Behnam Rahdari, Peter Brusilovsky, Daqing He, Khushboo Thaker, Zhimeng Luo, Young ji Lee
Recommending online resources to patients with ovarian cancer and their caregivers is a challenging task. On one hand, the recommended items must be relevant, recent, and reliable. On the other hand, they need to match the user’s levels of disease-specific health literacy. In this demonstration, we describe the overall architecture and key components of HELPeR, a knowledge-adaptive interactive recommender system for ovarian cancer patients and their caregivers.
{"title":"HELPeR: An Interactive Recommender System for Ovarian Cancer Patients and Caregivers","authors":"Behnam Rahdari, Peter Brusilovsky, Daqing He, Khushboo Thaker, Zhimeng Luo, Young ji Lee","doi":"10.1145/3523227.3551471","DOIUrl":"https://doi.org/10.1145/3523227.3551471","url":null,"abstract":"Recommending online resources to patients with ovarian cancer and their caregivers is a challenging task. On one hand, the recommended items must be relevant, recent, and reliable. On the other hand, they need to match the user’s levels of disease-specific health literacy. In this demonstration, we describe the overall architecture and key components of HELPeR, a knowledge-adaptive interactive recommender system for ovarian cancer patients and their caregivers.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"133121191","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}
Huiyuan Chen, Xiaoting Li, Kaixiong Zhou, Xia Hu, Chin-Chia Michael Yeh, Yan Zheng, Hao Yang
There has been an explosion of interest in designing various Knowledge Graph Neural Networks (KGNNs), which achieve state-of-the-art performance and provide great explainability for recommendation. The promising performance is mainly resulting from their capability of capturing high-order proximity messages over the knowledge graphs. However, training KGNNs at scale is challenging due to the high memory usage. In the forward pass, the automatic differentiation engines (e.g., TensorFlow/PyTorch) generally need to cache all intermediate activation maps in order to compute gradients in the backward pass, which leads to a large GPU memory footprint. Existing work solves this problem by utilizing multi-GPU distributed frameworks. Nonetheless, this poses a practical challenge when seeking to deploy KGNNs in memory-constrained environments, especially for industry-scale graphs. Here we present TinyKG, a memory-efficient GPU-based training framework for KGNNs for the tasks of recommendation. Specifically, TinyKG uses exact activations in the forward pass while storing a quantized version of activations in the GPU buffers. During the backward pass, these low-precision activations are dequantized back to full-precision tensors, in order to compute gradients. To reduce the quantization errors, TinyKG applies a simple yet effective quantization algorithm to compress the activations, which ensures unbiasedness with low variance. As such, the training memory footprint of KGNNs is largely reduced with negligible accuracy loss. To evaluate the performance of our TinyKG, we conduct comprehensive experiments on real-world datasets. We found that our TinyKG with INT2 quantization aggressively reduces the memory footprint of activation maps with 7 ×, only with 2% loss in accuracy, allowing us to deploy KGNNs on memory-constrained devices.
{"title":"TinyKG: Memory-Efficient Training Framework for Knowledge Graph Neural Recommender Systems","authors":"Huiyuan Chen, Xiaoting Li, Kaixiong Zhou, Xia Hu, Chin-Chia Michael Yeh, Yan Zheng, Hao Yang","doi":"10.1145/3523227.3546760","DOIUrl":"https://doi.org/10.1145/3523227.3546760","url":null,"abstract":"There has been an explosion of interest in designing various Knowledge Graph Neural Networks (KGNNs), which achieve state-of-the-art performance and provide great explainability for recommendation. The promising performance is mainly resulting from their capability of capturing high-order proximity messages over the knowledge graphs. However, training KGNNs at scale is challenging due to the high memory usage. In the forward pass, the automatic differentiation engines (e.g., TensorFlow/PyTorch) generally need to cache all intermediate activation maps in order to compute gradients in the backward pass, which leads to a large GPU memory footprint. Existing work solves this problem by utilizing multi-GPU distributed frameworks. Nonetheless, this poses a practical challenge when seeking to deploy KGNNs in memory-constrained environments, especially for industry-scale graphs. Here we present TinyKG, a memory-efficient GPU-based training framework for KGNNs for the tasks of recommendation. Specifically, TinyKG uses exact activations in the forward pass while storing a quantized version of activations in the GPU buffers. During the backward pass, these low-precision activations are dequantized back to full-precision tensors, in order to compute gradients. To reduce the quantization errors, TinyKG applies a simple yet effective quantization algorithm to compress the activations, which ensures unbiasedness with low variance. As such, the training memory footprint of KGNNs is largely reduced with negligible accuracy loss. To evaluate the performance of our TinyKG, we conduct comprehensive experiments on real-world datasets. We found that our TinyKG with INT2 quantization aggressively reduces the memory footprint of activation maps with 7 ×, only with 2% loss in accuracy, allowing us to deploy KGNNs on memory-constrained devices.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"14 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"130327199","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}
