The Study On Feature Engineering Construction And Application Based On User Behavior

Data mining and machine learning,as the frontier areas of artificial intelligence,are bringing innovation to the society at unimaginable speeds,and the advent of the Internet era has made the study of user behavior possible.Every operation on the user's online network may be an important message.The use of cutting-edge technology to dig out useful and hidden information values to better enhance user experience is exactly the direction in which major Internet companies continue to invest in resources.User behavior research is a study of the behavior pattern,thinking habits,etc.of users from a specific target group.Through the data generated by users during the process of network access,statistics and analysis are conducted on relevant data,and the rules for users to visit websites will be discovered.Combine these rules with user service experience,such as recommendation system,manual customer service,precise service,advertisement placement,etc.,as well as discover possible problems with the product or service,and provide the basis for further correction or re-definition of market strategies,will offer uses a better experience.This paper takes the user behavior data of the e-commerce platform as an entry point,introduces the related knowledge and concepts of feature engineering,and proposes a number of new theories and programs in the process of feature engineering and experimental verification,including the mathematical transformation feature construction method,the quadratic combined statistical feature construction method,variable length sliding window feature construction method,pseudo-elastic network feature dimensionality reduction method,pre-training feature dimensionality reduction method,and two different kinds of undersampling to resolve the unbalanced proportion of the sample.The paper analyzes the construction of major feature projects and combines three specific business scenarios such as abnormal user analysis,core user mining,and high-latency purchasing user mining to find out suitable feature methods and task resolution processes.The experiment uses different types of machine learning algorithms to cross-validate user behavior data and constructed feature combinations,and at the same time verifies the feasibility of various proposed theories and schemes in the experimental process.Finally,through the experimental results,the adaptability and improvement of the feature engineering and different models are analyzed.