Predicting User Behavior in Smart Spaces with LLM-Enhanced Logs and Personalized Prompts

TL;DR

This work tackles predicting user behavior in smart spaces under sparse, personalized logs. It introduces a dual-component framework that (i) constructs user-specific transition and co-occurrence graphs to generate soft prompts, and (ii) enriches event descriptions with LLM-generated semantic descriptions before decoding the next action with a transformer-based sequence model, trained via a pairwise ranking loss. Empirically, the approach achieves state-of-the-art results on four real-world datasets (two smart-vehicle and two smart-home) with pronounced gains for rare events, thanks to semantic augmentation and user-specific prompting. The method maintains low online computation by performing LLM augmentation offline and using efficient inference, supporting scalable deployment in real-world smart spaces with strong personalization and improved automation capabilities.

Abstract

Enhancing the intelligence of smart systems, such as smart home, and smart vehicle, and smart grids, critically depends on developing sophisticated planning capabilities that can anticipate the next desired function based on historical interactions. While existing methods view user behaviors as sequential data and apply models like RNNs and Transformers to predict future actions, they often fail to incorporate domain knowledge and capture personalized user preferences. In this paper, we propose a novel approach that incorporates LLM-enhanced logs and personalized prompts. Our approach first constructs a graph that captures individual behavior preferences derived from their interaction histories. This graph effectively transforms into a soft continuous prompt that precedes the sequence of user behaviors. Then our approach leverages the vast general knowledge and robust reasoning capabilities of a pretrained LLM to enrich the oversimplified and incomplete log records. By enhancing these logs semantically, our approach better understands the user's actions and intentions, especially for those rare events in the dataset. We evaluate the method across four real-world datasets from both smart vehicle and smart home settings. The findings validate the effectiveness of our LLM-enhanced description and personalized prompt, shedding light on potential ways to advance the intelligence of smart space.

Figures (5)

• Figure 1: Overview of the proposed method for smart space interaction prediction. Our method constructs graphs from the user's historical sequences to model individual preferences and generate personalized prompts. It also augments event descriptions using an LLM's common knowledge.
• Figure 2: Enhancing an event log with LLM.
• Figure 3: Impact of event frequency on prediction performance. The first and third panels categorize events into two groups based on whether their frequency exceeds 1%. The second and fourth panels further divide events with frequencies below 1% into ten groups using 0.1% intervals.
• Figure 4: Impact of different hyper-parameters.
• Figure 5: Event Distribution Across Time.