Research On Multi-modal Data Analysis And Influencer Recommendation Technology For Social Media Marketing

In recent years,with the close connection between social media and people’s lives,influencers derived from social media have become more and more concerned and trusted by people.As a result,the influencer marketing industry has developed rapidly.The marketing cooperation between brands and influencers has become a hot research topic.As a key sub-task of influencer marketing,micro-influencer recommendation has also entered its golden development period.Most of the existing micro-influencer recommendation methods focus on how to help the given brands find micro-influencers for better cooperation effect,which means these methods mainly focus on the marketing effect of the brands and micro-influencers.However,in influencer marketing,only considering marketing effect is not sufficient.Choosing partners will have a long-term impact on the image and development direction of brands and micro-influencers.Both of them need to consider whether the cooperation meet their own development needs.To this end,we propose a Conceptbased Micro-influencer Ranking Framework to solve the problems of marketing effect and self-development needs together.We design a concept-based social media account representation learning method and a micro-influencer ranking function to improve recommendation performance.Specifically,we introduce the social media concepts into micro-influencer recommendation,and then learn social media account representation from the perspective of historical activities and marketing direction.Two adaptive learned metrics are defined to model the micro-influencer ranking function.In order to meet the self-development needs,the concept-based interpretable parameters are used to help brands and micro-influencers make marketing decisions.Our major works are summarized as follows:（1）In order to solve the problems of marketing effect and self-development needs that brands and micro-influencers face in the task of micro-influencer recommendation,we propose a Concept-based Microinfluencer Ranking Framework（CAMERA）.CAMERA consists of three components,namely social media concept learning,social media account representation learning and micro-influencer ranking.（2）In social media concept learning,we introduce social media concepts into the task of micro-influencer recommendation for understanding the marketing intent of social media accounts at a fine-grained level.We propose a crossmodal Social Media Concept Learning Network（COSMIC）to learn social media concept representation.In social media account representation learning,social media concepts are used to learn social media account representation from the perspective of historical activities and marketing direction.Meanwhile,a bi-directional attention mechanism is used to focus on the marketing direction of brands and micro-influencers over social media concepts.According to the interpretable historical activities concept distribution and marketing direction concept weights,brands and micro-influencers can judge whether the marketing activities meet their self-development needs,and then make marketing decisions.（3）In micro-influencer ranking,the core is to learn a microinfluencer ranking function to calculate the ranking scores between the given brands and all micro-influencers.We model the endorsement information of brands and microinfluencers,and the influence information of micro-influencers,by which we define two adaptive learning indicators（endorsement effect score and micro-influencer influence score）.（4）We construct two datasets that can be used for future research in this field,namely the brand-micro-influencer dataset and the post-concept dataset.Our experiments are designed around these two datasets.The recommendation performance analysis and recommendation interpretability analysis prove the recommendation effectiveness and interpretability of CAMERA,respectively.