Activity Based Ranking Methods For Personal Dataspace Search

In today’s world,knowledge work plays a very vital role in organizations of all types.The effective use of the organizational knowledge leads to the difference between the success and failure of any organization.Therefore,to gain the competitive advantage and to get upgraded with the flow of the society towards a knowledge economy,the modern organizations have changed their way of working as they used to work earlier.In order to speed up the innovation and development cycles,now the focus is on increasing the productivity of the knowledge workers.However,the typical knowledge workers face with the dilemma of information overload due to the rapid growth of digital information.Despite of the fact that enough information is available but still it is difficult to obtain relevant and useful information when required.To ease this information overload problem,recent literatures have suggested a reduction in the duplication of information and the adaption of personal information management strategies together with intelligent software solutions and further the provisioning of value-added information.Previously,a number of efforts have been made to support the work of knowledge workers by employing different indexing and search tools that make information more accessible.However,recent analysis concludes that to be actually supportive all these systems have to take users context into account.Awareness of the users current context,i.e.,the ’activity’ that user resides in allows various ways of support their work.In this regard,so far many research works have focused on examining approaches that are capable of identifying and managing the activities of a user.However,the lack of correctness in the representation of the users’ activities as well as high set-up and maintenance costs in terms of training or learning to use these new systems are still a hurdle for their wide adoption.These findings among others motivated this research,the goal of which is to automatically recognize the user activities on a desktop computer system in order to enable an activity-specific support.Specifically by pointing to the open research questions,this thesis investigates the subareas of(i)User interaction logging,(ii)Searching context,(iii)Activity detection,(iv)Implicit feedback mechanism and(v)Activity based linkage and ranking of resources.Overall,this thesis proposes a novel activity assistance model which is:firstly,log system wide users’ desktop interactions in a completely automatic,unobtrusive and privacy preserving way.Second,present a novel approach to link and rank desktop resources by analyzing users’ activities over time,and exploited the associative links of resources from their implicit access patterns.Third,propose the multiple ranking methods,such as frequency of access,focus time,recency of access and connectivity of resources.Fourth,outline the basic framework of searching context as:Frequent users and infrequent users Context;and Working and Life Context.Fifth,propose the new different ranking algorithms by applying the implicit feedback mechanism for both searching contexts:Frequent users and infrequent users Context and Working and Life Context.Sixth,compare the performance of newly developed algorithm for four quadrants as:Life and frequent user,Work and infrequent user,Life and frequent user;and Life and infrequent user.Finally the ranked list which is now ready for re-use are proactively presented to the knowledge workers through a user interface.Specific contributions of this research effort are described below in the order they are presented and discussed in thesis:1)This thesis proposes an approach for logging user desktop interactions.The proposed mechanism employs implicit feedback method for recording the user actions and instead of being constrained to a limited set of applications,our data collection model logs cross application data from system wide user interactions.Moreover,the mechanism allows capturing desktop interactions in a completely automatic,unobtrusive and privacy preserving way;and users do not need to perform any explicit effort at any stage.We expect that effectiveness of this logging process will enable better insight into desktop user behaviour and will help discover guidelines for future personal information management systems.2)We introduce a searching context detection mechanism that takes advantage of temporal aspects of user activity behaviour and address the issues pertaining to automatic identification of user activities from their everyday desktop interactions.The proposed mechanism is based on classifying the type of users’desktop actions,and in contrast to previous approaches;it does not enforce to choose the type of searching context by hand by individual user,it automatically emerge from the logged activity data for that specific user.Moreover,we envision that this ability to distinguish between activities can be very useful in activity support and archival systems.3)Our work on machine learning based activity context detection presented first evidence that few features;particularly interaction based features are sufficient to accurately classify the context of desktop activities.The strong positive influence of specific interaction features indicates that it is not necessary to sense "everything"about the user’s desktop interactions but only some relevant elements.This finding can have an impact on what kind of sensors have to be developed for activity recognition purposes in general.Moreover,our results give a first impression on the fact that the environment in which the users perform their activities has no significant influence on the activity detection performance.A comparison among four type of context demonstrated that the performance of system is better for frequent user context than infrequent users.Despite the differences in desktop work behaviour of different users,standard machine learning techniques are able to accurately classify the context of user’s desktop activities.4)A major limitation in our proposed techniques for ARS(Activity Based Ranking System)searching system involves the decision related to retrieval performance which is based solely on the procedure of searching of queries as well as document collection.Due to this,information about the history of searching query and searching context as:actual user type and user activity type,is still lacking here.Hence,we study and propose solution to explore implicit feedback information incorporating all the previous queries history and click through information that helps in improving and enhancing the searching retrieval accuracy.To obtain the final unique combined ranked,we use the context sensitive retrieval algorithms proposed by(Shen et al.,2005)by merging all the clicked activity summaries and respective previous queries of the users with the present query to proceed better ranking of activity for our updated CARS(Context Based Activity Ranking System)searching system.More importantly,experiment results illustrates that retrieval performance can be substantially enriched and improved by using implicit feedback,especially the clicked document summaries.5)We provided new innovative and technical grounds for research into activity-specific desktop support,and presented an innovative solution using linkage and ranking of desktop resources.We conducted to alleviate desktop activity support deficiency,which occurs due to the lack of links between desktop resources.The research exploited information such as associations,contexts,and activity information about accesses to local resources,and translated this information into a personal linkage structure.More specifically,the research presented a novel approach to link and rank desktop resources by analyzing users’ activities over time,and exploited the associative links of resources from their implicit access patterns.Furthermore,multiple ranking methods,such as frequency of access,focus time,recency of access,and connectivity of resources,were proposed.Two prototype systems were also developed,and a user study was conducted to validate the effectiveness of the proposed methods.The results showed that ranking desktop resources by their relevance-as carried out in this research-could improve activity-specific support,as well as overall performance in the area of personal information management.Overall,we found our results highly encouraging and believe that they constitute a significant step towards solving the activity ranking and support problem in a completely unsupervised,unobtrusive and automatic manner.Moreover,current proof-of-concept implementation of our method delivers early indications that the system actually has the potential to assist the knowledge workers without implicating any additional burden on them.