Shulou(Shulou.com)11/24 Report--

Recently, the "vector database X cloud computing-driven large model landing e-commerce industry" organized by Zilliz and Amazon Cloud Technology was successfully concluded in Shanghai, which was widely praised by professionals in the industry.

As we all know, the development of large model technology is accelerating the reform and reshaping of thousands of industries. Vector database, as the massive memory of large models, and cloud computing, as the large computing platform of large models, is the cornerstone of large models to the industry. Because of its high degree of digitization, the e-commerce industry has become an excellent "arena" for polishing large models. Therefore, this activity arises at the historic moment.

A number of experts from vector database company Zilliz and Amazon Cloud Technology, and leading technology leaders of the e-commerce industry contributed wonderful speeches at the scene and shared with the audience the core topics such as application scenarios, solution analysis and best practices of the large model in the e-commerce industry. Among them, the speeches made by Shen Liang, senior solution architect of Zilliz, and Chen General, head of Zilliz Towhee open source project, are impressive.

Shen Liang starts with the basic knowledge of vector database and transitions from shallow to deep to the scene of landing vector database in e-commerce industry. He mentioned that typical application scenarios of vector database include: in the field of long text, it can carry out translation, question and answer, semantic retrieval, emotion analysis; in the field of images, it can carry out de-duplication, target detection, image retrieval, and multi-modal mutual search of image and text; recommendation, compliance detection, classification and so on in the video field. Of course, there are many emerging application scenarios, for example, in the field of autopilot, vector retrieval can be used to help find objects that have not been analyzed in the process of real-time judgment. The profit of e-commerce industry basically depends on "search wide push" three axes, vector database technology combined with deep learning can do higher quality "search wide push", one of the important technical directions is multimodal search. Shen Liang mentioned that the current scenarios where large models can be landed in e-commerce include recommendation system recall, intelligent customer service Q & A, UGC analysis and so on.

General Chen described the process of building an e-commerce industry knowledge base from zero to one from a technical level. He said that the current large language model has some limitations, such as lack of domain-specific information, easy to hallucinate, unable to obtain up-to-date information, pre-training data can not be changed, and knowledge base can improve this situation. Among them, we have to mention the new paradigm put forward by Zilliz-CVP Stack. Among them, C is ChatGPT, that is, using large language model to realize intelligent question answering represented by ChatGPT; V is Vector database, that is, providing large-scale and reliable knowledge base for ChatGPT through vector database; P is Prompt as code, even though prompts are used to match user questions and reference content from the knowledge base. The application scenario of CVP Stack represents the question and answer robot OSSChat (https://osschat.io/) for open source users and open source community operators, while in the e-commerce field, e-commerce users can use the question and answer robot to import product introduction into the knowledge base and build a knowledge base to enhance question and answer.

In addition, at the scene, Meng Lingong, an expert in AI technology from power flow e-commerce, shared the practice of a large model based on Milvus. According to Meng Linggong, hundreds of millions of vector data are currently stored on the open source vector database Milvus, and the Milvus-based large model training and reasoning platform can support many types of business to run smoothly. In the future, as the application of the large model moves towards scale and diversification, the product will also carry out more in-depth exploration in technology.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.