Large Language Models in Business Analytics : A Case Study
Heimonen, Henri (2024)
2024
Johtamisen ja tietotekniikan DI-ohjelma - Master's Programme in Management and Information Technology
Johtamisen ja talouden tiedekunta - Faculty of Management and Business
This publication is copyrighted. You may download, display and print it for Your own personal use. Commercial use is prohibited.
Hyväksymispäivämäärä
2024-02-27
Julkaisun pysyvä osoite on
https://urn.fi/URN:NBN:fi:tuni-202401121417
https://urn.fi/URN:NBN:fi:tuni-202401121417
Tiivistelmä
Artificial Intelligence (AI) became one of the most discussed topics in professional context soon after the launch of ChatGPT. Suddenly there was an approachable way to communicate with the AI. In business new disruptive moments tend to raise a question how this new thing could be embraced to advance business targets.
This study clarifies the possibilities of AI for business analytics. The goal is to understand if it is possible to extract summaries and detect sentiments from given, large texts. During the study, synthetic data is being used. Another target is to create a Proof-of-Concept (POC) software that demonstrates above-mentioned capabilities.
Reference software by Microsoft provided fast way forward, and enabled concentrating onto actual research questions, instead of building mere boilerplate. So called RAG-pattern (Retrieval Augmented Generation) was used. That enabled combining own data with capabilities provided by pre-trained AI model. Approach close to Design Science Research (DSR) was used to build the software iteratively and study the possibilities. In addition, narrative strategy was utilized to understand the DSR methodology itself deeper.
According to study AI can extract summaries and detect sentiments reasonably well. Recursion and handling larger texts in parts was used as a way around the limited size of the context window. Parallelism was utilized to increase efficiency. Few separate concepts were introduced to facilitate the handling of parts and maintaining the context awareness. One of these is a new idea coined as “base question”. For DSR process few observations were made, in addition it was proposed that aspect of agile development could be brought into process.
As future topics, one suggestion is advancing the idea of base question further and maturing it. It is also proposed to understand how to avoid the overemphasis of beginning of the larger text in summaries, and whether companies integrating AI into office tools could provide a way to use modified “system prompt” – behavioral guidance – for certain use cases. As a final managerial implication, the recommendation to use commercially available tools is made because of rapid development.
This study clarifies the possibilities of AI for business analytics. The goal is to understand if it is possible to extract summaries and detect sentiments from given, large texts. During the study, synthetic data is being used. Another target is to create a Proof-of-Concept (POC) software that demonstrates above-mentioned capabilities.
Reference software by Microsoft provided fast way forward, and enabled concentrating onto actual research questions, instead of building mere boilerplate. So called RAG-pattern (Retrieval Augmented Generation) was used. That enabled combining own data with capabilities provided by pre-trained AI model. Approach close to Design Science Research (DSR) was used to build the software iteratively and study the possibilities. In addition, narrative strategy was utilized to understand the DSR methodology itself deeper.
According to study AI can extract summaries and detect sentiments reasonably well. Recursion and handling larger texts in parts was used as a way around the limited size of the context window. Parallelism was utilized to increase efficiency. Few separate concepts were introduced to facilitate the handling of parts and maintaining the context awareness. One of these is a new idea coined as “base question”. For DSR process few observations were made, in addition it was proposed that aspect of agile development could be brought into process.
As future topics, one suggestion is advancing the idea of base question further and maturing it. It is also proposed to understand how to avoid the overemphasis of beginning of the larger text in summaries, and whether companies integrating AI into office tools could provide a way to use modified “system prompt” – behavioral guidance – for certain use cases. As a final managerial implication, the recommendation to use commercially available tools is made because of rapid development.