社群平台網紅與品牌雙向推薦模型

Abstract

隨著網紅行銷越來越受到歡迎，許多品牌雇用網紅，在社群平台上宣傳其產品和品牌形象，在社群平台上，品牌要找合適的網紅，且網紅也要找合適的品牌，但要找合適的合作對象，對雙方都是一個極大的挑戰。因此，提供雙向推薦模型為品牌推薦合適的網紅以及為網紅推薦合適的品牌是迫切且不可或缺的議題。許多研究顯示，Graves等人所提出的記憶體機制可有效學習複雜資料結構的表示式，但目前並未有研究利用記憶體機制從使用者產生內容中學習網紅和品牌的表示式。因此，基於記憶體機制，我們提出一個雙向推薦模型。實驗結果顯示，我們所提出的研究架構優於最先進的網紅推薦模型。我們的研究架構可幫助品牌接觸更多潛在的網紅來推廣他們的產品，也有利於網紅尋找合適的品牌，進而幫助品牌降低口碑行銷成本，亦可幫助網紅節省媒合費用。

Influencer marketing has become increasingly popular for brands to promote their products and brand images, and expand their reach on social media platforms. However, it is challenging for brands to find suitable influencers to work with, and for influencers to reach out proper brands to collaborate with. Therefore, it is desirable and essential to offer a two-way recommendation model for brands and influencers. It has been shown that the memory mechanism proposed by Graves et al. can effectively learn the representations of complex data structures. However, to the best of our knowledge, there is no study dedicated to applying the memory mechanism to learn the influencer and brand representations from user-generated contents. Therefore, in this study, based on the memory mechanism, we propose a two-way recommendation model to not only find suitable influencers for brands to promote their products or brand images, but also seek proper brands for influencers. The experiment results show that the proposed framework outperforms the state-of-the-art methods in terms of AUC, F1-score, MRR, mAP and MedR. Our recommender framework is beneficial for not only brands to reach potential influencers to promote their products but also influencers to seek prospective brands for their collaborative work, which in turn help brands implement word-of-mouth marketing strategies, and help influencers save money from the agency service charge.