Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Graphite: A Graph-based Extreme Multi-Label Short Text Classifier for Keyphrase Recommendation

Ashirbad Mishra, Soumik Dey, Jinyu Zhao, Marshall Wu, Binbin Li, Kamesh Madduri

TL;DR

Graphite introduces a deterministic, CPU-friendly graph-based approach for extreme multi-label short-text classification aimed at real-time keyphrase recommendation. By modeling two bipartite graphs that connect words to items and items to labels, Graphite performs inference through a clustering phase (identifying candidate labels by item similarity) followed by a ranking phase (using Word Match Ratio and label multiplicity). Across large-scale eBay datasets, Graphite outperforms a CPU baseline (fastText) and scales where DNN-based XML models (e.g., Astec) struggle due to memory constraints, while offering competitive AVP and strong inference speed. The method also demonstrates practical impact via increased keyphrase coverage and higher seller acceptance rates, with promising results on public XML datasets (KPTimes, KP20k) and extensive ablations supporting the value of the clustering-rank strategy and interpretability.

Abstract

Keyphrase Recommendation has been a pivotal problem in advertising and e-commerce where advertisers/sellers are recommended keyphrases (search queries) to bid on to increase their sales. It is a challenging task due to the plethora of items shown on online platforms and various possible queries that users search while showing varying interest in the displayed items. Moreover, query/keyphrase recommendations need to be made in real-time and in a resource-constrained environment. This problem can be framed as an Extreme Multi-label (XML) Short text classification by tagging the input text with keywords as labels. Traditional neural network models are either infeasible or have slower inference latency due to large label spaces. We present Graphite, a graph-based classifier model that provides real-time keyphrase recommendations that are on par with standard text classification models. Furthermore, it doesn't utilize GPU resources, which can be limited in production environments. Due to its lightweight nature and smaller footprint, it can train on very large datasets, where state-of-the-art XML models fail due to extreme resource requirements. Graphite is deterministic, transparent, and intrinsically more interpretable than neural network-based models. We present a comprehensive analysis of our model's performance across forty categories spanning eBay's English-speaking sites.

Graphite: A Graph-based Extreme Multi-Label Short Text Classifier for Keyphrase Recommendation

TL;DR

Graphite introduces a deterministic, CPU-friendly graph-based approach for extreme multi-label short-text classification aimed at real-time keyphrase recommendation. By modeling two bipartite graphs that connect words to items and items to labels, Graphite performs inference through a clustering phase (identifying candidate labels by item similarity) followed by a ranking phase (using Word Match Ratio and label multiplicity). Across large-scale eBay datasets, Graphite outperforms a CPU baseline (fastText) and scales where DNN-based XML models (e.g., Astec) struggle due to memory constraints, while offering competitive AVP and strong inference speed. The method also demonstrates practical impact via increased keyphrase coverage and higher seller acceptance rates, with promising results on public XML datasets (KPTimes, KP20k) and extensive ablations supporting the value of the clustering-rank strategy and interpretability.

Abstract

Keyphrase Recommendation has been a pivotal problem in advertising and e-commerce where advertisers/sellers are recommended keyphrases (search queries) to bid on to increase their sales. It is a challenging task due to the plethora of items shown on online platforms and various possible queries that users search while showing varying interest in the displayed items. Moreover, query/keyphrase recommendations need to be made in real-time and in a resource-constrained environment. This problem can be framed as an Extreme Multi-label (XML) Short text classification by tagging the input text with keywords as labels. Traditional neural network models are either infeasible or have slower inference latency due to large label spaces. We present Graphite, a graph-based classifier model that provides real-time keyphrase recommendations that are on par with standard text classification models. Furthermore, it doesn't utilize GPU resources, which can be limited in production environments. Due to its lightweight nature and smaller footprint, it can train on very large datasets, where state-of-the-art XML models fail due to extreme resource requirements. Graphite is deterministic, transparent, and intrinsically more interpretable than neural network-based models. We present a comprehensive analysis of our model's performance across forty categories spanning eBay's English-speaking sites.
Paper Structure (27 sections, 4 figures, 3 tables, 1 algorithm)

This paper contains 27 sections, 4 figures, 3 tables, 1 algorithm.

Table of Contents

  1. Related Work
  2. Graphite Model
  3. Problem Formulation and Notations
  4. Model Construction
  5. Inference Step
  6. Clustering Phase
  7. Ranking Phase
  8. Implementation Details
  9. Experimentation and Results
  10. Setup and Preprocessing
  11. Notations and Metrics
  12. Performance on eBay Datasets
  13. eBay Datasets
  14. Prediction Performance
  15. Execution performance
  16. ...and 12 more sections

Figures (4)

  • Figure 1: Illustration of Graphite's construction phase. Subfigure (a) shows a set of item's titles along with their associated keyphrase and subfigure (b) shows the graphs $G_{WI}$ and $G_{IL}$ that are constructed from the set in (a).
  • Figure 2: Comparison of Precision and Recall scores of fastText, Astec and Graphite with top $10$ predictions from each model.
  • Figure 3: Comparison of AVP, Model size and Inference time ratios of fastText, Astec and Graphite
  • Figure 4: Training times of fastText and Graphite for the top 6 categories.