Abstract
This study outlines the first stage of an in-depth analysis of nine years of Springshare LibChat transcripts. The Florida Gulf Coast University (FGCU) launched Springshare's LibChat service in 2015 to serve as a supplement to its in-person reference desk consultation model. Initially, staffing was informal, and librarians would log in from their offices. Later, chat was staffed by the librarian stationed at the physical reference desk. As chat traffic increased, the service evolved to regularly scheduled hours and staffing. The librarians decided to take advantage of chat technology and aggressively marketed the service to students as an alternative to in-person research assistance. The COVID-19 pandemic lockdown and subsequent closure of the FGCU campus greatly impacted the demand for virtual services across campus, including library services. Our LibChat transactions nearly tripled from 2019 – 2022. The purpose of the current investigation is multifaceted. We ultimately intend to: • quantify the changes in type of chat questions before, during, and after COVID-19. • determine what interventions are needed to improve our online chat service. • decide if chat should continue to be staffed by faculty librarians? • utilize chat transcripts and statistics to measure and improve upon student success. • investigate the role of artificial intelligence in the future of the service. Design and Methodology The team downloaded all the chat statistics from 2015-2023 using the Springshare dashboard. These statistics included the full, textual transcripts, along with a great deal of quantitative data including: • Chat ID • Name • Contact Info • IP • Browser • Answerer • Timestamp • Wait Time • Duration • Rating • Comment • Message Count Although our aspirations were lofty, the first step was to determine what information could be garnered by reviewing the quantitative data. Springshare provides useful data analysis and visualizations " out of the box, " but those tools could only be used for one year of data at a time. Because the raw transcript data is primarily textual and would require an exorbitant amount of work to manually tag and process, it was decided to utilize natural language processing to anonymize and categorize it.