface biases, it is crucial that recommender algorithms account for these biases. In this paper, we develop a principled approach for item-to-item recommendation based on causal inference and present a practical and highly effective method for estimating the causal parameters from a small annotated dataset. Empirically, we find that our approach substantially improves upon existing methods while requiring only small amounts of annotated data.

The widespread usage of e-commerce websites in daily life and the resulting wealth of implicit feedback data form the foundation for systems that train and test e-commerce search ranking algorithms. While convenient to collect, implicit feedback data inherently suffers from various types of bias since user feedback is limited to products they are exposed to by existing search ranking algorithms and impacted by how the products are displayed. In the literature, a vast majority of existing methods have been proposed towards unbiased learning to rank for list-based web search scenarios. However, such methods cannot be directly adopted by e-commerce websites mainly for two reasons. First, in e-commerce websites, search engine results pages (SERPs) are displayed in 2-dimensional grids. The existing methods have not considered the difference in user behavior between list-based web search and grid-based product search. Second, there can be multiple types of implicit feedback (e.g., clicks and purchases) on e-commerce websites. We aim to utilize all types of implicit feedback as the supervision signals. In this work, we extend unbiased learning to rank to the world of e-commerce search via considering a grid-based product search scenario. We propose a novel framework which (1) forms the theoretical foundations to allow multiple types of implicit feedback in unbiased learning to rank and (2) incorporates the row skipping and slower decay click models to capture unique user behavior patterns in grid-based product search for inverse propensity scoring. Through extensive experiments on real-world e-commerce search log datasets across browsing devices and product taxonomies, we show that the proposed framework outperforms the state of the art unbiased learning to rank algorithms. These results also reveal important insights on how user behavior patterns vary in e-commerce SERPs across browsing devices and product taxonomies.

Open world database management systems assume tuples not in the database still exist and are becoming an increasingly important area of research. We present Themis, the first open world database that automatically rebalances arbitrarily biased samples to approximately answer queries as if they were issued over the entire population. We leverage apriori population aggregate information to develop and combine two different approaches for automatic debiasing: sample reweighting and Bayesian network probabilistic modeling. We build a prototype of Themis and demonstrate that Themis achieves higher query accuracy than the default AQP approach, an alternative sample reweighting technique, and a variety of Bayesian network models while maintaining interactive query response times. We also show that Themis is robust to differences in the support between the sample and population, a key use case when using social media samples.

Narrative visualization is a popular style of data-driven storytelling. Authors use this medium to engage viewers with complex and sometimes controversial issues. A challenge for authors is to not only deliver new information, but to also overcome people’s biases and misconceptions. We study how people adjust their attitudes toward (or away from) a message experienced through a narrative visualization. In a mixed-methods analysis, we investigate whether eliciting participants’ prior beliefs, and visualizing those beliefs alongside actual data, can increase narrative persuasiveness. We find that incorporating priors does not significantly affect attitudinal change. However, participants who externalized their beliefs expressed greater surprise at the data. Their comments also indicated a greater likelihood of acquiring new information, despite the minimal change in attitude. Our results also extend prior findings, showing that visualizations are more persuasive than equivalent textual data representations for exposing con- tentious issues. We discuss the implications and outline future research directions.

The Web is the most powerful communication medium and the largest public data repository that humankind has created. Its content ranges from great reference sources such as Wikipedia to ugly fake news. Indeed, social (digital) media is just an amplifying mirror of ourselves, good or bad. Indeed, as all people has their own cultural and cognitive biases (e.g., confirmation bias, see Figure 1), web content as well as our web interactions are tainted with them. Data bias includes redundancy and spam, while interaction bias includes activity and presentation/exposure bias. In addition, sometimes algorithms add bias, particularly in the context of search and recommendation systems. As bias generates bias, we stress the importance of debiasing data as well as using the context and other techniques such as explore & exploit, to break filter bubbles.Our main goal is to make people aware of the different biases that affect all of us on the Web as well as stress that we should design inclusive content such that we help people with learning disabilities (e.g., dyslexia) or vision problems (e.g., daltonism), among others. This makes the Web more accessible for all people. Finally, we have to remark that awareness is the first step to be able to fight and reduce the vicious cycle of web bias. For more details see my article on this topic [1].