Ronay Ak, Benedikt D. Schifferer, Sara Rabhi, G. Moreira
Industrial recommender systems are made up of complex pipelines requiring multiple steps including feature engineering and preprocessing, a retrieval model for candidate generation, filtering, a feature store query, a ranking model for scoring, and an ordering stage. These pipelines need to be carefully deployed as a set, requiring coordination during their development and deployment. Data scientists, ML engineers, and researchers might focus on different stages of recommender systems, however they share a common desire to reduce the time and effort searching for and combining boilerplate code coming from different sources or writing custom code from scratch to create their own RecSys pipelines. This tutorial introduces the Merlin framework which aims to make the development and deployment of recommender systems easier, providing methods for evaluating existing approaches, developing new ideas and deploying them to production. There are many techniques, such as different model architectures (e.g. MF, DLRM, DCN, etc), negative sampling strategies, loss functions or prediction tasks (binary, multi-class, multi-task) that are commonly used in these pipelines. Merlin provides building blocks that allow RecSys practitioners to focus on the “what” question in designing their model pipeline instead of “how”. Supporting research into new ideas within the RecSys spaces is equally important and Merlin supports the addition of custom components and the extension of existing ones to address gaps. In this tutorial, participants will learn: (i) how to easily implement common recommender system techniques for comparison, (ii) how to modify components to evaluate new ideas, and (iii) deploying recommender systems, bringing new ideas to production- using an open source framework Merlin and its libraries.
{"title":"Training and Deploying Multi-Stage Recommender Systems","authors":"Ronay Ak, Benedikt D. Schifferer, Sara Rabhi, G. Moreira","doi":"10.1145/3523227.3547372","DOIUrl":"https://doi.org/10.1145/3523227.3547372","url":null,"abstract":"Industrial recommender systems are made up of complex pipelines requiring multiple steps including feature engineering and preprocessing, a retrieval model for candidate generation, filtering, a feature store query, a ranking model for scoring, and an ordering stage. These pipelines need to be carefully deployed as a set, requiring coordination during their development and deployment. Data scientists, ML engineers, and researchers might focus on different stages of recommender systems, however they share a common desire to reduce the time and effort searching for and combining boilerplate code coming from different sources or writing custom code from scratch to create their own RecSys pipelines. This tutorial introduces the Merlin framework which aims to make the development and deployment of recommender systems easier, providing methods for evaluating existing approaches, developing new ideas and deploying them to production. There are many techniques, such as different model architectures (e.g. MF, DLRM, DCN, etc), negative sampling strategies, loss functions or prediction tasks (binary, multi-class, multi-task) that are commonly used in these pipelines. Merlin provides building blocks that allow RecSys practitioners to focus on the “what” question in designing their model pipeline instead of “how”. Supporting research into new ideas within the RecSys spaces is equally important and Merlin supports the addition of custom components and the extension of existing ones to address gaps. In this tutorial, participants will learn: (i) how to easily implement common recommender system techniques for comparison, (ii) how to modify components to evaluate new ideas, and (iii) deploying recommender systems, bringing new ideas to production- using an open source framework Merlin and its libraries.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"36 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124962645","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}
Information access systems such as search engines and recommender systems often display results in a sorted ranked list based on their relevance. Fairness of these ranked list has received attention as an important evaluation criteria along with traditional metrics such as utility or accuracy. Fairness broadly involves both provider and consumer side fairness at both group and individual levels. Several fair ranking metrics have been proposed to measure group fairness for providers based on various “sensitive attributes”. These metrics differ in their fairness goal, assumptions, and implementations. Although there are several fair ranking metrics to measure group fairness, multiple open challenges still exist in this area to consider. In my thesis, I work on the area of fair ranking metrics for provider-side group fairness. I am interested in understanding the fairness concepts and practical applications of these metrics to identify their strength and limitations to aid the researchers and practitioner by pointing out the gaps. Moreover, I will contribute to this research area by focusing on some of the limitations like considering different browsing models and bias in relevance information.
