DiffNator: Generating Structured Explanations of Time-Series Differences

2025-09-24

AI Summary

To address the challenge of interpreting sensor time-series discrepancies in IoT applications—particularly for non-expert users—this paper introduces the first structured explanation framework specifically designed for temporal differences, automatically generating parseable JSON-formatted explanations. Methodologically, it employs a lightweight time-series encoder to extract discriminative features, while leveraging a frozen large language model (LLM) as a decoder to produce semantically coherent, structured outputs—thereby avoiding the computational and data overhead of LLM fine-tuning. Trained and optimized on the TORI dataset, our approach outperforms visual question-answering baselines and pretrained-encoder-based retrieval methods across three key metrics: explanation accuracy, structural compliance, and user interpretability. Key contributions include: (1) establishing and implementing a standardized, structured explanation paradigm for time-series discrepancies; and (2) empirically validating the efficacy of the frozen-LLM-plus-specialized-temporal-encoder architecture for low-resource explanatory tasks.

Technology Category

• Knowledge Representation and Reasoning: Geometric, Spatial, and Temporal Reasoning
• Machine Learning: Time-Series/Data Streams
• Humans and AI: Explainable AI (XAI) for Human Understanding

Application Category

• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Semantics and Knowledge: Data modeling to support human-machine intelligence, including LLMs agents, intelligent system behavior, explanations, and user-friendly interactions
• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization

Abstract

In many IoT applications, the central interest lies not in individual sensor signals but in their differences, yet interpreting such differences requires expert knowledge. We propose DiffNator, a framework for structured explanations of differences between two time series. We first design a JSON schema that captures the essential properties of such differences. Using the Time-series Observations of Real-world IoT (TORI) dataset, we generate paired sequences and train a model that combine a time-series encoder with a frozen LLM to output JSON-formatted explanations. Experimental results show that DiffNator generates accurate difference explanations and substantially outperforms both a visual question answering (VQA) baseline and a retrieval method using a pre-trained time-series encoder.

Problem

• Interpreting time-series differences requires expert knowledge
• Generating structured explanations for differences between time series
• Automating explanation of IoT sensor signal differences

Innovation

• JSON schema captures time-series difference properties
• Combines time-series encoder with frozen LLM
• Generates structured JSON explanations for differences