Numerous explanation methods have been recently developed to interpret the decisions made by deep neural network (DNN) models. For image classifiers, these methods typically provide an attribution score to each pixel in the image to quantify its contribution to the prediction. However, most of these explanation methods appropriate attribution scores to pixels independently, even though both humans and DNNs make decisions by analyzing a set of closely related pixels simultaneously. Hence, the attribution score of a pixel should be evaluated jointly by considering itself and its structurally-similar pixels. We propose a method called IProp, which models each pixel's individual attribution score as a source of explanatory information and explains the image prediction through the dynamic propagation of information across all pixels. To formulate the information propagation, IProp adopts the Markov Reward Process, which guarantees convergence, and the final status indicates the desired pixels' attribution scores. Furthermore, IProp is compatible with any existing attribution-based explanation method. Extensive experiments on various explanation methods and DNN models verify that IProp significantly improves them on a variety of interpretability metrics.

While recommendation systems enable users to find articles of interest, they can also create "filter bubbles" by presenting content that reinforces users' pre-existing beliefs. Users are often unaware that the system placed them in a filter bubble and, even when aware, they often lack direct control over it. To address these issues, we first design a political news recommendation system augmented with an enhanced interface that exposes the political and topical interests the system inferred from user behavior. This allows the user to adjust the recommendation system to receive more articles on a particular topic or presenting a particular political stance. We then conduct a user study to compare our system to a traditional interface and found that the transparent approach helped users realize that they were in a filter bubble. Additionally, the enhanced system led to less extreme news for most users but also allowed others to move the system to more extremes. Similarly, while many users moved the system from extreme liberal/conservative to the center, this came at the expense of reducing political diversity of the articles shown. These findings suggest that, while the proposed system increased awareness of the filter bubbles, it had heterogeneous effects on news consumption depending on user preferences.

We propose novel methods to identify tweets that criticize partisan news sources. Prior work suggests that criticism, ridicule, and distrust of news media all play important roles in hyperpartisanship, misinformation, and filter bubble formation. Thus, understanding the prevalence and temporal dynamics of media-targeted criticism can provide us with updated tools to assess the health of the information ecosystem. There is a scarcity of labeled data for this task, and we develop a weakly supervised learning approach that leverages multiple noisy labeling functions based on both the content of the tweet as well as the historical news sharing behavior of the user. Using this classifier, we explore how tweets expressing criticism vary by user, news source, and time, finding substantial spikes in media criticism during politically polarizing events, such as the investigation into Russian interference in the 2016 U.S. elections and the 2017 "unite the right" rally in Charlottesville. This type of media-targeting criticism is also more likely to occur after users have been exposed to unreliable and hyperpartisan media.

Understanding how political news consumption changes over time can provide insights into issues such as hyperpartisanship, filter bubbles, and misinformation. To investigate long-term trends of news consumption, we curate a collection of over 60M tweets from politically engaged users over seven years, annotating ~10% with mentions of news outlets and their political leaning. We then train a neural network to forecast the political lean of news articles Twitter users will engage with, considering both past news engagements as well as tweet content. Using the learned representation of this model, we cluster users to discover salient patterns of long-term news engagement. Our findings include the following: (1) hyperpartisan users are more engaged with news; (2) right-leaning users engage with contra-partisan sources more than left-leaning users; (3) topics such as immigration, COVID-19, Islamaphobia, and gun control are salient indicators of engagement with low quality news sources.

Content-based news recommenders learn words that correlate with user engagement and recommend articles accordingly. This can be problematic for users with diverse political preferences by topic — e.g., users that prefer conservative articles on one topic but liberal articles on another. In such instances, recommenders can have a homogenizing effect by recommending articles with the same political lean on both topics, particularly if both topics share salient, politically polarized terms like “far right” or “radical left.” In this paper, we propose attention-based neural network models to reduce this homogenization effect by increasing attention on words that are topic specific while decreasing attention on polarized, topic-general terms. We find that the proposed approach results in more accurate recommendations for simulated users with such diverse preferences.

Algorithmic personalization of news and social media content aims to improve user experience; however, there is evidence that this filtering can have the unintended side effect of creating homogeneous “filter bubbles,” in which users are over-exposed to ideas that conform with their preexisting perceptions and beliefs. In this paper, we investigate this phenomenon in the context of political news recommendation algorithms, which have important implications for civil discourse.

