Najmeh Forouzandehmehr, Nima Farrokhsiar, Ramin Giahi, Evren Korpeoglu, Kannan Achan
Personalized outfit recommendation remains a complex challenge, demanding both fashion compatibility understanding and trend awareness. This paper presents a novel framework that harnesses the expressive power of large language models (LLMs) for this task, mitigating their "black box" and static nature through fine-tuning and direct feedback integration. We bridge the item visual-textual gap in items descriptions by employing image captioning with a Multimodal Large Language Model (MLLM). This enables the LLM to extract style and color characteristics from human-curated fashion images, forming the basis for personalized recommendations. The LLM is efficiently fine-tuned on the open-source Polyvore dataset of curated fashion images, optimizing its ability to recommend stylish outfits. A direct preference mechanism using negative examples is employed to enhance the LLM's decision-making process. This creates a self-enhancing AI feedback loop that continuously refines recommendations in line with seasonal fashion trends. Our framework is evaluated on the Polyvore dataset, demonstrating its effectiveness in two key tasks: fill-in-the-blank, and complementary item retrieval. These evaluations underline the framework's ability to generate stylish, trend-aligned outfit suggestions, continuously improving through direct feedback. The evaluation results demonstrated that our proposed framework significantly outperforms the base LLM, creating more cohesive outfits. The improved performance in these tasks underscores the proposed framework's potential to enhance the shopping experience with accurate suggestions, proving its effectiveness over the vanilla LLM based outfit generation.
{"title":"Decoding Style: Efficient Fine-Tuning of LLMs for Image-Guided Outfit Recommendation with Preference","authors":"Najmeh Forouzandehmehr, Nima Farrokhsiar, Ramin Giahi, Evren Korpeoglu, Kannan Achan","doi":"arxiv-2409.12150","DOIUrl":"https://doi.org/arxiv-2409.12150","url":null,"abstract":"Personalized outfit recommendation remains a complex challenge, demanding\u0000both fashion compatibility understanding and trend awareness. This paper\u0000presents a novel framework that harnesses the expressive power of large\u0000language models (LLMs) for this task, mitigating their \"black box\" and static\u0000nature through fine-tuning and direct feedback integration. We bridge the item\u0000visual-textual gap in items descriptions by employing image captioning with a\u0000Multimodal Large Language Model (MLLM). This enables the LLM to extract style\u0000and color characteristics from human-curated fashion images, forming the basis\u0000for personalized recommendations. The LLM is efficiently fine-tuned on the\u0000open-source Polyvore dataset of curated fashion images, optimizing its ability\u0000to recommend stylish outfits. A direct preference mechanism using negative\u0000examples is employed to enhance the LLM's decision-making process. This creates\u0000a self-enhancing AI feedback loop that continuously refines recommendations in\u0000line with seasonal fashion trends. Our framework is evaluated on the Polyvore\u0000dataset, demonstrating its effectiveness in two key tasks: fill-in-the-blank,\u0000and complementary item retrieval. These evaluations underline the framework's\u0000ability to generate stylish, trend-aligned outfit suggestions, continuously\u0000improving through direct feedback. The evaluation results demonstrated that our\u0000proposed framework significantly outperforms the base LLM, creating more\u0000cohesive outfits. The improved performance in these tasks underscores the\u0000proposed framework's potential to enhance the shopping experience with accurate\u0000suggestions, proving its effectiveness over the vanilla LLM based outfit\u0000generation.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255155","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Rajaa El Hamdani, Thomas Bonald, Fragkiskos Malliaros, Nils Holzenberger, Fabian Suchanek
This paper investigates the factuality of large language models (LLMs) as knowledge bases in the legal domain, in a realistic usage scenario: we allow for acceptable variations in the answer, and let the model abstain from answering when uncertain. First, we design a dataset of diverse factual questions about case law and legislation. We then use the dataset to evaluate several LLMs under different evaluation methods, including exact, alias, and fuzzy matching. Our results show that the performance improves significantly under the alias and fuzzy matching methods. Further, we explore the impact of abstaining and in-context examples, finding that both strategies enhance precision. Finally, we demonstrate that additional pre-training on legal documents, as seen with SaulLM, further improves factual precision from 63% to 81%.
