Time Series Forecasting by Recommendation: An Empirical Analysis on Amazon Marketplace

This study proposes a forecasting methodology for univari ate time series (TS) using a Recommender System (RS). The RS is built from a given TS as only input data and following an item-based Collabo rative Filtering approach. A set of top-N values is recommended for this TS which represent the forec...

Descripción completa

Detalles Bibliográficos
Autores: Gómez Losada, Álvaro, Duch-Brown, Néstor
Tipo de recurso: artículo
Fecha de publicación:2019
País:España
Institución:Universidad Loyola Andalucía
Repositorio:Brújula
OAI Identifier:oai:repositorio.uloyola.es:20.500.12412/5446
Acceso en línea:https://hdl.handle.net/20.500.12412/5446
Access Level:acceso abierto
Palabra clave:Collaborative Filtering
Time series
Forecasting
Data science
Descripción
Sumario:This study proposes a forecasting methodology for univari ate time series (TS) using a Recommender System (RS). The RS is built from a given TS as only input data and following an item-based Collabo rative Filtering approach. A set of top-N values is recommended for this TS which represent the forecasts. The idea is to emulate RS elements (the users, items and ratings triple) from the TS. Two TS obtained from Italy’s Amazon webpage were used to evaluate this methodology and very promising performance results were obtained, even the difficult environ ment chosen to conduct forecasting (short length and unevenly spaced TS). This performance is dependent on the similarity measure used and suffers from the same problems that other RSs (e.g., cold-start). However, this approach does not require high computational power to perform and its intuitive conception allows for being deployed with any programming language.