Research on Extraction of Human Information Entity Relationship Based on Improved Capsule Network

Entity relation extraction is to learn the implicit semantic relations among entities from multiple entities of a single sentence. Extracting entity relationships from unstructured text information is a key step in building large-scale knowledge map, optimizing personalized search, machine translation and intelligent Q & A. At present, the more popular depth model of entity relationship extraction has a better effect on the relationship extraction of single entity pair, but the evaluation index data of the model is not high when it is extended to the situation of single sentence multi entity pair and document level complex semantics. In this paper, an improved capsule network model based on dynamic routing rules is introduced, and it is applied to the relationship extraction of multi entity pairs of unstructured human information in the field of literature. The capsule network uses the route iteration method to connect the capsules between different hidden layers, which makes the capsule network establish the position relationship between different features in the routing process. Therefore, the capsule network is more robust to the position and angle changes of the target than other neural networks, so as to avoid the loss of information. In the experiment, we use the improved capsule network model, transformer and CNN model to extract the entity relationship of human information. The experimental results show that the improved capsule network model can achieve high accuracy, recall rate and F1 value in the multi entity pair relation extraction of small language database in the field of literature.