{"title":"Fair Ranking Metrics","authors":"Amifa Raj","doi":"10.1145/3523227.3547430","DOIUrl":"https://doi.org/10.1145/3523227.3547430","url":null,"abstract":"Information access systems such as search engines and recommender systems often display results in a sorted ranked list based on their relevance. Fairness of these ranked list has received attention as an important evaluation criteria along with traditional metrics such as utility or accuracy. Fairness broadly involves both provider and consumer side fairness at both group and individual levels. Several fair ranking metrics have been proposed to measure group fairness for providers based on various “sensitive attributes”. These metrics differ in their fairness goal, assumptions, and implementations. Although there are several fair ranking metrics to measure group fairness, multiple open challenges still exist in this area to consider. In my thesis, I work on the area of fair ranking metrics for provider-side group fairness. I am interested in understanding the fairness concepts and practical applications of these metrics to identify their strength and limitations to aid the researchers and practitioner by pointing out the gaps. Moreover, I will contribute to this research area by focusing on some of the limitations like considering different browsing models and bias in relevance information.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"31 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"114646530","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}
Sequential recommendation has attracted a lot of attention from both academia and industry. Since item embeddings directly affect the recommendation results, their learning process is very important. However, most existing sequential models may introduce bias when updating the item embeddings. For example, in a sequence where all items are endorsed by a same celebrity, the co-occurrence of two items only indicates their similarity in terms of endorser, and is independent of the other aspects such as category and color. The existing models often update the entire item as a whole or update different aspects of the item without distinction, which fails to capture the contributions of different aspects to the co-occurrence pattern. To overcome the above limitations, we propose aspect re-distribution (ARD) to focus on updating the aspects that are important for co-occurrence. Specifically, we represent an item using several aspect embeddings with the same initial importance. We then re-calculate the importance of each aspect according to the other items in the sequence. Finally, we aggregate these aspect embeddings into a single aspect-aware embedding according to their importance. The aspect-aware embedding can be provided as input to a successor sequential model. Updates of the aspect-aware embedding are passed back to the aspect embeddings based on their importance. Therefore, different from the existing models, our method pays more attention to updating the important aspects. In our experiments, we choose self-attention networks as the successor model. The experimental results on four real-world datasets indicate that our method achieves very promising performance in comparison with seven state-of-the-art models.
{"title":"Aspect Re-distribution for Learning Better Item Embeddings in Sequential Recommendation","authors":"Wei Cai, Weike Pan, Jingwen Mao, Zhechao Yu, Congfu Xu","doi":"10.1145/3523227.3546764","DOIUrl":"https://doi.org/10.1145/3523227.3546764","url":null,"abstract":"Sequential recommendation has attracted a lot of attention from both academia and industry. Since item embeddings directly affect the recommendation results, their learning process is very important. However, most existing sequential models may introduce bias when updating the item embeddings. For example, in a sequence where all items are endorsed by a same celebrity, the co-occurrence of two items only indicates their similarity in terms of endorser, and is independent of the other aspects such as category and color. The existing models often update the entire item as a whole or update different aspects of the item without distinction, which fails to capture the contributions of different aspects to the co-occurrence pattern. To overcome the above limitations, we propose aspect re-distribution (ARD) to focus on updating the aspects that are important for co-occurrence. Specifically, we represent an item using several aspect embeddings with the same initial importance. We then re-calculate the importance of each aspect according to the other items in the sequence. Finally, we aggregate these aspect embeddings into a single aspect-aware embedding according to their importance. The aspect-aware embedding can be provided as input to a successor sequential model. Updates of the aspect-aware embedding are passed back to the aspect embeddings based on their importance. Therefore, different from the existing models, our method pays more attention to updating the important aspects. In our experiments, we choose self-attention networks as the successor model. The experimental results on four real-world datasets indicate that our method achieves very promising performance in comparison with seven state-of-the-art models.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"129 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124234163","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}
Recommender systems are central to online information consumption and user-decision processes, as they help users find relevant information and establish new social relationships. However, recommenders could also (unintendedly) help propagate misinformation and increase the social influence of the spreading it. In this context, we study the impact of friend recommender systems on the social influence of misinformation spreaders on Twitter. To this end, we applied several user recommenders to a COVID-19 misinformation data collection. Then, we explore what-if scenarios to simulate changes in user misinformation spreading behaviour as an effect of the interactions in the recommended network. Our study shows that recommenders can indeed affect how misinformation spreaders interact with other users and influence them.
