Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Multi-Label Zero-Shot Product Attribute-Value Extraction

Jiaying Gong, Hoda Eldardiry

TL;DR

HyperPAVE tackles unseen attribute-value extraction in e-commerce AVE under a multi-label zero-shot setting by constructing a heterogeneous hypergraph that encodes higher-order relations from user behavior and product inventory. It learns inductive, node-level representations through a fusion-augmented, multi-edge message-passing framework and performs inductive link prediction to identify unseen product–attribute value pairs, guided by a fine-tuned contextual encoder and a GPT-2 description generator. Across ten MAVE categories, HyperPAVE outperforms classification-based, generation-based, and other graph-based baselines in macro-F1 and mAP, with ablations confirming the contributions of hyperedge diversity and feature enrichments. The approach reduces labeling burdens for new products and offers scalable zero-shot AVE, with future work directed at incorporating multimodal signals and dynamic, timestamped graphs to better model evolving product catalogs.

Abstract

E-commerce platforms should provide detailed product descriptions (attribute values) for effective product search and recommendation. However, attribute value information is typically not available for new products. To predict unseen attribute values, large quantities of labeled training data are needed to train a traditional supervised learning model. Typically, it is difficult, time-consuming, and costly to manually label large quantities of new product profiles. In this paper, we propose a novel method to efficiently and effectively extract unseen attribute values from new products in the absence of labeled data (zero-shot setting). We propose HyperPAVE, a multi-label zero-shot attribute value extraction model that leverages inductive inference in heterogeneous hypergraphs. In particular, our proposed technique constructs heterogeneous hypergraphs to capture complex higher-order relations (i.e. user behavior information) to learn more accurate feature representations for graph nodes. Furthermore, our proposed HyperPAVE model uses an inductive link prediction mechanism to infer future connections between unseen nodes. This enables HyperPAVE to identify new attribute values without the need for labeled training data. We conduct extensive experiments with ablation studies on different categories of the MAVE dataset. The results demonstrate that our proposed HyperPAVE model significantly outperforms existing classification-based, generation-based large language models for attribute value extraction in the zero-shot setting.

Multi-Label Zero-Shot Product Attribute-Value Extraction

TL;DR

HyperPAVE tackles unseen attribute-value extraction in e-commerce AVE under a multi-label zero-shot setting by constructing a heterogeneous hypergraph that encodes higher-order relations from user behavior and product inventory. It learns inductive, node-level representations through a fusion-augmented, multi-edge message-passing framework and performs inductive link prediction to identify unseen product–attribute value pairs, guided by a fine-tuned contextual encoder and a GPT-2 description generator. Across ten MAVE categories, HyperPAVE outperforms classification-based, generation-based, and other graph-based baselines in macro-F1 and mAP, with ablations confirming the contributions of hyperedge diversity and feature enrichments. The approach reduces labeling burdens for new products and offers scalable zero-shot AVE, with future work directed at incorporating multimodal signals and dynamic, timestamped graphs to better model evolving product catalogs.

Abstract

E-commerce platforms should provide detailed product descriptions (attribute values) for effective product search and recommendation. However, attribute value information is typically not available for new products. To predict unseen attribute values, large quantities of labeled training data are needed to train a traditional supervised learning model. Typically, it is difficult, time-consuming, and costly to manually label large quantities of new product profiles. In this paper, we propose a novel method to efficiently and effectively extract unseen attribute values from new products in the absence of labeled data (zero-shot setting). We propose HyperPAVE, a multi-label zero-shot attribute value extraction model that leverages inductive inference in heterogeneous hypergraphs. In particular, our proposed technique constructs heterogeneous hypergraphs to capture complex higher-order relations (i.e. user behavior information) to learn more accurate feature representations for graph nodes. Furthermore, our proposed HyperPAVE model uses an inductive link prediction mechanism to infer future connections between unseen nodes. This enables HyperPAVE to identify new attribute values without the need for labeled training data. We conduct extensive experiments with ablation studies on different categories of the MAVE dataset. The results demonstrate that our proposed HyperPAVE model significantly outperforms existing classification-based, generation-based large language models for attribute value extraction in the zero-shot setting.
Paper Structure (33 sections, 13 equations, 4 figures, 7 tables, 1 algorithm)

This paper contains 33 sections, 13 equations, 4 figures, 7 tables, 1 algorithm.

Table of Contents

  1. Introduction
  2. Related Works
  3. Attribute Value Extraction
  4. Zero-shot Learning
  5. Heterogeneous Hypergraph
  6. Methodology
  7. Problem Definition
  8. Problem Statement.
  9. DEFINITION 1 (Heterogeneous Hypergraph):
  10. DEFINITION 2 (Zero-Shot Learning in Graph):
  11. Multi-Label Zero-Shot Data Sampling
  12. Overall Framework
  13. Heterogeneous Hypergraph Construction
  14. Heterogeneous Hypergraph Relation Learning
  15. Embedding Module.
  16. ...and 18 more sections

Figures (4)

  • Figure 1: An example of zero-shot product attribute-value extraction by semi-inductive link predictions.
  • Figure 2: Overall framework of our proposed model HyperPAVE. The framework includes three key components: (a) Hypergraph Construction (b) Heterogeneous Hypergraph Relation Learning and (c) Inductive Link Prediction.
  • Figure 3: Time Efficiency Performance (GPU Time of Model Learning in Seconds for One Training Epoch).
  • Figure 4: Effects on weights of different hyperedges on the category of giftcards.