Research On Container Portrait Modeling And Portrait Tag Generation

As lightweight virtualization technology evolves and software industry's demand for continuous deployment,continuous integration,and continuous delivery increases.Container,as a new software unit in mordern software system,has promoted software development with its "build once,run everywhere" feature.However,characteristics of containers are difficult to obtain and the container knowledge is not fully mined,which affects the application scenarios such as container management and container recommendation.Therefore,collecting container information and mining container knowledge is of great significance for future container-related application scenarios.In the field of Software Repository Mining,container repository mining has become an emerging research subfield.While existing methods focus on the analysis of container repositories at a lower level or in a single aspect,which lacks systematic cognition of containers,especially for High-dimensional semantic information mining of functional and non-functional characteristics of containers.The main goal of this paper is to analyze multiple types of description information of containers,build a container portrait based on a hierarchical tag architecture,and implement container portrait modeling.Driven by the design of container portrait,this paper studies the generation mechanism of container portrait tags,from the basic tags mining to find characteristics of the container itself,to the highdimensional semantic tags to describe the functional and non-functional characteristics of the container,which supports future application scenarios that require container knowledge such as cloud computing,container retrieval,container recommendation,etc.The main work of this paper includes the following three parts: First,a container portrait is proposed,a multi-layered container tag architecture is constructed,and the types of tags included in the tag structure of each layer are designed by combining the data information and data characteristics related to the container.Secondly,driven by the design of the container tag architecture,the container portrait tag generation mechanism is established,which analyzes container information from multiple perspectives in different methods.As for short descriptions of containers,a key tag extraction method based on TF-IDF(Term Frequency Inverse Document Frequency)is designed.This method is used to extract the tag information that characterizes the uniqueness of containers.A general tag extraction method based on hierarchical clustering is designed to achieve the general characteristics mining.As for at the processing of long description information,a long text-oriented LDA(Latent Dirichlet Allocation)topic model-based tag generation method is proposed to generate related tags for the container in the form of a topic model.In addition,this paper proposes a tag generation method based on Bayesian Rose Trees and Microsoft Concept Graph and a tag generation method based on Word2 Vec and Word Net knowledge base,which generate extended tags from the semantic vertical and horizontal perspectives respectively.The mechanism analyzes container information from simple keywords extraction and associated analysis,then to analysis of topic model and tags extension,which digs container knowledge at hierarchical semantic levels.Third,the container image tag extraction tool is designed and implemented,and methods in the container portrait tag generation mechanism are experimentally evaluated.Experiments show that the container portrait tag generation mechanism proposed in this paper can extract tags from container text description information,and supports the generation of extended tags based on existing tags.