As autonomous driving systems increasingly become part of daily transportation, the ability to accurately anticipate and mitigate potential traffic accidents is paramount. Traditional accident anticipation models primarily utilizing dashcam videos are adept at predicting when an accident may occur but fall short in localizing the incident and identifying involved entities. Addressing this gap, this study introduces a novel framework that integrates Large Language Models (LLMs) to enhance predictive capabilities across multiple dimensions-what, when, and where accidents might occur. We develop an innovative chain-based attention mechanism that dynamically adjusts to prioritize high-risk elements within complex driving scenes. This mechanism is complemented by a three-stage model that processes outputs from smaller models into detailed multimodal inputs for LLMs, thus enabling a more nuanced understanding of traffic dynamics. Empirical validation on the DAD, CCD, and A3D datasets demonstrates superior performance in Average Precision (AP) and Mean Time-To-Accident (mTTA), establishing new benchmarks for accident prediction technology. Our approach not only advances the technological framework for autonomous driving safety but also enhances human-AI interaction, making predictive insights generated by autonomous systems more intuitive and actionable.
论文检索
输入标题、作者或关键词，从 8,216 篇学术成果中精准定位
In the face of rising surface temperatures from climate change, impacting biodiversity, extreme weather events, and agricultural productivity, understanding the drivers behind temperature changes is imperative. Traditional global climate models (GCMs) are computationally expensive, limiting their applicability, while machine learning approaches, though promising, face interpretability challenges due to their "black box" nature, especially in a dynamic setting where the data is constantly evolving. We propose DUO, a framework to identify shifts in important features and feature combinations as the data distribution changes over time. Our model independently assesses the importance of features and their interactions while also evaluating their relevance when combined with additional features, contributing to the target class. As a case study, we apply DUO to assess the shifts in climate drivers for station-level temperatures in six locations across New Zealand from 1980 to 2020, we identify specific humidity, geopotential height, and air temperature at high atmospheric pressure levels as the most important features for describing temperature variability. By revealing how climate drivers change over time, DUO contributes to a deeper understanding of temperature change patterns, enabling practitioners to develop targeted and adaptive mitigation strategies.
Human vs ChatGPT: Effect of Data Annotation in Interpretable Crisis-Related Microblog Classification
PDF ↗Recent studies have exploited the vital role of microblogging platforms, such as Twitter, in crisis situations. Various machine-learning approaches have been proposed to identify and prioritize crucial information from different humanitarian categories for preparation and rescue purposes. In crisis domain, the explanation of models' output decisions is gaining significant research momentum. Some previous works focused on human annotations of rationales to train and extract supporting evidence for model interpretability. However, such annotations are usually expensive, require much effort, and are not always available in real-time situations of a new crisis event. In this paper, we investigate the recent advances in large language models (LLMs) as data annotators on informal tweet text. We perform a detailed qualitative and quantitative evaluation of ChatGPT rationale annotations over a few-shot setup. ChatGPT annotations are quite close to humans but less precise in nature. Further, we propose an active learning-based interpretable classification model from a small set of annotated data. Our experiments show that (a). ChatGPT has the potential to extract rationales for the crisis tweet classification tasks, but the performance is slightly less than the model trained on human-annotated rationale data (\sim3-6%), (b). active learning setup can help reduce the burden of manual annotations and maintain a trade-off between performance and data size.
In recent years, money laundering crimes on blockchain, especially on Ethereum, have become increasingly rampant, resulting in substantial losses. The unique features of money laundering on Ethereum, such as decentralization and pseudonymity, pose new challenges for Ethereum anti-money laundering. Specifically, the existence of dense and extensive laundering gangs and intricate multilayered laundering pathways makes it exceptionally challenging for regulators to identify suspicious accounts and trace money flows. To address this issue, we propose an innovative DenseFlow framework that effectively identifies and traces money laundering activities by finding dense subgraphs and applying the maximum flow idea. We conduct multiple experiments on four datasets from Ethereum to validate the effectiveness of our approach. The precision of our DenseFlow is 16.34% higher than the start-of-the-art comparison methods on average, highlighting its distinctive contribution to tackling money laundering issues on blockchain.
