Artificial Intelligence 12 min read

NLG Solutions for E‑Commerce: DAMO Academy’s XGeneration, Word2Text, KG2Text, and Metaphor Generation

This talk presents DAMO Academy’s end‑to‑end NLG pipeline for e‑commerce, detailing the XGeneration content‑reproduction system, controllable short‑text generation models (PKM and PSCN), knowledge‑graph‑to‑text generation, metaphor generation, and their extensions to short‑video script creation and intelligent video editing, together with performance results and practical applications.

DataFunTalk
DataFunTalk
DataFunTalk
NLG Solutions for E‑Commerce: DAMO Academy’s XGeneration, Word2Text, KG2Text, and Metaphor Generation

The presentation begins with a background on the challenges faced by e‑commerce merchants who must produce large volumes of promotional content quickly, often struggling to keep up with platform demands.

To address this, DAMO Academy proposes the XGeneration solution, an automated content‑reproduction system that offers high automation, low merchant effort, and adaptability to various e‑commerce scenarios.

In the NLG practice, two research questions are tackled: (RQ1) ensuring attribute‑controlled text generation, and (RQ2) guaranteeing logical consistency. Solutions include token‑to‑text generation for RQ1 and graph‑to‑text generation for RQ2.

Three model families are introduced:

Word2Text (PKM) : a BART‑based model fine‑tuned on a 300 k industry‑labeled corpus to generate product descriptions from keywords.

KG2Text (PSCN) : a knowledge‑graph‑driven model trained on a lexical‑controlled product description dataset, enabling logical reasoning over attribute and demand nodes.

Metaphor Generation : a BART‑based model with a continuation module that produces metaphorical sentences to enhance appeal.

Experimental results show high attribute accuracy (>80 %), logical coherence (>80 %), and strong adoption rates in real‑world deployments such as Taobao homepage posters, live‑stream assistants, and virtual hosts.

The talk then shifts to new‑media exploration, describing two short‑video innovations:

Short‑video script generation : an end‑to‑end system that, given a product, automatically produces hierarchical storyboards with scene descriptions, dialogues, and timing, reducing script‑writing effort by up to 60 % and achieving a 64 % script adoption rate.

Intelligent video clipping : a pipeline that combines product metadata, multi‑modal retrieval, and template‑based rendering to automatically edit and produce polished short videos, resulting in higher play rates and significantly increased conversion.

Finally, the presenter summarizes the two complementary solutions—NLG practice and new‑media exploration—and answers audience questions about content quality control, user preference evaluation, and cost, emphasizing that the services are currently free for merchants.

e-commerceartificial intelligenceKnowledge Graphtext generationNLGMetaphor GenerationVideo Script
DataFunTalk
Written by

DataFunTalk

Dedicated to sharing and discussing big data and AI technology applications, aiming to empower a million data scientists. Regularly hosts live tech talks and curates articles on big data, recommendation/search algorithms, advertising algorithms, NLP, intelligent risk control, autonomous driving, and machine learning/deep learning.

0 followers
Reader feedback

How this landed with the community

login Sign in to like

Rate this article

Was this worth your time?

Sign in to rate
Discussion

0 Comments

Thoughtful readers leave field notes, pushback, and hard-won operational detail here.