Survival of the Fittest – Utilization of Natural Selection Mechanisms for Improving PLE Behnam Taraghi Christian Stickel Martin Ebner Graz University of Technology Graz University of Technology Graz University of Technology Social Learning Social Learning Social Learning Steyrergasse 30/I, 8010 Graz, Austria Steyrergasse 30/I, 8010 Graz, Austria Steyrergasse 30/I, 8010 Graz, Austria +43 316 873 8542 +43 316 873 8541 +43 316 873 8540 b.taraghi@tugraz.at stickel@tugraz.at martin.ebner@tugraz.at ABSTRACT generation are terminated, while better adapted ones will have a In the current ongoing work we propose the use of tracking and higher probability to survive [7]. feedback mechanisms in order to improve our Personal Learning The interesting question is how evolution theory can help us in the Environment (PLE), officially launched in October 2010. The development of a Personal Learning Environment (PLE). First of approach can be seen as a necessary prerequisite similar to the all the concept of PLE is still a new and vaguely explored darwinistic model of evolution. This means the implemented concept. From an evolutionary point of view it could be widgets will be improved (variation) and removed (selection) considered as a new species conquering a still undetermined according to the observations. This paper will describe the territory in an eLearning environment. There is no guarantee of backgrounds, methods and some details of the technical success resp. survival of the species. More technical it could be implementation. considered as optimization process with undefined specifications how to solve the problem of helping the learner to overcome the challenge of managing distributed and potentially unknown but Categories and Subject Descriptors useful Web resources and Web applications. H.5.2 (D.2.2, H.1.2) [User Interfaces], D.2.4 [Software/Program Verification]. The biological evolution would approach this problem by choosing the r-strategy, which succeeds by a high (r)eproduction General Terms rate. This strategy can usually be found when a species conquers Measurement, Design, Experimentation. new space. In case of the PLE we need two different views on the evolution metaphor, in order to fully apply this strategy. The first view is macro evolutional, concerning the development of PLEs Keywords as a ‘species’. The question here is about finding the most PLE, Widget, User Experience, HCI. appropriate form, which includes the programming language, deployment, user interface metaphors and value within eLearning 1. INTRODUCTION environments (e.g. is it just a link list in an iPhone app or a full Variation and selection are important mechanisms in the grown web desktop). Therefore a long-time ‘survival’ of the evolutionary development of organismal life forms. These concept PLE would imply the development of many different mechanisms were extensively examined and described by Charles individual solutions in a short period time. The second view to Darwin in his famous book on the topic [6]. He argues that there adopt is the micro evolutional view. In this view the functional is an advantage in the probability to survive for these individuals elements of a single PLE solution are considered be individuals, and populations which are able to adapt better to their struggling to ‘survive’ within the PLE. This view solves the environment. This is described as fitness or ‘Survival of the question of adaptation to the user’s needs on a functional level. fittest’. Darwin’s theory was later used as base for the so called Which resources are really needed, which functions are necessary, evolutionary algorithms (EA), which represent a certain class of which are rarely used and which are never used? optimization algorithms, able to solve nonlinear, discontinuous and even multimodal problems. Evolution itself is a very efficient A first prototype of a Personal Learning Environment (PLE) has optimization process, which is able to adapt even pretty complex been developed and launched in October 2010 at Graz University organisms to a changing environment in a very short time. of Technology (TU Graz) [1]. Following the main PLE concept it Ernst Mayr, who developed the synthetic theory of evolution, aims to provide different learning and teaching resources, which states that the natural selection is rather a selection process but an can be personalized by each learner. Learners can decide if they elimination process. Thereby less adapted individuals of every like to use an application or not and build their own individual learning environment. This paper will outline our current research and development of a PLE. 4 2. Theoretical foundations determine which phenotypes reproduce at a higher rate. Phenotype describes the amount of all observable characteristics 2.1 Evolutional considerations applied of an individual, expressed by its genes and influence from its In order to apply evolutional thinking, it will be necessary to environment at a certain point of time. The natural selection is a establish the metaphorical links to the development of the PLE. non-deterministic process, as it’s disturbed and interrupted by The links will be mostly done on afore mentioned micro random events. Individuals can die, thereby the evolution loses evolutional level, as this is more important to the specific information which could have represented an optimum solution development, however they can be adapted to the macro (e.g. the Wikipedia widget is dismissed because the company evolutional view easily. Evolution theory of natural selection uses offering the service wasn’t able to raise enough funds). the following relevant factors: reproduction rate and mortality Environment