Section 3.1 presents the problem formulation. Section 3.2 describes an overview of the process of mining task-stage needs. The terms and notations used in this work are defined in Section 3.3.
3.1. Problem Formulation
This work broadly refers to a task as a unit of work in organizations, such as a project, research work, or activity. Moreover, this work uses profiling approach to model workers’
task-stage needs, namely information needs (profiles) on task-stages. Task-stage profiles are generated to model worker’s task-stage needs and are used further to provide task-relevant knowledge at various task stages. A task-stage profile specifies the key subjects of a task stage and can be represented as a feature vector of weighted terms. The profile is used to retrieve relevant codified knowledge in the repository. The key contents of codified knowledge (textual data; documents), namely document profiles, are also represented as a feature vector of weighted terms. Relevant documents can be retrieved to provide knowledge support for task execution according to the similarity measures (e.g. cosine measures) between task-stage profiles and document profiles.
A task class is defined to specify a type of tasks with similar properties (similar tasks, for brevity). Task classes are identified based on existing tasks. A task that belongs to a task class is called a task instance of the task class. Task class/instance resembles the concept of object class/instance. Tasks in different task classes generally have different task-stage needs, while tasks belonging to the same task class may have similar task-stage needs. Accordingly, task-stage profiles generated by analyzing existing similar tasks of the same task class are useful to provide needed relevant knowledge in supporting the execution of on-going tasks.
The term virtual task is used to represent a task class of similar tasks. A virtual task may have more than one task instances. For example, a virtual task “Research on recommender systems” has several task instances including “Hybrid approaches for product recommendations” and “Comparisons of collaborative filtering for recommendations”. Each existing task instance is associated with documents accessed/generated during task performance. Accordingly, the problem of identifying task-stage needs of a virtual task is described as follows.
Given a virtual task and a set of task instances with associated documents, find the number of task stages and the corresponding profile of each task-stage.
3.2. Process of Mining Task-stage Needs
This work employs information retrieval techniques to conduct text processing and data mining techniques to analyze workers’ task-needs, which are modeled as profiles at different task-stages. Figure 2 illustrates the process of mining task-stage needs. The operational procedure is described as follows.
The document database and the system log record the historical tasks with associated documents and usage data. The documents of a task are preprocessed and organized as a task document sequence (TDS) according to their accessed/generated time during task performance. Notably, each task has its own task document sequence.
Clustering and information retrieval techniques are then employed to cluster the task document sequences of similar tasks belonging to the same task class. A task-stage hierarchical clustering (TSHC) algorithm is proposed to cluster the task document sequences based on similarity measures and retrieval time of documents. Each cluster represents a task stage with associated documents of a task. More details are addressed in Section 4.2. Finally, the feature vector of each task stage is extracted from each cluster of documents to construct the task-stage profile. The task-stage knowledge retrieval module can then retrieve relevant codified knowledge (documents) to provide knowledge support for task execution according to the similarity measures (e.g. cosine measures) between task-stage profiles and document profiles.
Figure 2. Process of discovering task-stage needs
3.3. Term Definition
This section lists the definitions of terms used in this work.
(1) Definition I: Task
The task is the fundamental unit in business. This work broadly refers to a task as a unit of work in organizations, such as a project, research work, or activity.
T: task set A task is either an executing-task (on-going task) or an existing task in the task-based working environment. T = {t1, t2,…, tr, …, tn}
tr: existing-task An existing-task is a historical task accomplished within the organization.
tv: virtual task A virtual task represents a task class of similar tasks
task instance A task instance of tv is an existing-task that belongs to the class of tv.
Tv: set of task instances
A set of task instances of the virtual task tv
(2) Definition II: Task Document Sequence
A task document sequence is a sequence of documents retrieved (accessed or generated) during task performance. Documents are sorted according to their retrieval time. Each task instance has its own task document sequence.
TDS(tr): Task Document Sequence of tr
A sequence of documents accessed/generated while conducting a task tr. TDS(tr)=<d1, d2,…,dm>
(3) Definition III: Task Stage
Task document sequences of task instances are clustered into stages based on similarity measures and retrieval time of documents. The clustering result forms task stages of a task.
TS(tr): task-stages of a task instance
A task tr comprises several task stages. TS(tr)=<tsr[1], tsr[2],…, tsr[k]>, where tsr[i] denotes the task-stage i of tr .
TS(tv): task-stages of a virtual task
Task-stages of a virtual task tv which are derived from task stages of task instances in Tv. TS(tv)=<tsv[1], tsv[2],…, tsv[k] >
(4) Definition IV: Task Stage Documents and Profiles
Each cluster of documents represents a task stage with associated documents, named task stage documents. Documents in task stage k of task tr is denoted as tsr[k].docs, while documents in task stage k of virtual task tv is denoted as tsv[k].docs. Each task-stage profile can be derived from the feature vectors of documents in each task-stage. Let tsr[k].profile denote the profile of task stage k of task tr, and tsv[k].profile denote the profile of task stage k
of virtual task tv. tsr[k].docs: task-stage documents
Documents in task stage k of task tr.
d.profile The feature vector (profile) of document d.
tsr[k].profile: task-stage profile
The profile of task stage k of task tr. A task-stage profile is the vector obtained by averaging the feature vectors of documents in tsr[k].docs.
tsv[k].docs: task-stage documents
Documents in task stage k of virtual task tv.
tsv[k].profile: virtual task-stage profile
The profile of task stage k of virtual task tv. A virtual task-stage profile is the vector obtained by averaging the feature vectors of documents in tsv[k].docs.