Consistency Guided Knowledge Retrieval and Denoising in LLMs for Zero-shot Document-level Relation Triplet Extraction
PDF ↗Document-level Relation Triplet Extraction (DocRTE) is a fundamental task in information systems that aims to simultaneously extract entities with semantic relations from a document. Existing methods heavily rely on a substantial amount of fully labeled data. However, collecting and annotating data for newly emerging relations is time-consuming and labor-intensive. Recent advanced Large Language Models (LLMs), such as ChatGPT and LLaMA, exhibit impressive long-text generation capabilities, inspiring us to explore an alternative approach for obtaining auto-labeled documents with new relations. In this paper, we propose a Zero-shot Document-level Relation Triplet Extraction (ZeroDocRTE) framework, which Generates labeled data by Retrieval and Denoising Knowledge from LLMs, called GenRDK. Specifically, we propose a chain-of-retrieval prompt to guide ChatGPT to generate labeled long-text data step by step. To improve the quality of synthetic data, we propose a denoising strategy based on the consistency of cross-document knowledge. Leveraging our denoised synthetic data, we proceed to fine-tune the LLaMA2-13B-Chat for extracting document-level relation triplets. We perform experiments for both zero-shot document-level relation and triplet extraction on two public datasets. The experimental results illustrate that our GenRDK framework outperforms strong baselines.
Multimodal relation extraction is a fundamental task of multimodal information extraction. Recent studies have shown promising results by integrating hierarchical visual features from local regions, like image patches, to the broader global regions that form the entire image. However, research to date has largely ignored the understanding of how hierarchical visual semantics are represented and the characteristics that can benefit relation extraction. To bridge this gap, we propose a novel two-stage hierarchical visual context fusion transformer incorporating the mixture of multimodal experts framework to effectively represent and integrate hierarchical visual features into textual semantic representations. In addition, we introduce the concept of hierarchical tracking maps to facilitate the understanding of the intrinsic mechanisms of image information processing involved in multimodal models. We thoroughly investigate the implications of hierarchical visual contexts through four dimensions: performance evaluation, the nature of auxiliary visual information, the patterns observed in the image encoding hierarchy, and the significance of various visual encoding levels. Empirical studies show that our approach achieves new state-of-the-art performance on the MNRE dataset.
Knowledge tracing aims to estimate knowledge states of students over a set of skills based on students' past learning activities. Deep learning based knowledge tracing models show superior performance to traditional knowledge tracing approaches. Early works like DKT use skill IDs and student responses only. Recent works also incorporate questions IDs into their models and achieve much improved performance in the next question correctness prediction task. However, predictions made by these models are thus on specific questions, and it is not straightforward to translate them to estimation of students' knowledge states over skills. In this paper, we propose to replace question IDs with question difficulty levels in deep knowledge tracing models. The predictions made by our model can be more readily translated to students' knowledge states over skills. Furthermore, by using question difficulty levels to replace question IDs, we can also alleviate the cold-start problem in knowledge tracing as online learning platforms are updated frequently with new questions. We further use two techniques to smooth the predicted scores. One is to combine embeddings of nearby difficulty levels using the Hann function. The other is to constrain the predicted probabilities to be consistent with question difficulties by imposing a penalty if they are not consistent. We conduct extensive experiments to study the performance of the proposed model. Our experimental results show that our model outperforms the state-of-the-art knowledge tracing models in terms of both accuracy and consistency with question difficulty levels.
LARA: A Light and Anti-overfitting Retraining Approach for Unsupervised Time Series Anomaly Detection
PDF ↗Most of current anomaly detection models assume that the normal pattern remains the same all the time. However, the normal patterns of web services can change dramatically and frequently over time. The model trained on old-distribution data becomes outdated and ineffective after such changes. Retraining the whole model whenever the pattern is changed is computationally expensive. Further, at the beginning of normal pattern changes, there is not enough observation data from the new distribution. Retraining a large neural network model with limited data is vulnerable to overfitting. Thus, we propose a Light Anti-overfitting Retraining Approach (LARA) based on deep variational auto-encoders for time series anomaly detection. In LARA we make the following three major contributions: 1) the retraining process is designed as a convex problem such that overfitting is prevented and the retraining process can converge fast; 2) a novel ruminate block is introduced, which can leverage the historical data without the need to store them; 3) we mathematically and experimentally prove that when fine-tuning the latent vector and reconstructed data, the linear formations can achieve the least adjusting errors between the ground truths and the fine-tuned ones. Moreover, we have performed many experiments to verify that retraining LARA with even a limited amount of data from new distribution can achieve competitive performance in comparison with the state-of-the-art anomaly detection models trained with sufficient data. Besides, we verify its light computational overhead.