and other contextual conditions are ever changing. (cycle for update, replacement and new widgets), population size According to Solbrig [9][10] there are three different modes of (# of widgets), environmental capacity (max. # of widgets in the selection 1) stabilizing selection 2) disruptive selection and 3) system and # of users using the widget). directed selection. All these selection modes and evolutionary pressures aim at increasing the fitness of a population. In order to produce an evolutional pressure upon a population of 1) The stabilizing selection mode (as can be seen in fig.2) individuals, it is necessary to have a limited resource. In our case describes that the evolutionary pressure of the environmental there are actually two such resources driving the selection: a) the factors is directed at outliers, thus this mode favorites the average, limited space within the PLE UI and b) the limited number of which will result in a decrease of variability within the population. potential users. The first factor can also be described as growth regulated and limited by population density, which is depicted in fig.1. A population can’t grow unlimited, as there are limited resources. The environment has a capacity, which is in our PLE case represented by the maximum number of widgets. Figure 2. Stabilizing selection on the distribution of population Stabilizing selection on micro evolutional level can be done by analyzing which functions, respectively widgets in the system are hardly or never used. On a macro evolutional level it would mean to discontinue ‘excotic’ PLE solutions. 2) The disruptive selection mode (as can be seen in fig.3) is directed against the average, reinforcing the extremes, thus splitting a population into two new species. Since our population (on micro evolutional level) is the quantity of widgets, the Figure 1. Growth regulation by population density development path would split and result in two new different solutions for a PLE. On the macro evolutional level this would The second factor b) can be operationalized as selection criteria mean to dismiss the core idea of a PLE, while generating new by asking the questions: ‘Which widget draws the attention of the concepts. most users?’ and ‘Which widget has the biggest frequency of usage?’ [8]. 2.2 Selection So the individuals of a population are forced into a constant competition for a certain resource against each other and against potential harmful conditions of the environment, producing variations for better adaptation. The different probabilities for survival are the base of the selection mechanism. Indeed selection is the main controller for the search direction within the Figure 3. Disruptive selection on the distribution of population evolutional optimization process. In biological systems it would 3) The directed selection mode (as can be seen in fig.4) can be found in natural populations quite often. Thereby the selection Permission to make digital or hard copies of all or part of this work for works only against individuals on one side of the distribution, personal or classroom use is granted without fee provided that copies are moving the curve to a new optimum. This mode can also be found not made or distributed for profit or commercial advantage and that when the PLE developers define new functions and user copies bear this notice and the full citation on the first page. To copy requirements, resp. conceptual decisions (e.g. we will only otherwise, or republish, to post on servers or to redistribute to lists, support intranet applications). requires prior specific permission and/or a fee. Conference’10, Month 1–2, 2010, City, State, Country. Copyright 2010 ACM 1-58113-000-0/00/0010…$10.00. 5 In the example of our PLE we also use a mixed approach. With the beta ‘generation’ of the PLE a bunch of widgets was provided. These widgets were then tested for usability issues, corrected and deployed in the first generation. Last semester a class of students programmed additional widgets in order to produce a certain quantity of functions for the users. So the first and the second generation can be seen as mostly r-strategic. The update process will be repeated on an annual basis. Most of the students last Figure 4. Directed selection on the distribution of population semester chose to produce new widgets, while some reused and optimized existing code (which in turn can be considered K- 2.3 r/K selection theory strategic). The terminology of r/K-selection was defined by the ecologists Robert MacArthur and E. O. Wilson [11][12]. The r/K selection theory states that in the evolution of ecology two major strategic 2.4 Variation approaches for reproduction can be found, aiming to increase the The term variation in the evolutional context is usually described fitness of a species. The strategies are basically a tradeoff between as shift in the genotype or genomic sequence. These shifts occur quality and quantity of the offspring. through a) mutation and b) recombination and generate new Thereby increased quality come with a corresponding increase in phenotypes with different probabilities for survival. parental nurture, while a focus on quantity would decrease the amount of parental investment. Each of the strategies is designed a) Mutation is a random process, aiming only at the generation of for specific environmental constraints. It is also possible that a new alternatives. Mutation can result in different types of change species changes the strategy due to a change in the environment in DNA sequences. It can have either no effect, altering the (e.g. the ecosystem becoming stable for period of time). However product of a gene, or prevent the gene from