Despite the widespread adoption, there is a lack of research into how various critical aspects of pretrained language models (PLMs) affect their performance in hate speech detection. Through five research questions, our findings and recommendations lay the groundwork for empirically investigating different aspects of PLMs' use in hate speech detection. We deep dive into comparing different pretrained models, evaluating their seed robustness, finetuning settings, and the impact of pretraining data collection time. Our analysis reveals early peaks for downstream tasks during pretraining, the limited benefit of employing a more recent pretraining corpus, and the significance of specific layers during finetuning. We further call into question the use of domain-specific models and highlight the need for dynamic datasets for benchmarking hate speech detection.

Improving multilingual language models capabilities in low-resource languages is generally difficult due to the scarcity of large-scale data in those languages. In this paper, we relax the reliance on texts in low-resource languages by using multilingual lexicons in pretraining to enhance multilingual capabilities. Specifically, we focus on zero-shot sentiment analysis tasks across 34 languages, including 6 high/medium-resource languages, 25 low-resource languages, and 3 code-switching datasets. We demonstrate that pretraining using multilingual lexicons, without using any sentence-level sentiment data, achieves superior zero-shot performance compared to models fine-tuned on English sentiment datasets, and large language models like GPT--3.5, BLOOMZ, and XGLM. These findings are observable for unseen low-resource languages to code-mixed scenarios involving high-resource languages.

Vision-Language Models (VLMs) have emerged as the state-of-the-art representation learning solution, with myriads of downstream applications such as image classification, retrieval and generation. A natural question is whether these models memorize their training data, which also has implications for generalization. We propose a new method for measuring memorization in VLMs, which we call d\'ej\`a vu memorization. For VLMs trained on image-caption pairs, we show that the model indeed retains information about individual objects in the training images beyond what can be inferred from correlations or the image caption. We evaluate d\'ej\`a vu memorization at both sample and population level, and show that it is significant for OpenCLIP trained on as many as 50M image-caption pairs. Finally, we show that text randomization considerably mitigates memorization while only moderately impacting the model's downstream task performance.

LLM-based Automatic Prompt Optimization, which typically utilizes LLMs as Prompt Optimizers to self-reflect and refine prompts, has shown promising performance in recent studies. Despite the success, the underlying mechanism of this approach remains unexplored, and the true effectiveness of LLMs as Prompt Optimizers requires further validation. In this work, we conducted a comprehensive study to uncover the actual mechanism of LLM-based Prompt Optimization. Our findings reveal that the LLM optimizers struggle to identify the true causes of errors during reflection, tending to be biased by their own prior knowledge rather than genuinely reflecting on the errors. Furthermore, even when the reflection is semantically valid, the LLM optimizers often fail to generate appropriate prompts for the target models with a single prompt refinement step, partly due to the unpredictable behaviors of the target models. Based on the observations, we introduce a new "Automatic Behavior Optimization" paradigm, which directly optimizes the target model's behavior in a more controllable manner. We hope our study can inspire new directions for automatic prompt optimization development.

Graph Collaborative Filtering (GCF), one of the most widely adopted recommendation system methods, effectively captures intricate relationships between user and item interactions. Graph Contrastive Learning (GCL) based GCF has gained significant attention as it leverages self-supervised techniques to extract valuable signals from real-world scenarios. However, many methods usually learn the instances of discrimination tasks that involve the construction of contrastive pairs through random sampling. GCL approaches suffer from sampling bias issues, where the negatives might have a semantic structure similar to that of the positives, thus leading to a loss of effective feature representation. To address these problems, we present the \underline{Proto}typical contrastive learning through \underline{A}lignment and \underline{U}niformity for recommendation, which is called \textbf{ProtoAU}. Specifically, we first propose prototypes (cluster centroids) as a latent space to ensure consistency across different augmentations from the origin graph, aiming to eliminate the need for random sampling of contrastive pairs. Furthermore, the absence of explicit negatives means that directly optimizing the consistency loss between instance and prototype could easily result in dimensional collapse issues. Therefore, we propose aligning and maintaining uniformity in the prototypes of users and items as optimization objectives to prevent falling into trivial solutions. Finally, we conduct extensive experiments on four datasets and evaluate their performance on the task of link prediction. Experimental results demonstrate that the proposed ProtoAU outperforms other representative methods. The source codes of our proposed ProtoAU are available at \url{https://github.com/oceanlvr/ProtoAU}.

Mainstream cross-lingual task-oriented dialogue (ToD) systems leverage the transfer learning paradigm by training a joint model for intent recognition and slot-filling in English and applying it, zero-shot, to other languages. We address a gap in prior research, which often overlooked the transfer to lower-resource colloquial varieties due to limited test data. Inspired by prior work on English varieties, we craft and manually evaluate perturbation rules that transform German sentences into colloquial forms and use them to synthesize test sets in four ToD datasets. Our perturbation rules cover 18 distinct language phenomena, enabling us to explore the impact of each perturbation on slot and intent performance. Using these new datasets, we conduct an experimental evaluation across six different transformers. Here, we demonstrate that when applied to colloquial varieties, ToD systems maintain their intent recognition performance, losing 6% (4.62 percentage points) in accuracy on average. However, they exhibit a significant drop in slot detection, with a decrease of 31% (21 percentage points) in slot F1 score. Our findings are further supported by a transfer experiment from Standard American English to synthetic Urban African American Vernacular English.

