Video Quality Assessment (VQA) is a crucial component of broadscale video retrieval systems. Its goal is to accurately identify various quality issues in videos, thereby encouraging the video retrieval system to prioritize high-quality videos. In large-scale industrial video retrieval systems, we formulate the characteristics of low-quality videos into four categories: visual-related low-level quality problems such as mosaics and black boxes, textual-related low-level quality problems caused by video title and Optical Character Recognition (OCR) content, as well as semantic-level frame incoherence and frame-text mismatch caused by emerging AI-generated videos. These kinds of low-quality videos, which are widely present in industrial environments, have been overlooked in academic research before, and accurately identifying them is very challenging. In this paper, we introduce a Multi-Branch Collaborative learning Network (MBCN) to tackle the above issues. We carefully design four assessment branches for MBCN to adapt to the above four kinds of issues for industrial video retrieval systems. After obtaining independent scores for each branch, we perform a weighted aggregation of the various branches to dynamically address video quality issues in different scenarios with a squeeze-and-excitation mechanism. Finally, we integrate point-wise and pair-wise optimization objectives to ensure the predicted scores are stable and fall into a reasonable range. To demonstrate the effectiveness of our proposed MBCN, we conduct extensive offline and online experiments in a world-level video search engine. The experimental results show that due to the powerful ability of MBCN to identify video quality issues, the ranking ability of the video retrieval system has been significantly improved. We also conduct a series of detailed experimental analyses to verify that all four evaluation branches play a positive role. Besides that, for emerging low-quality AI-generated videos, the recognition accuracy of MBCN also improves significantly compared to the baseline.
论文检索
输入标题、作者或关键词，从 844 篇学术成果中精准定位
Struct-X: Enhancing the Reasoning Capabilities of Large Language Models in Structured Data Scenarios
Conducting reasoning tasks with large language models (LLMs) on structured and redundant data poses significant challenges, primarily due to the complexity introduced by the structured markdown tokens and the presence of extraneous contextual information. These elements can overburden and disrupt the generation process of LLMs, complicating the extraction of relevant insights and the production of coherent outputs. To address this, we propose Struct-X, a novel framework that operates through five key phases: ''read-model-fill-reflect-reason'' efficiently enabling LLMs to utilize structured data. It begins by encoding structured data into a topological space using graph embeddings, followed by filling in missing entity information with knowledge retrieval modules, and filtering out irrelevant tokens via a self-supervised module. The final phase involves constructing a topological network with selected tokens to further reduce the total token length for more effective LLM inference. Additionally, Struct-X includes an Auxiliary Module trained to generate prompts, aiding LLMs in analyzing structured data. Extensive experiments on open-source benchmarks, including the knowledge graph question-answer task and the long document reading comprehension task, show that Struct-X notably improves LLM reasoning in complex structured input context. Finally, we deployed Struct-X in a real-world financial report analysis task, where it exhibits enhanced reasoning capabilities when applied to authentic scenario. The code has been undergoing open-source development to facilitate easy replication.
In the task of fake news detection, ensuring authenticity and accuracy is of paramount importance. This task, however, is susceptible to the influence of confounders, necessitating effective confounder debiasing strategies. Conventional methods are typically designed to address specific confounders, resulting in frameworks that relatively lack generalization and overlook potential correlations among confounders. The presence of multiple confounders further escalates the complexity and challenges of debiasing learning. To tackle this issue, we introduce the Adversarial Multi-Deconfounded (AMD) Learning Paradigm, a generic training framework designed to eliminate biases from multiple confounders. Our approach leverages adversarial networks to extract confounder-invariant feature representations, guiding the model to ignore potential biases introduced by confounders and extract stable representations independent of these confounders, thereby enhancing generalization. Comprehensive experiments demonstrate that our approach outperforms state-of-the-art methods on the Weibo and GossipCop datasets, and significantly exceeds other methods in generalization evaluation on CHEF. Additionally, we validate that our AMD framework exhibits improved robustness against confounders.
