Although the LLaMA model has demonstrated impressive performance in the general field through instruction fine-tuning, its capabilities still need to be improved for the fields of self-media creation, live broadcast and operation due to the lack of professional training data. To solve this problem, we propose MediaGPT, a model specially trained for the self-media domain.
MediaGPT (formerly known as Media LLaMA) first conducts continuous pre-training on large-scale self-media corpus, and systematically learns the knowledge system of self-media. Then, with the help of ChatGPT, we collected a batch of analysis and answers on domain knowledge questions such as vibrato operation, short video creation, huge amount of thousands of streams, live broadcast operation and live broadcast skills, and used these data to fine-tune the model, so that Model learning how to apply self-media knowledge to actual scenarios.
Our model has the following capabilities:
Master the knowledge of self-media: be able to understand the core concepts of vibrato operations, short video creation, huge amount of thousands of streams, live broadcast operations, etc. and strategy.
Applicable to practical operation: Be able to explain the concept of self-media in easy-to-understand language, and conduct basic self-media operation consultation, covering content creation, platform operation, advertising and other fields.
In order to promote the open research of large Chinese self-media models, we will open source a series of instruction fine-tuning data in the field of self-media and the parameters of the large Chinese self-media models trained based on LLaMA.
Model effect Human:
Please introduce the operation strategy of Douyin with goods
Assistant: Douyin brings goods The operation strategy can be carried out from the following aspects: 1. Understand the target audience, choose suitable products according to the interests and needs of the audience; 2. Cooperate with high-quality anchors to increase the influence of the live broadcast room Power and attractiveness; 3. Utilize the characteristics of Douyin platform, such as short video drainage and live room interaction, etc., to bring product sales to users; 4. Regularly analyze data to understand users' attitudes toward products Response and preference, adjust operation strategy according to data; 5. Establish a good cooperative relationship with suppliers and brands to obtain more favorable commissions and commodity resources.
Human:
How to do a good stay model to supplement sales and conversion?
Assistant: To supplement the stay model in the live broadcast room, the following aspects can be considered: 1. Provide attractive content and bring more to the audience 2. Interact with the audience, answer their questions, and establish a good communication and interactive relationship; 3. Provide convenience for purchasing goods, Such as setting product links, reminding viewers to place orders, etc., to increase the conversion rate. 4. Provide other content related to the live broadcast content, such as tutorials, discount promotions, etc., to attract the attention and participation of the audience.
Visit Official Website
