Sustainable techniques to improve Data Quality for training image-based explanatory models for Recommender Systems

2024-07-09

AI Summary

Image-driven recommender systems suffer from low-quality explanations due to sparse training data and label noise, while conventional data augmentation exacerbates carbon footprint. Method: This paper proposes a low-carbon, efficient data quality enhancement paradigm that innovatively integrates positive-unlabeled (PU) learning to identify high-confidence negative samples, geometric/semantic transformations for augmentation, and CLIP-guided text-to-image generation. A multi-model ensemble evaluation framework is further introduced for automated quality verification. Contribution/Results: Evaluated on multiple real-world restaurant recommendation explanation datasets, the method improves average ranking metrics of three state-of-the-art explainable models by 5%. Crucially, it achieves zero additional carbon emissions and no computational overhead increase—marking the first approach to jointly optimize data quality improvement and sustainable model training.

Technology Category

• Computer Vision: Diffusion Models for Vision
• Machine Learning: Calibration & Uncertainty Quantification
• Knowledge Representation and Reasoning: Qualitative Reasoning

Application Category

• User Modeling, Personalization and Recommendation: Explainable and interpretable methods for personalization
• Search and Retrieval-Augmented AI: Web learning to rank, online learning, and counterfactual learning for ranking
• Economics, Online Markets and Human Computation: Data quality aspects of human-annotated datasets

Abstract

Visual explanations based on user-uploaded images are an effective and self-contained approach to provide transparency to Recommender Systems (RS), but intrinsic limitations of data used in this explainability paradigm cause existing approaches to use bad quality training data that is highly sparse and suffers from labelling noise. Popular training enrichment approaches like model enlargement or massive data gathering are expensive and environmentally unsustainable, thus we seek to provide better visual explanations to RS aligning with the principles of Responsible AI. In this work, we research the intersection of effective and sustainable training enrichment strategies for visual-based RS explainability models by developing three novel strategies that focus on training Data Quality: 1) selection of reliable negative training examples using Positive-unlabelled Learning, 2) transform-based data augmentation, and 3) text-to-image generative-based data augmentation. The integration of these strategies in three state-of-the-art explainability models increases 5% the performance in relevant ranking metrics of these visual-based RS explainability models without penalizing their practical long-term sustainability, as tested in multiple real-world restaurant recommendation explanation datasets.

Problem

• Improving data quality for image-based recommender system explainability models
• Addressing sparse and noisy training data in visual explanations
• Developing sustainable enrichment strategies without environmental cost

Innovation

• Positive-unlabelled Learning for reliable negatives
• Transform-based data augmentation technique
• Text-to-image generative data augmentation