In the e-commerce domain, the accurate extraction of attribute-value pairs (e.g., Brand: Apple) from product titles and user search queries is crucial for enhancing search and recommendation systems. A major challenge with neural models for this task is the lack of high-quality training data, as the annotations for attribute-value pairs in the available datasets are often incomplete. To address this, we introduce GenToC, a model designed for training directly with partially-labeled data, eliminating the necessity for a fully annotated dataset. GenToC employs a marker-augmented generative model to identify potential attributes, followed by a token classification model that determines the associated values for each attribute. GenToC outperforms existing state-of-the-art models, exhibiting upto 56.3% increase in the number of accurate extractions. Furthermore, we utilize GenToC to regenerate the training dataset to expand attribute-value annotations. This bootstrapping substantially improves the data quality for training other standard NER models, which are typically faster but less capable in handling partially-labeled data, enabling them to achieve comparable performance to GenToC. Our results demonstrate GenToC's unique ability to learn from a limited set of partially-labeled data and improve the training of more efficient models, advancing the automated extraction of attribute-value pairs. Finally, our model has been successfully integrated into IndiaMART, India's largest B2B e-commerce platform, achieving a significant increase of 20.2% in the number of correctly identified attribute-value pairs over the existing deployed system while achieving a high precision of 89.5%. We have released the code for GenToC model at https://github.com/KnowDisAI/GenToC.
CATER: A Cluster-Based Alternative-Term Recommendation Framework for Large-Scale Web Search at NAVER
Recently, searching for information by using search engines such as Google, Bing, and NAVER has become ubiquitous. While they attempt to provide information based on the search queries that users enter, it is not trivial to accurately capture the search intent of users. Motivated by this situation, NAVER Corp., the largest portal company in Korea, has developed a framework named as CATER (Cluster-based Alternative TErm Recommendation) framework that suggests alternative terms ("al-terms,'' in short) for better search outcomes relevant to a user's search intent. We introduce four design considerations (DCs) that were considered when designing and implementing CATER. Then, we describe how our CATER addresses the four DCs by using a clustering stage that dynamically maintains a pool of topic-oriented clusters containing terms, and a recommendation stage that identifies the top-k clusters (i.e., topics) and the top-k al-terms for each cluster. Furthermore, we present the scalable architecture adopted by CATER. Through various offline and online A/B tests using real-world datasets from NAVER, we validate that CATER successfully incorporates all DCs and that all design choices help improve the recommendation accuracy.
The rapid growth of health insurance and the rising incidence of fraudulent claims underscore the necessity for an efficient and professional claims management system. However, there is a noticeable lack of shared relevant experience from previous research in this field. In response to this challenge, we introduce AntAkso, a robust claims management system specifically designed for health insurance operations within Alipay. AntAkso incorporates a digital and professional management system, achieving a notable decrease in the volume of false claims, reduction in administrative costs, and heightened satisfaction among its policyholders. We begin by highlighting the core components of this system, including the case stratification, hospital recommendation, and case dispatch modules, along with the pivotal algorithms employed, i.e., the fraud detection, recommendation, and robust satisficing algorithms. We also detail the system's implementation and deployment. We substantiate the proposed system's effectiveness and efficiency with empirical evidence from experiments on a large set of real-world health insurance claims data.
In recent years, the data collected for artificial intelligence has grown to an unmanageable amount. Particularly within industrial applications, such as autonomous vehicles, model training computation budgets are being exceeded while model performance is saturating -- and yet more data continues to pour in. To navigate the flood of data, we propose a framework to select the most semantically diverse and important dataset portion. Then, we further semantically enrich it by discovering meaningful new data from a massive unlabeled data pool. Importantly, we can provide explainability by leveraging foundation models to generate semantics for every data point. We quantitatively show that our Semantic Selection and Enrichment framework (SSE) can a) successfully maintain model performance with a smaller training dataset and b) improve model performance by enriching the smaller dataset without exceeding the original dataset size. Consequently, we demonstrate that semantic diversity is imperative for optimal data selection and model performance.
Explore Further @ Bing is a webpage-to-webpage recommendation product, enhancing the search experience on Bing by surfacing engaging webpage recommendations tied to the search result URLs. In this paper, we present our approach for leveraging Large Language Models (LLMs) for enhancing our web-scale recommendation system. We describe the development and validation of our LLM-powered recommendation quality metric RecoDCG. We discuss our core techniques for utilizing LLMs to make our ranking stage quality-aware. Furthermore, we detail Q' recall, a recall path that enhances our system's candidate generation stage by leveraging LLMs to produce complementary and engaging recommendation candidates. We also address how we optimize our system for multiple objectives, balancing recommendation quality with click metrics. We deploy our work to production, achieving a significant improvement in recommendation quality. We share results from offline and online experiments as well as insights and steps we took to ensure our approaches scale effectively for our web-scale needs.