functioning properly in nature many different mixed forms of these strategies can be or completely [13]. According to the optimization theory, found. In long terms the k-strategy will always be superior, which mutation would be considered as a mechanism to overcome local means that quality succeeds in the long run over quantity. optima. Which means the evolution doesn’t stop if everything seems to be nicely adapted. There is still potential to explore new The r-selection strategy (also referred to as r-strategy) succeeds in variants. In case of PLE development mutation can be considered unpredictable, unstable environments. It is especially useful when as slight updates of existing code or UI elements. it comes to conquer a new unknown ecosystem. It would be a waste of energy and time to adapt to circumstances which are still b) Recombination unknown and will most likely change again. Therefore the r- Recombination is also referred to as cross-over. The process is strategy is characterized by a high reproduction rate and short working somewhere between mutation and selection, thereby lifespan (see fig.5). Transporting this to the PLE would mean to combining and distributing genetic material (DNA, RNA) in a provide a mass of functions (in our case widgets) without looking new way. There’s a random process determining the points where for quality in the first instance. crossovers occur, however recombination is not a random process like mutation, as the recombination itself is not random. This The K-selection strategy (also referred to a K-strategy) succeeds means that the probability is low to separate genes that are close in stable, predictable environments and describes a growth which together or functional linked. is ruled by population density, usually constant and close to the maximum capacity of the environment. The adaption process is The code to all widgets in our PLE is open source and so far all slower but the lifespan is longer and it fills more effective the widgets are open for variations by future developers. An open environmental niche. In case of a PLE the application of this source policy and continuous development, resp. variation are a strategy could mean the increase of quality of a single widget, due necessity for the ‘species’ PLE to finally succeed. to several update cycles, thus adapting optimal to the user’s needs. 2.5 Technical Implementation The basic architecture of the PLE is a mashup [4] of widgets. For each service a widget is provided that follows an extension of the W3C widget specifications [5]. The PLE, its requirements and its technological concept are described in detail in MUPPLE09 workshop [3]. Fig. 6 shows the general concept of the PLE as it is used at Graz University of Technology. The concept follows the idea to bring together university wide services with applications on the World Wide Web. The implemented first prototype of PLE offers centralized access to various University services [1], like administration system: TUGraz online, LMS: TU Graz TeachCenter (TUGTC) or blogospheres: TU Graz LearnLand (TUGLL) [14] in one overview. The users can personalize the PLE to their individual Figure 5. Depicting the relation between fitness and age for r/k information and learning needs. In addition, public services on 6 WWW are also offered in the PLE. For each of these services, a widget has been developed that can be integrated into the PLE. Widgets are small embeddable applications that can be included in an HTML-based web page or executed on the desktop. This client side code can be a simple JavaScript, Java-applets or what ever can be embedded in a valid HTML or XHTML document. It contains the functionality to build the GUI of the widget dynamically and the logic to retrieve or update data from services provided by the PLE server as well as remote servers. The mashup of widgets used in PLE can be classified to end-user mashups as described in [15]. The PLE contains a widget engine, implemented in Palette project [16] to load and handle the widgets according to the W3C widget specifications. While the data extraction is carried out on the server side, the data flow and presentation components are handled by the widget engine on the client side. Figure 7. User Interface Structure Fig.6 shows a conceptual view of the PLE first prototype that The PLE User Interface contains the following elements: integrates university portals and some other Internet services. 1) Sidebar elements contain widget topics. 2) Widget zone contains the widgets that belong to a widget topic. 3a and 3b) Widgets within the corresponding widget zone. 4) Hidden personal desktop containing a mash-up of widgets from different widget zones selected by the user. 5) Banner displays information in context of the active widget zone from the network. 2.6.1 Sidebar elements Widgets are categorized according to pre-defined topics. Each widget topic (category) has its own widget zones. The sidebar elements contain the main widget topics and help the user to switch between widget zones. The topics are easily extendible if the number of widgets is increasing. Furthermore, it is planned that the sidebar also updates the user on the status of the widgets Figure 6. PLE concept at TU Graz. Mashup of distributed by means of color and numerical indicators. The sidebar can be applications and resources from the university and the WWW. switched off in favor of the unfamiliar widget-based UI and replaced by another navigation element, which resembles the Mac Dock menu on the bottom, left, top or right part of widget zones. 