M. Volkovs, Anson Wong, Zhaoyue Cheng, Felipe Pérez, I. Stanevich, Y. Lu
Most online activity happens in the context of a session; to enable better user experience many online platforms aim to dynamically refine their recommendations as sessions progress. A popular approach is to continuously re-rank recommendations based on current session activity and past session logs. This motivates the 2019 ACM RecSys Challenge organised by Trivago. Using the session log dataset released by Trivago, the challenge aims to benchmark models for in-session re-ranking of hotel recommendations. In this paper we present our approach to this challenge where we first contextualize sessions in a global and local manner, and then train gradient boosting and deep learning models for re-ranking. Our team achieved 2nd place out of over 570 teams, with less than 0.3% relative difference in Mean Reciprocal Rank from the 1st place team. Code for our approach can be found here: https://github.com/layer6ai-labs/RecSys2019
{"title":"Robust contextual models for in-session personalization","authors":"M. Volkovs, Anson Wong, Zhaoyue Cheng, Felipe Pérez, I. Stanevich, Y. Lu","doi":"10.1145/3359555.3359558","DOIUrl":"https://doi.org/10.1145/3359555.3359558","url":null,"abstract":"Most online activity happens in the context of a session; to enable better user experience many online platforms aim to dynamically refine their recommendations as sessions progress. A popular approach is to continuously re-rank recommendations based on current session activity and past session logs. This motivates the 2019 ACM RecSys Challenge organised by Trivago. Using the session log dataset released by Trivago, the challenge aims to benchmark models for in-session re-ranking of hotel recommendations. In this paper we present our approach to this challenge where we first contextualize sessions in a global and local manner, and then train gradient boosting and deep learning models for re-ranking. Our team achieved 2nd place out of over 570 teams, with less than 0.3% relative difference in Mean Reciprocal Rank from the 1st place team. Code for our approach can be found here: https://github.com/layer6ai-labs/RecSys2019","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"53 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"115055475","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}
Edoardo D'Amico, Giovanni Gabbolini, Daniele Montesi, Matteo Moreschini, Federico Parroni, F. Piccinini, Alberto Rossettini, Alessio Russo Introito, Cesare Bernardis, Maurizio Ferrari Dacrema
In this paper we provide an overview of the approach we used as team PoliCloud8 for the ACM RecSys Challenge 2019. The competition, organized by Trivago, focuses on the problem of session-based and context-aware accommodation recommendation in a travel domain. The goal is to suggest suitable accommodations fitting the needs of the traveller to maximise the chance of a redirect (click-out) to a booking site, relying on explicit and implicit user signals within a session (clicks, search refinement, filter usage) to detect the users intent. Our team proposes a solution based on several new features, designed to capture specific types of information as well as some well-known models: gradient boosting, neural networks and a stacking-based ensemble.
{"title":"Leveraging laziness, browsing-pattern aware stacked models for sequential accommodation learning to rank","authors":"Edoardo D'Amico, Giovanni Gabbolini, Daniele Montesi, Matteo Moreschini, Federico Parroni, F. Piccinini, Alberto Rossettini, Alessio Russo Introito, Cesare Bernardis, Maurizio Ferrari Dacrema","doi":"10.1145/3359555.3359563","DOIUrl":"https://doi.org/10.1145/3359555.3359563","url":null,"abstract":"In this paper we provide an overview of the approach we used as team PoliCloud8 for the ACM RecSys Challenge 2019. The competition, organized by Trivago, focuses on the problem of session-based and context-aware accommodation recommendation in a travel domain. The goal is to suggest suitable accommodations fitting the needs of the traveller to maximise the chance of a redirect (click-out) to a booking site, relying on explicit and implicit user signals within a session (clicks, search refinement, filter usage) to detect the users intent. Our team proposes a solution based on several new features, designed to capture specific types of information as well as some well-known models: gradient boosting, neural networks and a stacking-based ensemble.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"131169672","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}
Being able to provide high quality search and recommendation services can be a decisive success factor for online applications, e.g., in today's competitive e-commerce environments. Context-adaptive and personalized item suggestions can help to both improve the user experience and the provider's short-term and long-term revenue. However, automating this form of adaptation can be challenging, when no long-term preference profiles are available. In these situations, the user's preferences and short-term intent must be derived from the last few observed interactions. In this work, we present a hybrid approach to rank hotels based on the user's most recent interactions and meta data about the available items. The developed recommendation approach can be used both for personalized search and session-based recommendation. Technically, we employed a combination of a gradient-boosted learning-to-rank model, Bayesian Personalized Ranking and an embedding model using Doc2Vec. The approach was successfully evaluated in the context of the ACM RecSys 2019 challenge, where it led our team letoh govatri to the fifth place on the leaderboard, with a ranking accuracy only 0.53% below the winning approach.