Wake vortices-strong, coherent air turbulences created by aircrafts-pose a significant risk to aviation safety and therefore require accurate and reliable detection methods. In this paper, we present an advanced, explainable machine learning method that utilizes Light Detection and Ranging (LiDAR) data for effective wake vortex detection. Our method leverages a dynamic graph CNN (DGCNN) with semantic segmentation to partition a 3D LiDAR point cloud into meaningful segments. Further refinement is achieved through clustering techniques. A novel feature of our research is the use of a perturbation-based explanation technique, which clarifies the model's decision-making processes for air traffic regulators and controllers, increasing transparency and building trust. Our experimental results, based on measured and simulated LiDAR scans compared against four baseline methods, underscore the effectiveness and reliability of our approach. This combination of semantic segmentation and clustering for real-time wake vortex tracking significantly advances aviation safety measures, ensuring that these are both effective and comprehensible.
In this work, we develop a graphical model to capture team dynamics. We analyze the model and show how to learn its parameters from data. Using our model we study the phenomenon of team collapse from a computational perspective. We use simulations and real-world experiments to find the main causes of team collapse. We also provide the principles of building resilient teams, i.e., teams that avoid collapsing. Finally, we use our model to analyze the structure of NBA teams and dive deeper into games of interest.
Products on e-commerce platforms are usually organized based on seller-provided product attributes. Customers looking for a product typically have certain needs or use cases in mind, such as headphones for gym classes, or a printer for school projects. However, they often struggle to map these use cases to product attributes, thereby failing to find the product they need. To help customers shop online confidently, we propose a Use case-Based Shopping (UBS) system that facilitates customer experiences based on use cases (Fig. 1). UBS consists of three steps: 1) use case phrase extraction from product reviews, 2) clustering the extracted use case phrases to identify the dominant ones, and 3) mapping products in a given category to one or more of these dominant use cases. In this work, we utilize instruction-tuned LMs to primarily focus on the first two steps. However, the way we design them also helps us to seamlessly solve the third step to complete the design of our proposed system. First, we define the novel task of joint Use Uase, Uentiment Uxtraction (UCSE) from product reviews which can be used for both steps 1 and 3. We harness the task adaptation capability of instruction-tuned FLAN-T5 models and gradually improve their zero-shot UCSE performance through instruction tuning, multi-task training, and few-shot iterative re-training for new categories, achieving around ~90% reduction in annotation bandwidth. We then employ Anthropic's Claude 2 LLM to propose an unsupervised approach for hierarchical use case phrase clustering that demonstrates better clustering and cluster naming capabilities when compared to K-Means and LDA. In an online experiment targeting the top 7 product categories, UBS recommendations on search, browse, and product pages resulted in a revenue lift of 0.77%, 0.94%, and 0.44% respectively, and an average click rate lift of 0.15%.
Roadside LiDAR (Light Detection and Ranging) sensors promise safer and faster traffic management and vehicular operations. However, occlusion and small view angles are significant challenges to widespread use of roadside LiDARs. We consider fusing data from multiple LiDARs at a traffic intersection to better estimate traffic parameters than one can estimate from a single LiDAR. The key challenge is to calibrate multiple LiDARs both in time and space. The problem is more complex when heterogeneous sensors differ in resolution and are positioned arbitrarily on a traffic intersection. We propose a calibration technique to fuse multiple LiDARs. We show that our technique works on various data granularity and enables real-time analytics for roadside traffic monitoring. We evaluate on a large number of simulated traffic scenarios and show that fusion improves accuracy of vehicle counting and near-collision detection. We apply our algorithm on real traffic data and demonstrate utility in classifying vehicles and detecting occluded traffic participants.
Session-level dynamic ad load optimization aims to personalize the density and types of delivered advertisements in real time during a user's online session by dynamically balancing user experience quality and ad monetization. Traditional causal learning-based approaches struggle with key technical challenges, especially in handling confounding bias and distribution shifts. In this paper, we develop an offline deep Q-network (DQN)-based framework that effectively mitigates confounding bias in dynamic systems and demonstrates more than 80% offline gains compared to the best causal learning-based production baseline. Moreover, to improve the framework's robustness against unanticipated distribution shifts, we further enhance our framework with a novel offline robust dueling DQN approach. This approach achieves more stable rewards on multiple OpenAI-Gym datasets as perturbations increase, and provides an additional 5% offline gains on real-world ad delivery data. Deployed across multiple production systems, our approach has achieved outsized topline gains. Post-launch online A/B tests have shown double-digit improvements in the engagement-ad score trade-off efficiency, significantly enhancing our platform's capability to serve both consumers and advertisers.
