面向电商内容安全风险管控的协同过滤推荐算法研究

A Collaborative Filtering Recommendation Algorithm for E-commerce Content Security Risk Control

[目的/意义]随着电商平台商家入驻要求降低以及商品上线审核流程简化,内容安全风险问题成为协同过滤推荐算法伦理审查的核心问题之一。[方法/过程]文章将内容安全风险问题纳入用户协同过滤推荐算法的优化过程,提出一种改进的推荐算法。首先,采用混合研究方法对内容安全风险商品的定义、外在表现形式、特点、分类和风险程度进行了界定;其次,利用图像增强和关键词提取技术构建识别内容安全风险商品的多模态特征库,用于训练不同模态深度学习识别模型;再次,利用深度学习、多模态融合和均值聚类等技术对经典CFR算法进行改进,提出面向电商内容安全风险管控的CSCFR算法;最后,基于3个新数据集设计并实施对照实验,证明该算法在内容安全风险、精度、召回率和稳定性上的优越性。[结果/结论]与最新推荐算法相比,所提算法不仅显著提升了内容安全性,而且在精度等性能指标上也略有提升。

[Purpose/significance] With the reduction of merchants’ entry requirements and the simplification of product online review process in e-commerce platforms,content security risk has become one of the core issues of ethical review of Collaborative Filtering Recommendation(CFR) Algorithms.[Method/process]For this reason,this paper incorporated the content security risk into the optimization process of user-based collaborative filtering recommendation algorithm and proposed an improved algorithm.Firstly,a mixed research method was identifying the external manifestations,characteristics,classification,and risk degree of products with content security risks.Secondly,this paper constructed a multi-modal feature base for products with content security risks by using image enhancement and keyword extraction technology to train deep learning recognition models of different modalities.Thirdly,by adopting deep learning,multimodal fusion,and mean clustering techniques,this paper proposed a Content Security-oriented Collaborative Filtering Recommendation(CSCFR) algorithm to reduce the content security risk of the algorithm recommendation.Finally,the superiority of the CSCFR algorithm in violation,precision,recall,and stability were demonstrated through three rigorous controlled experiments.[Result/conclusion] Compared with the state-of-the-art recommendation algorithms,the algorithm proposed in this paper does not only improve the content security significantly,but also slightly improves the performance of precision and other metrics.