Using neural net modeling for user assistance in HCI tasks
Ray E. Eberts, Leticia Villegas, Colleen L. Phillips, Cindelyn Eberts · International Journal of Human-Computer Interaction · 1992
In the UNIX operating system, many complex operations can be done using a single command line, in the most efficient method, or they can be done using several command lines of simple commands, in a less efficient method. Recognizing when and how to use these efficient commands is difficult for novice users and for many experts. Five UNIX‐based tasks were constructed that could be done using many simple commands, or they could be performed using one or two more complex and efficient commands. Subjects were asked to perform these tasks using the most efficient methods they could. Many different command sequences were generated from these subjects. These data were then used in a neural net model to map the commands to message markers for task assistance. An experiment was devised to test how well users could utilize the mappings of inefficient commands to help messages from the neural net. In the neural net assisted condition, subjects received assistance for the most efficient command whenever the neural net model detected that a command could be done more efficiently. This condition was compared to one in which the subjects could use off‐line help and to a control condition where subjects received no assistance. Results showed that the neural net assisted group was better able to find the most efficient commands, the variance of the tasks was reduced when compared to the other groups, and performance was related to the number of efficient commands needed, rather than the difficulty or uncertainty of the task.