Word embeddings are a widely used set of natural language processing techniques that map words to vectors of real numbers. These vectors are used to improve the quality of generative and predictive models. Recent studies demonstrate that word embeddings contain and amplify biases present in data, such as stereotypes and prejudice. In this study, we provide a complete overview of bias in word embeddings. We develop a new technique for bias detection for gendered languages and use it to compare bias in embeddings trained on Wikipedia and on political social media data. We investigate bias diffusion and prove that existing biases are transferred to further machine learning models. We test two techniques for bias mitigation and show that the generally proposed methodology for debiasing models at the embeddings level is insufficient. Finally, we employ biased word embeddings and illustrate that they can be used for the detection of similar biases in new data. Given that word embeddings are widely used by commercial companies, we discuss the challenges and required actions towards fair algorithmic implementations and applications.

Research in fairness, accountability, and transparency (FAT) in socio-technical systems needs to take into account how practices of computing are entrenched with power relations in complex and multi-layered ways. Trying to disentangle the way in which structural discrimination and normative computational concepts and methods are intertwined, this frequently raises the question of WHO are the actors that shape technologies and research agendas---who gets to speak and to define bias, (un)fairness, and discrimination? "Deconstructing FAT" is a CRAFT workshop that aims at complicating this question by asking how "we" as researchers in FAT (often unknowingly) mobilize implicit assumptions, values and beliefs that reflect our own embeddedness in power relations, our disciplinary ways of thinking, and our historically, locally, and culturally-informed ways of solving computational problems or approaching our research. This is a vantage point to make visible and analyze the normativity of technical approaches, concepts and methods that are part of the repertoire of FAT research. Inspired by a previous international workshop [1], this CRAFT workshop engages an interdisciplinary panel of FAT researchers in a deconstruction exercise that traces the following issues: (1) FAT research frequently speaks of social bias that is amplified by algorithmic systems, of the problem of discriminatory consequences that is to be solved, and of underprivileged or vulnerable groups that need to be protected. What does this perspectivity imply in terms of the approaches, methods and metrics that are being applied? How do methods of debiasing and discrimination-awareness enact the epistemic power of a perspective of privilege as their norm? (2) FAT research has emphasized the need for multi- or interdisciplinary approaches to get a grip on the complex intertwining of social power relations and the normativity of computational methods, norms and practices. Clearly, multi- and interdisciplinary research includes different normative frameworks and ways of thinking that need to be negotiated. This is complicated by the fact that these frameworks are not fully transparent and ready for reflection. What are the normative implications of interdisciplinary collaboration in FAT research? (3) While many problems of discrimination, marginalization and exploitation can be similar across places, they can also have specific local shapes. How can FAT research e.g. consider historically grown specifics such as the effects of different colonial histories? If these specifics make patterns of discrimination have different and more nuanced dimensions than clear-cut 'redlining', what does this imply? To explore these questions, we use the method of 'mind scripting' which is based in theories of discourse, ideology, memory and affect and aims at investigating hidden patterns of meaning making in written memories of the panelists [2]. The workshop strives to challenge some of the implicit norms and tensions in FAT research and to trigger future directions.

The recent interest in identifying and mitigating bias in computational systems has introduced a wide range of different---and occasionally incomparable---proposals for what constitutes bias in such systems. This tutorial introduces the language of measurement modeling from the quantitative social sciences as a framework for examining how social, organizational, and political values enter computational systems and unpacking the varied normative concerns operationalized in different techniques for measuring "bias." We show that this framework helps to clarify the way unobservable theoretical constructs---such as "creditworthiness," "risk to society," or "tweet toxicity"---are turned into measurable quantities and how this process may introduce fairness-related harms. In particular, we demonstrate how to systematically assess the construct validity and reliability of these measurements to detect and characterize specific types of harms, which arise from mismatches between constructs and their operationalizations. We then take a critical look at existing approaches to examining "bias" in NLP models, ranging from work on embedding spaces to machine translation and hate speech detection. We show that measurement modeling can help uncover the implicit constructs that such work aims to capture when measuring "bias." In so doing, we illustrate the limits of current "debiasing" techniques, which have obscured the specific harms whose measurements they implicitly aim to reduce. By introducing the language of measurement modeling, we provide the FAT* community with a framework for making explicit and testing assumptions about unobservable theoretical constructs embedded in computational systems, thereby clarifying and uniting our understandings of fairness-related harms.

