In the last decade, the sudden development of the world wide web and of the computational power of computer has granted to big companies like Ama- zon, Netflix or Spotify the ability to offer more extensive catalogs. These companies have based their business model on the capability to offer the user the chance of finding each product easily on their platforms. The cata- log size increment, on the other hand, has highlighted the intense need for a tool that can guide the user in the platform exploration considering his/her tastes and needs. The need for this tool gave birth to the research in the Recommender Systems field. The primary objective of these new systems is to create a personalized experience for each user through the offered service. The idea behind this thesis was born from the ACM RecSys Competition of 2017 offered by XING. A social-network thought to create a job-related network. (Similar to Linkedin) The competition request to use some of the most modern recommendation techniques in the context of ”cold-start” items, which are items that have been inserted recently inside the catalog and so there aren’t enough data to build a reliable model of the intercon- nection between item and users. The lack of this data has brought to light the need for creating ad-hoc algorithms to be able to solve the problem satisfactorily.
Negli ultimi dieci anni il repentino sviluppo delle connessioni web e della potenza computazionale delle macchine ha permesso alle grandi aziende come Amazon, Netflix e Spotify di offrire cataloghi di prodotti sempre piu` vasti. Queste aziende hanno basato il loro modello di business proprio sulla capacit`a di offrire al cliente la possibilit`a di trovare ogni prodotto rapi- damente tramite le loro piattaforme. L’incremento della dimensione del catalogo d’altra parte ha creato la forte necessit`a di uno strumento che possa guidare l’utente nella scoperta dello stesso considerando i suoi gusti e desideri. Da questa necessit`a nacque la ricerca nel campo dei Sistemi di Raccomandazione. L’obbiettivo di questi nuovi sistemi risiede proprio nella creazione di un’esperienza personalizzata per ogni utente all’interno del servizio offerto. L’idea alla base della nostra tesi `e nata dalla compe- tizione ACM RecSys 2017 offerta da XING, un social network utilizzato per lo sviluppo di una rete di contatti lavorativi. (Simile a Linkedin) La competizione richiedeva di sfruttare tecniche moderne di raccomandazione nel contesto di prodotti ”cold-start” ovvero coloro che, essendo appena stati inseriti nella piattaforma, non hanno ancora interazioni sufficienti ad elab- orare un modello delle loro relazioni con gli utenti. La mancanza di tali informazioni ha portato alla luce la necessit`a di creare algoritmi ad-hoc per la risoluzione del problema in questo contesto.
Automated feature engineering for recommender systems
INAJJAR, ILYAS;LOCCI, GIACOMO
2017/2018
Abstract
In the last decade, the sudden development of the world wide web and of the computational power of computer has granted to big companies like Ama- zon, Netflix or Spotify the ability to offer more extensive catalogs. These companies have based their business model on the capability to offer the user the chance of finding each product easily on their platforms. The cata- log size increment, on the other hand, has highlighted the intense need for a tool that can guide the user in the platform exploration considering his/her tastes and needs. The need for this tool gave birth to the research in the Recommender Systems field. The primary objective of these new systems is to create a personalized experience for each user through the offered service. The idea behind this thesis was born from the ACM RecSys Competition of 2017 offered by XING. A social-network thought to create a job-related network. (Similar to Linkedin) The competition request to use some of the most modern recommendation techniques in the context of ”cold-start” items, which are items that have been inserted recently inside the catalog and so there aren’t enough data to build a reliable model of the intercon- nection between item and users. The lack of this data has brought to light the need for creating ad-hoc algorithms to be able to solve the problem satisfactorily.File | Dimensione | Formato | |
---|---|---|---|
2018_12_Locci_Inajjar.pdf
accessibile in internet per tutti
Descrizione: Testo della tesi 2
Dimensione
19.69 MB
Formato
Adobe PDF
|
19.69 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/10589/144786