Transparency - Activity Theory-Guided Support for Information Work Benedikt Schmidt, Sebastian Döweling SAP Research Bleichstr. 8, 64283 Darmstadt, Germany firstname.lastname@sap.com ABSTRACT is agreed that information work is characterized by two Information work is characterized by non-routine problem features [7]: 1) information work does not follow pre- solving and a highly context-dependent execution of work defined processes, 2) information in any form is both the processes. Moreover, frequent switches between work items central work resource and the outcome of information are typical for this kind of work, making frequent work. adaptation and reevaluation of the work context necessary. While the focus on (often digital) information seems to As the work structure emerges ad-hoc, it is difficult for the work well with IT support systems, both the lack of a clear information worker to keep track of all ongoing activities. definition and of pre-established work structure make it Yet, the non-standardized execution of work processes difficult to build tools that support information workers. As prevents the use of common workflow management a result, there is a tendency in most current tools (e.g. [6]) systems. For this very reason, existing tools for information to focus on the management of information resources only, work support often focus on support for the organization of effectively providing little to no support for structuring information objects, but neglect the work process work or improving work process awareness for self- perspective. optimization. In this paper, we employ a systems design method based on We argue that this lack of support stems at least partly from activity theory to elicit requirements for knowledge work the lack of adequate theoretical underpinning, and that it tools, focusing on interruptions and a loss of work can be addressed using activity theory, a framework that overview. We address these requirements with a tool named has recently received increasing attention by the HCI and Transparency that improves information work process CSCW communities, to analyze knowledge work. Activity awareness. We present two design cycles with real world theory has features that make it particularly useful for such evaluations. an endeavor. In particular, it provides a contextualized understanding of HCI that stresses the interdependencies Author Keywords between humans, the tools they use and the artifacts they Activity Theory; Systems Design; Context; Knowledge use these tools on. Work In this paper, we analyze information work with a recently ACM Classification Keywords developed activity-theory based system design method [10] H.5.2 User Interfaces: Graphical user interfaces (GUI) which extends the human centered design cycle [4]. Based on a literature review and the results of two previous studies INTRODUCTION on information workers [8], we derive requirements for an Information work, despite its acknowledged relevance for information work support tool via the named method. The today’s economy [7], is a concept that is hard to define tool Transparency tackles interruptions and a loss of work precisely. More than often, it is used rather as an abstract overview by improving information work process label, encompassing a large variety of job roles (e.g. awareness. We report on two design cycles with engineers and construction managers, but also accountants accompanied evaluations. Finally, we conclude with a and countless other workers in modern enterprises). Yet, it discussion of the evaluation results and the implications on the employed design method, and identify directions for Permission to make digital or hard copies of all or part of this work for further research. personal or classroom use is granted without fee provided that copies are not made or distributed for profit or commercial advantage and that copies DESIGN PROCESS bear this notice and the full citation on the first page. To copy otherwise, or republish, to post on servers or to redistribute to lists, requires prior The following section describes and underpins the specific permission and/or a fee. application of the activity theory based system design model (AT-SDM) to the domain of information work at the computer workplace. Most existing human-centered design design is proposed which is evaluated. If the evaluation methods (e.g. ISO 9241-210) apply task analysis to identify shows deficits the upcoming knowledge enriches the requirements for a system. Focusing on task analysis is a system model and a new cycle starts until the evaluation is convenient method for work tasks that have stable successful. As the initial AT-SDM is not product of a workflows and can be standardized well. For information design evaluation, one starts with a data grounding that is work we see problems, as occurring workflows highly deduced from literature or initial studies. differ and work emerges as two factors interplay: 1) context demand and personal decisions, e.g. an individual works on TRANSPARENCY two projects, recognizes demands to write reports to certain The data grounding for the design process is provided by deadlines and then autonomously structures the personal two sources: a literature review and two field studies. work accordingly. Such an open setting demands the Initially, we reviewed literature about knowledge work integration of factors like the individual, the autonomy and execution and organization. We saw a gap with respect to the work context (e.g. projects) into the design. workflows, context and individual awareness of work Additionally it seems to be important to identify the effects structure. Therefore, we have conducted studies focusing on of changes to the complex work setting that result from a the gap, using interaction histories and questionnaires [8]. newly introduced tool. As Schultze put it: when a new The resulting structure can be understood as follows. Based system is developed, one needs to identify “what the doing on a network of motives, the subject anticipates goals which does” [9]. are objectified by an outcome. To realize the outcome solution techniques, so called knowledge actions are applied which are executed by fine grained interaction with Motive network the computer, so called desktop operations (for details, see [8]). 