Word embeddings, trained through models like skip-gram, have shown to be prone to capturing the biases from the training corpus, e.g. gender bias. Such biases are unwanted as they spill in downstream tasks, thus, leading to discriminatory behavior. In this work, we address the problem of prior sentiment associated with names in word embeddings where for a given name representation (e.g. "Smith"), a sentiment classifier will categorize it as either positive or negative. We propose DebiasEmb, a skip-gram based word embedding approach that, for a given oracle sentiment classification model, will debias the name representations, such that they cannot be associated with either positive or negative sentiment. Evaluation on standard word embedding benchmarks and a downstream analysis show that our approach is able to maintain a high quality of embeddings and at the same time mitigate sentiment bias in name embeddings.

Word-embedding is a major machine learning technique for computational applications of languages. For a given corpus, the process of word-embedding is to embed each word onto multi-dimensional space such that semantic similarities between similar words are retained. While learning the similarity as encapsulated in the training corpus, the embedding process inadvertently captures many other inherent features present in the corpus. One such thing is the bias arising out of stereotyping present in almost all the corpus no matter how extensively used and trusted they are. We study this aspect of word-embedding in the context of Hindi language. We show that many gender-neutral words in Hindi are mapped to vectors which are inclined towards one gender or the other in multi-dimensional space. We propose a new algorithm of debiasing and demonstrate its efficacy in the context of Hindi language. Further, we build a SVM-based classifier that determines whether a gender-neutral word is classified as neutral or otherwise. We corroborate our claim with experimental results on large number of individual words. This work is first ever result on debiasing in Hindi Language and our new debiasing algorithm can be applicable in the context of any language.

Community detection is an important task in social network analysis, allowing us to identify and understand the communities within the social structures provided by the network. However, many community detection approaches either fail to assign low-degree (or lowly connected) users to communities, or assign them to trivially small communities that prevent them from being included in analysis. In this work we investigate how excluding these users can bias analysis results. We then introduce an approach that is more inclusive for lowly connected users by incorporating them into larger groups. Experiments show that our approach outperforms the existing state-of-the-art in terms of F1 and Jaccard similarity scores while reducing the bias towards low-degree users.

One of the foundational models for the study of inventory management is the newsvendor problem. Since the newsvendor problem involving perishable goods, one application that might be very concerning nowadays is food inventory management. Particularly, the food and culinary industries face the problem associated with the setting: supply--demand mismatch which causes business performance reduction due to profit loss. Thus, the developing of mathematical models in the newsvendor problem could be the solution to the problem since it can provide a good insight to determine optimal order quantities. However, inventory managers' order decision might deviate from the assumption in newsvendor setting which claims that individuals would make a rational decision that can maximize their utility and profit as well. Schweitzer & Cachon [1] is one of the earliest works that provides evidence of this deviation and concludes that there is a mismatch between newsvendor theory and experimental observations which causes non-optimal decisions due to the decision biases that occur in the newsvendor context. Thereafter, a growing number of studies in newsvendor problem have started to move toward experimental studies. However, most of the existing studies only involve students as the subject, leaving an important question of how the result of such studies can be implemented in the real world where the manager really works. In this study, we conduct an experiment to investigate the inventory managers' order decision in newsvendor settings in small fast-food restaurants in Yogyakarta, Indonesia. Afterward, we conduct the same experiment with students to provide a structured comparison between manager and student on decision making in the newsvendor problem. After obtaining the order decision pattern, which is not optimal due to anchoring and insufficient adjustment bias that occur, this study will also come up with a debiasing strategy in the form of Decision Support System (DSS). The DSS we propose aims to provide an alternative order for the inventory manager so that the overall inventory performance can be improved. To prove the effectiveness of the DSS we propose, we will also conduct an experimental work to compare the result of the order decisions with and without DSS provided.

