LLM-Based System for Estimating Timing to Recommend Information Related to Web Discussion

The Internet has enabled the exchange of massive amounts of information and discussion across the globe. Utilizing large language models (LLM) to recommend relevant information to participants during discussions could be beneficial in aiding their understanding of the content and facilitating the progress of the discussion. However, LLM typically generates one output for each input, which could hinder the flow of discussion. Therefore, we developed a system that estimates the appropriate time to recommend information. Compared to conventional systems that recommend information at set intervals, our system demonstrated a tendency to make information recommendations at more appropriate times and to improve the ease of user participation in the discussion, though these results were not statistically significant.