Dynamic pricing algorithms have been widely studied to manage hotel and platform revenue over online travel platforms (OTPs). For better dynamic pricing, the accurate estimation of the market demand and the market competitiveness are crucial. However, the existing approaches obtain a pricing strategy tailored to each specific scenario using data only from that scenario. They are not considering the shared information between different scenarios, i.e., the data from different scenarios are not fully utilized. So we propose a Multi Scenario Pricing model (MSP) with a novel sharing structure design that leverages cross-scenario and specific information to capture more accurate market demand and competitiveness. Specifically, the model structure explicitly separates information into shared components as market demand and specific information as scenario-wise price competitiveness to prevent domain seesaw. To capture the inherent correlation between listings in different scenarios, an attention network named Price Competitiveness Representation Extraction (PCRE) is well-designed. Meanwhile, traditional metrics are skewed towards model that tends to reduce the price regardless of sample distribution. Thus we propose new offline evaluation metrics that shift attention with sample distribution to avoid biased pricing strategies, which is proved to be more closely related to actual business revenue. Our proposed MSP shows superiority under both offline and online experiments on real-world datasets. The multi-scenario industry dataset and our code are available. To the best of our knowledge, it will be the first real-industry multi-scenario pricing data.
To recommend the points of interest (POIs) that a user would check-in next, most deep-learning (DL)-based existing studies have employed random negative (RN) sampling during model training. In this paper, we claim and validate that, as the training proceeds, such an RN sampling in reality performs as sampling easy negative (EN) POIs (i.e., EN sampling) that a user was highly unlikely to check-in at her check-in time point. Furthermore, we verify that EN sampling is more disadvantageous in improving the accuracy than sampling hard negative (HN) POIs (i.e., HN sampling) that a user was highly likely to check-in. To address this limitation, we present the novel concept of the Degree of Positiveness (DoP), which can be formulated by two factors: (i) the degree to which a POI has the characteristics preferred by a user; (ii) the geographical distance between a user and a POI. Then, we propose a new model-training scheme based on HN sampling by using DoP. Using real-world datasets (i.e., NYC, TKY, and Brightkite), we demonstrate that all the state-of-the-art models trained by our scheme showed dramatic improvements in accuracy by up to about 82.8%.
In recent years, the video game industry has experienced substantial growth, presenting players with a vast array of game choices. This surge in options has spurred the need for a specialized recommender system tailored for video games. However, current video game recommendation approaches tend to prioritize accuracy over diversity, potentially leading to unvaried game suggestions. In addition, the existing game recommendation methods commonly lack the ability to establish strict connections between games to enhance accuracy. Furthermore, many existing diversity-focused methods fail to leverage crucial item information, such as item category and popularity during neighbor modeling and message propagation. To address these challenges, we introduce a novel framework, called CPGRec, comprising three modules, namely accuracy-driven, diversity-driven, and comprehensive modules. The first module extends the state-of-the-art accuracy-focused game recommendation method by connecting games in a more stringent manner to enhance recommendation accuracy. The second module connects neighbors with diverse categories within the proposed game graph and harnesses the advantages of popular game nodes to amplify the influence of long-tail games within the player-game bipartite graph, thereby enriching recommendation diversity. The third module combines the above two modules and employs a new negative-sample rating score reweighting method to balance accuracy and diversity. Experimental results on the Steam dataset demonstrate the effectiveness of our proposed method in improving game recommendations. The dataset and source codes are anonymously released at: https://github.com/CPGRec2024/CPGRec.git.
GNN-based recommendation systems have been successful in capturing complex user-item interactions using multi-hop message passing. However, these methods often struggle to handle the dynamic nature of user-item interactions, making it challenging to adapt to changes in user preferences and new data distributions. This limits their scalability and performance in real-world dynamic scenarios. In our study, we propose a framework called GraphPro that combines dynamic graph pre-training with prompt learning in an efficient way. This unique approach allows GNNs to effectively capture both long-term user preferences and short-term behavior changes, resulting in accurate and up-to-date recommendations. To address the issue of changing user preferences, we integrate a temporal prompt mechanism and a graph-structural prompt learning mechanism into the pre-trained GNN architecture. The temporal prompt mechanism incorporates time-related information into user-item interactions, enabling the model to naturally incorporate temporal dynamics. The graph-structural prompt learning mechanism allows the model to apply pre-trained insights to new behavior dynamics without the need for continuous retraining. We also introduce a dynamic evaluation framework for recommendations that better reflects real-world scenarios and reduces the offline-online discrepancy. Through comprehensive experiments, including deployment in a large-scale industrial scenario, we demonstrate the seamless scalability of GraphPro with various leading recommenders. Our results highlight the superiority of GraphPro in terms of effectiveness, robustness, and efficiency. We release the model implementation at the link: https://github.com/HKUDS/GraphPro.