Crowdsourced knowledge bases like Wikidata suffer from low-quality edits and vandalism, employing machine learning-based approaches to detect both kinds of damage. We reveal that state-of-the-art detection approaches discriminate anonymous and new users: benign edits from these users receive much higher vandalism scores than benign edits from older ones, causing newcomers to abandon the project prematurely. We address this problem for the first time by analyzing and measuring the sources of bias, and by developing a new vandalism detection model that avoids them. Our model FAIR-S reduces the bias ratio of the state-of-the-art vandalism detector WDVD from 310.7 to only 11.9 while maintaining high predictive performance at 0.963 ROC and 0.316 PR.

Recommender Systems (RSs) are widely used to help online users discover products, books, news, music, movies, courses, restaurants, etc. Because a traditional recommendation strategy always shows the most relevant items (thus with highest predicted rating), traditional RS’s are expected to make popular items become even more popular and non-popular items become even less popular which in turn further divides the haves (popular) from the have-nots (unpopular). Therefore, a major problem with RSs is that they may introduce biases affecting the exposure of items, thus creating a popularity divide of items during the feedback loop that occurs with users, and this may lead the RS to make increasingly biased recommendations over time. In this paper, we view the RS environment as a chain of events that are the result of interactions between users and the RS. Based on that, we propose several debiasing algorithms during this chain of events, and evaluate how these algorithms impact the predictive behavior of the RS, as well as trends in the popularity distribution of items over time. We also propose a novel blind-spot-aware matrix factorization (MF) algorithm to debias the RS. Results show that propensity matrix factorization achieved a certain level of debiasing of the RS while active learning combined with the propensity MF achieved a higher debiasing effect on recommendations.

Machine learning is a tool for building models that accurately represent input training data. When undesired biases concerning demographic groups are in the training data, well-trained models will reflect those biases. We present a framework for mitigating such biases by including a variable for the group of interest and simultaneously learning a predictor and an adversary. The input to the network X, here text or census data, produces a prediction Y, such as an analogy completion or income bracket, while the adversary tries to model a protected variable Z, here gender or zip code. The objective is to maximize the predictor's ability to predict Y while minimizing the adversary's ability to predict Z. Applied to analogy completion, this method results in accurate predictions that exhibit less evidence of stereotyping Z. When applied to a classification task using the UCI Adult (Census) Dataset, it results in a predictive model that does not lose much accuracy while achieving very close to equality of odds (Hardt, et al., 2016). The method is flexible and applicable to multiple definitions of fairness as well as a wide range of gradient-based learning models, including both regression and classification tasks.

In this position paper we try to extend the discussion about the human augmentation of the mind through reasoning. Memory seems favored but both of them play a major role in the mind, we want to equilibrate it. Those two components of our cognitive system are actually quite intertwined with sometimes similar properties and goals. Both are failing, in many ways, that is why we are interested in it, a slight augmentation to those abilities would have huge impact. Therefore, we made a first attempt to propose an approach of a system capable of augmenting the reasoning of his user. First by sensing his context and detect a kind of cognitive bias for many domains through wearable devices and natural language processing. Then we propose a radical method to debiasing that could be used in different scenarios thanks to augmented reality. Finally, those propositions are a beginning of a much needed bigger work that confronts many challenges discussed at the end.

