Research And Application For Website Accessibility Evaluation Oriented Group Sparse Feature Selection

The Internet is the main route for the majority of disabled people to obtain information,participate in employments and entertainments.However,most of the existing websites have only focused on the user experience of ordinary people in the initial design and implementation process,and have ignored the accessibilities for disabled people,making it difficult for disabled persons to visit.After knowing this problem,it is necessary to do the accessibility modification for these websites.Thus,the website accessibility evaluation comes into being and helps web content provider to discover the inaccessible parts and provide support for the following accessibility modification.Many accessibility evaluation regulations can not be evaluated automatically and requires certain human involvement,which greatly limits the efficiency of the entire evaluation process.In order to solve this problem,sampling methods have be widely introduced into the process of website accessibility evaluation.Through sampling,the scale of webpage data that needs to be manually evaluated is greatly reduced,so that evaluation can be completed in a short period of time.But this also brings another problem that some critical pages may not be sampled,and the evaluation results may be inaccurate.How to make use of the small amount of the manual evaluation results,try not to lose the information contained in those unsampled data and obtain a more accurate accessibility evaluation result,has become the main target for this article.In order to achieve this goal,we first try to find out the features that best preserve the structure of the web page,and pay more attention on the anti-noise performance of the feature subsets.Finally,an entire set of sparse feature selection methods for website accessibility evaluation is designed and related system applications are implemented.The specific tasks include:(1)A group sparse feature selection algorithm based on local learning is designed to extract the HTML tags that best preserves the structure information of web pages.Since the structure information of the webpage is mainly reflected on the HTML tags,it is possible to use tags and the number of its occurrences to approximate the webpage.Base on this,we design a group sparse feature selection algorithm for web page structure extraction,analyzing the web page distribution through local learning,and using group sparse regression to evaluate the importance of features so as to select the features that best reflected the web page structure information,and provide support for the following accessibility evaluation process.(2)A joint local learning and group sparse regression based feature selection algorithm is designed to easily eliminate the influences of the redundant and unrelated tags for the website accessibility evaluation.With the development of front-end web design technology,the content and form of web pages have become more and more complex.A web page may contain hundreds of tags,and these tags also inevitably introduce a lot of redundancy and noise for the web page structure analysis.In order to better deal with the redundancy and noise in the tags,we design a feature selection algorithm that combines local learning and sparse grouping regression.By alternately performing data distribution analysis and feature weight analysis,we gradually eliminated the negative effects of redundancy and noise to guaranteed the quality of the selected HTML tags.(3)A semi-supervised group sparse regression algorithm for website accessibility evaluation was designed.Website accessibility evaluation introduces sampling techniques to reduce labor costs,but due to the randomness of sampling,some critical web pages may be lost,resulting in inaccurate results.To address this problem,we design a semi-supervised group sparse regression algorithm.Based on the manual evaluation results of the sampled web pages and the similarity constraint of unevaluated web pages which is introduced to retain the information contained in all web pages as much as possible,the quality of the website accessibility evaluation results is significantly improved.The feature selection experiments on standard datasets and website accessibility evaluation datasets fully verifies the advantages of our two feature selection methods compared to other algorithms in extracting the structure of web pages and suppressing redundancy and noise interference.The final website accessibility evaluation experiment proves that the semi-supervised group sparse regression algorithm can efficiently and accurately give the detection results of each web page in the website,and help obtain the accessibility evaluation results of the entire website.