Developing accurate off-policy estimators is crucial for both evaluating and optimizing for new policies. The main challenge in off-policy estimation is the distribution shift between the logging policy that generates data and the target policy that we aim to evaluate. Typically, techniques for correcting distribution shift involve some form of importance sampling. This approach results in unbiased value estimation but often comes with the trade-off of high variance. Furthermore, importance sampling relies on the common support assumption, which becomes impractical when the action space is large. To address these challenges, we introduce the Policy Convolution (PC) family of estimators for the contextual bandit setting. These methods leverage latent structure within actions---made available through action embeddings---to strategically convolve the logging and target policies. This convolution introduces a unique bias-variance trade-off, that can be controlled via the amount of convolution. Our experiments on synthetic and benchmark datasets demonstrate remarkable mean squared error (MSE) improvements when using PC, especially when either the action space or policy mismatch becomes large, with gains of up to 5-6 orders of magnitude over existing estimators.
Click-Through Rate (CTR) prediction holds paramount significance in online advertising and recommendation scenarios. Despite the proliferation of recent CTR prediction models, the improvements in performance have remained limited, as evidenced by open-source benchmark assessments. Current researchers tend to focus on developing new models for various datasets and settings, often neglecting a crucial question: What is the key challenge that truly makes CTR prediction so demanding? In this paper, we approach the problem of CTR prediction from an optimization perspective. We explore the typical data characteristics and optimization statistics of CTR prediction, revealing a strong positive correlation between the top hessian eigenvalue and feature frequency. This correlation implies that frequently occurring features tend to converge towards sharp local minima, ultimately leading to suboptimal performance. Motivated by the recent advancements in sharpness-aware minimization (SAM), which considers the geometric aspects of the loss landscape during optimization, we present a dedicated optimizer crafted for CTR prediction, named Helen. Helen incorporates frequency-wise Hessian eigenvalue regularization, achieved through adaptive perturbations based on normalized feature frequencies. Empirical results under the open-source benchmark framework underscore Helen's effectiveness. It successfully constrains the top eigenvalue of the Hessian matrix and demonstrates a clear advantage over widely used optimization algorithms when applied to seven popular models across three public benchmark datasets on BARS. Our code locates at github.com/NUS-HPC-AI-Lab/Helen.
The Point-of-Interest (POI) recommendation system, designed to recommend potential future visits of users based on their check-in sequences, faces the challenge of data scarcity. This challenge primarily stems from the data sparsity issue, namely users interact with only a small number of POIs. Most existing studies attempt to solve this problem by focusing on POI check-in sequences, without considering the substantial multi-modal content information (e.g. textual and image data) commonly associated with POIs. In this paper, we propose a novel multi-modal content-aware framework for POI recommendation (MMPOI). Our approach addresses the issue of data sparsity by incorporating multi-modal content information about POIs from a new perspective. Specifically, MMPOI leverages pre-trained models for inter-modal conversion and employs a unified pre-trained model to extract modal-specific features from each modality, effectively bridging the semantic gap between different modalities. We propose to build a Multi-Modal Trajectory Flow Graph (MTFG) which combines the multi-modal semantic structure with check-in sequences. Moreover, we design an adaptive multi-task Transformer that models users' multi-modal movement patterns and integrates them for the next POI recommendation tasks. Extensive experiments on four real-world datasets demonstrate that MMPOI outperforms state-of-the-art POI recommendation methods. To facilitate reproducibility, we have released both the code and the multi-modal POI recommendation datasets we collect https://github.com/zzmylq/MMPOI
Short- and long-term outcomes of an algorithm often differ, with damaging downstream effects. A known example is a click-bait algorithm, which may increase short-term clicks but damage long-term user engagement. A possible solution to estimate the long-term outcome is to run an online experiment or A/B test for the potential algorithms, but it takes months or even longer to observe the long-term outcomes of interest, making the algorithm selection process unacceptably slow. This work thus studies the problem of feasibly yet accurately estimating the long-term outcome of an algorithm using only historical and short-term experiment data. Existing approaches to this problem either need a restrictive assumption about the short-term outcomes called surrogacy or cannot effectively use short-term outcomes, which is inefficient. Therefore, we propose a new framework called Long-term Off-Policy Evaluation (LOPE), which is based on reward function decomposition. LOPE works under a more relaxed assumption than surrogacy and effectively leverages short-term rewards to substantially reduce the variance. Synthetic experiments show that LOPE outperforms existing approaches particularly when surrogacy is severely violated and the long-term reward is noisy. In addition, real-world experiments on large-scale A/B test data collected on a music streaming platform show that LOPE can estimate the long-term outcome of actual algorithms more accurately than existing feasible methods.
