Since the advent of ChatGPT, there have been document-based question-and-answer projects such as ChatPDF and ChatDOC abroad, but domestic research and development of dialogue products has been in progress.
Guizhou Yuanchuang Technology has developed a ChatDoc knowledge base system based on the local vector model, which can directly upload Doc, Docx, PDF, txt, web links, etc. for question and answer.
Experience address: ChatDoc | ChatDoc
Today, let's actually test a wave of effects and see how the experience is.
Official e-commerce customer service Q&A
Q: Is there still stock?
Answer: According to the known information, the baby you fancy is in stock.
Q: Can the price be cheaper?
Answer: Dear, our pricing is already the lowest selling price, there is no way to make any more discounts~~~
Q: Is free shipping available?
Answer: Dear, I am very sorry, the postage is collected by the courier company, we only collect it on behalf of you, and we cannot provide you with free shipping~~~
EMS, EMS is accessible all over the country, but EMS does not include postage, you need to pay 10 yuan for postage, (same as sending with the wind, you need to pay 20 yuan for postage)
Dear, the default is to send Yunda express, can you receive Yunda's goods here?
Dear, if you want a discount, join our group purchase, we can give you free shipping!
Q: How long will it take to ship?
Answer: Dear, we can arrange delivery for you within 42 hours after you take pictures.
After experiencing the question and answer of e-commerce customer service, the effect can be said to be amazingly accurate.
Finally, I tested the official introduction question and answer, and the effect is not bad.
Q: What problems can ChatDoc solve?
A: ChatDoc can solve the following problems:
1. Knowledge management: ChatDoc can help you create, organize and manage knowledge bases, making your knowledge materials more orderly and easy to find.
2. Knowledge retrieval: By asking questions, ChatDoc can intelligently return the most relevant and accurate answers to help you quickly find what you need
Q: What is the principle of ChatDOC?
A: The principle of ChatDoc is based on natural language processing and AI deep learning technology. It uses advanced language models, such as ChatGPT, to understand user questions and generate corresponding answers. ChatDoc can understand context, reason and generate natural and fluent answers by training and learning from a large amount of text data.
Q: How does the ChatDOC knowledge base charge?
A: ChatDoc offers both free and paid versions. The free version allows users to create and manage a basic knowledge base and enjoy some core functions. The paid version allows users to create and manage multiple advanced knowledge bases and enjoy all advanced functions. The specific charging standards and package details can be obtained on the ChatDoc official website.
I tested two official examples, and overall it is quite acceptable. Next, I will upload a test result myself.
According to the official guidelines, the first step is to create a knowledge base, and then upload the document for training. After uploading the document, there will be a prompt in training, and the question and answer can only be started after the training is completed.
I tested it after uploading, and the effect fully meets the needs. Of course, there are some deficiencies, but it is not a big problem.
The overall test feeling is pretty good. I asked the official team. In order to avoid data leakage, openai was not used for training, but a local vector model was used.
Some application scenarios of enterprise knowledge base
Read papers, literature, research reports. Support various languages. Even if the uploaded document is in English, you can still ask questions in Chinese to overcome language barriers.
Intelligent customer service. It can simulate human customer service and give accurate and fast answers based on the knowledge learned. It can simulate the tone and speaking style of real customer service, and it is not easy for users to find that it is an AI robot.
personal knowledge base. You can upload your own various documents, and AI will automatically index the corresponding answers according to the questions, and tell the documents cited by the answers and the corresponding page numbers.
contract management. Enterprise users have a large number of contracts, and the software can efficiently index documents to help you quickly extract useful information.
Industry knowledge base. Upload industry knowledge, and the AI Learning Club will become an industry expert. For example, you can upload various knowledge such as programming, literature, and law.
When it comes to application scenarios, I suddenly thought about whether it is possible to combine the knowledge base with the digital human, so that the digital human can not only generate memory, but also solve the digital human's feeling of being stiff.
The question and answer section is currently based on Baidu's Wenxin and openai's ChatGPT. It is said that the open source llama2 will be supported for privatized deployment in the future.
Looking forward to open source later, so that more companies can use the enterprise knowledge base.