We present LinkSAGE, an innovative framework that integrates Graph Neural Networks (GNNs) into large-scale personalized job matching systems, designed to address the complex dynamics of LinkedIn's extensive professional network. Our approach capitalizes on a novel job marketplace graph, the largest and most intricate of its kind in industry, with billions of nodes and edges. This graph is not merely extensive but also richly detailed, encompassing member and job nodes along with key attributes, thus creating an expansive and interwoven network. A key innovation in LinkSAGE is its training and serving methodology, which effectively combines inductive graph learning on a heterogeneous, evolving graph with an encoder-decoder GNN model. This methodology decouples the training of the GNN model from that of existing Deep Neural Network (DNN) models, eliminating the need for frequent GNN retraining while maintaining up-to-date graph signals in near real-time, allowing for the effective integration of GNN insights through transfer learning. The subsequent nearline inference system serves the GNN encoder within a real-world setting, significantly reducing online latency and obviating the need for costly real-time GNN infrastructure. Validated across multiple online A/B tests in diverse product scenarios, LinkSAGE demonstrates marked improvements in member engagement, relevance matching, and member retention, confirming its generalizability and practical impact.
Multi-scenario recommendation (MSR) is a popular training paradigm in industrial platforms for uniformly integrating information from multiple scenarios and serving them simultaneously. A key challenge in MSR research is accurately identifying the commonalities and distinctive information between scenarios. Currently, most existing MSR methods focus on implicitly extracting this information from the architectural level. However, this continues to increase the complexity and training overhead of MSR. Furthermore, the custom components responsible for extracting implicit information in each MSR method are too dependent on the specific MSR architecture and are not easily reused in other methods. Given these challenges, we first show in a motivating experiment that it may be beneficial to explicitly select a reasonable set of shared instances that can affect parameter optimization in all scenarios during the training of MSR, i.e., to explicitly obtain the critical information required for MSR from the data level. Then, this paper proposes SSIM with an adaptive selection network. Specifically, SSIM can be integrated with existing MSR methods in a lightweight way to adaptively select an informative and shareable subset of instances from each scenario to improve recommendations. In particular, the selected multi-scenario shared subset has extraordinary reusability and can be easily saved to benefit model training of various future MSR models. Finally, we evaluate SSIM and demonstrate its effectiveness through experiments on two public multi-scenario benchmarks and an online A/B test.
Intensity-modulated radiation therapy (IMRT) is a crucial radiotherapy technique, which is often formulated as an optimization problem. However, when the constraints are too restrictive to provide a feasible solution, human planners resort to relaxing the optimization parameters and re-evaluating the problem until an acceptable solution is obtained. However, this process is laborious and time-consuming, which has prompted attempts to automate radiotherapy through inverse planning studies using reinforcement learning. Unfortunately, these studies face two major limitations. First, a separate sub-network must be designed for each organ, rendering it difficult to apply to patients with an inconsistent number of structures. Second, the low signal-to-noise input and discrete action space result in low training efficiency. To address these issues, we propose an organ-sharing network that contains a functional embedding layer to extract curve features of the dose-volume histogram. It outputs continuous actions that can adjust the optimization parameters, thereby automating the radiotherapy planning process. The results from a cervical cancer dataset demonstrate the feasibility and efficiency of the proposed model in real-world radiotherapy. The code is available on https://github.com/Cissise/FatPIN.
While representation learning has been used to boost the performance of Learning-to-Rank (LTR) models through distilling key features for webpage ranking, the weak supervision signals extracted from users' sparse click-through data lead to inadequate representation of query-webpage pairs for ranking score prediction. Recent studies in generative LTR pre-training demonstrate the feasibility of incorporating reconstruction loss for enhanced ranking score prediction. However, LTR is afterall a regression task and it might be reasonable to find an alternate route that pre-trains LTR models with discriminative losses. Following the success of Electra in representation learning for natural language processing (NLP), this work proposes RankElectra that pre-trains the LTR model as a discriminator module inside a generative learning framework. Specifically, RankElectra first structures sparsely-annotated query-webpage pairs into a bipartite graph, with query and webpage feature vectors as node types and ranking scores as the connecting edges, and then leverages positive and negative extension strategies to densify the graph by link predictions. Later, this work proposes a novel Electra module that pre-trains the LTR model as a discriminator module for node reconstruction tasks, where node features of selected edges would be randomly masked and reconstructed by a generator, and the discriminator learns to classify whether the reconstructed features are the original or replaced as well as perform correct ranking. Finally, the pre-trained discriminator module, rather than the generator, would be fine-tuned on the labeled graph. We carried out extensive offline and online evaluations using the real-world web traffic of Baidu search engine. The results show that RankElectra could significantly boost the ranking performance of Baidu Search compared with numbers of competitor systems.
