What Data Does ChatGPT Use?
At the end of November 2022, OpenAI, an American artificial intelligence research laboratory, newly launched a natural language processing tool driven by artificial intelligence technology-ChatGPT chat robot. Once launched, it quickly became popular on social media and became the hottest topic in the field of AI, setting off a new wave of artificial intelligence.
ChatGPT’s human-like dialogue process is the biggest highlight, and the dialogue semantic technology behind it is indispensable. ChatGPT uses a large-scale language model GPT-3.5, and its core technology covers the understanding of user intentions during multiple rounds of dialogue, as well as advanced content generation technologies such as machine translation, information extraction, copy generation, code generation, and email writing. It has language understanding and text generation capabilities.
However, ChatGPT is not a disruptive innovation of technology, but why is this application so “out of the circle”? In the final analysis, the underlying technology that supports this set of artificial intelligence technology training language models is becoming more and more mature. In fact, if you want to complete human-computer interaction such as ChatGPT or even more advanced, you need to process, analyze and train massive amounts of data behind it.
As the world’s leading data service provider, Datatang has designed and produced a large number of multi-round dialogue text training datasets covering multiple fields for dialogue semantics. The following are related NLP datasets of Datatang:
203,029 groups of medical questions and answers
More than 200,000 groups, each containing multiple rounds of conversations between doctors and patients.
830,276 groups of multi-round dialogue text data
More than 830,000 groups, each containing multiple rounds of conversations between two people.
47,811 sentences with single-sentence intent annotation data in interactive scenarios
Intent labeling data covering 15 fields including phone calls, navigation, translation, affiliated intents, alarm clocks, photos, schedules, settings, videos, reminders, weather, information, page control, music, and applications.
84,516 English single-sentence intent annotation data in interactive scenes
Intent labeling data covering 16 fields including phone calls, navigation, translation, affiliated intents, alarm clocks, photos, schedules, settings, videos, reminders, weather, information, page control, music, applications, and voice assistants.
687,694 sentences with open domain intent annotation data
Cover travel, travel by car, by plane, call a car, rent a car, purchase tickets for a trip, book air tickets, rebook air tickets, book train tickets, rebook train tickets, book hotels, watch movies, inquire about movies, order movie tickets, watch variety shows, Watching concerts, querying locations, contacting, making calls, sending messages, sending couriers, picking up couriers, querying couriers, recharging phone charges, recharging traffic, meeting, sending people off, picking up people, ordering restaurants, eating food, watching anime, etc. 60 domain intent labeling data.
In addition, Datatang also provides text data customization services and text data labeling platform services.
Datatang’s data customization service can support the collection of multi-language and multi-field dialogue text data, and can perform tasks such as sentiment analysis, topic classification, and question-and-answer annotation on different types of text data according to different business objectives.
Datatang’s data labeling platform covers entity, entity relationship, reading comprehension, interaction intent, text attribute, document attribute, text question and answer and other labeling tools. It is built by Datatang based on years of experience in labeling implementation. Test, and strive to optimize the operating experience to the extreme.
Datatang will continue to produce new dialogue semantic training datasets to support the implementation of the ChatGPT model.
If you want to know more details about the datasets or how to acquire, please feel free to contact us: firstname.lastname@example.org.