In the past two years, the application of artificial intelligence has become increasingly “competitive”, with a new technology emerging every so often that makes people exclaim “respect”.
AI seems to understand humans better and is getting closer to “human intelligence”.
For instance, during this year’s college entrance examination, the case of AI participating in the exam was frequently discussed by major tech media. Today’s “AI exam-takers” can not only take the exam but also challenge undergraduates and even postgraduates. For example, OpenAI’s Codex achieved an accuracy rate of 81.1% on advanced mathematics problems at MIT, which is not an exaggeration to say it is on par with top undergraduate students.
Despite previous failures, the expectations for the commercialization of capabilities like real-time AI translation are very clear in the industry. The scene of participants from different countries communicating without barriers in multinational conferences is no longer an unattainable imagination.
The performance of chatbots is no longer just mechanical Q&A some can perceive emotions, analyze sentiments, and provide corresponding feedback. Enabling virtual idols to have emotional companionship is also one of the popular directions.
All these applications are inseparable from the continuous advancement of Natural Language Processing (NLP) technology.
Let’s also talk about the recently discussed “Eastern Masterpiece”—the tiger wearing VR. Creative foreign netizens fed Google’s AI painting system, Imagen, the instruction: give a Song Dynasty tiger VR. The result was a series of artworks that felt completely harmonious, making anyone who saw it exclaim: Google, amazing!
Imagen is currently a culmination of AI’s multimodal understanding and creation, but the precise recognition and understanding of human instructions behind this model also rely on NLP.
It can be said that to make artificial intelligence more like “human intelligence”, we cannot avoid the hot direction of language understanding and processing.
What pressing technical issues need breakthroughs in the field of NLP? What are the technical challenges in processing long texts? How can adaptive information retrieval for open-domain Q&A be refined? How can NLP technology be applied and implemented in complex UGC content communities like Xiaohongshu?
To get answers to these questions, you definitely cannot miss the third episode of the technical live broadcast series “REDtech Is Here” produced by the Xiaohongshu technical team. On August 5 at 19:00, Xiaohongshu will invite top industry experts to discuss the “Frontiers and Practices of Natural Language Processing“.
In addition to the rich sharing from academic heavyweights Zhang Yue and Pang Liang, the first-hand industrial case sharing on the NLP issues and practices faced by Xiaohongshu from Zeng Shu, the head of community search text understanding, and Wang Shusen, the head of technical models, will also be unmissable.
As a unique content community in China, as of October 2021, Xiaohongshu APP has exceeded 200 million monthly active users. Xiaohongshu’s biggest feature is that its community content is mainly UGC, and the content is mostly presented in a multimodal format of audio and video + text, which poses higher demands for text and content understanding.
Xiaohongshu’s decentralized distribution mechanism for user content also determines that understanding the content itself occupies a larger proportion in the algorithm, rather than simply relying on the analysis of user behavior.
The massive UGC content also brings non-standardization issues, the most common case being the understanding of metaphorical and extended meanings.
When a user sees a refreshing grassy field, takes a photo, and records, “as if in The Legend of Zelda: Breath of the Wild.” How should the algorithm understand the core content of this note and determine which users would like this note?
For example, in a travel note, if a user refers to a seaside fishing village as “Little Greece in Fujian”, when other users search with the keyword “Greece”, should this note appear, and when—this is all complex NLP issues.
What better solutions will Xiaohongshu try for these problems? The business scenarios and data of Xiaohongshu can also give rise to exciting technical issues. I believe that answers will also be found in this live broadcast on the evening of August 5.
Live Broadcast Viewing Guide:
Time: August 5, 2022 (Friday) 19:00-21:00
Live Broadcast Platform: Follow the 【Xiaohongshu Technology REDtech】 video account to reserve the live broadcast, see you there. (The live broadcast will also be simultaneously conducted on Douyin and Bilibili, search “Xiaohongshu Technology REDtech”)
👇 Scan the QR code below to enter the live broadcast group, and you will receive the live link and reminders in real-time.
We will release guest speech highlights and lottery activities in the WeChat group, and there is a chance to have your questions answered by the guests through interactive Q&A.