The unbiasedness of online product ratings, an important property to ensure that users' ratings indeed reflect their true evaluations to products, is vital both in shaping consumer purchase decisions and providing reliable recommendations. Recent experimental studies showed that distortions from historical ratings would ruin the unbiasedness of subsequent ratings. How to "discover" the distortions from historical ratings in each single rating (or at the micro-level), and perform the "debiasing operations" in real rating systems are the main objectives of this work.Using 42 million real customer ratings, we first show that users either "assimilate" or "contrast" to historical ratings under different scenarios: users conform to historical ratings if historical ratings are not far from the product quality (assimilation), while users deviate from historical ratings if historical ratings are significantly different from the product quality (contrast). This phenomenon can be explained by the well-known psychological argument: the "Assimilate-Contrast" theory. However, none of the existing works on modeling historical ratings' influence have taken this into account, and this motivates us to propose the Historical Influence Aware Latent Factor Model (HIALF), the first model for real rating systems to capture and mitigate historical distortions in each single rating. HIALF also allows us to study the influence patterns of historical ratings from a modeling perspective, and it perfectly matches the assimilation and contrast effects we previously observed. Also, HIALF achieves significant improvements in predicting subsequent ratings, and accurately predicts the relationships revealed in previous empirical measurements on real ratings. Finally, we show that HIALF can contribute to better recommendations by decoupling users' real preference from distorted ratings, and reveal the intrinsic product quality for wiser consumer purchase decisions.

Software is playing an increasingly important role in supporting human decision-making. Previous HCI research on decision support systems (DSS) has improved the information visualization aspect of DSS information design, but has somewhat overlooked the cognitive aspect of decision-making, namely that human reasoning is heuristic and reflects systematic errors or cognitive biases. We report on an empirical study of two cognitive biases: conservatism and loss aversion. Two remediation techniques recommended by previous research were tested: the expected return method, an actuarial-inspired approach presenting objective metrics; and bootstrapping, a technique successful in improving judgment consistency. The results show that the two biases can occur simultaneously and can have a huge impact on decision-making. The results also show that the two debiasing techniques are only partly effective. These findings suggest a need for more research on debiasing, and indicate some directions for exploring debiasing techniques and building decision support systems.

Information about quantitative characteristics in local businesses and services, such as the number of people waiting in line in a cafe and the number of available fitness machines in a gym, is important for informed decision, crowd management and event detection. In this paper, we investigate the potential of leveraging crowds as sensors to report such quantitative characteristics and investigate how to recover the true quantity values from noisy crowdsourced information. Through experiments, we find that crowd sensors have both bias and variance in quantity sensing, and task difficulties impact the sensing accuracy. Based on these findings, we propose an unsupervised probabilistic model to jointly assess task difficulties, ability of crowd sensors and true quantity values. Our model differs from existing categorical truth finding models as ours is specifically designed to tackle quantitative truth. In addition to devising an efficient model inference algorithm in a batch mode, we also design an even faster online version for handling streaming data. Experimental results in various scenarios demonstrate the effectiveness of our model.

With the explosive growth of social networks, many applications are increasingly harnessing the pulse of online crowds for a variety of tasks such as marketing, advertising, and opinion mining. An important example is the wisdom of crowd effect that has been well studied for such tasks when the crowd is non-interacting. However, these studies don't explicitly address the network effects in social networks. A key difference in this setting is the presence of social influences that arise from these interactions and can undermine the wisdom of the crowd [17]. Using a natural model of opinion formation, we analyze the effect of these interactions on an individual's opinion and estimate her propensity to conform. We then propose efficient sampling algorithms incorporating these conformity values to arrive at a debiased estimate of the wisdom of a crowd. We analyze the trade-off between the sample size and estimation error and validate our algorithms using both real data obtained from online user experiments and synthetic data.

We consider the problem of identifying the consensus ranking for the results of a query, given preferences among those results from a set of individual users. Once consensus rankings are identified for a set of queries, these rankings can serve for both evaluation and training of retrieval and learning systems. We present a novel approach to collecting the individual user preferences over image-search results: we use a collaborative game in which players are rewarded for agreeing on which image result is best for a query. Our approach is distinct from other labeling games because we are able to elicit directly the preferences of interest with respect to image queries extracted from query logs. As a source of relevance judgments, this data provides a useful complement to click data. Furthermore, the data is free of positional biases and is collected by the game without the risk of frustrating users with non-relevant results; this risk is prevalent in standard mechanisms for debiasing clicks. We describe data collected over 34 days from a deployed version of this game that amounts to about 18 million expressed preferences between pairs. Finally, we present several approaches to modeling this data in order to extract the consensus rankings from the preferences and better sort the search results for targeted queries.