Automatic generation of task scripts from web browsing interaction history

Abstract

Embodiments of the invention relate to automatically identifying web browsing tasks based on a web browsing interaction history. According to one embodiment of the invention, a web browsing interaction history of a user is analyzed to identify web browsing actions associated with web sites. Abstracted action sequences for the web browsing actions that are identified are generated, and action subsequences for the abstracted action sequences are generated. A similarity between each of the action subsequences is determined, and similar action subsequences are designated as a web browsing task.

1 . A method comprising:
 analyzing a web browsing interaction history of a user to identify a plurality of web browsing actions associated with a plurality of web sites;   generating a plurality of abstracted action sequences for the web browsing actions that are identified;   generating a plurality of action subsequences for the abstracted action sequences;   determining a similarity between each of the action subsequences in the plurality of action subsequences; and   designating similar action subsequences as a web browsing task.   
     
     
         2 . The method of  claim 1 , wherein generating the plurality of abstracted action sequences comprises, for each of the web browsing actions that is identified:
 extracting a web page on which the web browsing action was performed;   analyzing a document object model associated with the web page that is extracted; and   identifying a node in the document object model for a web object accessed by the web browsing action.   
     
     
         3 . The method of  claim 2 , wherein generating the plurality of abstracted action sequences further comprises, for each node that is identified:
 extracting a first set of features from the node; and   extracting a second set of features from contextual nodes that have substantially similar textual content as the node.   
     
     
         4 . The method of  claim 3 , wherein generating the plurality of abstracted action sequences further comprises:
 constructing a plurality of feature vectors for each of the web browsing actions that is identified based on the first set of features and the second set of features that are extracted;   creating a plurality of clusters, each of the clusters comprising web browsing actions with similar feature vectors;   generating an action-class label for each of the clusters; and   replacing each of the web browsing action in each of the clusters with the action-class label for that cluster.   
     
     
         5 . The method of  claim 4 , wherein determining the similarity between each of the action subsequences comprises:
 determining if one of the action subsequences comprises substantially identical sequences of action-class labels as one or more other of the action subsequences; and   determining if the one action subsequence is a generalization of the one or more other action subsequences.   
     
     
         6 . The method of  claim 1 , further comprising:
 automatically generating a set of executable scripts based on the similar action subsequences designated as the web browsing task.   
     
     
         7 . The method of  claim 1 , further comprising:
 generating a web browsing task model based on the similar action subsequences designated as the web browsing task, the web browsing task model comprising the similar action subsequences, a plurality of feature classifiers that classify features associated with each of the web browsing actions in the similar action subsequences, and a plurality of generalization heuristics that compute similarity of abstracted action sequences.   
     
     
         8 . The method of  claim 7 , further comprising:
 identifying a set of web browsing actions from the web browsing interaction history;   comparing the set of web browsing actions to the similar action subsequences in the web browsing task model; and   designating the set of web browsing actions as an instance of the web browsing task if the set of web browsing actions is substantially similar to the similar action subsequences in the web browsing task model.   
     
     
         9 . A computer program product comprising a computer readable storage medium having computer readable program code embodied therewith, the computer readable program code comprising:
 computer readable program code configured to:
 analyze a web browsing interaction history of a user to identify a plurality of web browsing actions associated with a plurality of web sites; 
 generate a plurality of abstracted action sequences for the web browsing actions that are identified; 
 generate a plurality of action subsequences for the abstracted action sequences; 
 determine a similarity between each of the action subsequences in the plurality of action subsequences; and 
 designate similar action subsequences as a web browsing task. 
   
     
     
         10 . The non-transitory computer readable medium of  claim 9 , wherein generating the plurality of abstracted action sequences comprises, for each of the web browsing actions that is identified:
 extracting a web page on which the web browsing action was performed;   analyzing a document object model associated with the web page that is extracted; and   identifying a node in the document object model for a web object accessed by the web browsing action.   
     
     
         11 . The non-transitory computer readable medium of  claim 10 , wherein generating the plurality of abstracted action sequences further comprises, for each node that is identified:
 extracting a first set of features from the node; and   extracting a second set of features from contextual nodes that have substantially similar textual content as the node.   
     
     
         12 . The non-transitory computer readable medium of  claim 11 , wherein generating the plurality of abstracted action sequences further comprises:
 constructing a plurality of feature vectors for each of the web browsing actions that is identified based on the first set of features and the second set of features that are extracted;   creating a plurality of clusters, each of the clusters comprising web browsing actions with similar feature vectors;   generating an action-class label for each of the clusters; and   replacing each of the web browsing action in each of the clusters with the action-class label for that cluster.   
     
     
         13 . The non-transitory computer readable medium of  claim 9 , wherein the computer readable program code is further configured to:
 automatically generate a set of executable scripts based on the similar action subsequences designated as the web browsing task.   
     
     
         14 . The non-transitory computer readable medium of  claim 9 , wherein the computer readable program code is further configured to:
 generate a web browsing task model based on the similar action subsequences designated as the web browsing task, the web browsing task model comprising the similar action subsequences, a plurality of feature classifiers that classify features associated with each of the web browsing actions in the similar action subsequences, and a plurality of generalization heuristics that compute similarity of abstracted action sequences.   
     
     
         15 . A system comprising:
 a task manager for:
 analyzing a web browsing interaction history of a user to identify a plurality of web browsing actions associated with a plurality of web sites; 
 generating a plurality of abstracted action sequences for the web browsing actions that are identified; 
 generating a plurality of action subsequences for the abstracted action sequences; 
 determining a similarity between each of the action subsequences in the plurality of action subsequences; and 
 designating similar action subsequences as a web browsing task. 
   
     
     
         16 . The system of  claim 15 , wherein in generating the plurality of abstracted action sequences, the task manager, for each of the web browsing actions that is identified:
 extracts a web page on which the web browsing action was performed;   analyzes a document object model associated with the web page that is extracted; and   identifies a node in the document object model for a web object accessed by the web browsing action.   
     
     
         17 . The system of  claim 16 , wherein in generating the plurality of abstracted action sequences, the task manager further, for each node that is identified:
 extracts a first set of features from the node; and   extracts a second set of features from contextual nodes that have substantially similar textual content as the node.   
     
     
         18 . The system of  claim 17 , wherein in generating the plurality of abstracted action sequences, the task manager further:
 constructs a plurality of feature vectors for each of the web browsing actions that is identified based on the first set of features and the second set of features that are extracted;   creates a plurality of clusters, each of the clusters comprising web browsing actions with similar feature vectors;   generates an action-class label for each of the clusters; and   replaces each of the web browsing action in each of the clusters with the action-class label for that cluster.   
     
     
         19 . The system of  claim 15 , wherein the task manager further:
 automatically generates a set of executable scripts based on the similar action subsequences designated as the web browsing task.   
     
     
         20 . The system of  claim 15 , wherein the task manager further:
 generates a web browsing task model based on the similar action subsequences designated as the web browsing task, the web browsing task model comprising the similar action subsequences, a plurality of feature classifiers that classify features associated with each of the web browsing actions in the similar action subsequences, and a plurality of generalization heuristics that compute similarity of abstracted action sequences.