1…* 1…* For information work, the outcome generally is an Multiple task activity system model information artifact or a change in the world which is A) organizational A) internal or external triggered by information. The information artifact is an context information state B) spatial context C) social context B) information artefact obvious work result: the externalization and organization of Information Worker D) environmental context A) mental model information in encoding systems which are stored or E) work context B) relevant information as artefacts or knowledge disseminated using a medium. The change triggered by the information is more complex to understand. It is the effect A) corporate culture of information in a subject which is closely related to the B) expert culture A) process autonomy B) process heterenomy effects of speech acts on real world activity (referred to as C) deadline A) higher mental functions perlocutionary and illocutionary acts) described by Austin (incl. mental model) B) computer to access, [1]. As the information is consumed by subjects it interacts transform, create and disseminate information with the cognitive processes which encode it and may 1 change the behavior directly or indirectly. 1…* Knowledge actions as work proactices To realize the outcome anticipations, an activity system is constructed for each task, which shows the relation of the 1 subject to the context of the task, the mediators (rules, tools, 1…* workflow) and the object. Each system produces the Desktop operations outcome based on the object. For the information worker the object generally is a mental model of involved domains Figure 1: The AT-SDM for knowledge work. and the relevant information existing as artifacts of knowledge. All other elements involved in the activity AT-SDM builds upon activity theory which provides a system are visible in figure 1. Information workers use vocabulary to model connected systems and the interplay of higher mental functions as well as information and the different system elements. For AT-SDM the system communication technology (tool mediator) in work consists of subject, instrument, object, context, workflow processes which are largely autonomously identified and rules. AT-SDM covers the first two elements of the (workflow mediator) following rules like the corporate human-centred design cycle [4]: the context of use analysis culture and the expert culture (rule mediator) to realize the and the requirement engineering. To design a tool using the outcome based on the object. The context is a complex human-centred design cycle with AT-SDM as method for combination different context factors (including but not context of use analysis and requirement engineering follows limited to organizational, spatial, social, environmental and this process: first an activity system is constructed to model work context). the context of use, second a tension analysis helps to identify required modifications which result in We have conducted a conflict analysis on the structure and requirements. Third, based on the requirements a system identified different threads that emerge in a structure of many parallel maintained task activity systems. Here, we Thus, decisions for interruptions as well as the task only report on two main aspects: prospective and switching process itself can be supported, as the user retrospective memory failures and show their relation to directly knows what he has done, can recreate the required interruptions. working environment and keep track of the spent time. We do not follow the idea of avoiding interruptions by hiding Interruptions occur frequently for information workers. information, as most interruptions come from the individual Interruptions are “the process of coordinating abrupt and are not externally triggered [5]. changes in people’s activities” [5]. This statement already shows that interruptions are required. Information work is The Transparency tool logs the user-system interaction challenged by multitasking. Therefore they have to based on software APIs and hooks, i.e. . The collected data coordinate many coexisting task activity systems which are is input to a pipeline that creates activity data as sets of used executed in parallel or in rapid succession. The switches applications and the accessed information objects. among the different activities (as represented by AT-SDM) Heuristics classify the activity data sets (e.g. identify are largely caused by interruptions. Even before a goal is authoring or consuming). The system uses a clustering completed, an information worker might be interrupted. He algorithm to identify activities that belong together (based decides on the parallel activation of another activity system on temporal and semantic relatedness). or on a switch of the activity system. One can distinguish between internal and external interruptions. Internal interruptions are decisions of the subjects while external interruptions are events that interrupt the work process. Studies have shown that interruptions are evenly distributed among internal and external interruptions (Czerwinsky [2] talks about 40 % self-initiated interruptions). Each interruption triggers the identification and the activation of a new task activity system. The task activity system which is active when the interruption occurs needs to be kept active additionally or needs to be rehearsed. As a result, the likelihood of prospective and retrospective memory failures is increased. Prospective memory failures stand for forgetting outcomes and the respective systems. Retrospective memory failures stand for problems in the Figure 2: Graph representation of the executed work. activation of an activity system as its status is not remembered anymore. The large amount of tools and Design: The Transparency 1.0 system offers two information objects involved in activities and the