Research On Distant Supervision Relation Extraction Based On Deep Learning

Relation extraction is mainly responsible for identifying entities from unstructured text and extracting semantic relationships between entities.It has extensive theoretical value and practical significance in the fields of machine translation,intelligent question answering and information retrieval.The distant supervision method reduces the workload of labeling data by automatically aligning the knowledge base with the corpus,but due to its strong assumptions and introducing a lot of noise,the classifier is not effective.Aiming at the problem of noise data,this paper studies the reinforcement learning distant supervision relation extraction combined with noise network and the distant supervision relation extraction combined with adversarial training and semi supervision learning.Compared with other methods,it has higher classification performance.The main research contents are as follows:Firstly,the reinforcement learning distant supervised relationship extraction combined with noise network is studied.The reinforcement learning strategy is used to design the noise indicator.By interacting with the environment composed of relationship classifier and noise data,the false positive and false negative cases of each relationship category are dynamically identified,and the correct relationship labels are reassigned,so as to convert the noise data into useful training samples,It is helpful to improve the performance of distant supervised relationship extraction model;In addition,in the training process,the exploration and utilization of the strategy network are balanced by adding noise to the weight of the strategy network,so as to enhance the exploration ability of the noise indicator and make the noise indicator more accurately select sentences that can correctly express entity relationship.Then,the distant supervised relation extraction combined with adversarial training and semi supervised learning is studied.By assigning correct labels to noisy data,useful examples are added on the basis of reducing noisy sentences,so as to improve the performance of the model.Using adversarial training to learn a generator to capture the characteristics of the data set,so as to identify the noise data and divide it into unlabeled data;Then,semi supervised learning is used to interpolate the labeled and unlabeled data,and the information of unlabeled sentences is used to train a robust relational classifier.Finally,the development of distant supervised relationship extraction system is realized,and the distant supervised relationship extraction algorithm is encapsulated,so that users can intuitively understand and use it through the interface.