{"title":"Learning to rank hotels for search and recommendation from session-based interaction logs and meta data","authors":"Malte Ludewig, D. Jannach","doi":"10.1145/3359555.3359561","DOIUrl":"https://doi.org/10.1145/3359555.3359561","url":null,"abstract":"Being able to provide high quality search and recommendation services can be a decisive success factor for online applications, e.g., in today's competitive e-commerce environments. Context-adaptive and personalized item suggestions can help to both improve the user experience and the provider's short-term and long-term revenue. However, automating this form of adaptation can be challenging, when no long-term preference profiles are available. In these situations, the user's preferences and short-term intent must be derived from the last few observed interactions. In this work, we present a hybrid approach to rank hotels based on the user's most recent interactions and meta data about the available items. The developed recommendation approach can be used both for personalized search and session-based recommendation. Technically, we employed a combination of a gradient-boosted learning-to-rank model, Bayesian Personalized Ranking and an embedding model using Doc2Vec. The approach was successfully evaluated in the context of the ACM RecSys 2019 challenge, where it led our team letoh govatri to the fifth place on the leaderboard, with a ranking accuracy only 0.53% below the winning approach.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"54 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"125104825","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}
Jaehoon Oh, Sangmook Kim, Seyoung Yun, Seungwoo Choi, M. Yi
In a session-based recommendation service, currently offered by many online companies including trivago, it is important to effectively incorporate user interactions into recommendations. However, a major challenge lies in the fact that both inter-session and intra-session contexts should be considered at the same time for recommendations to become effective. To address this issue, we propose a pipelined hybrid recommender system that considers the two contexts simultaneously via weighted summation of loss functions designed for the combination of a recurrent neural network (RNN) and a convolutional neural network (CNN). With the hybrid system, our team, OSI LAB, achieved the final score of 0.670167 and reached the 16th place in the RecSys Challenge 2019. Our source code is available from https://github.com/jhoon-oh/recsys2019challenge.
{"title":"A pipelined hybrid recommender system for ranking the items on the display","authors":"Jaehoon Oh, Sangmook Kim, Seyoung Yun, Seungwoo Choi, M. Yi","doi":"10.1145/3359555.3359565","DOIUrl":"https://doi.org/10.1145/3359555.3359565","url":null,"abstract":"In a session-based recommendation service, currently offered by many online companies including trivago, it is important to effectively incorporate user interactions into recommendations. However, a major challenge lies in the fact that both inter-session and intra-session contexts should be considered at the same time for recommendations to become effective. To address this issue, we propose a pipelined hybrid recommender system that considers the two contexts simultaneously via weighted summation of loss functions designed for the combination of a recurrent neural network (RNN) and a convolutional neural network (CNN). With the hybrid system, our team, OSI LAB, achieved the final score of 0.670167 and reached the 16th place in the RecSys Challenge 2019. Our source code is available from https://github.com/jhoon-oh/recsys2019challenge.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"63 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"126311746","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}
Sara Rabhi, Wenbo Sun, Julio Perez, M. R. B. Kristensen, Jiwei Liu, Even Oldridge
In this paper we present the novel aspects of our 15th place solution to the RecSys Challenge 2019 which are focused on the acceleration of feature generation and model training time. In our final solution we sped up training of our model by a factor of 15.6x, from a workflow of 891.8s (14m52s) to 57.2s, through a combination of the RAPIDS.AI cuDF library for preprocessing, a custom batch dataloader, LAMB and extreme batch sizes, and an update to the kernel responsible for calculating the embedding gradient in PyTorch. Using cuDF we also accelerated our feature generation by a factor of 9.7x by performing the computations on the GPU, reducing the time taken to generate the features used in our model from 51 minutes to 5. We demonstrate these optimizations on the fastai tabular model which we relied on extensively in our final ensemble. With training time so drastically reduced the iteration involved in generating new features and training new models is much more fluid, allowing for the rapid prototyping of deep learning based recommender systems in hours as opposed to days.
{"title":"Accelerating recommender system training 15x with RAPIDS","authors":"Sara Rabhi, Wenbo Sun, Julio Perez, M. R. B. Kristensen, Jiwei Liu, Even Oldridge","doi":"10.1145/3359555.3359564","DOIUrl":"https://doi.org/10.1145/3359555.3359564","url":null,"abstract":"In this paper we present the novel aspects of our 15th place solution to the RecSys Challenge 2019 which are focused on the acceleration of feature generation and model training time. In our final solution we sped up training of our model by a factor of 15.6x, from a workflow of 891.8s (14m52s) to 57.2s, through a combination of the RAPIDS.AI cuDF library for preprocessing, a custom batch dataloader, LAMB and extreme batch sizes, and an update to the kernel responsible for calculating the embedding gradient in PyTorch. Using cuDF we also accelerated our feature generation by a factor of 9.7x by performing the computations on the GPU, reducing the time taken to generate the features used in our model from 51 minutes to 5. We demonstrate these optimizations on the fastai tabular model which we relied on extensively in our final ensemble. With training time so drastically reduced the iteration involved in generating new features and training new models is much more fluid, allowing for the rapid prototyping of deep learning based recommender systems in hours as opposed to days.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"1 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"128923503","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}
Paweł Jankiewicz, Liudmyla Kyrashchuk, Pawel Sienkowski, Magdalena Wójcik
To keep up with a highly competitive the online hotel booking sector, it is necessary to develop fast and robust recommender systems. The 2019 RecSys Challenge is focused on ways we may use session-based and context-aware signals from users to improve the quality of hotel booking recommendations. In this paper, we present our approach to the challenge. We focus on the proper problem representation, feature extraction, and model blending. Our team achieved the 1st place out of 500 teams in the challenge, with the final MRR score of 0.685711.