2.6 User Interface Structure There are many e-Learning services that are already provided by 2.6.2 Widget Zone the TU Graz, including course administrations in TUGraz online, The widget topics include different areas related to formal and course learning materials such as e-books, podcasts etc. in informal learning, i.e. ”Communication Center” for emails, chats TUGTC and user generated contents as well as user contributions and news groups, ”TeachCenter” for all services related to the TU such as blogs, bookmarks and files posts in TUGLL. Graz LMS system TUGTC, such as course materials, podcasts All these services are going to be integrated in the PLE as etc., ”LearnLand” for services related to the TU Graz blogosphere widgets. Therefore it was necessary to design a coherent GUI to system TUGLL social bookmarking, file sharing, etc. and ”Help avoid the possible usability and consistency problems that may and Support” for the help desk as well as Frequently Asked occur [3]. The PLE GUI (see fig.7) is a combination of a Questions (FAQ). These areas are called widget zones. Widget traditional UI with a sidebar element and banner for orientation zones contain widgets and are structured in columns. The users and navigation. In addition, it offers a widget-based UI with the can switch between widget zones, add, open, close, customize, so-called ”widget zones”, which require an adjustment by the position and arrange the widgets in different columns according to user. their personal learning preferences. 2.6.3 Widgets The widgets consist of a front side and a rear side, where the rear side contains the widget preferences that can be modified by the user. If preferences must be changed, the desired widget can be flipped. By this applied flip-animation the users spatial perception is undisturbed and makes the GUI more understandable. There are two kinds of widgets a) system widgets and b) standard widgets. 7 2.6.4 Personal Desktop At the current state of the PLE development there are 912 users in The users are able to create a mash-up of the most frequently used the system, whereof almost 30% can be said to use the PLE. In the interesting widgets from different widget zones in a special last semester a group of students developed new widgets, in order interface called ”personal desktop”. The personal desktop is to provide additional functionality as well as improving widgets always available to the user and can be activated at any time. from the previous beta stage. The system was introduced to the When the user activates the personal desktop it overlays the whole students in October 2010. The Tracking module was active since screen from the bottom of the page upwards (see figure 4.2 part 1st of November 2010. At the current date this is 102 days. 4). The user can add or remove widgets from all widget zones to his personal desktop and arrange them in columns according to his 5. Discussion personal taste. First the acquired data seem not sufficient to draw any clean conclusions for improvement. As the feedback module wasn’t From the very beginning, an appropriate and good usability of the implemented yet, there is no chance of getting qualitative TU Graz PLE interface was one of the main objectives in the feedback, without performing another usability test. The analyzed development process. Therefore during the implementation of the data are purely quantitative. Nevertheless from the number of first prototype several usability tests were conducted, including users, who have installed a certain widget, we are able to heuristic evaluation and thinking aloud tests. The results were determine to top 5 used widgets out of the 30 provided. Actually integrated and deployed in the current version. theses top 5 are about the universities eLearning services, a mail widget and a system widget for changing the color styles of the 3. Hypothesis interface (tugWidget, tccourses, tugllBlogs, mail, Tracking user behavior, respectively the usage of individual changeThemecolor). Within the top10 we find further a widgets in combination with a feedback mechanism will provide newsgroup reader, a game, google maps, facebook and the leo empirical evidence for adaptive development. dictionary. From an educational point of view these choices make perfectly sense as these services are well known and frequently Following an evolutional model of developing the PLE, this will used even without the PLE. mean a stepwise improvement and rejection of individual widgets in further iterations of the development cycle. Interestingly the most installed widgets are not necessarily the most used ones. The top 5 with the highest usage rate include weather forecast, rss reader, twitter, TUG library widget and again 4. Methods and Materials the leo dictionary. Within the top 10 we find here again google maps, Facebook and tugllBlogs, beside another dictionary and a In order to improve the PLE we needed to consider different currency converter. parameters that influence the attractiveness and effectiveness of the whole system in general as well as individual widgets. To Within the last update cycle, resp. the time when the students meet this goal a tracking module was implemented to measure course developed new widgets, the weather widget was replaced quantitatively how often the widgets are used and by how many by a new version. Actually this can be seen our current update users. The measurement was operationalized by the means of strategy. If the outcome of the variation is a widget that fulfills a tracking individual and overall usage of widgets. In order to function better, then the old one will be replaced. measure the usage of widgets a hidden module in the background tracked the users' active widgets. 