{"title":"Do Recommender Systems Make Social Media More Susceptible to Misinformation Spreaders?","authors":"Antonela Tommasel, F. Menczer","doi":"10.1145/3523227.3551473","DOIUrl":"https://doi.org/10.1145/3523227.3551473","url":null,"abstract":"Recommender systems are central to online information consumption and user-decision processes, as they help users find relevant information and establish new social relationships. However, recommenders could also (unintendedly) help propagate misinformation and increase the social influence of the spreading it. In this context, we study the impact of friend recommender systems on the social influence of misinformation spreaders on Twitter. To this end, we applied several user recommenders to a COVID-19 misinformation data collection. Then, we explore what-if scenarios to simulate changes in user misinformation spreading behaviour as an effect of the interactions in the recommended network. Our study shows that recommenders can indeed affect how misinformation spreaders interact with other users and influence them.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"18 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124945252","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}
Though recommender systems fundamentally rely on human input and feedback, human-centered research in the RecSys discipline is lacking. When recommender systems aim to treat users more fairly, misinterpreting user objectives could lead to unintentional harm, whether or not fairness is part of the aim. When users seek to understand recommender systems better, a lack of transparency could act as an obstacle for their trust and adoption of the platform. Human-centered machine learning seeks to design systems that understand their users, while simultaneously designing systems that the users can understand. In this work, I propose to explore the intersection of transparency and user-system understanding through three phases of research that will result in a Human-Centered Framework for Transparency in Fairness-Aware Recommender Systems.
{"title":"Developing a Human-Centered Framework for Transparency in Fairness-Aware Recommender Systems","authors":"Jessie J. Smith","doi":"10.1145/3523227.3547428","DOIUrl":"https://doi.org/10.1145/3523227.3547428","url":null,"abstract":"Though recommender systems fundamentally rely on human input and feedback, human-centered research in the RecSys discipline is lacking. When recommender systems aim to treat users more fairly, misinterpreting user objectives could lead to unintentional harm, whether or not fairness is part of the aim. When users seek to understand recommender systems better, a lack of transparency could act as an obstacle for their trust and adoption of the platform. Human-centered machine learning seeks to design systems that understand their users, while simultaneously designing systems that the users can understand. In this work, I propose to explore the intersection of transparency and user-system understanding through three phases of research that will result in a Human-Centered Framework for Transparency in Fairness-Aware Recommender Systems.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"129521210","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}
While personalised recommendations are successful in domains like retail, where large volumes of user feedback on items are available, the generation of automatic recommendations in data-sparse domains, like insurance purchasing, is an open problem. The insurance domain is notoriously data-sparse because the number of products is typically low (compared to retail) and they are usually purchased to last for a long time. Also, many users still prefer the telephone over the web for purchasing products, reducing the amount of web-logged user interactions. To address this, we present a recurrent neural network recommendation model that uses past user sessions as signals for learning recommendations. Learning from past user sessions allows dealing with the data scarcity of the insurance domain. Specifically, our model learns from several types of user actions that are not always associated with items, and unlike all prior session-based recommendation models, it models relationships between input sessions and a target action (purchasing insurance) that does not take place within the input sessions. Evaluation on a real-world dataset from the insurance domain (ca. 44K users, 16 items, 54K purchases, and 117K sessions) against several state-of-the-art baselines shows that our model outperforms the baselines notably. Ablation analysis shows that this is mainly due to the learning of dependencies across sessions in our model. We contribute the first ever session-based model for insurance recommendation, and make available our dataset to the research community.
{"title":"Learning Recommendations from User Actions in the Item-poor Insurance Domain","authors":"Simone Borg Bruun, Maria Maistro, C. Lioma","doi":"10.1145/3523227.3546775","DOIUrl":"https://doi.org/10.1145/3523227.3546775","url":null,"abstract":"While personalised recommendations are successful in domains like retail, where large volumes of user feedback on items are available, the generation of automatic recommendations in data-sparse domains, like insurance purchasing, is an open problem. The insurance domain is notoriously data-sparse because the number of products is typically low (compared to retail) and they are usually purchased to last for a long time. Also, many users still prefer the telephone over the web for purchasing products, reducing the amount of web-logged user interactions. To address this, we present a recurrent neural network recommendation model that uses past user sessions as signals for learning recommendations. Learning from past user sessions allows dealing with the data scarcity of the insurance domain. Specifically, our model learns from several types of user actions that are not always associated with items, and unlike all prior session-based recommendation models, it models relationships between input sessions and a target action (purchasing insurance) that does not take place within the input sessions. Evaluation on a real-world dataset from the insurance domain (ca. 44K users, 16 items, 54K purchases, and 117K sessions) against several state-of-the-art baselines shows that our model outperforms the baselines notably. Ablation analysis shows that this is mainly due to the learning of dependencies across sessions in our model. We contribute the first ever session-based model for insurance recommendation, and make available our dataset to the research community.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"2004 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"116847302","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}
In this talk I will introduce the principles of data feminism and give a first-person report from a large participatory-action-research-design project where we are co-designing technology with data activists. The “we” in question is myself, Silvana Fumega, and Helena Suárez Val, and we work in collaboration and solidarity with activist groups producing data to challenge feminicide – fatal gender-related violence against women – across the Americas. As all practitioners know, practice is messy and rarely adheres cleanly to pleasing principles. Throughout the talk, I will highlight resonances and tensions between our design process and the principles of data feminism, showing how we tried to operationalize these principles in interactive digital tools and machine learning models. I hope to surface my aspirations for more participatory technology design processes as well as raise lingering questions to the community so that we may think together about the limitations of co-designing for justice.
