| More and more employees are free to express their views on the company they have worked for or are working for on the online platform.These reviews contain a lot of valuable information,which can make it clearer for job seekers to get an inside view of the company and understand all aspects of the company in advance.It also allows the company to further understand the needs and opinions of its employees,which is important for improving the competitiveness of the company in all aspects.Many previous efforts focused on the polarity of employees’ emotions towards the company(i.e.,whether employees’ emotions towards the company are positive,neutral or negative),without comparing the differences and mining the connections between former and current employees’ emotions toward the company,and without considering the emphasis on the different aspects(i.e.,aspect weights)of Work/Life Balance,Culture & Values,Career Opportunities,Salary & Benefits and Senior Management.In view of the above two,this thesis uses a distributed headless browser web crawler method to collect online anonymous review data about the employees of the Fortune Global 100 in 2020 from the Glassdoor,totaling more than 200,000.Among them,the review data as up to March 2021 includes textual and numerical data.The latest company employee review dataset has been constructed.Then,we analyze the textual and numerical data of employees’ online reviews with current and former employees as the research subjects to reveal the differences between the former and current employees’ emotions.The experimental results show that the current employees value Work/Life Balance and Senior Management more than former employees,while former employees value Senior Management most.Both former and current employees are willing to recommend the company,but current employees are more active.Further based on the collected textual and numerical data of more than 200,000 company reviews,it is company’s Aspect Based Sentiment Analysis that we apply for the first time the “Boot-strapping(semi-supervised)+ LRR” method to,while most of the previous related work is catering industry,hotel,film and other fields.The experimental results show that: the “Boot-strapping(semi-supervised)+ LRR” method is more effective than the “LDA(unsupervised)+ LRR” method in analyzing aspect ratings and aspect weights based on the dataset of this experiment,and the “Bootstrapping(semi-supervised)+ LRR” also outperforms the “LDA(unsupervised)+ LRR”method in both qualitative and quantitative evaluation.The research provides useful information to both groups of job seekers and the companies,which not only facilitates the companies to know its strengths and weaknesses,but also allows job seekers to find the companies they need. |
