Influencer fraud, which can significantly damage authentic influencers and companies, has become one of the most important problems that can adversely affect the influencer marketing industry. Fraudulent influencers obtain fake engagements on their posts by purchasing engagement bots that automatically generate likes and comments. To identify bots that make fake engagements to influencers, we perform an in-depth analysis on the social network of influencer engagements, which consists of 14,221 influencers, 9,290,895 users, and 65,848,717 engagements. We find that bots tend to have low local clustering coefficients and write short comments which are similar to each other. Based on the analysis results of the unique engagement behavior of bots, we propose a neural network-based model that learns text, behavior, and graph representations of social media users to detect the engagement bots from audiences of influencers. The experimental results show that the proposed model outperforms well-known baseline methods by achieving 80% accuracy.