A SURVEY ON USER MODELING IN HCI

Computer Applications: An International Journal (CAIJ), Vol.5, No.1, February 2018
DOI:10.5121/caij.2018.5102 21
A SURVEY ON USER MODELING IN HCI
Mohammad Sajib Al Seraj1
, Robert Pastel2
, Md. Al-Hasan3
1,2
Department of Computer Science,
Michigan Technological University Houghton, MI 49931, USA
3
Department of Computer Science & Engineering,
Bangladesh Army University of Science &
Technology, Nilphamari, Bangladesh
ABSTRACT
In the last few years user modeling has become an important research area in Human Computer
Interaction. A large amount of research has been conducted in this field where different approaches on
user modeling are shown. In this paper, we provide an overview of the field of user modeling and describes
the different user model namely, GOMS family of models, cognitive architecture, grammar-based model,
and application specific models. We have discussed a few examples of user models in each category. This
paper also discusses the future challenges of this research area.
Index Terms—User Model, Human Computer Interaction (HCI), GOMS model, Cognitive model,
Grammar based model.
1.INTRODUCTION
From user experience and meticulous research, we find that computer systems are not easy to
learn and once learnt, may be easily forgotten. Software industry updating their software in
regularity basis with different interface functionality which sometimes creates difficulties for
even for the learned users to cope with. Users changing their perception of and proficiency with
different software systems which is another problem. The range of skills, knowledge and
preferences required of users’ means that any computer system which offers a fixed interface will
be better suited to some users than to others.
Different types of user use the computer system in different way and their perspective is also
different. An effective way of dealing with system complexity for the novice user is to provide a
functionally simple system. Intermittent or discretionary computer users have to master different
application packages as the need arises, and seldom have any real choice in the purchase,
selection or use conditions of the software which their management decides upon. Discretionary
users are a particularly important class as they often have to be encouraged to make use of a
system and will generally benefit from easy to use, intuitive and functionally simple interfaces.
Another familiar class of user is the committed or so called ‘expert’ user who may choose to
employ different packages for different activities
Addressing a large variety of users is always a challenge to designers due to diverse range of
abilities and differences in task, prior knowledge and situation. A user model is a representation
of the knowledge and preferences of users [Benyon & Murray, 1993]. It is not a mandatory part

22
of the software but it helps to get the system serve the user better. Any information stored about
the user or usage pattern is not a user model unless it can be used to get some explicit assumption
about the user.
In the following section, the history of the user model will be presented. Different types of user
model will be presented in the next section and the future challenge of the user model will be
presented in the last section.
1.1 History Of User Modeling
User modeling is traced back to the works of Allen, Cohen and Perrault [Perrault et al., 1978;
Cohen and Perrault, 1979] and Elaine Rich [Rich, 1979a, 1979b]. For a ten-year period following
this seminal research, numerous application systems were developed that collected different types
of information about, and exhibited different kinds of adaptation to, their current users. In these
early modeling systems, the modeling components were not distinct from the rest of the
application, but as the field grew user modeling systems were made more modular.
The Command Language Grammar [Moran, 1981] developed by Moran at Xerox Parc could be
considered as the first HCI model. It took a top-down approach to decompose an interaction task
and gave a conceptual view of the interface before its implementation. However it completely
ignored the human aspect of the interaction and did not model the capabilities and limitations of
users. Card, Moran and Newell’s Model Human Processor (MHP) [Card & Newell, 1983] was an
important milestone in modeling HCI since it introduced the concept of simulating HCI from the
perspective of users. It gave birth to the GOMS family of models [Card & Newell, 1983] that are
still the most popular modeling tools in HCI.
Outside the domain of HCI, recent researches on cognitive modeling address a wide range of
topics such as investigating mental processes for new idea generation. However, the domain of
cognitive modeling is currently overwhelmed by the cognitive architectures and models
developed using them. This kind of models does not only work for HCI but also aims to establish
a unified theory of cognition.
These two main approaches of user modeling: the GOMS family of models was developed only
for HCI while the models involving cognitive architectures took a more detailed view of human
cognition. Based on the accuracy, detail and completeness of these models, Kieras classified them
as low fidelity and high-fidelity models respectively [Kieras, 2005]. These two types of model
can be roughly mapped to two different types of knowledge representation. The GOMS family of
models is based on goal-action pairs and corresponds to the Sequence/Method representation
while cognitive architectures aim to represent the users’ mental model. The Sequence/Method
representation assumes that all interactions consist of a sequence of operations or generalized
methods, while the mental model representation assumes that users have an underlying model of
the whole system.
There is a third kind of model in HCI that evaluates an Interface by predicting users’, rather than
their performance. These models represent an interaction by using formal grammar where each .
action is modelled by a sentence. They can be used to compare users’ performance based on
standard sentence complexity measures; however, they have not yet been used and tested as
extensively as users’ behavior simulator