Active learning methods aim to choose the most informative instances to effectively learn a good classifier. Uncertainty sampling, arguably the most frequently utilized active learning strategy, selects instances which are uncertain according to the model. In this paper, we propose a framework that distinguishes between two types of uncertainties: a model is uncertain about an instance due to strong and conflicting evidence (most-surely uncertain) vs. a model is uncertain about an instance because it does not have conclusive evidence (least-surely uncertain). We show that making a distinction between these uncertainties makes a huge difference to the performance of active learning. We provide a mathematical formulation to distinguish between these uncertainties for naive Bayes, logistic regression and support vector machines and empirically evaluate our methods on several real-world datasets.

To date, many active learning techniques have been developed for acquiring labels when training data is limited. However, an important aspect of the problem has often been neglected or just mentioned in passing: the curse of dimensionality. Yet, the curse of dimensionality poses even greater challenges in the case of limited data, which is precisely the setup for active learning. Reducing the dimensions is not a trivial task, however, as the correct number of dimensions depends on a number of factors including the training data size, the number of classes, the discriminative power of the features, and the underlying classification model. Moreover, active learning is typically applied in an iterative manner where the number of labels is smaller in the earlier iterations compared to the later ones. We propose an adaptive dimensionality reduction technique that determines the appropriate number of dimensions for each active learning iteration, utilizing the labeled and unlabeled data effectively to learn more accurate models. Extensive experiments comparing various approaches and parameter settings show that the proposed method improves performance drastically on three real-world text classification tasks.

Methods that reduce the amount of labeled data needed for training have focused more on selecting which documents to label than on which queries should be labeled. One exception to this (Long et al. 2010) uses expected loss optimization (ELO) to estimate which queries should be selected but is limited to rankers that predict absolute graded relevance. In this work, we demonstrate how to easily adapt ELO to work with any ranker and show that estimating expected loss in DCG is more robust than NDCG even when the final performance measure is NDCG.

Information diffusion, viral marketing, graph-based semi-supervised learning, and collective classification all attempt to model and exploit the relationships among nodes in a network to improve the performance of node labeling algorithms. However, sometimes the advantage of exploiting the relationships can become a disadvantage. Simple models like label propagation and iterative classification can aggravate a misclassification by propagating mistakes in the network, while more complex models that define and optimize a global objective function, such as Markov random fields and graph mincuts, can misclassify a set of nodes jointly. This problem can be mitigated if the classification system is allowed to ask for the correct labels for a few of the nodes during inference. However, determining the optimal set of labels to acquire is intractable under relatively general assumptions, which forces us to resort to approximate and heuristic techniques. We describe three such techniques in this article. The first one is based on directly approximating the value of the objective function of label acquisition and greedily acquiring the label that provides the most improvement. The second technique is a simple technique based on the analogy we draw between viral marketing and label acquisition. Finally, we propose a method, which we refer to as reflect and correct , that can learn and predict when the classification system is likely to make mistakes and suggests acquisitions to correct those mistakes. We empirically show on a variety of synthetic and real-world datasets that the reflect and correct method significantly outperforms the other two techniques, as well as other approaches based on network structural measures such as node degree and network clustering.

Information diffusion, viral marketing, and collective classification all attempt to model and exploit the relationships in a network to make inferences about the labels of nodes. A variety of techniques have been introduced and methods that combine attribute information and neighboring label information have been shown to be effective for collective labeling of the nodes in a network. However, in part because of the correlation between node labels that the techniques exploit, it is easy to find cases in which, once a misclassification is made, incorrect information propagates throughout the network. This problem can be mitigated if the system is allowed to judiciously acquire the labels for a small number of nodes. Unfortunately, under relatively general assumptions, determining the optimal set of labels to acquire is intractable. Here we propose an acquisition method that learns the cases when a given collective classification algorithm makes mistakes, and suggests acquisitions to correct those mistakes. We empirically show on both real and synthetic datasets that this method significantly outperforms a greedy approximate inference approach, a viral marketing approach, and approaches based on network structural measures such as node degree and network clustering. In addition to significantly improving accuracy with just a small amount of labeled data, our method is tractable on large networks.