Search for: All records

Producing high-quality labeled data is a challenge in any supervised learning problem, where in many cases, human involvement is necessary to ensure the label quality. However, human annotations are not flawless, especially in the case of a challenging problem. In nontrivial problems, the high disagreement among annotators results in noisy labels, which affect the performance of any machine learning model. In this work, we consider three noise reduction strategies to improve the label quality in the Article-Comment Alignment Problem, where the main task is to classify article-comment pairs according to their relevancy level. The first considered labeling disagreement reduction strategy utilizes annotators' background knowledge during the label aggregation step. The second strategy utilizes user disagreement during the training process. In the third and final strategy, we ask annotators to perform corrections and relabel the examples with noisy labels. We deploy these strategies and compare them to a resampling strategy for addressing the class imbalance, another common supervised learning challenge. These alternatives were evaluated on ACAP, a multiclass text pairs classification problem with highly imbalanced data, where one of the classes represents at most 15% of the dataset's entire population. Our results provide evidence that considered strategies can reduce disagreement between annotators. However, data quality improvement is insufficient to enhance classification accuracy in the article-comment alignment problem, which exhibits a high-class imbalance. The model performance is enhanced for the same problem by addressing the imbalance issue with a weight loss-based class distribution resampling. We show that allowing the model to pay more attention to the minority class during the training process with the presence of noisy examples improves the test accuracy by 3%. 

Many online news outlets, forums, and blogs provide a rich stream of publications and user comments. This rich body of data is a valuable source of information for researchers, journalists, and policymakers. However, the ever-increasing production and user engagement rate make it difficult to analyze this data without automated tools. This work presents MultiLayerET, a method to unify the representation of entities and topics in articles and comments. In MultiLayerET, articles' content and associated comments are parsed into a multilayer graph consisting of heterogeneous nodes representing named entities and news topics. The nodes within this graph have attributed edges denoting weight, i.e., the strength of the connection between the two nodes, time, i.e., the co-occurrence contemporaneity of two nodes, and sentiment, i.e., the opinion (in aggregate) of an entity toward a topic. Such information helps in analyzing articles and their comments. We infer the edges connecting two nodes using information mined from the textual data. The multilayer representation gives an advantage over a single-layer representation since it integrates articles and comments via shared topics and entities, providing richer signal points about emerging events. MultiLayerET can be applied to different downstream tasks, such as detecting media bias and misinformation. To explore the efficacy of the proposed method, we apply MultiLayerET to a body of data gathered from six representative online news outlets. We show that with MultiLayerET, the classification F1 score of a media bias prediction model improves by 36%, and that of a state-of-the-art fake news detection model improves by 4%.