Section 4.1 describes how our system captures and stores users' access behavior (document accessed). Section 4.2 illustrates the proposed self profile adaptation approach that considers the effect of time factor and the user's behavior (document accessed) to adjust the corresponding task profile with the aid of task-based topic taxonomy.
4.1. Capturing users' access behavior
Our K-support system records workers' knowledge activities during the execution of their works in previous research [19]. Whenever a worker performs an action about accessing any document, the system creates a new record to store the information of corresponding knowledge activity. In the following, an example is shown to explain how to capture and store users' access behavior.
Example: Assuming that a worker "Mrkid" is searching for knowledge
documents in K-support system, and he finds that a document "Learning User Interest Dynamics with a Three-Descriptor Representation" may help his task. "Mrkid"performs a "reading" action at time "2005-10-31 21:05:03" accordingly, and with the help of K-support system, the information about the "reading" action is recorded in the system.
In the above example, the stored information is {"Mrkid", "2005-10-31 21:05:03", "reading", "Learning User Interest Dynamics with a Three-Descriptor Representation"}. All attributes are converted properly into identifiable number except the 'time' attribute.
Hereafter, we use the word 'event' to denote an action performed by some user about accessing any document. In this research, only four kinds of event are adopted, including "download documents", "download reports of documents", "read documents on-line", and "upload documents".
4.2. Self profile adaptation
Whenever an event of worker's access behavior is detected, the system captures and records the document accessed by the worker. The event triggers the self profile adaptation process to adjust the worker's task profile according to the information of the corresponding event.
A document/task/topic profile specifies the weighted concept terms of the document/task/topic. Vector-based approach is adopted to represent a document/task/topic profile. A modified relevance feedback technique, adopted from the techniques proposed by Rocchio (1971), is used to adjust the workers' task profiles based on the profiles of documents accessed by the worker and the topic profiles of identified relevant topics. The adjustment considers the effect of time factor.
The proposed profiling technique is given in E.q. 4.1, and E.q. 4.2. The associated definitions of symbols and parameters used in the equations are listed in Table 4.1. Let T denote the index of the actual time when the worker performs the latest action of document access.
S
rT+1denotes the worker’s task profile generated at time T, which can be used to model his/her task-needs at time T+1. The equation includes the decay of previous task-profile SrT
, which models the worker’s task needs at time T, and the current information needs derived from the document accessed at time T.
) (
S
TDecay
vrepresents the accumulated task needs from the beginning to the current time T by considering the time decay of previous task-profile.
[
T T]
Table 4.1 Definitions of symbols and parameters used in the equations T
the index of the actual time when the worker performs the latest actionof document access
+1
S
rTthe task profile generated at time T, which can be used to model the worker’s task-needs at time T+1
D
vTthe document profile of the document accessed by the worker at time T
OvT
the aggregate topic profile derived from the topic profiles of relevant topics and irrelevant topics
The self-adapted task profile
S
rT+1is generated from previous task profile
S
rT applied with a decay function and is refined by using the current information needs derived from the document accessed at time T. The current information needs consists of two parts: the document profile and the aggregate topic profile. The documentprofile D
vTintuitively is the profile (feature vector) of the document accessed at time T.
Task-based topics play as important references of past experience to adjust task profiles according to their relevance (similarity) to the document accessed by the workers. The relevance degree of a topic Oi to the document DT is obtained by calculating the similarity (cosine measure) between Ovi and
D
vT. The aggregate topic
profile is derived from the topic profiles of relevant topics in the positive topic set and
irrelevant topics in the negative topic set. The positive topic set reflects the positive information needs of the worker, and is obtained by selecting the topics with relevance degree higher than a defined threshold. The negative topic set reflects the negative information needs of the worker on the topic taxonomy and is obtained by selecting the topics with relevance degree lower than a defined threshold. Our previous research shows that topic profiles are more important to adjust task profiles than documentO
i the topic i in the topic taxonomyT
Opos the positive topic set derived at time T base on the relevance degrees of the topics to the document accessed at time T
T
O
pos the number of topics in OTposT
Oneg the negative topic set derived at time T based on the relevance degrees of the topics to the document accessed at time T
T
O
neg the number of topics in OnegTt
the index of an actual time when the user performs an action of document access; t=0 denotes the time when the task startsT
TW
t, the time weight of the event that occurred at time t (with respect to timeT)
ST
the starting time when the worker's task starts (in milliseconds) α the tuning parameter used to adjust the weight of previous task profile λ the tuning parameter used to adjust the weights of topic profile OvT anddocument profile
D
vTβ,γ the tuning parameters used to adjust the weights of positive topic set and negative topic set
profiles during the early phase of task executions [41], so a parameter λ is used here to adjust the weights of the document profile and the aggregate topic profile. Fig. 4.3 illustrates the given technique.
The profile adaptation also takes the effect of time factor into consideration.
) (
S
TDecay
vrepresents the accumulated task needs from the beginning to the current time T by considering the time decay of previous task-profile, as given in E.q. 4.2.
S
rT denotes the previous task profile generated at time T-1, and plays the role of previous task needs.S
rTis the summation of aggregate topic profiles and document profiles derived from time ST to T-1. Generally, the more recent the document accessed the more important it is to reflect a work’s current task needs. Thus, a time weight is employed to reflect the decay of previous aggregate topic profiles and document profiles towards their contribution to worker’s current task needs.
TW
t,T is the time weight of an event occurred at time t with respect to T, and is defined as the ratio of time difference t-ST to T-ST. Thus,Decay
(S
vT)can reflect the effect of previous task profile towards current task profile more accurately with TW than just using
S
rT.
Fig. 4.3 Example of modeling task needs
accessed document:
D0000000001
positive topic set:
T0000000001 T0000000002 T0000000003
negative topic set:
T0000000007
accessed document:
D0000000547
positive topic set:
T0000000011 T0000000004 T0000000002
negative topic set:
T0000000006
accessed document:
D0000000046
positive topic set:
T0000000002 T0000000007 T0000000005
negative topic set:
T0000000004
Time d Time e Time f
[
2 7 5 4 46]
1 Decay(S ) [ (O O O )/3 O ] (1 )D Svf+ =α× vf + λ β v + v + v −γv + −λ v