With the development of web technology, social media texts are becoming a rich source for automatic mental health analysis. As traditional discriminative methods bear the problem of low interpretability, the recent large language models have been explored for interpretable mental health analysis on social media, which aims to provide detailed explanations along with predictions. The results show that ChatGPT can generate approaching-human explanations for its correct classifications. However, LLMs still achieve unsatisfactory classification performance in a zero-shot/few-shot manner. Domain-specific finetuning is an effective solution, but faces 2 challenges: 1) lack of high-quality training data. 2) no open-source LLMs for interpretable mental health analysis were released to lower the finetuning cost. To alleviate these problems, we build the first multi-task and multi-source interpretable mental health instruction (IMHI) dataset on social media, with 105K data samples. The raw social media data are collected from 10 existing sources covering 8 mental health analysis tasks. We use expert-written few-shot prompts and collected labels to prompt ChatGPT and obtain explanations from its responses. To ensure the reliability of the explanations, we perform strict automatic and human evaluations on the correctness, consistency, and quality of generated data. Based on the IMHI dataset and LLaMA2 foundation models, we train MentalLLaMA, the first open-source LLM series for interpretable mental health analysis with instruction-following capability. We also evaluate the performance of MentalLLaMA on the IMHI evaluation benchmark with 10 test sets, where their correctness for making predictions and the quality of explanations are examined. The results show that MentalLLaMA approaches state-of-the-art discriminative methods in correctness and generates high-quality explanations.

Effectively leveraging multimodal information from social media posts is essential to various downstream tasks such as sentiment analysis, sarcasm detection or hate speech classification. Jointly modeling text and images is challenging because cross-modal semantics might be hidden or the relation between image and text is weak. However, prior work on multimodal classification of social media posts has not yet addressed these challenges. In this work, we present an extensive study on the effectiveness of using two auxiliary losses jointly with the main task during fine-tuning multimodal models. First, Image-Text Contrastive (ITC) is designed to minimize the distance between image-text representations within a post, thereby effectively bridging the gap between posts where the image plays an important role in conveying the post's meaning. Second, Image-Text Matching (ITM) enhances the model's ability to understand the semantic relationship between images and text, thus improving its capacity to handle ambiguous or loosely related modalities. We combine these objectives with five multimodal models across five diverse social media datasets, demonstrating consistent improvements of up to 2.6 points F1. Our comprehensive analysis shows the specific scenarios where each auxiliary task is most effective.

We study the problem of agent selection in causal strategic learning under multiple decision makers and address two key challenges that come with it. Firstly, while much of prior work focuses on studying a fixed pool of agents that remains static regardless of their evaluations, we consider the impact of selection procedure by which agents are not only evaluated, but also selected. When each decision maker unilaterally selects agents by maximising their own utility, we show that the optimal selection rule is a trade-off between selecting the best agents and providing incentives to maximise the agents' improvement. Furthermore, this optimal selection rule relies on incorrect predictions of agents' outcomes. Hence, we study the conditions under which a decision maker's optimal selection rule will not lead to deterioration of agents' outcome nor cause unjust reduction in agents' selection chance. To that end, we provide an analytical form of the optimal selection rule and a mechanism to retrieve the causal parameters from observational data, under certain assumptions on agents' behaviour. Secondly, when there are multiple decision makers, the interference between selection rules introduces another source of biases in estimating the underlying causal parameters. To address this problem, we provide a cooperative protocol which all decision makers must collectively adopt to recover the true causal parameters. Lastly, we complement our theoretical results with simulation studies. Our results highlight not only the importance of causal modeling as a strategy to mitigate the effect of gaming, as suggested by previous work, but also the need of a benevolent regulator to enable it.

Fox News chief national security correspondent Jennifer Griffin has the latest on the retaliation against the drone attack on U.S. servicemembers in Jordan on'Your World.' The White House stressed Friday evening that the United States is "not looking for a war with Iran," saying the retaliatory strikes carried out in Syria and Iraq were designed to "de-escalate" tensions and "put an end" to attacks on U.S. troops in the region. The United States began retaliatory strikes on more than 85 targets in Iraq and Syria against Iran's Islamic Revolutionary Guards Corps (IRGC) Quds Force and affiliated militia groups and proxies. The strikes come in response to the deaths of three U.S. service members last Sunday on a U.S. base in Jordan. White House National Security Council Coordinator for Strategic Communications John Kirby stressed that the United States is not seeking conflict with Iran or in the Middle East, but explained that the strikes that began Friday evening "will not end" tonight.