{"title":"The Factuality of Large Language Models in the Legal Domain","authors":"Rajaa El Hamdani, Thomas Bonald, Fragkiskos Malliaros, Nils Holzenberger, Fabian Suchanek","doi":"arxiv-2409.11798","DOIUrl":"https://doi.org/arxiv-2409.11798","url":null,"abstract":"This paper investigates the factuality of large language models (LLMs) as\u0000knowledge bases in the legal domain, in a realistic usage scenario: we allow\u0000for acceptable variations in the answer, and let the model abstain from\u0000answering when uncertain. First, we design a dataset of diverse factual\u0000questions about case law and legislation. We then use the dataset to evaluate\u0000several LLMs under different evaluation methods, including exact, alias, and\u0000fuzzy matching. Our results show that the performance improves significantly\u0000under the alias and fuzzy matching methods. Further, we explore the impact of\u0000abstaining and in-context examples, finding that both strategies enhance\u0000precision. Finally, we demonstrate that additional pre-training on legal\u0000documents, as seen with SaulLM, further improves factual precision from 63% to\u000081%.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142269089","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yifan Sun, Rang Liu, Zhiping Lu, Honghao Luo, Ming Li, Qian Liu
Synthetic Aperture Radar (SAR) utilizes the movement of the radar antenna over a specific area of interest to achieve higher spatial resolution imaging. In this paper, we aim to investigate the realization of SAR imaging for a stationary radar system with the assistance of active reconfigurable intelligent surface (ARIS) mounted on an unmanned aerial vehicle (UAV). As the UAV moves along the stationary trajectory, the ARIS can not only build a high-quality virtual line-of-sight (LoS) propagation path, but its mobility can also effectively create a much larger virtual aperture, which can be utilized to realize a SAR system. In this paper, we first present a range-Doppler (RD) imaging algorithm to obtain imaging results for the proposed ARIS-empowered SAR system. Then, to further improve the SAR imaging performance, we attempt to optimize the reflection coefficients of ARIS to maximize the signal-to-noise ratio (SNR) at the stationary radar receiver under the constraints of ARIS maximum power and amplification factor. An effective algorithm based on fractional programming (FP) and majorization minimization (MM) methods is developed to solve the resulting non-convex problem. Simulation results validate the effectiveness of ARIS-assisted SAR imaging and our proposed RD imaging and ARIS optimization algorithms.
{"title":"Active Reconfigurable Intelligent Surface Empowered Synthetic Aperture Radar Imaging","authors":"Yifan Sun, Rang Liu, Zhiping Lu, Honghao Luo, Ming Li, Qian Liu","doi":"arxiv-2409.11728","DOIUrl":"https://doi.org/arxiv-2409.11728","url":null,"abstract":"Synthetic Aperture Radar (SAR) utilizes the movement of the radar antenna\u0000over a specific area of interest to achieve higher spatial resolution imaging.\u0000In this paper, we aim to investigate the realization of SAR imaging for a\u0000stationary radar system with the assistance of active reconfigurable\u0000intelligent surface (ARIS) mounted on an unmanned aerial vehicle (UAV). As the\u0000UAV moves along the stationary trajectory, the ARIS can not only build a\u0000high-quality virtual line-of-sight (LoS) propagation path, but its mobility can\u0000also effectively create a much larger virtual aperture, which can be utilized\u0000to realize a SAR system. In this paper, we first present a range-Doppler (RD)\u0000imaging algorithm to obtain imaging results for the proposed ARIS-empowered SAR\u0000system. Then, to further improve the SAR imaging performance, we attempt to\u0000optimize the reflection coefficients of ARIS to maximize the signal-to-noise\u0000ratio (SNR) at the stationary radar receiver under the constraints of ARIS\u0000maximum power and amplification factor. An effective algorithm based on\u0000fractional programming (FP) and majorization minimization (MM) methods is\u0000developed to solve the resulting non-convex problem. Simulation results\u0000validate the effectiveness of ARIS-assisted SAR imaging and our proposed RD\u0000imaging and ARIS optimization algorithms.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255157","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Yuening Zhou, Yulin Wang, Qian Cui, Xinyu Guan, Francisco Cisternas
Next Basket Recommendation (NBR) is a new type of recommender system that predicts combinations of items users are likely to purchase together. Existing NBR models often overlook a crucial factor, which is price, and do not fully capture item-basket-user interactions. To address these limitations, we propose a novel method called Basket-augmented Dynamic Heterogeneous Hypergraph (BDHH). BDHH utilizes a heterogeneous multi-relational graph to capture the intricate relationships among item features, with price as a critical factor. Moreover, our approach includes a basket-guided dynamic augmentation network that could dynamically enhances item-basket-user interactions. Experiments on real-world datasets demonstrate that BDHH significantly improves recommendation accuracy, providing a more comprehensive understanding of user behavior.
