DEEP LEARNING BASED CHATBOT IN FINTECH APPLICATIONS
Links to Files
Permanent Link
Author/Creator
Author/Creator ORCID
Date
Type of Work
Department
Computer Science and Electrical Engineering
Program
Engineering, Electrical
Citation of Original Publication
Rights
This item may be protected under Title 17 of the U.S. Copyright Law. It is made available by UMBC for non-commercial research and education. For permission to publish or reproduce, please see http://aok.lib.umbc.edu/specoll/repro.php or contact Special Collections at speccoll(at)umbc.edu
Distribution Rights granted to UMBC by the author.
Distribution Rights granted to UMBC by the author.
Abstract
Recent advancements in artificial intelligence, particularly in natural language processing (NLP) and large language models (LLM), have utilized deep neural networks trained with vast amounts of data. These networks have been applied to text generation and question-answering systems, with the goal of achieving human-like intelligence. The next wave of deep network application involves deploying these networks on more creative tasks and common-sense reasoning. Cryptocurrency is a relatively new but rapidly growing digital or virtual currency, and information about it is widely dispersed. In this proposal, we explore a new approach to implementing a deep learning-based chatbot in fintech applications. Discovering causality from the abundance of available data, with increasing dimensionality, presents significant computing challenges for most existing approaches. In addition, it can be difficult to manipulate the data to achieve reasonable time and budget constraints. The complexity of data sources and content often makes it challenging for users to locate and analyze relevant information. Building a chatbot in the fintech industry poses three main challenges: tool reproducibility, data manipulation, and result uncertainty and usability. To address the challenge of tool reproducibility in chatbots, we propose a hybrid architecture for natural language processing that combines Bidirectional Encoder Representations from Transformer (BERT) and the cutting-edge language model GPT-3. This customized solution allows us to effectively sort through the knowledge in the cryptocurrency and blockchain industries. In addition, we propose multimodal methods, including text generation, image creation, text-to-voice, and voice-to-text, as well as knowledge graphs, to generate multi-faceted outputs. To optimize prompt generation and image processing, we introduce a simple image-to-prompt algorithm and utilize word embedding and GPT-3 to optimize the prompts and create realistic images from natural language descriptions. Leveraging these approaches, we design and implement a novel chatbot system with multimodal features, including real-time information retrieval. Our experiments and discussions with both synthetic and real-world datasets have demonstrated the effectiveness and efficiency of the proposed methods as solutions to the challenges.