Research On Location-referenced Web Textual Information Extraction And Cartographic Visualization

Space is an important organizational unit of information.Research shows that nearly 57% of information in human society is related to spatial location.Especially with the development of GPS,sensor network,mobile internet etc,the ubiquitous geographic infor mation transmitted and shared through the network grows exponentially,which constitutes a dynamic image of the objective world with multiple scales,large depth and full coverage.In the ubiquitous geographic information on the Internet,web text is the most important form of its existence.Nearly 20% of the web text contains the geographical location information,and more than a quarter of the network retrieval is related to geographical location.The large number of geographical location description in web text and people's general demand for geographic location information make spatial content become the core factor in the process of information extraction and analysis.In the face of the complex network texts containing rich geographical location descriptions,how to perceive and resolve web text information from the perspective of space is a major challenge facing the field of geographic information science.On the one hand,the web text is mainly formed in unstructured natural language,how to effectively extract structured information content from the spatial,temporal and semantic perspectives is an important issue which needs to be explored;On the other hand,the geographical location content provides an important perspective for the spatial cognition of web textual information.How to make effective use of this perspective to transmit and express the extracted location-referenced information based on the map is also an important topic to be explored.Driven by the practical problems of information extraction and visual expression of the web text,this study focuses on the spatial,temporal and semantic dimensions to carry out structured extraction and formal modeling of web textual information.On this basis,the map as a carrier of spatial information,is combined with cartographic processing and information visualization methods,to achieve the visual expression of location-referenced web textual information.The research is carried out from the following aspects:(1)A model of location-referenced web textual information based on the spatial,temporal and semantic dimensions is proposed,from which the web text information formed in natural language is formally expressed.On the basis of the model,the web text is extracted and analyzed by natural language processing,named entity recognition,geocoding.Furthermore,the location descriptions in the model are further analyzed,such as the elimination of geographic ambiguity and the acquisition of geographic focus,so as to complete the modeling of location-referenced web textual information.The construction of information model provides support for the visualization of information in the map space.(2)The specific forms and methods for map-based information indexing and browsing are explored,by taking map mashup as the form of location-referenced information presentation.In this process,in order to solve the problem of information overload and visual confusion in the map mashup results,the spatial,temporal and semantic similarity between information items was measured to cluster and integrate the information set.At the same time,the theories and methods in cartography and information visualization are explored and expanded,and two kinds of information presentation modes,namely,traditional labeling and boundary labeling,are applied to optimize the map mashup effect.By exploring the content organization and presentation mode of location-referenced information,an effective framework is provided for map-based information browsing and spatial cognition.(3)The entity level knowledge contained in the location-referenced information set is explored in this research.Specifically,the entities involved in the information are taken as the analysis objects,and co-occurrences of entities in the information are analyzed to construct the entity-relationship graph.Based on the constructed entity-relationship graph,the importance of entities are measured and edge bundling algorithm is used to visually present the entity-relationship.Thus,an effective strategy for the entity level knowledge discovery and expression for the information set is provided in this research.