321, Jingdong Yanxi × Nlpcc 2022 challenge starts!

With the rapid development of the Internet, it brings the problem of information overload , It brings obstacles to people's access to and understanding of the information they need . Therefore, there are many essential summary service models. , For example, Luo Zhenyu's 60 Second voice 、 Quick Book Telling 、 Fast talking about movies, etc , All through refining the core information to serve the audience . In recent years AI Technology has made great progress , Especially in the field of text generation in natural language processing , It has a wide range of industrial application prospects , Not only text-based , It can also be based on multimodal information such as pictures , Concise and beautiful copywriting .2020 year OpenAI released 1750 Billion parameter GPT-3, Its writing level can be comparable to that of human beings .

Jingdong found good goods - Siemens Copywriter

Jingdong found good goods - Sea blue mystery copy

JD has applied text generation technology on a large scale in its own business , It can write exquisite product summary copy for users in a short time , And recommend it to users according to their preferences “ Want to buy ”, JD Yanxi domain based pre training K-PLUG The commodity copywriting generation model of has covered JD's 3000+ Three categories , Pass rate of manual review 90% above , Accumulated generated copy 30 Billion words , Applied to Jingdong find good goods channel 、 Go with it 、AI Live delivery, etc , The commodity Abstract copy reflects the most unique differentiated value of commodities , Maximize user purchase , Accumulated over 3 One hundred million yuan GMV. Besides , Yanxi can also write poetry 、 Write spring couplets 、 Calligraphy creation, etc AI Ability .

Jingdong found good goods - Horizontal refrigerator Copywriter

The challenge of commodity Abstract copywriting mainly comes from three aspects , First , The sources of information are rich , contain title 、 Commodity specifications 、 Product posters and other information ; secondly , Need to deal with multimodal and structured information , Contains text and image information , as well as Structured information such as product specifications and parameters ; Again ,AI System Must be smart enough , Be able to fully explore commodities Selling point information of , And at the right time , Recommend thousands of people to the right users .

Based on this , JD Yanxi has opened the desensitization data of JD's real scene , Hand in hand NLPCC 2022 Hold a multimodal commodity summary challenge . The purpose of this task is to generate a concise text summary for a specified product , The given input includes a detailed text description of the product 、 Product knowledge map and product picture . The following describes the definition of this task 、 Data sets and evaluation methods .