Recently, Ping An Life participated in the 3rd International Face Expression Challenge (ABAW) held by the European International Conference on Computer Vision (ECCV) 2022, and stood out among 51 internationally renowned universities, scientific research institutions and large companies, and won the best solution. The good results of the runner-up and the first place of the industrial team have once again demonstrated to the world Ping An Life's technical strength in the fields of artificial intelligence and computer vision.
It is reported that the technology is widely used in the recruitment, training, customer management and other activities of the agent team. The strength of AI technology escorts the digital transformation of life insurance and injects strong impetus into the reform of life insurance.
The track of Ping An Life Insurance in this competition is “Learning from Synthetic Data (LSD)”. This track requires the training on synthetic face data to obtain an expression recognition model and use it in real Validated and tested on face data.
Because the distribution of generated data and real data is very different, the difficulty of the problem is greater than that of previous facial expression recognition competitions.
The representative team of Ping An Life proposed an integrated model with Graph Neural Network (GCN) as the backbone, combining mid-level feature enhancement (SPS) and non-deterministic data mining (DMUE) methods, and finally achieved an F1 score of 36.51% on this track. It is only 0.6 points away from the first-place Moscow HSE University, and the method proposed by the team will be included in the ECCV2022 conference as a workshop paper.
The relevant person in charge of Ping An Life introduced that the technical solutions used in this competition are mainly based on the long-term algorithm experience accumulated by the AI R&D team in Ping An Life’s business video understanding and generation applications. Through the virtual human generation technology including expression generation, virtual explanations are generated. The big coffee and the training virtual human are used for agent training video production, which greatly reduces the cost of video production and improves the production efficiency and training experience.
In addition, through the intelligent video post-production technology including the expression recognition generation technology, intelligent special effects are triggered and generated in the video, which is used for the post-production of the short video shot by the agent, which greatly enriches the video effect and improves the video quality. This technology is widely used In terms of agent recruitment and client management.
In the agent training session, intelligent sparring simulates real sales scenarios, supports agents and machines in man-machine training, and reviews the training results to help agents quickly improve their sales skills.
As of the first quarter of 2022, smart sparring has supported agent sparring for a total of 83.05 million times.
In the agent recruitment process, the AI interview tool developed by Ping An Life can identify multi-dimensional information of prospective agents through intelligent dialogue and generate interview reports, saving a lot of interview manpower and costs.
As of the end of 2021, AI interviews have supported prospective recruits to complete 17.04 million interviews, with a cumulative interview time of 2.58 million hours.
In the aspect of customer management, Ping An Life has created tools such as "smart visit assistant" and "AI follow-up short video production tool" to build an efficient digital marketing model to support the online business development of the agent team.
AI follow-up short video production tool, easy to create high-quality short videos, assist agents to create personal IP, and manage private domain traffic; in 2021, a total of 500,000 short videos will be produced by assisting agents throughout the year, covering insurance knowledge, wealth management, healthy living, etc.
Ping An Life said that technology research and development will inject strong impetus into life insurance reform.
In the future, Ping An will continue to deepen research and development in artificial intelligence, computational vision and other fields, continuously promote the development of cutting-edge technologies, continue to accelerate the integration and application of new technologies and life insurance business scenarios, enable high-quality business development, improve operation and management efficiency, and facilitate the digital transformation of life insurance. .
According to reports, ECCV is one of the top three international computer vision conferences and has a very high influence in Eurasia.
The ABAW competition, namely the facial expression analysis competition in natural environment scenes, is one of the most authoritative international competitions in the field of facial expression recognition. The 4th competition will be held in 2022.
Other teams that scored the top 10 included HSE University in Russia (1st place), Seoul University of Science and Technology (3rd place), Institute of Computing Technology, Chinese Academy of Sciences (4th place), Huake-Ant Financial Team (5th place). No. 1), University of Science and Technology of China (No. 6), SSSI Institute of India (No. 7) and other domestic and foreign enterprises and scientific research institutions.