logo SBA

ETD

Archivio digitale delle tesi discusse presso l’Università di Pisa

Tesi etd-11202019-110832


Tipo di tesi
Tesi di laurea magistrale
Autore
NEBIU, FATJON
URN
etd-11202019-110832
Titolo
A Recommendation System for Personal Financial Management via Deep Learning-based Categorization
Dipartimento
INGEGNERIA DELL'INFORMAZIONE
Corso di studi
COMPUTER ENGINEERING
Relatori
relatore Cimino, Mario Giovanni Cosimo Antonio
relatore Vaglini, Gigliola
tutor Giannini, Maurizio
tutor Monaco, Manilo
Parole chiave
  • Convolutional Neural Networks
  • Deep Learning
  • Personal Financial Management
  • Recommendation System
  • Transaction Categorization
Data inizio appello
09/12/2019
Consultabilità
Non consultabile
Data di rilascio
09/12/2089
Riassunto
Nowadays, Personal Financial Management (PFM) is a fundamental part of banking services. It allows clients to get a holistic view of their transactions and overall financial situation, and banks to dynamically collect data about their clients’ movements and mine these data for useful information that would help to build financial profiles, suggest the clients personalized services and/or predict their future needs.
The work of this thesis consists of two parts. At first, it is proposed a Deep Learning solution to the Transaction Categorization problem, leveraging a Convolutional Neural Network (CNN). Each transaction contains enough information to allow for an efficient categorization. State-of-the-art technologies and methods of Natural Language Processing (NLP) were used in order to classify clients’ transactions in 7 different spending areas. The CNN model proposed is able to classify with very high accuracy the transactions in the pre-defined areas. Transaction Categorization provides a useful insight to understand where clients spend their money, what are their needs and habits, and what possible banking services/products could be worth to them, thus offering the possibility to create a personalized experience to each customer.
In the second part, techniques of clustering and similarity in vector space were used in order to associate items to the users based on their interests, which were analysed by getting statistics from the history of their past categorized transactions. Similar users were grouped in clusters using the K-means algorithm in order to set the path for a collaborative-filtering RS, where the behaviour of a user within the cluster would help to recommend similar users the same services.
File