cognitive visualizations. The situation view and the management work on executing and controlling the system status once view. The situation view shows a graph representation of work was interrupted threats successful work executions. user activities (see Figure 2). Nodes represent work on an information object with an application (e.g. authoring Due to deadlines and priorities interruptions are used to “casfe.doc” with Microsoft Word). Based on the node, a organize work executions. Overall, interruptions are user is working on, and its neighbors, recommendations are necessary to coordinate information work, but they increase generated. the threat of prospective and retrospective memory failures. Support has been proposed: systems that have information The management view allows the user to identify relevant about the user activities to 1) block certain information (to related activity data (e.g. all activity data around a travel restrict external interruption) [3] and 2) manage relevant planning), and to manage it, i.e. to have access to the time resources [6]. As these tools require additional manual spent with the activities, the work sequence and to simplify effort during work execution, the main problem of such accessing the elements. tools is that they themselves create interruptions and Setup: For the evaluation of Transparency 1.0, eight users generate a conflict between subject, process and object. To were recruited using convenience sampling. 6 were male, 2 solve the conflict, we see a demand for tools that improve female, their age between 26 and 38. Users were either awareness of personal activities without requiring manual researchers or managers and had significant IT experience. effort during task execution. Their work included a high degree of self-organization, involvement in multiple projects and commitment to an First design iteration expert culture, thus fitting the profile of the information For the first design iteration of a tool to support knowledge worker very well. None of them had used Transparency work, we focused on improving awareness of the personal before. Users tested Transparency 1.0 for two weeks during work process, to support managing work related their daily work activities (i.e. for 10 work days). information and switching between different work items. At the beginning of the study, they received a interacting with transparency; this may, however, be the demonstration of Transparency’s features and were asked to effect of transparency being perceived as complementary to fill out one questionnaire regarding their personal working existing work organization procedures. A complementary style and one regarding their impression of Transparency. tool that requires additional work efforts (especially due to They were asked to complete the latter again, after the the complex visualization) tends to be rejected due to a study was completed. Additionally, an unstructured conflict in the AT-SDM between subject, workflow and interview was conducted after the study. We evaluated the instruments. We integrated this additional knowledge into interviews using topic coding. the AT-SDM, concluding that a tool that aims to support focus in work and an improved insight into personal work Result: The questionnaire on working styles found that 2 not only needs to improve the data collection, but also users consider their personal work style as unstructured, 3 needs quickly consumable visualizations that seamlessly as partly structured, and 4 as completely structured. 2 are integrate with the existing efforts of structuring work (6 of rarely confronted with external interruptions, 4 are 9 persons already did task management). occasionally confronted w. external interruptions, 2 are frequently confronted w. interruptions. Most (6 of 8) state Second design iteration that they sometimes have difficulties to remember all things The major goal of Transparency 2.0, the second iteration of they worked on. 6 users already used tools to organize our design, was to improve the integration into existing work, mostly using sticky notes and Outlook tasks activities in users’ daily work. Therefore, we focused on The initial questionnaire on Transparency showed a general one function in particular: activity data enriched task appreciation for the presented features. Most participants management. The tracking of user interaction and the expected the graph representation of the work to be useful automated clustering of related activities remained or very useful (6 of 9). The attached recommendations were unchanged from Transparency 1.0, but were combined with rated moderately useful or useful (8 of 9). The participants features and a user interface for task management. especially expected an improved insight into their personal Design: To provide a convenient task management UI, we work style (moderate impact (2 of 9), relevant impact (1 of extended an existing system called tasks.show1 with 9), strong impact (1 of 9)). Most did not assume that the capabilities to enrich tasks with activity data, to access tool will help them to focus their work (small impact (5 of information objects, and to get detailed information about 9), no impact (1 of 9)). Activity mining and the possibility the individual work process. The main entry is a task list w. to save collections of related activities to keep track of the standard features (tasks belong to categories, have due dates time spent on these activities were considered an interesting and time estimates - see Figure 3.1). Additionally, each task feature; yet, most users did not see it as very relevant (no can carry activity data. The activity data is presented in the one chose the management view as preferred functionality). form of related information objects, statistics on the time After two weeks of using the tool in a normal work context, spent on activities (see Figure 3.2), and a graph different problems became apparent. The appreciation of representation of the interactions. the graph representation for the work decreased (not useful 1 (2 of 9), partly useful (2 of 9), moderately useful (2 of 9), 1 useful (3 of 9)). 