{"about":{"site":"https://codewithpapers.app","non_affiliation":"Code with Papers and Syntology are not affiliated with, endorsed by, or sponsored by Papers with Code, Meta, or the pwc-archive mirror.","licence":"CC BY-SA 4.0","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","attribution":"https://codewithpapers.app/attribution","modified":"archive material modified by Syntology; see the attribution page"},"url":"/paper/multivariate-arrival-times-with-recurrent","title":"Multivariate Arrival Times with Recurrent Neural Networks for Personalized Demand Forecasting","arxiv_id":"1812.11444","date":"2018-12-29","proceeding":null,"authors":["Tianle Chen","Brian Keng","Javier Moreno"],"abstract":"Access to a large variety of data across a massive population has made it\npossible to predict customer purchase patterns and responses to marketing\ncampaigns. In particular, accurate demand forecasts for popular products with\nfrequent repeat purchases are essential since these products are one of the\nmain drivers of profits. However, buyer purchase patterns are extremely diverse\nand sparse on a per-product level due to population heterogeneity as well as\ndependence in purchase patterns across product categories. Traditional methods\nin survival analysis have proven effective in dealing with censored data by\nassuming parametric distributions on inter-arrival times. Distributional\nparameters are then fitted, typically in a regression framework. On the other\nhand, neural-network based models take a non-parametric approach to learn\nrelations from a larger functional class. However, the lack of distributional\nassumptions make it difficult to model partially observed data. In this paper,\nwe model directly the inter-arrival times as well as the partially observed\ninformation at each time step in a survival-based approach using Recurrent\nNeural Networks (RNN) to model purchase times jointly over several products.\nInstead of predicting a point estimate for inter-arrival times, the RNN outputs\nparameters that define a distributional estimate. The loss function is the\nnegative log-likelihood of these parameters given partially observed data. This\napproach allows one to leverage both fully observed data as well as partial\ninformation. By externalizing the censoring problem through a log-likelihood\nloss function, we show that substantial improvements over state-of-the-art\nmachine learning methods can be achieved. We present experimental results based\non two open datasets as well as a study on a real dataset from a large\nretailer.","url_abs":"http://arxiv.org/abs/1812.11444v1","url_pdf":"http://arxiv.org/pdf/1812.11444v1.pdf","source":{"archive":"pwc-archive (Hugging Face), CC BY-SA 4.0","snapshot":"2025-07-28","licence_url":"https://creativecommons.org/licenses/by-sa/4.0/legalcode","row_kind":"abstracts"},"code_links":[{"paper_slug":"multivariate-arrival-times-with-recurrent","repo_url":"https://github.com/rubikloud/matrnn","is_official":1,"mentioned_in_paper":1,"mentioned_in_github":0,"framework":"tf","reach":null}],"tasks":[{"task_slug":"demand-forecasting","task_name":"Demand Forecasting"},{"task_slug":"marketing","task_name":"Marketing"},{"task_slug":"survival-analysis","task_name":"Survival Analysis"}],"methods":[],"datasets_introduced":[],"methods_introduced":[],"results":[],"syntology":{"atlas_url":null,"mcp":null,"developers":"https://syntology.ai/developers"},"arxiv_metadata":null,"syntology_extracted_results":null}