With recent generative models facilitating photo-realistic image synthesis, the proliferation of synthetic images has also engendered certain negative impacts on social platforms, thereby raising an urgent imperative to develop effective detectors. Current synthetic image detection (SID) pipelines are primarily dedicated to crafting universal artifact features, accompanied by an oversight about SID training paradigm. In this paper, we re-examine the SID problem and identify two prevalent biases in current training paradigms, i.e., weakened artifact features and overfitted artifact features. Meanwhile, we discover that the imaging mechanism of synthetic images contributes to heightened local correlations among pixels, suggesting that detectors should be equipped with local awareness. In this light, we propose SAFE, a lightweight and effective detector with three simple image transformations. Firstly, for weakened artifact features, we substitute the down-sampling operator with the crop operator in image pre-processing to help circumvent artifact distortion. Secondly, for overfitted artifact features, we include ColorJitter and RandomRotation as additional data augmentations, to help alleviate irrelevant biases from color discrepancies and semantic differences in limited training samples. Thirdly, for local awareness, we propose a patch-based random masking strategy tailored for SID, forcing the detector to focus on local regions at training. Comparative experiments are conducted on an open-world dataset, comprising synthetic images generated by 26 distinct generative models. Our pipeline achieves a new state-of-the-art performance, with remarkable improvements of 4.5% in accuracy and 2.9% in average precision against existing methods. Our code is available at: https://github.com/Ouxiang-Li/SAFE.
Effective traffic signal control (TSC) is crucial in mitigating urban congestion and reducing emissions. Recently, reinforcement learning (RL) has been the research trend for TSC. However, existing RL algorithms face several real-world challenges that hinder their practical deployment in TSC: (1) Sensor accuracy deteriorates with increased sensor detection range, and data transmission is prone to noise, potentially resulting in unsafe TSC decisions. (2) During the training of online RL, interactions with the environment could be unstable, potentially leading to inappropriate traffic signal phase (TSP) selection and traffic congestion. (3) Most current TSC algorithms focus only on TSP decisions, overlooking the critical aspect of phase duration, affecting safety and efficiency. To overcome these challenges, we propose a robust two-stage fuzzy approach called FuzzyLight, which integrates compressed sensing and RL for TSC deployment. FuzzyLight offers several key contributions: (1) It employs fuzzy logic and compressed sensing to address sensor noise and enhances the efficiency of TSP decisions. (2) It maintains stable performance during training and combines fuzzy logic with RL to generate precise phases. (3) It works in real cities across 22 intersections and demonstrates superior performance in both real-world and simulated environments. Experimental results indicate that FuzzyLight enhances traffic efficiency by 48% compared to expert-designed timings in the real world. Furthermore, it achieves state-of-the-art (SOTA) performance in simulated environments using six real-world datasets with transmission noise. The code and deployment video are available at the Github.
Online Travel Platforms (OTPs) serve as crucial bridges between hotels and users, hotel staff can synchronize room inventory information with OTPs through manual and auto modes. In the manual mode, the hotel staff must manually maintain the inventory information on the OTPs. This mode often leads to the "inventory synchronization delay'' phenomenon where OTPs show no availability while hotels still have available rooms, seriously affecting the competitiveness of OTPs and hotel sales. To address this issue, Fliggy uses inventory add prediction (IAP) to determine whether to add an inventory for the sold-out room type. However, in practice, accurate modeling of IAP faces significant challenges due to the data sparsity. In this paper, we propose a Contrastive Learning framework for Inventory Add Prediction at Fliggy (CL4IAP), which consists of the Joint Pay-Accept Prediction Module, the Data Augmentation Module, and the Contrastive Learning Module. Specifically, the Joint Pay-Accept Prediction Module aims to predict the likelihood of generating an order and the hotel acceptance after adding an inventory. It also includes a specially designed correlation enhancement component that facilitates the expert prediction network's learning through knowledge transfer based on inter-task correlation. In the Data Augmentation Module, we design three novel data augmentation strategies for the first time based on the correlation and importance of features. In the Contrastive Learning Module, we design instance-level and cluster-level contrastive losses, which aim to minimize the distance between positive sample pairs and mitigate the negative impact of false negative sample pairs, respectively. Both offline and online experiments demonstrate the effectiveness of CL4IAP, and CL4IAP has been successfully deployed on Fliggy.