Generate a public opinion analysis report such as "Ant Punished" within 2 minutes, and publish a large public opinion analysis model in the language | Public Opinion | Analysis Report
The artificial intelligence big language model has brought changes to many industries, and in the field of news and public opinion analysis, big models also have great potential. Today, Shanghai enterprise Mido released the "Honey Nest" intelligent public opinion analysis big language model at the 2023 World Artificial Intelligence Conference. After entering keywords, it can automatically generate a "Hot News Quick Report" of relevant news in about 2 minutes, including four sections: event overview, data overview, public opinion views, and analysis suggestions. It is expected to become a powerful assistant for the government and relevant enterprises and institutions.
In the Honey Degree exhibition area, staff demonstrated how to use "Honey Nest" through a large screen. After entering the keywords "Ant Group" and "Punished", Ant Group and its subsidiaries were fined 7.123 billion yuan. The news that occurred on July 7th appeared on the big screen, with an overview of the event of 200-300 words at a glance.
The second section of the public opinion analysis report is the data overview: "From 0:00 on the 6th to 23:00 on the 8th, there were 46799 pieces of information about 'Ant Group' on the entire network, and 20503 pieces of information on Weibo, accounting for 43.81%, making it the main platform for information dissemination." In this section, the report provides an information trend chart, pointing out which Weibo accounts increased the information on the entire network after publishing content.
Institutional media is also within the scope of analysis. According to the report statistics, there were 673 relevant media reports from 0:00 on the 6th to 23:00 on the 8th, of which commercial media accounted for 61.07% of the coverage, provincial media accounted for 21.1%, and municipal media accounted for 9.81%.
On this basis, the report analyzed the public opinion situation, showing the proportion of sensitive, non sensitive, and neutral public opinion, and displaying keyword clouds such as "finance", "Tenpay", "regulation", and "rectification".
The third section of the report is public opinion viewpoints. The big model classifies public opinion into two categories through semantic and emotional judgments, namely, "some public opinion expressed support for the work of the financial management department" and "some public opinion expressed dissatisfaction with Alipay's' interest rolling 'interest calculation method", and gave representative netizen comments under these two categories.
The fourth section is analysis and recommendations. The content generated by the big model based on public opinion is: "Suggestions to relevant departments: first, closely monitor the network information and respond promptly to issues raised by netizens; second, do a good job in legal publicity, help consumers understand relevant financial product knowledge, and enhance consumer self-protection awareness."
According to Wang Fang, Vice President of Midu Micro Hotspot Research Institute, the development of this vertical field big model began in April this year when a central enterprise client proposed: "The era of big models has arrived. Can we simply input a few words like ChatGPT to generate public opinion monitoring and analysis reports?"? After research by this language intelligence technology company, it is believed that this is feasible and has great practical value.
It is reported that it takes about 1 hour to manually write a public opinion analysis report. Although software has many automation functions, there are many steps that need to be manually completed, such as manually identifying official reports related to events; Manually edit, organize, and extract key information; Manually observe changes in information trends and identify key points within them; Manually view comments from netizens and cluster a large number of comments to extract the core elements of each viewpoint. After the emergence of the big language model, these tasks can be delegated to artificial intelligence, greatly saving manpower and improving efficiency.
To this end, Mido initiated a large-scale model pre training work, using trillion level word elements for unsupervised pre training, which enabled the large-scale model to have strong general knowledge capabilities. Subsequently, the R&D team conducted professional data fine-tuning and inputted about 6000 high-quality public opinion analysis reports, enabling the large model to learn and master the ability to generate reports, becoming an intelligent public opinion analysis model.
"Welcome everyone to experience the 'Honey Nest' intelligent public opinion analysis model." Wang Fang said at the press conference that users can log in to the company's website, register online, and apply for product trials.