Automatic Recognition of Exploratory and Lookup Intents Based on Berry Picking Model
[Objective]This paper selects several new classification features to improve the accuracy of automatic recognition of exploratory and lookup intents.[Methods]Firstly,we collected 1805 queries from the AOL search log and manually labelled them.Then,we proposed classification features from three aspects:query nature,search process,and information source inspired by the Berry Picking model.Third,we evaluated the performance of the proposed features in Naive Bayes,SVM,Decision Tree,Random Forest,and Neural Network.Finally,we explored the classification performance of individual features and feature sets.[Results]The three types of classification features can effectively distinguish exploratory and lookup intentions,with query nature-based features achieving the best performance.Among the five classification models,the neural network algorithm-based model performed the best(Accuracy=0.817 2,Precision=0.849 4,Recall=0.774 7,F1 Score=0.810 3).[Limitations]We did not examine the performances of newly proposed classification features with multiple datasets.User searching behaviors need to be fully explored to form more effective classification features.Moreover,the dataset applied to exploratory/lookup intent recognition was limited due to the high time consumption and labor cost of manual labelling.[Conclusions]The proposed features based on the Berry Picking model can effectively distinguish between exploratory and lookup intents.
Query Intent RecognitionExploratory IntentLookup IntentBerryPicking Model
NETL
NSTL
万方数据
