Research On Construction And Algorithm Of Knowledge Map For Urban Environment

With the rapid development of modern information technology,the Internet has been widely used in all walks of life.Massive information is often presented to people in the form of semi-structured or unstructured.How to deeply analyze these explosive growth data and provide people with higher quality and more accurate knowledge has become a hot research issue in the era of big data.As a branch of the field of artificial intelligence,knowledge mapping essentially links entities with attributes into a network knowledge base through relationships.People can obtain knowledge from massive and diverse data more quickly and accurately through knowledge mapping.In this context,knowledge mapping and its related technologies have been widely concerned by scholars.Domain oriented knowledge mapping construction and algorithm research has important academic and application value.This thesis analyzes the knowledge map construction and related algorithms,focuses on the related algorithms of entity relationship extraction and entity link,and carries out in-depth research on the algorithm in the field of urban environmental data,improves and optimizes the existing traditional classical algorithms,and designs and studies the related improved model.The algorithm is applied to the urban environmental data,the situation knowledge map for urban environment is constructed,and the intelligent platform based on the knowledge map of environmental data is built.The main work of this thesis includes the following aspects(1)Entity relation extraction is the core task of knowledge mapping.In the traditional pipeline extraction method,entity extraction and relationship extraction are separated,and the correlation between them is not considered.Based on the multi head selection joint entity relationship extraction model,this thesis improves it from three aspects.Firstly,in order to solve the problem that clause information between entity and its head is not taken into account in relation feature calculation of multi head selection joint extraction model,attention and word order are used to strengthen the representation clause information and add relation feature calculation to obtain more semantic relation selection scores and improve the performance of relation classification.Secondly,aiming at the problem of class imbalance in the named entity recognition task of the multi head selection joint extraction model,a loss function which can adjust and balance the positive and negative samples is introduced on the basis of the original cross entropy loss function to improve the loss training performance of the recognition task.In addition,different from the original structure,this thesis introduces a pre trained encoder based on transformer(BERT)to enhance the semantic representation ability and better extract the context features of words in the text to enhance the performance of the model.(2)Entity linking is widely used in the fields of atlas updating and intelligent question answering.The focus of entity linking work is mainly in the stage of candidate entity ranking,and the research of entity linking method in this thesis is mainly aimed at the problem of entity ranking.The traditional entity link model is usually based on the similarity between the entity reference and the candidate entity character,and the context semantic information is not fully considered.This thesis combines the text level and semantic related information to obtain the feature information based on the entity link model of conjoined network.For the candidate entities in the knowledge map,the translation representation learning model of joint entity description text is used to enhance the acquisition of structural context features,and finally the entity link is completed by similarity matching.(3)At present,there are few researches on Knowledge Mapping in the field of environment.Firstly,for the multi-source heterogeneous environmental data,the crawler and other tools are used to obtain the environment related data on the network,and then the processed data is converted,and the entity relationship extraction algorithm improved in this thesis is applied for extraction operation.Finally,the knowledge map facing the urban environmental field is constructed,and the intelligent platform based on the knowledge map is built,The platform includes visualization query and display module,entity relationship extraction module,knowledge question answering module and multidimensional information statistics module,which provides help for the application of urban environmental data.