Chapter 5 Collaborative Task-Relevance Assessment26
5.4 Task-based K-Support based on assessment
5.4.1 Task profile generation by relevance feedback technique
The task profile of an executing task is specifies key subjects of the executing task, which represented as a feature vector of weighted terms. The task profile is initially derived via analyzing the task contents, or alternatively using the corresponding task corpus. Moreover, the collaborative task-assessment identifies the relevance degrees of referring tasks to the executing task. The result is used to further construct or refine the task profile of the executing task based on the relevance feedback (RF) techniques introduced in Chapter 2.
The RF technique employs the process of reformulating or expanding the original query based on partial relevance judgments, i.e. feedbacks on part of the evaluation set. The RF technique is adopted in this work to construct and refine the task profile of the executing task. Two kinds of relevance judgment on referring tasks are considered: positive feedback and negative feedback. The standard RF technique employs binary feedback without considering the degrees of relevance, as shown in Eq. 5.1. Relevant tasks with positive feedback give positive influence on the weights of terms occurring in relevant tasks. The irrelevant tasks with negative feedback give negative influence on the weights of terms occurring in irrelevant tasks. A refined task profile can be generated by adding the term weights of relevant tasks and subtracting the term weights of irrelevant tasks. Consequently, the feature vector of new term weights derived based on the RF technique forms a new task profile for further knowledge retrieval. The idea of relevance feedback shifts the new profile closer to the relevant task set and away from the irrelevant task set. The parameters β and γ are used to determine the relative amount of influence of the relevant task set to the irrelevant task set.
This work modifies the standard Rocchio and Ide_Dec_Hi methods by considering the relevance degrees of referring tasks obtained from fuzzy linguistic assessment. The modification considers the relative importance of relevant (positive) and irrelevant (negative) tasks from the perspective of users. The feature vectors of referring tasks are multiplied with their relevance degrees to reflect their relative contributions in the refinement of the task profile, as expressed in Eq. 5.2. Detailed formulations are described as follows.
Two approaches are proposed for constructing the task profile
s G
eof executing task te based on the relevance assessment introduced in the previous section. The binary relevance assessment method, denoted as B-RA, conducts binary (relevant and irrelevant) assessment. The fuzzy linguistic relevance assessment method, denoted as F-RA, considers the relevance degree based on user perceptions.
B-RA: initial
represents the initial profile derived from analyzing the collected relevant documents for the executing task, if available. Moreover, Tr denotes the set of relevant tasks selected from positive referring tasks according to collaborative assessment of experts and workers. Tn represents the set of the last-M irrelevant tasks that are selected by the system automatically (t2 and t6, in the previous given example, Table 3). Furthermore, tj
G
is the task corpus of task tj with an associated weight wtjrepresenting the relevance degree of tj to the executing task. wtj is set to AE(tj), which is the aggregated relevance rating of task tj to the executing task. AE(tj) is derived from the task assessment procedure illustrated in Section 5.1.2, and α, β and γ are tuning constants.
The task profile of the executing task te, derived from Eq. 5.1 or 5.2 can be expressed as a feature vector of weighted terms, SGe
= <w(k1, te), w(k2, te), …, w(kn,
t
e)>, where w(ki, te) is the weight of a term ki in representing the main subjects of te; n denotes the number of discriminating terms. Meanwhile, SGeis used to retrieve relevant codified knowledge from the repository.
5.4.2 K -Support: task-based knowledge retrieval
A task-based knowledge support system can be realized with the proposed systematic profile modeling approach. The generated task profile is the system kernel that streamlines knowledge retrieval activity for further realizing task-based knowledge support. A task profile specifies key subjects of the executing task, and is constructed to model the information needs of knowledge workers during task execution. Based on task profiles, the system can recommend/retrieve relevant
knowledge from the repository to assist knowledge workers. Workers conduct further search activity are assisted by the highly correlated term set presented in the system interface. The relevant knowledge includes relevant tasks, associated peer groups, relevant documents, and highly correlated term set.
The similarity measures between the executing task and the codified knowledge items can be calculated to select Top-N relevant tasks or documents from the knowledge repository. The key contents of a codified knowledge item (task or document) are represented as a feature vector of weighted terms. The task profile is also expressed as a feature vector of weighted terms. The cosine measure of feature vectors described in Section 2.2.1 can be used to derive the similarity measure.
Moreover, the task profile can be further adjusted during task performance by monitoring the workers’ feedback behavior. The most task-relevant codified knowledge can be retrieved based on the adjusted task profile to fit the worker’s current information needs. We also presented an adaptive task-based profiling approach to model workers’ dynamic task needs. The task-based peer-groups can be analyzed from the retrieved relevant task set to provide knowledge sharing. Details for identifying task-based peer groups to support knowledge sharing are given in next chapter and can be found in our recent work [47].
Relevant Tasks Recommendation. As the task profile
SGe(feature vector) of the executing task te has been derived by B-RA or F-RA method, retrieving relevant tasks for references will be helpful. The cosine measure of SGe
and tGj
, namely,
cosine(
SGe,
tGj), is calculated as the similarity measure between the executing task
and task tj. Notably, SGeand tGj
are the feature vectors of te
and t
j, respectively. Tasks with top-N similarity measures are selected as the relevant tasks for recommendation.The relevant tasks and associated knowledge workers engaged in these relevant tasks are recommended for consultation. Effectively codifying tacit knowledge may be difficult. However, the system can locate valuable knowledge sources such as knowledge workers engaged in relevant tasks, providing a knowledge support platform for gathering and exchanging task-relevant knowledge among workers.
Relevant Documents and Term Recommendation. The relevant documents are
retrieved using the profile of the executing-task. Similarity measurement is also adopted to select top-N relevant documents. Let tGjare the feature vector of document dj. The cosine measure of SGe
and dGj
, namely, cosine(SGe
,
dGj), is
calculated as the similarity measure between the executing task and document dj. Documents with top-N similarity measures are selected as the relevant documents for recommendation. Meanwhile, the important term set representing the main subjects of the executing task is derived from the constructed task profileSGe
. The system displays the discriminating terms and their associated weights to assist knowledge workers with further retrieval. The term set forms the task corpus of the executing task, and can be modified during the subsequent stages of task execution.
R elevent D ocum ents
R elevent T asks
K eyw ords for R etrieval
P eer G roup T a sk -relev a n t
K n o w led g e S o u rce
Fig. 4. Task-relevant knowledge source (explicit or tacit knowledge source)