While detecting offensive language in online spaces remains an important societal issue, there is still a significant gap in existing research and practial datasets specific to chatbots. Furthermore, many of the current efforts by service providers to automatically filter offensive language are vulnerable to users’ deliberate text manipulation tactics, such as misspelling words. In this study, we analyze offensive language patterns in real logs of 6,254,261 chat utterance pairs from the commercial chat service Simsimi, which cover a variety of conversation topics. Based on the observed patterns, we introduce a novel offensive language detection method—a contrastive learning model that embeds chat content with a random masking strategy. We show that this model outperforms existing models in detecting offensive language in open-domain chat conversations while also demonstrating robustness against users’ deliberate text manipulation tactics when using offensive language. We release our curated chatbot dataset to foster research on offensive language detection in open-domain conversations and share lessons learned from mitigating offensive language on a live platform.

Two types of topic modeling predominate: generative methods that employ probabilistic latent models and clustering methods that identify semantically coherent groups. This paper newly presents UTopic (Unified neural Topic model via contrastive learning and term weighting) that combines the advantages of these two types. UTopic uses contrastive learning and term weighting to learn knowledge from a pretrained language model and discover influential terms from semantically coherent clusters. Experiments show that the generated topics have a high-quality topic-word distribution in terms of topic coherence, outperforming existing baselines across multiple topic coherence measures. We demonstrate how our model can be used as an add-on to existing topic models and improve their performance.

The potential social harms that large language models pose, such as generating offensive content and reinforcing biases, are steeply rising. Existing works focus on coping with this concern while interacting with ill-intentioned users, such as those who explicitly make hate speech or elicit harmful responses. However, discussions on sensitive issues can become toxic even if the users are well-intentioned. For safer models in such scenarios, we present the Sensitive Questions and Acceptable Response (SQuARe) dataset, a large-scale Korean dataset of 49k sensitive questions with 42k acceptable and 46k non-acceptable responses. The dataset was constructed leveraging HyperCLOVA in a human-in-the-loop manner based on real news headlines. Experiments show that acceptable response generation significantly improves for HyperCLOVA and GPT-3, demonstrating the efficacy of this dataset.

Quotes are critical for establishing credibility in news articles. A direct quote enclosed in quotation marks has a strong visual appeal and is a sign of a reliable citation. Unfortunately, this journalistic practice is not strictly followed, and a quote in the headline is often “contextomized.” Such a quote uses words out of context in a way that alters the speaker’s intention so that there is no semantically matching quote in the body text. We present QuoteCSE, a contrastive learning framework that represents the embedding of news quotes based on domain-driven positive and negative samples to identify such an editorial strategy. The dataset and code are available at https://github.com/ssu-humane/contextomized-quote-contrastive.

This paper presents a time-topic cohesive model describing the communication patterns on the coronavirus pandemic from three Asian countries. The strength of our model is two-fold. First, it detects contextualized events based on topical and temporal information via contrastive learning. Second, it can be applied to multiple languages, enabling a comparison of risk communication across cultures. We present a case study and discuss future implications of the proposed model.

This study analyzes the political slants of user comments on Korean partisan media. We built a BERT-based classifier to detect political leaning of short comments via the use of semi-unsupervised deep learning methods that produced an F1 score of 0.83. As a result of classifying 21.6K comments, we found the high presence of conservative bias on both conservative and liberal news outlets. Moreover, this study discloses an asymmetry across the partisan spectrum in that more liberals (48.0%) than conservatives (23.6%) comment not only on news stories resonating with their political perspectives but also on those challenging their viewpoints. These findings advance the current understanding of online echo chambers.