How Multi‑Task Learning Can Shrink E‑Commerce Product Titles Without Losing Sales

Abstract

On platforms such as Taobao and Tmall, merchants often write overly redundant product titles for SEO, which become too long to display fully on mobile apps, harming user experience. This paper introduces a multi‑task learning method that compresses original titles into short titles without affecting conversion rates.

Research Background

Product titles are a crucial communication medium between sellers and buyers. With the rapid shift of online shopping to mobile devices, the limited screen size demands concise yet informative titles. Excessively long titles are truncated on search result pages, reducing visibility of key product attributes.

Existing Methods

Traditional text summarization methods fall into extractive and generative categories. Extractive approaches select words directly from the source, while generative methods produce new sentences. Prior e‑commerce solutions focus solely on length reduction, ignoring impacts on click‑through and conversion rates.

Methodology

The proposed framework consists of two Sequence‑to‑Sequence tasks sharing a common encoder. The primary task uses a Pointer Network to generate a short title from the original title. The auxiliary task employs an attention‑based encoder‑decoder to generate the corresponding user search query. Joint optimization aligns the attention distributions of both tasks, encouraging the short title to retain core information and reflect high‑frequency query terms.

Main Contributions

The multi‑task learning approach outperforms traditional extractive summarization methods in offline automatic, offline human, and online evaluations.

An end‑to‑end training pipeline eliminates the need for extensive manual preprocessing and feature engineering.

Attention distribution consistency ensures that generated short titles highlight keywords that drive transactions, benefiting broader e‑commerce scenarios.

Experimental Results

Experiments on a Taobao women’s clothing dataset compare five methods: Truncation, Integer Linear Programming (ILP), Pointer Network (Ptr‑Net), vanilla multi‑task learning (Vanilla‑MTL), and the proposed attention‑consistent multi‑task learning (Agree‑MTL). Agree‑MTL achieves the highest ROUGE scores and superior human evaluation metrics (readability, informativeness, accuracy). Online A/B testing shows a 2.58% increase in CTR and a 1.32% rise in CVR over the baseline ILP method.

Conclusion

By leveraging abundant user search queries and transaction data, the proposed multi‑task learning method compresses long e‑commerce titles while preserving essential information and enhancing conversion‑related keywords. Both offline and online evaluations confirm improvements in readability, informativeness, and core product word accuracy without sacrificing conversion rates.