Chapter 7 Task-Stage Knowledge Support
7.2 Changes of task-stage
In this section, a correlation analysis method is proposed to identify the changes of worker’s task stage is proposed in this section. The objective of task stage identification is to identify the worker’s task stage and then deliver task-relevant knowledge according to the worker’s task-needs based on different stages. In general, three task stages based on the previous pilot studies: task pre-focus, task focus formulation, and task post-focus stages, are identified to differentiate the worker’s three types of information needs during task performance [76][77].
7.2.1 Stage identification process
The on-line task-stage identifier analyzes and determines the worker’s task stage based on his/her access pattern. The task temporal profile in each timeframe is the basis for identifying the worker’s task stage. A change of task-stage is inferred by analyzing the correlation of task temporal profile of the worker’s consecutive transactions. We now discuss the three steps that are executed to analyze and determine the worker’s task stage.
Step 1. Task-need pattern calculation: As we have addressed in Section 7.1.2, a worker’s task-need pattern can be expressed as a set of topics and associated
relevance degree. The task-need pattern of a session j in transaction i , j
i s T r a n s
P a tt , is expressed as a set of topics with the associated relevance degree (topicj
, rd
j). The relevance degree of a topic j, rdj, is derived by the cosine measure,( iSj, j) sim TransJJJJJJJJG JJJJJJGtopic
. The cosine measure of feature vectors is used as similarity measure.
•
Similarity measure: The cosine formula is a widely used similarity measure
to assess the degree of similarity between two items x and y by computing the cosine of the angle between their corresponding feature vectors xGand yG
, which is given by Eq. 2.2. The degree of similarity is higher if the cosine similarity is close to 1.0.
y
denotes the temporal profile (feature vector of weighted terms) derived from the documents accessed in session j of transaction i. Appendix A shows an example of j
P a tt is defined and generated similarly by considering the whole transaction.
Step 2. Correlation calculation: Once the similarity pattern has been derived, the correlation of the worker’s task-need patterns across transactions can be calculated by Pearson’s correlation coefficient. A reasonable assumption is that the worker’s task-needs will not dramatically change during consecutive sessions of the same transaction. Accordingly, we calculate the correlation between the previous transaction, Transi-1, and the start session of current transaction, Transis1, as shown in the equation below. Note that the time interval of a transaction is based on the characteristics of our research domain.
2 2
The changes of worker’s task stage are based on the correlation results of similarity pattern calculation.
Step 3. Task stage determination: The rationale behind the proposed correlation
analysis method is to identify the worker’s task stage based on the changes of
task-needs for topics, i.e., the task-need pattern in our work. In other words, some task-relevant topics within the DO may have a high degree of relevance to the temporal profile of the previous transaction, whereas some task-relevant topics may have a low degree of relevance to the temporal profile at the beginning of the current transaction. Because the correlation values are within the range [-1,1], it is easy to track the worker’s access pattern based on the correlation value between transactions.
We took around one to two years to observe the worker’s access behavior based on the correlation analysis method. And three correlation ranges are set based on the result of our sample analysis, which are “low”, “moderate“, and “high" correlation.
Low correlation: If a worker is in the early stage of executing a task, the correlation value between transactions will be within the interval [-1, 0.2), which indicates that he/she is in the task pre-focus stage and is uncertain about the perceived task.
Moderate correlation: If a worker has decided the research area, but feels uncertain about the research topic of a specific area, the correlation value between transactions will be within the interval [0.2, 0.5), which indicates that he/she is towards or in the task focus formulation stage.
High correlation: Once a worker has focused on a specific topic, the correlation value between transactions will be within the interval [0.5, 1.0], which indicates the worker dedicates his/her task-needs for specific topics. In other words, the worker accessed and read similar documents belonging to the related topics. Hence, the worker is within task post-focus stage.
Table 14. Task stage identification rule
Input: 1
: Correlation values between transactions
Transi-1.stage: Task-stage of i-1th transaction (previous transaction)
Output: Transi.stage：Task-stage of ith transaction (current transaction)
Case of Transi-1.stage
“pre-focus stage”: If 1
( 1, )
Transi .stage = task pre-focus stage
Else Transi .stage = task formulation stage;
“formulation stage”: If 1
( 1, )
Transi .stage = task post-focus stage Else Transi .stage = task formulation stage;
“post-focus stage”: Transi .stage = task post-focus stage;
Return (Trans .stage)
In this work, the time point to decide the worker’s task stage is the beginning of the current transaction, Transi, which is the worker’s current session. Assume we know the worker’s task stage for the previous transaction, Transi-1, we can then infer the worker’s current task stage from the correlation value. The task stage determination rule is given in the Table 14. In the following, we will explain the worker’s evolution pattern based on this rule.
7.2.2 Sample analysis
Example 3: In our domain, there are 36 task-level topics (historical task corpora).
Continuing with Examples 1 and 2 in Section 7.1, we analyze the four transactions of the access pattern of the executor “PoTsun”. Trans2, Trans3, Trans4 and Trans5 are sampled to explain how the worker’s task stage can be determined, as shown in Figure 11.
(1) Trans
2and Trans
3: The correlation between Trans
2 and the start session of Trans3 is -0.024 (low), which indicates there is a very different pattern between transactions. Because the worker’s task stage of Trans2 is the pre-focus stage, we can infer that the worker’s task stage of Trans3 is also the pre-focus stage.(2) Trans
3and Trans
4: The correlation between Trans
3 and the start session of Trans4 is 0.467 (moderate), which indicates that there are moderately similar patterns between transactions. Therefore, we can infer that the worker’s task stage of Trans4is the task formulation stage.
(3) Trans
4and Trans
5: The correlation between Trans
4 and the start session ofTrans
5 is 0.705 (high), which indicates that there are very similar patterns between transactions. Because the worker’s previous task stage is in task formulation stage and the correlation between Trans4 and Trans5 is high, we can infer that the worker’s task stage of Trans5 is in the post-focus stage.Trans 3
(S31, S32, S33) Trans 4
(S41, S42, S43, S44) Trans 5 (S51, …,…,) Trans 2
(S21, S22, S23) Corr=
-0.024 Corr=
0.467
Corr=
0.705
Pre-focus stage focus-formulation stage post-focus stage Fig. 11. Changes of task stages