{"title":"Basket-Enhanced Heterogenous Hypergraph for Price-Sensitive Next Basket Recommendation","authors":"Yuening Zhou, Yulin Wang, Qian Cui, Xinyu Guan, Francisco Cisternas","doi":"arxiv-2409.11695","DOIUrl":"https://doi.org/arxiv-2409.11695","url":null,"abstract":"Next Basket Recommendation (NBR) is a new type of recommender system that\u0000predicts combinations of items users are likely to purchase together. Existing\u0000NBR models often overlook a crucial factor, which is price, and do not fully\u0000capture item-basket-user interactions. To address these limitations, we propose\u0000a novel method called Basket-augmented Dynamic Heterogeneous Hypergraph (BDHH).\u0000BDHH utilizes a heterogeneous multi-relational graph to capture the intricate\u0000relationships among item features, with price as a critical factor. Moreover,\u0000our approach includes a basket-guided dynamic augmentation network that could\u0000dynamically enhances item-basket-user interactions. Experiments on real-world\u0000datasets demonstrate that BDHH significantly improves recommendation accuracy,\u0000providing a more comprehensive understanding of user behavior.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255159","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Kasra Hosseini, Thomas Kober, Josip Krapac, Roland Vollgraf, Weiwei Cheng, Ana Peleteiro Ramallo
Evaluating production-level retrieval systems at scale is a crucial yet challenging task due to the limited availability of a large pool of well-trained human annotators. Large Language Models (LLMs) have the potential to address this scaling issue and offer a viable alternative to humans for the bulk of annotation tasks. In this paper, we propose a framework for assessing the product search engines in a large-scale e-commerce setting, leveraging Multimodal LLMs for (i) generating tailored annotation guidelines for individual queries, and (ii) conducting the subsequent annotation task. Our method, validated through deployment on a large e-commerce platform, demonstrates comparable quality to human annotations, significantly reduces time and cost, facilitates rapid problem discovery, and provides an effective solution for production-level quality control at scale.
{"title":"Retrieve, Annotate, Evaluate, Repeat: Leveraging Multimodal LLMs for Large-Scale Product Retrieval Evaluation","authors":"Kasra Hosseini, Thomas Kober, Josip Krapac, Roland Vollgraf, Weiwei Cheng, Ana Peleteiro Ramallo","doi":"arxiv-2409.11860","DOIUrl":"https://doi.org/arxiv-2409.11860","url":null,"abstract":"Evaluating production-level retrieval systems at scale is a crucial yet\u0000challenging task due to the limited availability of a large pool of\u0000well-trained human annotators. Large Language Models (LLMs) have the potential\u0000to address this scaling issue and offer a viable alternative to humans for the\u0000bulk of annotation tasks. In this paper, we propose a framework for assessing\u0000the product search engines in a large-scale e-commerce setting, leveraging\u0000Multimodal LLMs for (i) generating tailored annotation guidelines for\u0000individual queries, and (ii) conducting the subsequent annotation task. Our\u0000method, validated through deployment on a large e-commerce platform,\u0000demonstrates comparable quality to human annotations, significantly reduces\u0000time and cost, facilitates rapid problem discovery, and provides an effective\u0000solution for production-level quality control at scale.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255156","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Hybrid recommender systems, combining item IDs and textual descriptions, offer potential for improved accuracy. However, previous work has largely focused on smaller datasets and model architectures. This paper introduces Flare (Fusing Language models and collaborative Architectures for Recommender Enhancement), a novel hybrid recommender that integrates a language model (mT5) with a collaborative filtering model (Bert4Rec) using a Perceiver network. This architecture allows Flare to effectively combine collaborative and content information for enhanced recommendations. We conduct a two-stage evaluation, first assessing Flare's performance against established baselines on smaller datasets, where it demonstrates competitive accuracy. Subsequently, we evaluate Flare on a larger, more realistic dataset with a significantly larger item vocabulary, introducing new baselines for this setting. Finally, we showcase Flare's inherent ability to support critiquing, enabling users to provide feedback and refine recommendations. We further leverage critiquing as an evaluation method to assess the model's language understanding and its transferability to the recommendation task.