{"title":"Boosting algorithms for a session-based, context-aware recommender system in an online travel domain","authors":"Paweł Jankiewicz, Liudmyla Kyrashchuk, Pawel Sienkowski, Magdalena Wójcik","doi":"10.1145/3359555.3359557","DOIUrl":"https://doi.org/10.1145/3359555.3359557","url":null,"abstract":"To keep up with a highly competitive the online hotel booking sector, it is necessary to develop fast and robust recommender systems. The 2019 RecSys Challenge is focused on ways we may use session-based and context-aware signals from users to improve the quality of hotel booking recommendations. In this paper, we present our approach to the challenge. We focus on the proper problem representation, feature extraction, and model blending. Our team achieved the 1st place out of 500 teams in the challenge, with the final MRR score of 0.685711.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"20 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121742083","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 RecSys Challenge 2019 seeks a better solution for item recommendation on short session-based data with limited user history. This paper describes the team PVZ's approach to this challenge, which won the 3rd place in the contest. Our solution consists of the following components. Firstly, we cast the hotel recommendation task as a binary classification problem. Secondly, we spend most of the time doing feature engineering and mining a series of useful features in various aspects. Then we train individual models with a different set of features and blend them with some important features using stacking method. At last, we create other new pair-wise features based on the existing model predictions and train a stacking model again which generates our final result. Our final solution achieved a public score of 0.685929 and a private score of 0.684071, ranking the third place on both sides.
{"title":"Session-based item recommendation with pairwise features","authors":"Zhe Wang, Yangbo Gao, Huan Chen, Peng Yan","doi":"10.1145/3359555.3359559","DOIUrl":"https://doi.org/10.1145/3359555.3359559","url":null,"abstract":"The RecSys Challenge 2019 seeks a better solution for item recommendation on short session-based data with limited user history. This paper describes the team PVZ's approach to this challenge, which won the 3rd place in the contest. Our solution consists of the following components. Firstly, we cast the hotel recommendation task as a binary classification problem. Secondly, we spend most of the time doing feature engineering and mining a series of useful features in various aspects. Then we train individual models with a different set of features and blend them with some important features using stacking method. At last, we create other new pair-wise features based on the existing model predictions and train a stacking model again which generates our final result. Our final solution achieved a public score of 0.685929 and a private score of 0.684071, ranking the third place on both sides.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"64 4 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"121790916","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 the RecSys Challenge 2019 the participants were asked to predict which items, from a presented list of items/accommodations of a search result on trivago, had been clicked-on during the last part of a user's session. Here we present the 7th place solution1. It consists of a neural network designed to learn interactions between session, context, sequence features, and the features of the displayed items at the time of a click. Our approach uses well established deep learning techniques, such as Recurrent Neural Networks, Attention and self-Attention mechanisms to deal with the different aspects of the information available, and it predicts a (categorical) probability distribution over the list of presented items. In addition to the model structure we also describe the somewhat heavy feature engineering, data augmentation and other decisions/observations made a long the way.
{"title":"An attentive RNN model for session-based and context-aware recommendations: a solution to the RecSys challenge 2019","authors":"Ricardo Gama, Hugo L. Fernandes","doi":"10.1145/3359555.3359757","DOIUrl":"https://doi.org/10.1145/3359555.3359757","url":null,"abstract":"In the RecSys Challenge 2019 the participants were asked to predict which items, from a presented list of items/accommodations of a search result on trivago, had been clicked-on during the last part of a user's session. Here we present the 7th place solution1. It consists of a neural network designed to learn interactions between session, context, sequence features, and the features of the displayed items at the time of a click. Our approach uses well established deep learning techniques, such as Recurrent Neural Networks, Attention and self-Attention mechanisms to deal with the different aspects of the information available, and it predicts a (categorical) probability distribution over the list of presented items. In addition to the model structure we also describe the somewhat heavy feature engineering, data augmentation and other decisions/observations made a long the way.","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"331 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"2019-09-20","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"124466365","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}
{"title":"Proceedings of the Workshop on ACM Recommender Systems Challenge","authors":"","doi":"10.1145/3359555","DOIUrl":"https://doi.org/10.1145/3359555","url":null,"abstract":"","PeriodicalId":255213,"journal":{"name":"Proceedings of the Workshop on ACM Recommender Systems Challenge","volume":"35 1","pages":"0"},"PeriodicalIF":0.0,"publicationDate":"1900-01-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"120944605","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：