Scalable Approach for Normalizing E-commerce Text Attributes (SANTA)

TL;DR

SANTA is a scalable framework that combines optimized syntactic matching with self-supervised token embeddings to improve normalization of e-commerce attribute values, outperforming traditional methods.

Contribution

The paper introduces a novel self-supervised embedding learning approach using a twin network with triplet loss for attribute normalization, surpassing existing string similarity and unsupervised embedding methods.

Findings

Cosine similarity outperforms Jaccard index by 2.7%.

Self-supervised token embeddings improve normalization accuracy.

Proposed method achieves 19.3% better results than unsupervised embeddings.

Abstract

In this paper, we present SANTA, a scalable framework to automatically normalize E-commerce attribute values (e.g. "Win 10 Pro") to a fixed set of pre-defined canonical values (e.g. "Windows 10"). Earlier works on attribute normalization focused on fuzzy string matching (also referred as syntactic matching in this paper). In this work, we first perform an extensive study of nine syntactic matching algorithms and establish that 'cosine' similarity leads to best results, showing 2.7% improvement over commonly used Jaccard index. Next, we argue that string similarity alone is not sufficient for attribute normalization as many surface forms require going beyond syntactic matching (e.g. "720p" and "HD" are synonyms). While semantic techniques like unsupervised embeddings (e.g. word2vec/fastText) have shown good results in word similarity tasks, we observed that they perform poorly to distinguish between close canonical forms, as these close forms often occur in similar contexts. We propose to learn token embeddings using a twin network with triplet loss. We propose an embedding learning task leveraging raw attribute values and product titles to learn these embeddings in a self-supervised fashion. We show that providing supervision using our proposed task improves over both syntactic and unsupervised embeddings based techniques for attribute normalization. Experiments on a real-world attribute normalization dataset of 50 attributes show that the embeddings trained using our proposed approach obtain 2.3% improvement over best string matching and 19.3% improvement over best unsupervised embeddings.