6. Conclusion and future works According to the hypothesis we expected to get more knowledge The widgets that are used in PLE can be classified to three about user behavior, user preferences and derive data, which categories depending on how they interact with other services and would help us to differentiate user behaviors, for instance between applications on World Wide Web (WWW). students of first and last semesters or students of different major • Widgets that have no interactions with WWW such as of studies, and finally to improve the system in a natural way by widgets representing learning objects. variation and selection. However due to lacking qualitative data we are not able to falsify the hypothesis. • Widgets that have a server side component to preprocess the data on PLE server such as widgets that In order to gather qualitative measures of the user experience integrate university services in PLE. (UX) in future versions, a rating system will be implemented. This will be done either by a 5 star rating system or alternatively by a • Widgets that use the PLE built-in proxy to request data small feedback questionnaire contained in every widget, which from remote services such as RSS FEED reader widget. consists of less than ten items of semantic differentials inspecting The client-side tracking module is added to the PLE widget the UX quality of the widget, respectively important variables of engine to provide widgets including the possibility to offer evolvability. These would be attractiveness, dependability and information about user behavior on the client side. In periodic perceived effectiveness. The semantic differentials will be taken intervals the information (if any) is captured from all activated from the reliable UEQ inventor constructed by [2]. The fig. 2 widgets in PLE and sent to the server-side tracking module for depicts the questionnaire integrated into the widgets backside. further processing. The server-side tracking module is used also for second and third widget types to capture information related to the user behavior in widgets depending on the data traffic on the server side. 8 Mensch & Computer 2006, 125–134. München: Oldenbourg Verlag. [3] Taraghi, B., Ebner, M. & Schaffert, S. 2009. Personal Learning Environments for Higher Education: A Mashup Based Widget Concept, Proceedings of the Second International Workshop on Mashup Personal Learning Environments (MUPPLE09), Nice, France (2009), ISSN 1613-0073, Vol-506 http://ceur-ws.org/Vol-506/. [4] Tuchinda, R., Szekely, P. & Knoblock, C. 2008. Building Mashups By Example. ACM Proceedings of IUI 2008, Maspalomas, Spain (2008). Figure 2. Mockup of questionnaire integrated into the widget GUI [5] Widgets 1.0 Packaging and Configuration. 2008. for qualitative measures of the user experience. http://www.w3.org/TR/widgets/ (2008). In our PLE users can select some widgets from a widget pool and [6] Darwin, C. 1911. On the origin of species by means of activate them for personal use. However if the user activates some natural selection. Hurst widgets it does not necessarily mean that these widgets are [7] Mayr, E. 2005, Das ist Evolution.(pp. 18, 250) actively used. In future versions the tracking module might be able to detect an active widget usage und track the usage in detail [8] Hastings, A. (1996) Population Biology. Springer Verlag, as deeply as possible. Berlin. [9] Solbrig O.T. (1970). Principles and methods of Plant In future works it would also be interesting to classify users Biosystematics. The Mac-Millan Company. Collier-Mac according to their individual needs, for instance users who use Millan Limited, London. more often only widgets with a strong focus on communication or users who use PLE more for learning issues, etc. [10] Solbrig O.T. & D.J. Solbrig. (1979). Population biology and evolution. Addison-Wesley. Publ. Co. Reading Mass. In order to meet data privacy considerations, we will implement a disclaimer, or terms of service (TOS) which needs to be agreed by [11] Pianka, E.R. (1970). On r and K selection. American the users once in order to use the PLE. Naturalist 104, 592–597. [12] MacArthur, R. and Wilson, E.O. (1967). The Theory of The tracking module provides sufficient quantitative data about Island Biogeography, Princeton University Press (2001 the usage of the widgets. Bearing in mind that more knowledge reprint), ISBN 0-691-08836-5M. about the learner will help in designing didactical models for providing learning courses, data gathering must be seen as a first [13] Sniegowski P, Gerrish P, Johnson T, Shaver A (2000). "The valuable step. Furthermore these data combined with user profiles evolution of mutation rates: separating causes from will be a precondition for building a recommender system on consequences". Bioessays 22 (12): 1057–66. learning objects within PLE. [14] Ebner, Martin and Behnam Taraghi [2008]. A Blog Sphere for Higher Education. In Luca, Joseph and Edgar R.Weippl (Editors), Proceedings of World Conference on Educational Multimedia, Hypermedia and Telecommunications 2008, pages 5618–5625. AACE, Vienna, Austria. 7. REFERENCES [15] Gamble,M.T. and R. Gamble [2008]. Monoliths to Mashups: [1] Ebner, M. and Taraghi, B. 2010. Personal Learning Increasing Opportunistic Assets. Software, IEEE, 25(6), Environment for Higher Education – A First Prototype. In pages 71–79. Proceedings of World Conference on Educational Multimedia, Hypermedia and Telecommunications 2010 (pp. PALETTE [2008]. Pedagogically sustained Adaptive 1158-1166). Chesapeake, VA: AACE (2010). Learning through the Exploitation of Tacit and Explicit Knowledge. http://palette.ercim.org/. [2] Laugwitz, B. Schrepp, M. & Held, T. 2006. Konstruktion eines Fragebogens zur Messung der User Experience von Softwareprodukten. In: A.M. Heinecke & H. Paul (Hrsg.): 9