This paper promotes the automated creation of hybrid personal plans, comprising web services and real human activities, to be supported by the next generation of intelligent calendar applications. A prototype work is present-ed, utilizing both atomic web services and composite ones, the latter having been previously generated from a contingent web service composition module, aiming at in-creasing the likelihood of achieving their intended goal. A metric planner generates a plan, giving priority to web service calls over human activities. Then, a scheduler schedules the human activities into the user's calendar, taking into account the ordering constraints that result from the plan. The resulting schedules substitute human activities with web services, thus increasing the user's capacity, his free time, as well as the scheduling options. As a proof of concept we present a case study implementation utilizing existing state-of-theart components.
Although in the recent past the popularity of paper calendars has steadily declined in favor of web-based calendar applications like Google Calendar, such applications still do not provide any means for automated activity scheduling. In that way, users are forced to decide for themselves whether a particular activity has enough time to be scheduled alongside another, or whether the distance between the places where the two activities occur is a prohibiting factor.
This problem was efficiently tackled in our previous work, by SELFPLANNER [
The obvious next step in intelligent calendar applications is to employ planning, instead of pure scheduling, to achieve the user’s goals. Existing systems, like SELFPLANNER, rely on the user to select the activities to be included in his plan, with the system providing only scheduling functionalities. We envision a situation where intelligent calendar applications (a) support action ontologies, with action descriptions containing pre-conditions and effects, (b) allow the user to set his goals, (c) know the user’s state, and (d) employ auto-mated planning to create plans that achieve the user’s goals.
An even further step would be to extend intelligent calendar applications so as to take web services into account. Web services can be considered as normal actions that can be used to achieve users’ goals or other actions’ preconditions. Incorporating web services into the action ontology enables the substitution of human activities by web service calls, thus allowing for more flexible plans, more goals to achieve or just more free time. As an example of such a setting that is very common in our daily routine consider this: a person may wish to attend a concert, and for that reason he may insert a personal activity in his schedule so as to reserve time and remember to buy tickets for it. To achieve this goal the user is required to physically go to a brick-and-mortar shop or buy his tickets online: both options re-quire some of his time (obviously, buying the tickets online requires less time). A more efficient electronic calendar, on the other hand, would have searched for an alternative - automated - way of achieving such goals, so as to relieve the user from the burden of manually executing the necessary actions.
In this work we propose such an approach; it is based on the integration of web services with an existing metric planner and an intelligent calendar application, namely SELFPLANNER. Web services may be simple or composite; in the latter case, a contingent web service composition module has been used, trying to reduce the nondeterminism underlying their execution. From the planning perspective, though, web services are considered deterministic. As a result of the above process, the user’s schedule contains only the human activities that cannot be completed through the use of web services, thus relieving him of the extra burden of achieving the rest.
The rest of the paper is structured as follows: First we present related work; next, we give a motivating example; then we present our approach and, finally, we conclude the paper and pose future directions of research. 2
SELFPLANNER was the first system to tackle the problem of automated scheduling
personal activities into electronic calendars, using a combination of greedy
optimization algorithm, namely a modified version of the Squeaky Wheel Optimization
(SWO) [
Bank et al. [
La Placa, Pigot and Kabanza [
Finally, Berry et al. [
In regard to non-deterministic web service composition, Kuzu and Cicekli [
Dacosta et al. [
In our previous work [
Let us imagine a usual week of a Bob, who uses a web-based calendar application to organize his time. Due to being self-employed, Bob needs to spend all working hours at his office, to which he commutes every day from his home. Once every week, he watches a movie, and although he prefers to go to the cinema, sometimes he watches the movie at home, depending on his work schedule. However, if the movie’s duration is such that it will end later than 11 pm, he does not desire to watch a movie at all, as he has to sleep early. Moreover, this week, he will travel on a business trip abroad; a day before the trip, the day he usually watches a movie, he has to book his airplane tickets and hotel, as well as to buy a travel guide for the city he will visit.
In order to schedule these activities and insert them into a calendar, the user has to define their temporal domain and, if they are interruptible, the minimum and maximum allowed duration for their parts. Moreover, for each activity, the user has to declare whether it is periodic or not, as well as if it is bound to specific locations. For example, the user should define his daily work as a periodic task, with a temporal domain from 8 a.m. to 5 p.m., and a minimum and maximum duration of its parts – as he cannot work continuously, e.g., 30 minutes and 2 hours respectively. Watching a movie is also periodic but non-interruptible, and has a temporal domain set late in the evening, with a minimum and maximum duration of 90 minutes and 180 minutes respectively.
