Family education is an important part of children’s education, which has a great influence on children’s character and mental health. The intelligent system of family education, which is oriented by home-school-society collaborative parenting, is analyzed for its importance and functional performance in children’s psychological education. A personalized hybrid recommendation model for children’s mental health education has been established based on the ConvMF model and combined with the WMF model. Based on the consideration of children’s implicit feedback learning data in the intelligent system of family education, the Set2setRank framework is proposed to mine children’s implicit feedback data, and an adaptive importance resampling strategy is designed to optimize their sampling data. The HR and NDCG of the proposed method on the LastFM dataset are 0.882 and 0.787, respectively, and the recommended performance P@10 index reaches a peak of 0.538 when the number of samples is 40. Based on the children’s mental health education book list obtained through the method of this paper, the students’ liking and perceived difficulty scores were 4.55 and 4.44, respectively. The establishment of a children’s implicit feedback optimization strategy in the family education system can enhance the accuracy of personalized recommendations for children’s mental health education resources and help children develop their mental health.