In the rental market, buyers(tenants) and sellers (landlords) express their preferences while making a deal. For example, a Landlord wants a well-behaved and reliable Tenant, and the Tenant wants an apartment that suits its tastes and budget. This scenario sets us apart from a regular market, where buyers have all the freedom to choose and sellers just want to sell their products. We will try to exploit these differences with an ensembling technique for Learning-To-Rank tasks, learning multiple concurrent preferences in a single rank. This idea was tested and developed jointly with HousingAnywhere, a world-wide online rental marketplace for mid to long-term accommodations, ranking Landlords and Tenants to build a Recommender System improving the user experience of their online marketplace. We found a good solution through performance-driven research, showing significant improvements over a single tree-based model over different market scenarios.
Nel mercato degli affitti, acquirenti (inquilini) e venditori (proprietari) esprimono le loro preferenze mentre concludono un accordo. Ad esempio, un locatore desidererebbe un inquilino ben educato e affidabile, invece un inquilino vorrebbe un appartamento che si adatti ai suoi gusti e al suo budget. Questo scenario ci distanzia da un mercato tradizionale, dove gli acquirenti hanno libertà di scelta e i venditori vogliono semplicemente vendere i loro prodotti. Proveremo a sfruttare queste differenze attraverso una tecnica di aggregazione (ensembling) di modelli di apprendimento automatico, nell'ambito del Learning-To-Rank, in grado di imparare più preferenze simultanee in un unica classifica (rank). Questa idea è stata testata e sviluppata in collaborazione con HousingAnywhere, marketplace online internazionale specializzato negli affitti a medio e lungo termine, per classificare Proprietari e Inquilini con l'obiettivo di costruire un sistema di raccomandazione (Recommender System) e migliorare l'esperienza degli utenti sulla loro piattaforma. Una ricerca qualitativa ci ha portato a trovare una valida soluzione, mostrando miglioramenti significativi rispetto a un singolo modello tree-based in diversi scenari di mercato.
Matching buyers with sellers : a learning-to-rank approach to the online rental market
Peresson, Tommaso
2020/2021
Abstract
In the rental market, buyers(tenants) and sellers (landlords) express their preferences while making a deal. For example, a Landlord wants a well-behaved and reliable Tenant, and the Tenant wants an apartment that suits its tastes and budget. This scenario sets us apart from a regular market, where buyers have all the freedom to choose and sellers just want to sell their products. We will try to exploit these differences with an ensembling technique for Learning-To-Rank tasks, learning multiple concurrent preferences in a single rank. This idea was tested and developed jointly with HousingAnywhere, a world-wide online rental marketplace for mid to long-term accommodations, ranking Landlords and Tenants to build a Recommender System improving the user experience of their online marketplace. We found a good solution through performance-driven research, showing significant improvements over a single tree-based model over different market scenarios.File | Dimensione | Formato | |
---|---|---|---|
Tommaso_Peresson___Thesis (1).pdf
non accessibile
Dimensione
3.67 MB
Formato
Adobe PDF
|
3.67 MB | Adobe PDF | Visualizza/Apri |
I documenti in POLITesi sono protetti da copyright e tutti i diritti sono riservati, salvo diversa indicazione.
https://hdl.handle.net/10589/173270