Abstract: 【Purpose/significance】Entity relation extraction is the base work of constructing domain ontology, knowledge graph, devel?
oping question answering system. Distant supervision method aligns large-scale unstructured text with existing knowledge base enti?
ties and automatically annotates training samples. This solves the time-consuming and laborious problem of manually labeling training
corpus with supervised machine learning methods but brings data noise.【Method/process】This paper sorts methods of distant supervi? sion for entity relation extraction combined with deep learning in recent years.【Result/conclusion】The convolutional neural network can better capture the sentence parts and key features, the long and short-term memory network can better deal with the long-distance dependence of sentence entities. The model automatically extracts sentence lexical and syntactic features, the attention mechanism gives the sentence key context and words with greater weight, the integration of prior knowledge into the neural network can enrich the semantic information of sentence entity pairs and significantly improve the performance of relation extraction.【Innovation/limitation】The next step of research should consider the processing methods of entity pairs overlapping relation and entity pairs long-tail seman? tic relation, and solve the problem of entity pairs relation noise more comprehensively.