5 users considered reading the graph to be very complex. In the interview, all participants stated that they found the graph view interesting, but did not find a 2 connection to their daily work, as it was time consuming to 1 interact with the visualization, esp. due to its size (after 8 hours of work a graph sometimes contained more than 100 nodes). Few participants saw the possibility to improve insight into personal work – and, as stated in interviews, this was directly connected to the complexity of the graph. Some stated, however, that they started to be more focused in work due to the visualization (5 of 9). Saving performed Figure 3: Transparency 2.0: 1) Task List with activity data for activities to keep track of the time spent and the improved one task 2) Statistic about time spent with the resource. quick access to information access was an aspect that was mentioned frequently in the interview as an “unfinished To relate tasks and activity data two paths exist: 1) Users feature”. The participants saw it close to task management can browse clusters of related activities - extracted from but they missed functionalities like due dates and task their personal work history - and create new tasks based classification. from activity data. 2) Users can manually add information The participants showed a general appreciation of the design ideas and features. The resentment shown by the 1 participants was directly related to the complexity of http://code.msdn.microsoft.com/windowsdesktop/TasksShow-1bf01c8d objects to tasks; the system automatically enriches the Most participants explained in the interview that they would respective activity data and proposes related information start working with the tool on a daily basis, given objects based on extracted activity clusters. integration with the existing tool landscape was improved (synchronization of tasks with Outlook, support for the detection of more/other accessed information objects as e.g. Setup: For a first evaluation of Transparency 2.0, 4 users sites visited in Google Chrome). were recruited, again using convenience sampling. 3 were Furthermore, in the interviews two participants stated that male, 1 female, their age in the between 25 and 40. Users they miss additional information retrieval functionalities. had similar characteristics to the participants of the first We plan to address this in a third design iteration. study. In fact, 3 of them had already participated in the first study. Users tested Transparency 2.0 for four days during REFERENCES their daily work activities. While there was a demonstration 1. Austin, J.L. How to do Things with Words. of Transparency’s features again, no questionnaire was Harvard University Press, Cambridge, Mass., handed out initially, this time, i.e. there was only post-study 1962. questionnaire and interview. 2. Czerwinski, M., Horvitz, E. and Wilhite, S. A Result: Both interview and questionnaire indicate that most diary study of task switching and interruptions. participants consider task management as useful/very useful Proceedings of the SIGCHI. (3 of 4) or moderately useful (1 of 4). The integration of 3. Horvitz, E. and Apacible, J., Learning and activity data in task management is considered beneficial by reasoning about interruption. in, (2003), ACM, 20- all participants. The features deemed most helpful are 27. access information objects and time information. However, 4. Maguire, M. Methods to support human-centred one participant saw it as time consuming, to keep an eye on design. International journal of human-computer the completeness of the activity data associated with a task. studies, 55. 587-634. We will investigate this in more detail in the future. 5. McFarlane, D.C. Interruption of people in human- computer interaction: A general unifying The interviews revealed that both described methods of definition of human interruption and taxonomy, using transparency were applied (starting with tasks and DTIC Document, 1997. attached information objects as well as starting with activity 6. Morteo, R., Gonzalez, V.M., Favela, J. and Mark, data to identify tasks). G. Sphere juggler: fast context retrieval in support With regard to the results of the first evaluation, the most of working spheres. Proceedings of the Fifth important outcome of the second study were an (self- Mexican International Conference in Computer reported) improved awareness of the personal work process Science, 2004. ENC 2004. 361-367. (3 of 4 useful/very useful, 1 moderately useful) and the 7. Pyöriä, P. The concept of knowledge work perception of being supported in organizing the personal revisited. Journal of Knowledge Management, 9. work process (2 moderately useful, 2 useful). 116-127. 8. Reinhardt, W., Schmidt, B., Sloep, P. and CONCLUSION Drachsler, H. Knowledge Worker Roles and The two design cycles have shown that supporting Actions— Results of Two Empirical Studies. knowledge work is complex, as support solutions need to be Knowledge and Process Management, 18. 150- integrated into the individual work processes w.o. requiring 174. additional effort. The design of a knowledge work support 9. Schultze, U. and Borland, R.J.J. Knowledge system needs to reflect this. management technology and the reproduction of knowledge work practices. The Journal of In particular, we argue that the application of AT-SDM has Strategic Information Systems, 9. 193-212. shown the strength of activity theory-based approaches for 10. Sebastian, D., Schmidt, B. and Göb, A., A Model complex domains like information work. The systemic for the Design of Interactive Systems based on approach to requirements in the AT-SDM supports the Activity Theory. in Forthcoming: Proceedings of reflection of the effects of a newly developed solution, and CSCW 2012, (2011), ACM. helps to integrate evaluation results into the design. The presented Transparency tool is the result of two design iteration and two respective evaluations. Drawbacks of the initial design have been addressed in Transparency 2.0 by integrating activity data and task management.