23
2. DIFFERENT MODELS
2.1. THE GOMS FAMILY OF MODELS
GOMS (Goals, Operators, Method and Selection) was in-spired by the GPS system developed by
Newell. The method is composed of methods that are used to achieve specific goals. It enables a
designer to simulate the sequence of actions of a user while undertaking a task by decomposing
the task into goals and sub goals. There exist four different GOMS model.
2.1. KLM
Keystroke-Level Model (KLM) [Card & Newell, 1983] is the first and simplest GOMS technique
by eliminating the goals, methods, and selection rules, leaving only six primitive operators. The
KLM model [Card & Newell, 1983] simplifies the GOMS model by eliminating the goals,
methods, and selection rules, leaving only six primitive operators. They are:
1) K keystroking/ keypressing
2) P pointing with a mouse to a target
3) H homing the hand on the keyboard
4) M performing mental preparation,
5) D drawing a line segment on a grid
6) R waiting for the computer to execute a command.
The durations for each of these six operations have been empirically determined. The task
completion time is predicted by the number of times each type of operation must occur to
accomplish the task.
2.2. CMN-GOMS
CMN-GOMS is the original GOMS model proposed by Stuart Card, Thomas P. Moran and Allen
Newell. It takes the KLM as its bases and adds sub goals and selection rules. This model can
predict operator sequence as well as execution time. A CMN-GOMS model can be represented in
pro-gram form, making it amenable to analysis as well as execution. CMN-GOMS has been used
to model word processors and CAD systems for ergonomic design. The CMN method can predict
the operator sequence and the execution time of a task on a quantitative level and can focus its
attention on methods to accomplish goals on a qualitative level.
2.3. NGOMSL
A structured language representation of GOMS model, called NGOMSL (Natural GOMS
Language) [Kieras, 1994] developed by Kieras. NGOMSL builds on CMN-GOMS by providing a
natural-language notion for representing GOMS models, as well as a procedure for constructing
the models. Under NGOMSL, methods are represented in terms of an underlying cognitive theory
known as cognitive complexity theory, or CCT model at a higher level of notation. CCT is a rule-
based system developed by Bovair and colleagues to model the knowledge of users of an
interactive computer system. Kieras also developed a modeling tool, GLEAN (GOMS Language

24
Evaluation and Analysis), to execute NGOMSL. It simulates the interaction between a simulated
user with a simulated device for undertaking a task.
2.4. CPMGOMS
John and Kieras [John & Kieras, 1996] proposed a new version of the GOMS model, called
CPMGOMS, to explore the parallelism in users’ actions. This model decomposes a task into an
activity network (instead of a serial stream) of basic operations (as defined by KLM) and predicts
the task completion time based on the Critical Path Method.
3. COGNITIVE ARCHITECTURES
Allen Newell developed the Soar architecture [Newell, 1990] as a possible candidate for his
unified theories of cognition. According to Newell [Newell, 1990] and Johnson-Laird [Johnson-
Laird, 1988], the vast variety of human response functions for different stimuli in the
environment can only be explained by a symbolic system. So, the Soar system models human
cognition as a production-rule based system and any task is carried out by a search in a problem
space. The heart of the Soar system is its chunking mechanism. Chunking is “a way of converting
goal-based problem solving into accessible long-term memory (productions)” [Newell, 1990]. It
operates in the following way. During a problem-solving task, whenever the system cannot
determine a single operator for achieving a task and thus cannot move to a new state, an impasse
is said to occur. An impasse models a situation where a user does not have sufficient knowledge
to carry out a task. At this stage Soar explores all possible operators and selects the one that
brings it nearest to the goal. It then learns a rule that can solve a similar situation in future.
However, there are certain aspects of human cognition that can be better explained by a
connectionist approach than a symbolic one. It is believed that initially conscious processes
control our responses to any situation while after sufficient practice; automatic processes are in
charge for the same set of responses. Lallement and Alexandre have classified all cognitive
processes into synthetic or analytical processes. Synthetic operations are concerned with low-
level, non-decomposable, unconscious, perceptual tasks. In contrast, analytical operations signify
high level, conscious, decomposable, reasoning tasks. From the modeling point of view, synthetic
operations can be mapped on to connectionist models while analytic operations correspond to
symbolic models. Considering these facts, the ACT-R system does not follow the pure symbolic
modeling strategy of the Soar, rather it was developed as a hybrid model, which has both
symbolic and sub symbolic levels of processing. At the symbolic level, ACT-R operates as a rule-
based system. It divides the long-term memory into declarative and procedural memory.
Declarative memory is used to store facts in the form of ‘chunks’ and the procedural memory
stores production rules. The system works to achieve a goal by firing appropriate productions
from the production memory and retrieving relevant facts from the declarative memory.
However, the variability of human behavior is modelled at the sub-symbolic level. The long-term
memory is implemented as a semantic network. Calculation of the retrieval time of a fact and
conflict resolution among rules is done based on the activation values of the nodes and links of
the semantic network. The EPIC (Executive-Process/Interactive Control) architecture pioneers to
incorporate separate perception and motor behavior modules in a cognitive architecture. It Mainly
concentrates on modeling the capability of simultaneous multiple task performance of users. It
also inspired the ACT-R architecture to install separate perception and motor modules and

