Deep Learning Based Chatbot Adapted to the Electronic Funds Transfer Process of Turkish Banking
Küçük Resim Yok
Tarih
2023
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Ismail Saritas
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Advanced applications of Natural Language Processing require understanding the semantic of the language. If traditional machine learning techniques are used, the models built for conversations, called chatbots, are unable to be truly generic. On the other hand, deep learning allows us to extract the complexities within language and makes it easier to model. It can also leverage for building a chatbot which has a real conversation with human. In the study, Electronic Funds Transfer process of Turkish bank operations has been designed. A dictionary of terms used in this process has been created in order to train dialog model. Language descriptor layer first checks the language of command. Named Entity Recognition layer later, classifies the words according to their asset structures, especially the amount and account number information in the Electronic Funds Transfer process. LSTM architecture is used to keep the other stages of the dialog, so that order of dialog is in control. The performance evaluation of the designed model was calculated separately for 3 different EFT processes. According to the results obtained, a success rate of 70% was achieved in EFT with account number, 90% in EFT with IBAN number, and 90% in EFT with credit card number.
Açıklama
Anahtar Kelimeler
hatbot system; intent classification; long short term memory; natural language processing; recurrent neural networks
Kaynak
International Journal of Intelligent Systems and Applications in Engineering
WoS Q Değeri
Scopus Q Değeri
N/A
Cilt
11
Sayı
1