Since the user’s work requires his physical presence, only a human activity can be placed in his calendar. However, for the rest of the aforementioned tasks, a combination of human and web services’ activities can be inserted. The user may have to drive to the cinema, choose among the available movies there and buy the tickets himself. Alternatively, he may ask for a list of the available movies to be emailed to him, book the tickets online and then, having saved considerable time, travel to the cinema later. Another option altogether would be to rent a movie online and watch it at home. As to the user’s business trip, again, there are various alternative activities; the user may visit a single tourist agent to book his tickets and hotel; or prefer an online reseller. He may also require another trip to a bookstore to purchase his travelling guide or purchase it online and have it sent at home.
Since these options create a complex problem, containing a multitude of constraints and preferences, a schedule manually created by a human is usually highly inefficient. Applications such as SELFPLANNER tackle this problem; however, they only deal with human activities and, as such, they cannot take advantage of the opportunities that are offered by the use of web services. In the aforementioned scenario, since a human activity requires the user to first purchase a movie ticket himself, in certain situations he may not have had the time to do so, and he would have had to watch it at home instead. Even worse, if he had to visit a travel agent and a bookstore, he may not have even had the time to watch a movie at home.
With the introduction of web services, the user saves the time needed to perform the actual action of purchasing these services and to travel between locations. 4
This work assumes that web services are semantically annotated and that their
descriptions are present in an online registry; in previous work, we presented such a
registry [
Fig. 1 presents the proposed system’s architecture. Initially, we employ an existing
metric planner, namely LPG-td [
This problem comprises of the translation of the web services from the registry, as well as a simplified version of the human activities; that is, the locations, durations and temporal domains of the activities, along with any preferences and constraints in regard to them are removed. The metric planner does not differentiate between the two types of activities; as a result, it also treats all web services as deterministic ones, i.e., as if their intended output (the most probable one) is always outputted. Moreover, the planner has to take into account that web services are preferred to their human activities’ counterpart. This is achieved by setting the cost of human activities higher than that of web services.
LPG-td is capable of generating a sequence of alternative plans, each being an improvement – in terms of the specified metric – compared to the previous one; the plans generated by LPG-td allow for parallel actions and, thus, are very similar to partial order ones.
We feed the best generated plan by LPG-td to SELFPLANNER so as to create a detailed schedule. In case a feasible plan does not exist, the rest of the previously generated alternative partial order plans are attempted to be scheduled in a similar process. In order to schedule the activities of the partial order plan, SELFPLANNER employs the information concerning temporal constraints and preferences (loaded from a separate activity definition file for a given problem instance), as well as the actual distances between the activities’ locations as returned by the Google Maps Distance Matrix API.1 Web services are considered to have an open temporal domain and are not related to a specific location, as they can be executed at any time and place. Moreover, they can be scheduled in parallel to human activities. SELFPLANNER can generate multiple alternative plans; the human activities of the best schedule are uploaded into the user’s Google calendar. 5
This paper presents the first steps towards the next generation of intelligent calendar applications. We propose an approach comprising a contingent web-service composition system, a partial order metric planner and a scheduler to insert human activities into a user’s web calendar. The contingent planner is used to create composite web services, able to achieve complex goals with high probability of success; the metric planner is used to select an optimized set of activities to achieve a user’s goal, favoring web service calls than real activities; and, finally, the scheduler automatically produces an optimized plan based on the user’s constraints and preferences.
The paper also presents a motivating example along with a – still under work – implementation. We aim to further improve this implementation, first by providing a graphical interface, as well as by integrating it with a web service execution platform. Also, we propose to perform an exploratory evaluation, by providing the - users with two schedules, one consisting solely of human activities and an equivalent one comprising both human and web service activities, and having them rate each one online.
Finally, various research and implementation issues are still open; for example, we assume that the user’s schedule should only contain the human activities that cannot be substituted by web services; this is achieved through setting the cost of human activities higher than that of web services. However, in some cases this may not be true; employing a web service may incur a (financial) cost that the user does not prefer over the benefit of not employing the respective human activity. Moreover, a web service activity may not always provide exactly the same functionality or satisfaction as its human counterpart; such a case is present in our motivating example, in which a user prefers to go to the cinema to watch a movie, than watching it at home through streaming. In this case, however, if the problem is modeled so as to favor the human activity, it could be impossible to include it in the plan, thus providing the use of a web service as an alternative plan. Such problems require further investigation. 1 https://developers.google.com/maps/documentation/distancematrix/ Acknowledgment. This research has been co-financed by the European Union (European Social Fund – ESF) and Greek national funds through the Operational Program “Education and Lifelong Learning” of the National Strategic Reference Framework (NSRF) - Research Funding Program: Heracleitus II. Investing in knowledge society through the European Social Fund.