{"title":"FLARE: Fusing Language Models and Collaborative Architectures for Recommender Enhancement","authors":"Liam Hebert, Marialena Kyriakidi, Hubert Pham, Krishna Sayana, James Pine, Sukhdeep Sodhi, Ambarish Jash","doi":"arxiv-2409.11699","DOIUrl":"https://doi.org/arxiv-2409.11699","url":null,"abstract":"Hybrid recommender systems, combining item IDs and textual descriptions,\u0000offer potential for improved accuracy. However, previous work has largely\u0000focused on smaller datasets and model architectures. This paper introduces\u0000Flare (Fusing Language models and collaborative Architectures for Recommender\u0000Enhancement), a novel hybrid recommender that integrates a language model (mT5)\u0000with a collaborative filtering model (Bert4Rec) using a Perceiver network. This\u0000architecture allows Flare to effectively combine collaborative and content\u0000information for enhanced recommendations. We conduct a two-stage evaluation, first assessing Flare's performance\u0000against established baselines on smaller datasets, where it demonstrates\u0000competitive accuracy. Subsequently, we evaluate Flare on a larger, more\u0000realistic dataset with a significantly larger item vocabulary, introducing new\u0000baselines for this setting. Finally, we showcase Flare's inherent ability to\u0000support critiquing, enabling users to provide feedback and refine\u0000recommendations. We further leverage critiquing as an evaluation method to\u0000assess the model's language understanding and its transferability to the\u0000recommendation task.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255158","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Despite the popularity of the two-tower model for unbiased learning to rank (ULTR) tasks, recent work suggests that it suffers from a major limitation that could lead to its collapse in industry applications: the problem of logging policy confounding. Several potential solutions have even been proposed; however, the evaluation of these methods was mostly conducted using semi-synthetic simulation experiments. This paper bridges the gap between theory and practice by investigating the confounding problem on the largest real-world dataset, Baidu-ULTR. Our main contributions are threefold: 1) we show that the conditions for the confounding problem are given on Baidu-ULTR, 2) the confounding problem bears no significant effect on the two-tower model, and 3) we point to a potential mismatch between expert annotations, the golden standard in ULTR, and user click behavior.
{"title":"Understanding the Effects of the Baidu-ULTR Logging Policy on Two-Tower Models","authors":"Morris de Haan, Philipp Hager","doi":"arxiv-2409.12043","DOIUrl":"https://doi.org/arxiv-2409.12043","url":null,"abstract":"Despite the popularity of the two-tower model for unbiased learning to rank\u0000(ULTR) tasks, recent work suggests that it suffers from a major limitation that\u0000could lead to its collapse in industry applications: the problem of logging\u0000policy confounding. Several potential solutions have even been proposed;\u0000however, the evaluation of these methods was mostly conducted using\u0000semi-synthetic simulation experiments. This paper bridges the gap between\u0000theory and practice by investigating the confounding problem on the largest\u0000real-world dataset, Baidu-ULTR. Our main contributions are threefold: 1) we\u0000show that the conditions for the confounding problem are given on Baidu-ULTR,\u00002) the confounding problem bears no significant effect on the two-tower model,\u0000and 3) we point to a potential mismatch between expert annotations, the golden\u0000standard in ULTR, and user click behavior.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142269087","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Peng Liu, Jiawei Zhu, Cong Xu, Ming Zhao, Bin Wang
As the last key stage of Recommender Systems (RSs), Multi-Task Fusion (MTF) is in charge of combining multiple scores predicted by Multi-Task Learning (MTL) into a final score to maximize user satisfaction, which decides the ultimate recommendation results. In recent years, to maximize long-term user satisfaction within a recommendation session, Reinforcement Learning (RL) is widely used for MTF in large-scale RSs. However, limited by their modeling pattern, all the current RL-MTF methods can only utilize user features as the state to generate actions for each user, but unable to make use of item features and other valuable features, which leads to suboptimal results. Addressing this problem is a challenge that requires breaking through the current modeling pattern of RL-MTF. To solve this problem, we propose a novel method called Enhanced-State RL for MTF in RSs. Unlike the existing methods mentioned above, our method first defines user features, item features, and other valuable features collectively as the enhanced state; then proposes a novel actor and critic learning process to utilize the enhanced state to make much better action for each user-item pair. To the best of our knowledge, this novel modeling pattern is being proposed for the first time in the field of RL-MTF. We conduct extensive offline and online experiments in a large-scale RS. The results demonstrate that our model outperforms other models significantly. Enhanced-State RL has been fully deployed in our RS more than half a year, improving +3.84% user valid consumption and +0.58% user duration time compared to baseline.
