Research On Noun Visualization Oriented To Text-to-scene Conversion

With the development of information visualization,the visualization of natural language has gradually entered people's vision.Text-to-scene conversion is the process of displaying the scenes described by natural language in the form of pictures or animations,which may have an important impact on the development of cultural exchange,design and other fields.The visual information of text-to-scene conversion comes from text,and nouns are the main expressions of visual information such as entities,events,and actions in the text,but not all nouns are suitable for visualization.Therefore,how to distinguish the visibility of nouns has become a key problem to solve in the text-to-scene conversion.Because the research of noun visibility is a new problem,currently there is no unified basic theory and technology method of visibility discrimination.Facing this situation,this thesis makes a deep research on the definition,classification,influencing factors and discrimination methods of noun visibility.The main work includes the following:1.Aiming at the lack of theoretical basis for noun visibility.Based on the existing research on nouns,this thesis qualitatively analyzes the visibility of nouns,gives the definition,classification of noun visibility and characteristics of each category,analyzes the influencing factors of noun visibility,and lays a foundation for further research.2.Aiming at the lack of technical methods for the visibility discrimination of nouns,this thesis proposes a noun filtering method based on constraint rules and noun visibility discrimination based on dictionary.First,analyzes the sentence components of nouns,and extracts the nouns that need to be visualized by making constraint rules.Then,distinguish the visibility of these nouns.In the case of lack of data base,this thesis establishes the rules of constructing a dictionary based on the analysis of noun category and visibility,and then constructs a dictionary of noun visibility to realize the basic discrimination of noun visibility.The experimental results show that the method based on constraint rules can effectively extract the nouns that need to be visualized,and the method based on dictionary can distinguish the visibility of some nouns,but there is a problem that new words cannot be recognized.3.Aiming at the problem that dictionary can't recognize new words,a method based on semantic classification is adopted.Among them,the problem of low recognition rate of non-visual nouns in classification due to the imbalance of dictionary data.This thesis proposes an over sampling based on data distribution and density(ODD),which improves the SMOTE algorithm.The experimental results show that the method based on semantic classification can effectively recognize the visibility of new words.After processing the dictionary data by the ODD method,the classification model based on the semantic classification method is more stable,and the accuracy,recall rate and F1 value are significantly improved.4.Finally,the experimental prototype system is established.Combining the natural language processing technology,the noun visibility annotation system and the discrimination method proposed in this thesis,the common text is processed to form the noun visibility discrimination and annotation.At the same time,the validity and feasibility of the theory and method proposed in this thesis are proved.