Research On Prediction Of Health Information Query Reformulation Strategy Based On User Behavior

Objective:As people's living standards improve,the demand for health information also grows.The current web is the main channel for health information query,in the web query,it is very difficult to get the desired result through one query.Usually,it is necessary to adjust the query formula based on the result feedback,and obtain satisfactory results through query reformulation.The purpose of this study is to build a health information query reformulation strategy prediction model,with a view to helping the query system optimize query recommendation services and improve the interactive effect of the web query system.First,analyze the relationship between behavior variables before query reformulation and query reformulation strategies,then select the input feature variables of the health information query reformulation strategy prediction model;second,use support vector machines,neural networks,and decision trees to construct health information prediction model of query reformulation strategy;finally,the situation factors are introduced to discuss the optimization strategy of the prediction model.Methods:In this study,the user's experiment method and questionnaire survey method were used to collect the user's query behavior variables and situational factors when completing the online health information task.A total of 67 users were invited to participate in the experiment.After screening,61 valid questionnaires were finalized,and 366 health information task queries were completed,including 171 query reformulations;using statistical methods such as chi-square test,nonparametric rank sum test to discover the correlation of behavior variables before reformulation and situational factors with query reformulation strategies;using data mining methods of decision trees,neural networks and support vector machines to establish prediction models,and make comparisons and improvements.Results:(1)Taking the user's query reformulation behavior when completing the health information query task as the object,statistical analysis is performed on the behavior variables before query reformulation and query reformulation strategies,and 8 kinds of behavior variables before query reformulation and query reformulation strategies are selected.There was a significant correlation between them,p <0.05.(2)Use the selected user behavior variables before reformulation as input feature variables to predict the query reformulation strategy,use decision trees,neural networks and support vector machine methods to build prediction models respectively,and the overall prediction accuracy is 49.75%,54.98 % and 55.06%.(3)Screen the situation variables that have an impact on the query reformulation strategy,add the prediction model input feature items,and optimize it by integrating the three binary classification support vector machine models.The optimized model prediction accuracy is 64.63 %,An increase of 9.57% from the initial forecast accuracy.Conclusion:(1)The length of the query,browsing the summary of results,the use of query recommendations,and historical characteristics of the query can be used to predict query reformulation strategies.(2)Among the models constructed by decision trees,neural networks and support vector machines,the best prediction effect of health information query reformulation strategies is the support vector machine model(3)The prediction accuracy can be improved after adding the situational factor variables to the prediction model.