Research on Intelligent Recommendation Technology for Complex Tasks

The value of personalized recommendation lies in mining potential association rules from massive data. In order to provide users with high-quality and personalized items. Due to the complex and changeable tasks, scenarios, and environments in the military field, traditional recommendation algorithms based on collaborative filtering have poor recommendation accuracy in the face of such highly variable and highly sparse data. Therefore, this paper proposes a design idea based on events/tasks. It builds a heterogeneous information network consisting of four types of nodes and seven types of links through many users historical behavior records. Then use heterogeneous network representation learning technology to extract the relationship features between users and items, thus retaining the complex topological relationship between users and items. Finally, the obtained user embedding vector and item embedding vector are combined with user characteristics and item characteristics, and then input into the neural network to predict the degree of preference.