How can we recommend cold-start bundles to users? The cold-start problem in bundle recommendation is crucial because new bundles are continuously created on the Web for various marketing purposes. Despite its importance, existing methods for cold-start item recommendation are not readily applicable to bundles. They depend overly on historical information, even for less popular bundles, failing to address the primary challenge of the highly skewed distribution of bundle interactions. In this work, we propose CoHeat (Popularity-based Coalescence and Curriculum Heating), an accurate approach for cold-start bundle recommendation. CoHeat first represents users and bundles through graph-based views, capturing collaborative information effectively. To estimate the user-bundle relationship more accurately, CoHeat addresses the highly skewed distribution of bundle interactions through a popularity-based coalescence approach, which incorporates historical and affiliation information based on the bundle's popularity. Furthermore, it effectively learns latent representations by exploiting curriculum learning and contrastive learning. CoHeat demonstrates superior performance in cold-start bundle recommendation, achieving up to 193% higher nDCG@20 compared to the best competitor.
Many existing recommender systems (RSs) assume user behavior is governed solely by their interests. However, the peer effect often influences individual decision-making, which leads to conformity behavior. Conventional solutions that eliminate indiscriminately such bias may cause RSs to neglect valuable information and depersonalize the recommendation results. Also, conformity can transform into user interest, e.g., discovering new tastes after a glance at popular music. By better representing different forms of conformity influence, we can do a better job at interest mining and debiasing. In certain extreme circumstances, the herd effect may be exacerbated by user anxiety with uncertainty (e.g., panic buying during the COVID-19 pandemic). RSs may thus fail to respond in time due to sudden and dramatic changes. Moreover, many existing studies potentially conflate conformity bias with popularity bias and lump together various factors responsible for differences in popularity. In this paper, we identify two distinct types of conformity behavior: informational conformity and normative conformity. To address this, we introduce the TCHN model, which utilizes attentional Hawkes processes to disentangle user self-interest and conformity in a personalized manner. Our approach incorporates temporal graph attention networks to capture users' stable and volatile dynamics. We conduct experiments on three real-world datasets, which uncover diverse levels of conformity among users. The results show that TCHN excels in recommendation accuracy, diversity, and fairness across various user groups.
Empowered by the widespread adoption of Internet of Things (IoT) devices and smartphones, last-mile delivery services have evolved to accommodate both delivery and pickup tasks. An essential challenge in last-mile delivery is efficiently planning routes for couriers to handle pre-scheduled delivery requests as well as stochastic pickup requests. Existing work approaches this problem by either adjusting routes on the fly when new requests arise or preplanning routes based on predicted future pickup requests. However, these methods either compromise the optimality of planned routes or heavily rely on the accuracy of predictions. In this work, we take conformal prediction as an opportunity to address the issue of prediction uncertainty. We design ROPU, a novel courier route planning framework for logistics systems that incorporates conformal prediction into reinforcement learning. Our work advances the existing work from two aspects: (i) Pickup request prediction utilizes spatial-temporal conformal prediction to capture historical pickup request patterns, providing a unified spatial-temporal conformal interval with high confidence (ii) A spatial-temporal attention network assesses location importance from various perspectives and enables the actor to perceive time and integrate the spatial-temporal conformal interval. We implement and evaluate ROPU on one of the largest logistics platforms. Extensive experiment results demonstrate that our method outperforms other state-of-the-art methods with improvements of at least 30.49% in the pickup overdue rate, 25.00% in the delivery overdue rate, and 5.49% in the traveling distance metric.
The global deployment of the 5G network has led to a substantial increase in the deployment of edge servers to host web applications, catering to the growing demand for low service latency by edge web users. Yet, running edge servers 24/7 leads to enormous energy consumption and excessive carbon emissions. Energy-efficient edge resource provision is desired to achieve sustainable development goals in the new multi-access edge computing (MEC) architecture. Recently, several approaches have been proposed to solve the demand response problem for energy saving in cloud computing and MEC. However, accurate location information of edge web users should always be provided, which sacrifices users' privacy. To protect edge web users' location privacy while saving energy in MEC, we systematically formulate this location privacy-preserving edge demand response (LEDR) problem. To solve the LEDR problem effectively and efficiently, we propose a system named GEES by incorporating differential geo-obfuscation to secure user privacy while maximizing system utility and energy efficiency through inferences with theoretical analysis. Extensive and comprehensive experiments are conducted based on a synthetic real-world dataset, and the results demonstrate that GEES outperforms representative approaches by 23.02%, 31.47%, and 17.29% on average in terms of energy efficiency, user privacy and system utility.