Research On The Techniques Of Semantic-based Internet Information Analysis

Massive and unstructured information in the Internet contains rich semantic contents. So it is ofgreat significance to do semantic analysis on this kind of information. The main objects of semanticanalysis contain texts, images and the texts in the images. Currently, microblog is a popular socialnetwork, which contains a large number of texts and images. This thesis has done research onsemantic analysis technology to text information in microblog, and treats images as a separate kind ofstudying object to research on image character extraction and semantic image classificationtechnology respectively. The contributions of this thesis are as follows:(1)In order to analyze the value of users in promoting microblog events, we study two evaluationindexes: Event Impact Degree and Event Potential Value. We design the algorithms, and carry on theconfirmation through the experiment.(2)This thesis proposes an image text extraction method based on morphology and color layering.Firstly, we extract the image edge, and then extract rectangular regions and non-rectangular regionsaccording to text features by using morphological methods. Finally we handle these two types ofregions respectively wherein Color Layering Algorithm is implemented when dealing withnon-rectangular regions. The experimental results show the proposed method has a high accuracy rate.(3)This thesis proposes an image semantic classification method based on SVM and HSVlayering-based local features. The method extracts visual features of image, including the proposedlocal features based on HSV layering, and then uses SVM to map semantic information. The methodthat fuses local regional features and global features improves the effect of semantic classification.The experimental results prove the effectiveness of this method.(4)We implement an image retrieval system based on semantic and image text extraction. Thesystem extracts and recognizes text from textual images, and classifies all kinds of images accordingto image semantic classification. So the system can be used for retrieving both of textual images andnormal images.