Research Area “Virtual Assistants”: Leveraging AI to build virtual assistants
Motivation
The rapid advancement and widespread adoption of artificial intelligence (AI) technologies have significantly contributed to the rise of virtual assistants as essential tools in various domains. Virtual assistants enhance efficiency and quality by streamlining workflows, reducing response times, and automating routine tasks. They assist with daily activities, handle complex queries, and provide personalized recommendations, making processes faster, smarter, and more intuitive for individuals and businesses. However, the effectiveness of virtual assistants hinges on their design and implementation. Addressing challenges such as personalization and contextual understanding is vital to creating virtual assistants that are functional, genuinely useful, and reliable.
Background
Leveraging AI technology to build virtual assistants is a topic of heavy discussion in research. The main AI technologies relevant to that field are:
- Large language models (LLM) like ChatGPT[1] are often used to create virtual assistants because they can process textual information and interact in natural speech.
- Reinforcement Learning (RL) is a form of ML where an agent interacts with its environment and learns from it[2]. Thus, RL could create recommendations in complex situations, e.g. in domains like healthcare[3] or manufacturing[4].
- Geometric Deep Learning (GDL), especially Graph Learning[5] allows for representing data in a non-euclidian form and thus modelling the relationships within the data. Such data representations created using GDL could be used for regression, generation, or classification tasks, e.g. on relational databases [6] or over textual documents[7].
When integrated into virtual assistants, these advanced AI technologies can significantly expand their capabilities, enabling them to handle complex tasks, adapt to dynamic environments, and leverage structured and relational data.
Research Gaps & focus
The integration and combination of ML technologies, such as LLMs, GDL, and RL, to develop virtual assistants capable of making processes faster, smarter, and more intuitive remains a significant research challenge. Our work explores and applies these advanced technologies to design domain-specific virtual assistants tailored to fields like healthcare, finance (fintech), and manufacturing, aiming to deliver innovative, efficient, personalized and context-aware solutions.
References
[1] https://arxiv.org/abs/2303.08774
[2] https://web.stanford.edu/class/psych209/Readings/SuttonBartoIPRLBook2ndEd.pdf
[3] https://arxiv.org/abs/1907.09475
[4] https://onlinelibrary.wiley.com/doi/10.1155/2021/5880795
[5] https://arxiv.org/abs/2104.13478
[6] https://proceedings.mlr.press/v235/fey24a.html
[7] https://www.microsoft.com/en-us/research/project/graphrag/