Slot Filling Via Deep Learning

How to convert a large amount of unstructured information into structured information,which is easy to be understood and to extract semantic information has become a hotspot of research in recent years.Entity slot filling method is an important method to complete this conversion,and is an important sub-task of natural language processing tasks.Although there are many methods proposed to complete the task of slot filling,these methods still exist some problems to be solved.In this paper,we focus on the problems existing in these methods,and propose a multi-instance and multi-label learning algorithm based on long-short-term memory network,mainly work in the following order:Firstly,several methods of slot filling are presented,which are rule-based method,supervised and unsupervised machine learning method.These methods have a great deal of dependence on artificial and other natural language processing tools,can not be applied to the target relationship,performance has yet to be improved and so on.To solve these problems,the method proposed in this paper uses distant supervision to generate the samples needed for model training and testing,and reduce the dependence on labor.Multi-instance and multi-label model is used to solve the problem of multi-instance and multi-label in distant supervision generation.The multi-instance multi-label model uses a graph model with hidden variables to model an entity pair with multiple instances,and different instances may correspond to different labels and thus have multiple labels.Meanwhile,deep learning methods are used to automatically learn the internal rules of the vast number of sample data.The usage of word vector as features of a model can express the similarity between words,reduce the dependence on other natural language processing tools and prevent the propagation and accumulation of errors.Using the long short term memory network to train the model,makes full use of the time sequence information of the sentence and learns the grammatical relations and semantic relations existing in the sentence through the context.Long short term memory network can selectively discard some useless information,while retaining important useful information,is very effective for learning the inner relations in long sentences.In addition,this article also uses the entity’s type information,which is used to distinguish between different entity pair’s different relationships.In this paper,we use the slot filling algorithm’s commonly used datasets,to do comparison experiments with the current popular slot filling model.The experimental results show that the proposed method outperforms other contrastive models in most evaluation metrics.It proves that it has some improvement in performance and verifies its validity.The algorithm is also applied to the 973 cross-media computing demonstration application platform,and constructs a knowledge map about disease.The practicality of the method is proved by the application.