What question did this study set out to answer?

This research aims to optimize the session creation process for conferences by using automated clustering algorithms and large language models.

April 12, 2026

Optimizing ASABE AIM Session Creation with Text Clustering and LLMs

Key Points

This research aims to optimize the session creation process for conferences by using automated clustering algorithms and large language models.
Developed an agglomerative clustering algorithm to group similar presentation submissions.
Utilized text embedding models to determine similarity based on titles and abstracts.
Tested the system on AIM 2025 submissions to identify coherent session topics and ensure minimum size requirements were met.
Employed large language models to generate session titles and keywords for automatically created sessions.
The automated system created more coherent sessions compared to traditional methods.
All generated sessions exceeded the minimum size of eight participants.
Emergent topic clusters were successfully identified that were missed by traditional organization methods.

Abstract

Highlights An agglomerative clustering-based algorithm groups similar presentations to reach minimum session size. The automated procedure creates more coherent initial sessions than traditional presenter self-selection. The automated algorithm can identify session topics not realized by human session creators. LLMs can create session title suggestions and keywords for automatically generated sessions. ABSTRACT. The traditional manual session organization at the ASABE Annual International Meeting (AIM) is a time-consuming process that is frustrating for organizers. It is also frustrating for attendees when sessions are not thematically coherent. The proposed system leverages text embedding and large language models (LLMs) to thematically cluster presentation submissions into sessions and generate relevant session titles and keywords. Text embedding models determined similarity between submissions based on their titles and abstracts. This was followed by an agglomerative clustering algorithm to form sessions based on topic similarity and size constraints. This process was tested on the AIM 2025 submissions for the Information Technology, Sensors and Controls (ITSC) community to create 21 sessions to replace the initial set of 21 sessions created using the traditional presenter self-selection process. These automatically generated initial sessions were more coherent and all above the minimum size of eight. The system also successfully identified emergent topic clusters that were not previously identified using the traditional process. Experiments using LLMs to generate title suggestions and keywords used Google’s Gemini-2.0-Flash and Meta’s Llama-3.2-3b with three prompts, which included either no example, a minimal example, or a full example. The LLMs could identify appropriate session titles and create keywords to describe the topics of presentations within each session. More complete examples improved LLM outputs, but it also created unexpected and undesirable style mimicry. While the LLMs are helpful, human oversight is necessary to refine these suggestions, ensure variety, and accurately capture the full thematic breadth of sessions. Keywords: ASABE annual international meeting, Automation, Conference organization, Natural language processing, Session creation.

Bookmark

Cite This Study

Joe Dvorak (Thu,) studied this question.

synapsesocial.com/papers/69db37b04fe01fead37c5b00 https://doi.org/https://doi.org/10.13031/ja.16435

Bookmark