Research On Data-driven User Requirements Acquisition And Smart Service Improvement Methods

With the rapid growth of mobile Internet users,research on smart services in mobile applications is in full swing.Different from the traditional "request-response" service mode,smart services in mobile applications should be proactive and personalized.However,the use scenarios of mobile applications are complex,the managers' observation views are diversified,and users often only focus on the use of smart services and do not care about the improvement of smart services,which brings challenges to the realization of intelligent services.Therefore,vectorization of user feathres to support automatic user features analysis and requirements acquisition,discover user roles from user features to enrich the user descriptiuon,expand data sources to discover new services and realize continual smart services improvement are important problems to be solved in current research on smart service in mobile applications.Based on the above analysis,the research of this thesis mainly includes the following contents:(1)A data-driven personalized user requirement elicitation method is proposed.This method first uses information entropy gain to realize the quantitative analysis of the influence of context on the view.User view feature construction algorithm and user context view feature construction algorithm are proposed to realize the vectorization of user features through data analysis.On this basis,algorithm based on difference value is designed to realizes the applicability analysis of the user to different features,and uses multi-channel CNN to learn this applicability.Finally,fuzzy logic is used to generate the description of the user's requirements in natural language.Experiments show that after the applicability analysis of users to different features,the user requirements can be elicitate more accurate.This method provides a feasible method and technical route for the discovery of user's personalized requirements in complex mobile applications,and has a certain promotion value.(2)A enhancement-learning based user role discovery and optimization method named is proposedThe method first uses K-Means++ algorithm to discover user roles by clustering the user feature matrix.On this basis,in view of the randomness of the results of the K-Means++ algorithm,the method of reinforcement learning is used to strengthen the clustering results.The experimental results show that through the process of reinforcement learning,the silhouette coefficient of the cluster can be increased by about 1.5 to 2 times;the randomness of the user role classification by running the K-Means++ algorithm for multiple times is reduced to about 1/4 to 1/ 3.This method is the first time that the reinforcement learning method is applied to user role discovery in mobile applications,and it solves the problem of randomness when the automated method discovers user roles.(3)A smart service framework and improvement method is proposed.This method includes two parts: a multi-source fusion smart service framework and an active learning-based smart service improvement method.The two parts complement each other.The first part realizes the discovery of new services from a wider range of data sources,and the second part uses the results in the first part to improve smart services.The first part of the method is found from three data sources.The first data source is user behavior data,services are discovered through the construction and analysis of user features.The second source is user role data,services are discovered through investigation of experts;and the third source is user comments,the VSM+BTM method and the set operation is used to discover the service.The second part is the smart service improvement method based on active learning.In this section,the personalized service list provided for users is used as the data to be labeled,and the actual situation of users using services is used as the oracle.Through continuous learning of the service process of users,the continuous improvement of personalized user smart services is realized.The experimental results show that after using this method,the click-through rate of users on the service has increased,and compared with the relevant latest Kano model method and crowdsourcing-based methods,this method has certain advantages.