{"title":"An Enhanced-State Reinforcement Learning Algorithm for Multi-Task Fusion in Large-Scale Recommender Systems","authors":"Peng Liu, Jiawei Zhu, Cong Xu, Ming Zhao, Bin Wang","doi":"arxiv-2409.11678","DOIUrl":"https://doi.org/arxiv-2409.11678","url":null,"abstract":"As the last key stage of Recommender Systems (RSs), Multi-Task Fusion (MTF)\u0000is in charge of combining multiple scores predicted by Multi-Task Learning\u0000(MTL) into a final score to maximize user satisfaction, which decides the\u0000ultimate recommendation results. In recent years, to maximize long-term user\u0000satisfaction within a recommendation session, Reinforcement Learning (RL) is\u0000widely used for MTF in large-scale RSs. However, limited by their modeling\u0000pattern, all the current RL-MTF methods can only utilize user features as the\u0000state to generate actions for each user, but unable to make use of item\u0000features and other valuable features, which leads to suboptimal results.\u0000Addressing this problem is a challenge that requires breaking through the\u0000current modeling pattern of RL-MTF. To solve this problem, we propose a novel\u0000method called Enhanced-State RL for MTF in RSs. Unlike the existing methods\u0000mentioned above, our method first defines user features, item features, and\u0000other valuable features collectively as the enhanced state; then proposes a\u0000novel actor and critic learning process to utilize the enhanced state to make\u0000much better action for each user-item pair. To the best of our knowledge, this\u0000novel modeling pattern is being proposed for the first time in the field of\u0000RL-MTF. We conduct extensive offline and online experiments in a large-scale\u0000RS. The results demonstrate that our model outperforms other models\u0000significantly. Enhanced-State RL has been fully deployed in our RS more than\u0000half a year, improving +3.84% user valid consumption and +0.58% user duration\u0000time compared to baseline.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255160","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
The ID-free recommendation paradigm has been proposed to address the limitation that traditional recommender systems struggle to model cold-start users or items with new IDs. Despite its effectiveness, this study uncovers that ID-free recommender systems are vulnerable to the proposed Text Simulation attack (TextSimu) which aims to promote specific target items. As a novel type of text poisoning attack, TextSimu exploits large language models (LLM) to alter the textual information of target items by simulating the characteristics of popular items. It operates effectively in both black-box and white-box settings, utilizing two key components: a unified popularity extraction module, which captures the essential characteristics of popular items, and an N-persona consistency simulation strategy, which creates multiple personas to collaboratively synthesize refined promotional textual descriptions for target items by simulating the popular items. To withstand TextSimu-like attacks, we further explore the detection approach for identifying LLM-generated promotional text. Extensive experiments conducted on three datasets demonstrate that TextSimu poses a more significant threat than existing poisoning attacks, while our defense method can detect malicious text of target items generated by TextSimu. By identifying the vulnerability, we aim to advance the development of more robust ID-free recommender systems.
无 ID 推荐范式的提出是为了解决传统推荐系统难以对冷启动用户或具有新 ID 的项目进行建模的限制。尽管无 ID 推荐系统很有效，但本研究发现它很容易受到旨在推广特定目标项目的文本模拟攻击（TextSimu）的攻击。作为一种新型文本中毒攻击，TextSimu 利用大型语言模型（LLM），通过模拟流行项目的特征来改变目标项目的文本信息。它在黑盒和白盒环境下都能有效运行，利用了两个关键组件：一个是统一的流行度提取模块，它能捕捉流行项目的基本特征；另一个是 N 人一致性模拟策略，它能创建多个角色，通过模拟流行项目来协作合成目标项目的精炼促销文本描述。为了抵御类似 TextSimu 的攻击，我们进一步探索了识别 LLM 生成的促销文本的检测方法。在三个数据集上进行的广泛实验表明，TextSimu 比现有的中毒攻击构成了更大的威胁，而我们的防御方法可以检测到由 TextSimu 生成的目标项目的恶意文本。通过识别该漏洞，我们旨在推动更强大的无 ID 推荐系统的开发。
{"title":"LLM-Powered Text Simulation Attack Against ID-Free Recommender Systems","authors":"Zongwei Wang, Min Gao, Junliang Yu, Xinyi Gao, Quoc Viet Hung Nguyen, Shazia Sadiq, Hongzhi Yin","doi":"arxiv-2409.11690","DOIUrl":"https://doi.org/arxiv-2409.11690","url":null,"abstract":"The ID-free recommendation paradigm has been proposed to address the\u0000limitation that traditional recommender systems struggle to model cold-start\u0000users or items with new IDs. Despite its effectiveness, this study uncovers\u0000that ID-free recommender systems are vulnerable to the proposed Text Simulation\u0000attack (TextSimu) which aims to promote specific target items. As a novel type\u0000of text poisoning attack, TextSimu exploits large language models (LLM) to\u0000alter the textual information of target items by simulating the characteristics\u0000of