Artificial Intelligence and Big Data Analytics Modules

Underwriting is an essential link in the traditional insurance business chain. It requires a lot of manpower and is time-consuming. Policyholder is often required to provide comprehensive background information and wait for more than 30 days for medical examination.

The insChain platform adopts a completely different approach. We first import user data into the AI model and then store only its results on the blockchain, effectively eliminating the possibility that data breach would lead to loss of confidential client information. The results of the AI model could be a risk level assignment between 0 and 10. Such data could be easily shared with the underwriting or claim teams/functions as an important basis for pricing and claim processing.


Diaglogflow Basic framework

The interactive flow is as follows:

Step1: User enter a query (Intent Classification).

Step2: The chatbot knows the user intent and ask users about related information. User provides the information by text or voice or medical report in an interactive way. Chatbot keeps asking until all the required information is obtained.

Step3: A decision tree is constructed in advance according to underwriting rules. Chatbot will give personalized reply according to the rules and user information.

Natural language processing (NLP)

The key problem to solve in smart underwriting is determining the applicant’s true intent, which also poses a similar challenge in current natural language processing. With respect to the English language, he most advanced models include Dynamic Memory Networks and End-to-End Memory Networks. Both types of network are capable of storing long-term memory and capturing the context. They have shown accurate prediction results in the bAbI project of Facebook AI Research. We haven’t identified an appropriate NLP model for the Chinese language due to its dynamic nature. To overcome this challenge, our solution is to classify the users’ intents in accordance with underwriting rules, and to provide flexible responses not based on predefined templates but by different classification of intents. Such kind of interactive dialogue brings better user experience.

Optical character recognitionOCR

Optical character recognition is the mechanical or electronic conversion of images of typed, handwritten or printed text into machine-encoded text, whether from a scanned document, a photo of a document, or from subtitle text superimposed on an image. We applied the state-of-art model CTPN (connectionist text proposal network) to detect text regions. We further trained the CRNN (Convolutional Recurrent Neural Network) on Chinese characters and use the trained model to convert the regions into texts. An example is given as follows:

New user
Forgot password