Collecting the activity time data is very crucial for the Personal Software Process (PSP) users. For the user to collect his activity time as accurate as possible, PSP directs the user to manually record the activity right at starting or ending any activity. Research have found that due to the burden of data collection, users are reluctant to adopt PSP in daily life and manual activity time data entry leads to the problems of 'recording overhead' and 'context switching'. In order to solve these problems, Sensor based automatic time data collection approach has been proposed but it has its own limitations i.e. Sensors can only record the user activity as long as the user is using the particular application for which that Sensor was designed. As soon as the user stop using that application and switches to some other activity (which cannot be captured by the Sensor), the sensor cannot record that information. Moreover, the Sensors for some applications such as 'Microsoft Office' can only tell that the user work on 'Microsoft Word' from time t1 to time t2. It cannot tell what was the user activity while he was using 'Microsoft Word'? Was he using it for 'coding' or 'designing' or 'architecture' purposes?
Our approach address the problems of 'recording overhead' and 'context switching' and limitation of existing approaches from a different prospective by asking the user to record the activity any time while doing that activity using Speech sensor. The two problems that arise in the Speech sensor time log collected using this way is that later one cannot infer when the activity started or ended and also for the activities which a user forget to record. To solve these problems, we then investigated how we can infer best information about user activity duration without asking the user to provide further information. We proposed using additional sensors to infer information about user activity duration. In order to utilize all the available information collected from the sensors (sensor fusion), we then proposed Time Log Processing algorithms which use temporal reasoning to calculate the activity duration. We also provided an example for an imaginary user 'Alice' and how 'Alice' can get benefit by using our approach.
