Speakers
Description
Throughout history, the overall society and economy have undergone various transformations intended to streamline performed actions and processes by minimizing the efforts and resources involved. Countless changes determined by digital evolution have been gradually embraced by both individuals and organizations, once their beneficial effects have been demonstrated and, consequently, made aware. However, technological progress in the last decade has laid the foundation for the beginning of a new era, in which artificial intelligence (AI) is becoming an important pillar of increasing competitiveness.
Natural Language Processing (NLP), a branch of AI, represents the technology behind some of the fastest-adopted digital tools, facilitating human-machine interaction by interpreting and generating human language. Given the explosion in the number of active users of such tools, an organizational approach focused exclusively on the use of human resources, to the detriment of NLP tools, becomes counterproductive.
The increasing accessibility of natural language processing technologies undoubtedly opens up new directions for their integration into organizational processes, especially through the use of intelligent chatbots. Under these circumstances, following a natural path of research, with an extensive prior foundation focused on observing and analyzing relevant scientific literature, the present research aims to explore and highlight potential use of open-source NLP technologies in the development of custom conversational systems for organizations. Even though such tools already exist, and their number continues to grow, a micro-level approach could be more effective, by facilitating the implementation of existing technologies in an accessible and efficient way for small organizations.
Thus, by using the Python programming language and relevant key libraries, such as Sentence-Transformers, the demonstrative component of the study was focused on testing the applicability of such technologies in real scenarios, as well as on identifying the main advantages and possible limitations. Encompassing a mode of operation that can be easily adapted and applied in other organizational settings, the tested model served as an assistant for frequently asked questions in an educational environment.
The obtained results highlighted, among other important issues, several key directions for organizations that want to adopt such personalized technologies, using open-source instruments, including the following: the need for well-structured training data, the balance between model complexity and response speed, and the need for domain-specific tuning. Furthermore, the paper also discusses limitations that can occur in the development and usage of such systems.