popular items. It operates effectively in both black-box and white-box\u0000settings, utilizing two key components: a unified popularity extraction module,\u0000which captures the essential characteristics of popular items, and an N-persona\u0000consistency simulation strategy, which creates multiple personas to\u0000collaboratively synthesize refined promotional textual descriptions for target\u0000items by simulating the popular items. To withstand TextSimu-like attacks, we\u0000further explore the detection approach for identifying LLM-generated\u0000promotional text. Extensive experiments conducted on three datasets demonstrate\u0000that TextSimu poses a more significant threat than existing poisoning attacks,\u0000while our defense method can detect malicious text of target items generated by\u0000TextSimu. By identifying the vulnerability, we aim to advance the development\u0000of more robust ID-free recommender systems.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255161","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Morgan E. Prior, Thomas Howard III, Emily Light, Najib Ishaq, Noah M. Daniels
The Big Data explosion has necessitated the development of search algorithms that scale sub-linearly in time and memory. While compression algorithms and search algorithms do exist independently, few algorithms offer both, and those which do are domain-specific. We present panCAKES, a novel approach to compressive search, i.e., a way to perform $k$-NN and $rho$-NN search on compressed data while only decompressing a small, relevant, portion of the data. panCAKES assumes the manifold hypothesis and leverages the low-dimensional structure of the data to compress and search it efficiently. panCAKES is generic over any distance function for which the distance between two points is proportional to the memory cost of storing an encoding of one in terms of the other. This property holds for many widely-used distance functions, e.g. string edit distances (Levenshtein, Needleman-Wunsch, etc.) and set dissimilarity measures (Jaccard, Dice, etc.). We benchmark panCAKES on a variety of datasets, including genomic, proteomic, and set data. We compare compression ratios to gzip, and search performance between the compressed and uncompressed versions of the same dataset. panCAKES achieves compression ratios close to those of gzip, while offering sub-linear time performance for $k$-NN and $rho$-NN search. We conclude that panCAKES is an efficient, general-purpose algorithm for exact compressive search on large datasets that obey the manifold hypothesis. We provide an open-source implementation of panCAKES in the Rust programming language.
{"title":"Generalized compression and compressive search of large datasets","authors":"Morgan E. Prior, Thomas Howard III, Emily Light, Najib Ishaq, Noah M. Daniels","doi":"arxiv-2409.12161","DOIUrl":"https://doi.org/arxiv-2409.12161","url":null,"abstract":"The Big Data explosion has necessitated the development of search algorithms\u0000that scale sub-linearly in time and memory. While compression algorithms and search algorithms do exist independently,\u0000few algorithms offer both, and those which do are domain-specific. We present panCAKES, a novel approach to compressive search, i.e., a way to\u0000perform $k$-NN and $rho$-NN search on compressed data while only decompressing\u0000a small, relevant, portion of the data. panCAKES assumes the manifold hypothesis and leverages the low-dimensional\u0000structure of the data to compress and search it efficiently. panCAKES is generic over any distance function for which the distance between\u0000two points is proportional to the memory cost of storing an encoding of one in\u0000terms of the other. This property holds for many widely-used distance functions, e.g. string edit\u0000distances (Levenshtein, Needleman-Wunsch, etc.) and set dissimilarity measures\u0000(Jaccard, Dice, etc.). We benchmark panCAKES on a variety of datasets, including genomic, proteomic,\u0000and set data. We compare compression ratios to gzip, and search performance between the\u0000compressed and uncompressed versions of the same dataset. panCAKES achieves compression ratios close to those of gzip, while offering\u0000sub-linear time performance for $k$-NN and $rho$-NN search. We conclude that panCAKES is an efficient, general-purpose algorithm for\u0000exact compressive search on large datasets that obey the manifold hypothesis. We provide an open-source implementation of panCAKES in the Rust programming\u0000language.","PeriodicalId":501281,"journal":{"name":"arXiv - CS - Information Retrieval","volume":null,"pages":null},"PeriodicalIF":0.0,"publicationDate":"2024-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"142255194","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：