25
developing the ACT-R/PM system. A few examples of their usage in HCI are the modeling of
menu searching and icon searching tasks.
The CORE system (Constraint-based Optimizing Reaso ing Engine) [Lewis et al., 2006; Howes
et al., 2004] takes a different approach to model cognition. Instead of a rule-based system, it
models cognition as a set of constraints and an objective function. Constraints are specified in
terms of the relationship between events in the environment, tasks and psychological processes.
Unlike the other systems, it does not execute a task hierarchy; rather prediction is obtained by
solving a constraint satisfaction problem. The objective function of the problem can be tuned to
simulate the flexibility in human behavior.
There exist additional cognitive architectures, but they are not yet as extensively used as the
previously discussed systems.
2.1. Grammar-Based Models
The grammar based model simulates an interaction in the form of grammatical rules. As for
example, Task Action Language models
• Operations by Terminal symbols
• Interaction by a Set of rules
• Knowledge by Sentences
This type of modeling is quite useful to compare different interaction techniques. However, they
are more relevant to model knowledge and competence of a user than performance.
2.2 Application Specific Models
A lot of works has been done on user modeling for developing customizable applications. These
models have the following generic structure. They maintain a user profile and use different types
of AI systems to predict performance. These types of models are particularly popular in online
adaptable systems (like personalized search engines or portals). We discuss a few more examples
besides online systems in the next sub-section.
The Generative User Model [Motomura, 2000] is developed for personalized information
retrieval. In this model user given query words are related to user’s mental state and retrieved
object using latent probabilistic variables. Norcio [Norcio, 1989] used fuzzy logic to classify
users of an intelligent tutoring system. The fuzzy groups are used to derive certain characteristic
of the user and thus deriving new rules for each class of users. Norcio and Chen also used an
artificial neural network for the same purpose as their previous work [Norcio, 1989]. The user’s
characteristic is stored as user image and neural networks are used as pattern associates or pattern
classifiers to get user’s knowledge, detect user’s goal etc. Lumiere convenience project used
influence diagram in modeling users. Lumiere project is the background theory of the Office
Assistant shipped with Microsoft Office application. The influence diagram models the
relationships among users’ needs, goals, user background etc. However, all these models are
developed by keeping only a single application in mind and so they are hardly usable to model
human performance in general.