{"title":"Co-designing ML Models with Data Activists","authors":"C. D’Ignazio","doi":"10.1145/3523227.3556646","DOIUrl":"https://doi.org/10.1145/3523227.3556646","url":null,"abstract":"In this talk I will introduce the principles of data feminism and give a first-person report from a large participatory-action-research-design project where we are co-designing technology with data activists. The “we” in question is myself, Silvana Fumega, and Helena Suárez Val, and we work in collaboration and solidarity with activist groups producing data to challenge feminicide – fatal gender-related violence against women – across the Americas. As all practitioners know, practice is messy and rarely adheres cleanly to pleasing principles. Throughout the talk, I will highlight resonances and tensions between our design process and the principles of data feminism, showing how we tried to operationalize these principles in interactive digital tools and machine learning models. I hope to surface my aspirations for more participatory technology design processes as well as raise lingering questions to the community so that we may think together about the limitations of co-designing for justice.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"141 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"117064475","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}
Shuchang Liu, Yingqiang Ge, Shuyuan Xu, Yongfeng Zhang, A. Marian
Achieving fairness over different user groups in recommender systems is an important problem. The majority of existing works achieve fairness through constrained optimization that combines the recommendation loss and the fairness constraint. To achieve fairness, the algorithm usually needs to know each user’s group affiliation feature such as gender or race. However, such involved user group feature is usually sensitive and requires protection. In this work, we seek a federated learning solution for the fair recommendation problem and identify the main challenge as an algorithmic conflict between the global fairness objective and the localized federated optimization process. On one hand, the fairness objective usually requires access to all users’ group information. On the other hand, the federated learning systems restrain the personal data in each user’s local space. As a resolution, we propose to communicate group statistics during federated optimization and use differential privacy techniques to avoid exposure of users’ group information when users require privacy protection. We illustrate the theoretical bounds of the noisy signal used in our method that aims to enforce privacy without overwhelming the aggregated statistics. Empirical results show that federated learning may naturally improve user group fairness and the proposed framework can effectively control this fairness with low communication overheads.
{"title":"Fairness-aware Federated Matrix Factorization","authors":"Shuchang Liu, Yingqiang Ge, Shuyuan Xu, Yongfeng Zhang, A. Marian","doi":"10.1145/3523227.3546771","DOIUrl":"https://doi.org/10.1145/3523227.3546771","url":null,"abstract":"Achieving fairness over different user groups in recommender systems is an important problem. The majority of existing works achieve fairness through constrained optimization that combines the recommendation loss and the fairness constraint. To achieve fairness, the algorithm usually needs to know each user’s group affiliation feature such as gender or race. However, such involved user group feature is usually sensitive and requires protection. In this work, we seek a federated learning solution for the fair recommendation problem and identify the main challenge as an algorithmic conflict between the global fairness objective and the localized federated optimization process. On one hand, the fairness objective usually requires access to all users’ group information. On the other hand, the federated learning systems restrain the personal data in each user’s local space. As a resolution, we propose to communicate group statistics during federated optimization and use differential privacy techniques to avoid exposure of users’ group information when users require privacy protection. We illustrate the theoretical bounds of the noisy signal used in our method that aims to enforce privacy without overwhelming the aggregated statistics. Empirical results show that federated learning may naturally improve user group fairness and the proposed framework can effectively control this fairness with low communication overheads.","PeriodicalId":443279,"journal":{"name":"Proceedings of the 16th ACM Conference on Recommender Systems","volume":"26 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2022-09-18","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"127720808","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：