# A Heterogeneous Information Network based Cross Domain Insurance Recommendation System for Cold Start Users

**Authors:** Ye Bi, Liqiang Song, Mengqiu Yao, Zhenyu Wu, Jianming Wang, Jing Xiao  
**arXiv:** [2007.15293](https://arxiv.org/abs/2007.15293)  
**Date:** 2020-07-30  
**Summary page:** https://www.summarizepaper.com/en/paper/2007.15293v1/

## Key points

- The internet has revolutionized the way businesses operate, including traditional insurance companies.
- Adapting to internet sales can bring significant revenue to insurance companies.
- Online insurance is still in its early stages of development and faces the challenge of attracting customers with little or no prior interaction with the company (cold start problem).
- Cross-domain recommendation (CDR) methods have been studied in traditional e-commerce to infer preferences of cold start users based on their preferences in other domains.
- Directly applying CDR methods to the insurance domain is not feasible due to its specific properties.
- A novel framework called Heterogeneous Information Network based Cross Domain Insurance Recommendation (HCDIR) system is proposed for cold start users in the insurance domain.
- The framework aims to learn more effective user and item latent features in both source and target domains.
- Gated recurrent unit (GRU) is employed to model user dynamic interests in the source domain.
- An insurance heterogeneous information network (IHIN) is constructed in the target domain, connecting users, agents, insurance products, and product properties for richer information for recommendation purposes.
- Three-level attention aggregations are used to obtain user and insurance product representations from IHIN.
- Latent features of overlapping users between two domains are obtained, and a feature mapping between them is learned using a multi-layer perceptron (MLP).
- The proposed HCDIR system outperforms existing approaches when applied on Jinguanjia dataset.

## Summary

The internet has revolutionized the way businesses operate, and traditional insurance companies are no exception. Adapting to the trend of internet sales can bring significant revenue to these companies. However, online insurance is still in its early stages of development, and one of the biggest challenges it faces is the cold start problem, which refers to attracting prospective customers who have little or no prior interaction with the company. In the field of traditional e-commerce, several cross-domain recommendation (CDR) methods have been studied to infer preferences of cold start users based on their preferences in other domains. However, directly applying these CDR methods to the insurance domain is not feasible due to its specific properties. To address this issue, this paper proposes a novel framework called a Heterogeneous Information Network based Cross Domain Insurance Recommendation (HCDIR) system for cold start users. The framework aims to learn more effective user and item latent features in both source and target domains. In the source domain, the authors employ a gated recurrent unit (GRU) to model user dynamic interests. In the target domain, where insurance products are complex and data sparsity is a problem, they construct an insurance heterogeneous information network (IHIN) based on data from PingAn Jinguanjia. This IHIN connects users, agents, insurance products, and insurance product properties together providing richer information for recommendation purposes. To obtain user and insurance product representations from this IHIN three-level attention aggregations (relational node and semantic) are employed. After obtaining latent features of overlapping users between two domains a feature mapping between them is learned using a multi-layer perceptron (MLP). The proposed HCDIR system is applied on Jinguanjia dataset and compared against state-of-the-art solutions. The results show that HCDIR significantly outperforms these existing approaches. Overall this paper presents a comprehensive framework that tackles the cold start problem in the insurance domain by leveraging a heterogeneous information network and employing advanced attention mechanisms.