26
3. FUTURE CHALLENGES
3.1. Support different modeling techniques
Most of the user modeling approaches failed because of relying on one specific technique. A
substantial leverage can be gained by integrating modeling with implicit modeling. This enriched
synthesis can be further complemented by asking users (in otherwise open environments, such as
HFAs and design environments) to solve specific problems in which the selection of the problem
is driven by specific needs of the user modeling component.
3.2. Payoff Of User Modeling
There is little to be gained if expensive mechanisms against minimal improvements in usability
and usefulness. The payoff or utility of cognitive artifacts can be characterized by the quotient of
“value / effort.” To increase the payoff, we have two options: (1) increase the value by showing
that future systems relying on user models are more usable and more useful, or (2) decrease the
effort associated with creating a user model.
3.3. Wrong, Outdated, And Inadequate Information
User models represent a world that is outside the computational environment. The mapping of
external information to the internal model may be wrong to start with, but even under the
assumption that it is an adequate representation at some point of time, it may become outdated by
external changes of which the model is unaware [Allen, 1997]. How, when, and by whom can a
wrong user model be identified? Who will have the authority and the knowledge to change the
model, and which modification mechanisms will be available to do so?
3.4. Criteria For Different Domains
Assuming that user modeling is useful in some domains but not in others, which criteria do we
have to distinguish these domains?
4. CONTROL
A consequence of any smart behavior of systems is that agents (humans or computers) can guess
wrong and per-form hidden changes that users do not like. Current systems often lack the
possibility or at least the transparency for users to turn off these “smart” features, which can get
more in the way than help. As argued above, systems, even smart ones, are aware of only a
fraction of the total problem-solving process their human partners undergo [Hollan, 1990], and
they cannot share an understanding of the situation or state of problem-solving of a human.
Whereas these drawbacks of smart systems may be only annoying in HFAs such as word
processors, they are unacceptable in other collaborative human-computer systems, such as airline
cockpit computers serving as intelligent agents. Billings [Billings, 1991] argues convincingly that
in computerized cockpit design each intelligent agent in a human-computer system must have
knowledge of the intent and the rationale of the actions of the other agents. To avoid these
drawbacks, intelligent systems should pro-vide malleable tools that empower rather than diminish
users, giving them control over tasks necessary for everyday life. There are situations in which

27
we desire automation and intelligence (for example, few people will have the desire to compile
their programs themselves) — but the decision as to what should be automated and what not
should be under the control of the people affected by the system.
5. PRIVACY
Security is a major concern in this computational era. Numerous organizations compile user
models of our behavior and actions — and there is the great danger that this in-formation can be
misused. It will be a major challenge to find ways to avoid misuse, either by not allowing
companies to collect this information at all or by finding ways that the individual users have
control over these user models.
6. CONCLUSION
In this paper, different types of user models have been elaborately discussed. We have classified
user model into four categories. We have also analyzed user modeling from a broader
perspective. It should be evident that the use of modeling and the type of model to be used
depend on many factors like the application, the designers, availability of time and cost for design
etc. However, we hope this paper will give system analysts an understanding of different
modeling paradigms, which in turn may help them to select the proper type of model for their
applications.
REFERENCES
[1] Benyon D., Murray D (August 1993)., Applying User Modeling to Human Computer Interaction
Design, Artificial Intelligence Review, Volume 7, Numbers 3-4, pp. 199 – 225.
[2] Perrault C. R., Allen J. F., & Cohen P. R. (1978). Speech Acts As a Basis for Understanding Dialogue
Coherence, In Proceedings of the 1978 Workshop on Theoretical Issues in Natural Language
Processing (pp. 125–132). Stroudsburg, PA, USA: Association for Computational Linguistics.
doi:10.3115/980262.980282.
[3] Cohen P. R., & Perrault C. R. (1979). Elements of a plan-based theory of speech acts, Cognitive
Science, 3(3), 177–212. doi:10.1016/S0364-0213(79)80006-3
[4] Rich, E. (1979a). Building and exploiting user models. In Proceedings of the 6th international joint
conference on Artificial intelligence-Volume 2, pp. 720–722.
[5] Rich, E. (1979b). User modeling via stereotypes*. Cognitive Science, 3(4), 329–354.
[6] Moran T.P. (1981) Command Language Grammar: A Representation For The User Interface of
Interactive Computer Systems, International Journal of Man-Machine Studies 15.1, pp. 3-50.
[7] Card S., Moran T. and Newell A.(1983), the Psychology of Human-Computer Interaction, Hillsdale,
NJ, USA: Lawrence Erlbaum Associates.
[8] Kieras D. E.(2005), Fidelity Issues In Cognitive Architectures For HCI Modelling: Be Careful What
You Wish For, International Conference On Human Computer Interaction.
[9] Kieras D.(1994), GOMS Modeling of User Interfaces Using NGOMSL, Conference Companion On
Human Factors In Computing Systems,Boston, Massachusetts, USA, pp.371-372.

28
[10] Newell A.(1990), Unified Theories of Cognition, Cambridge, MA, USA: Harvard University Press.
[11] Johnson-Laird P.A.(1988), The Computer and The Mind, Cambridge, MA, USA: Harvard University
Press.
[12] Eng K., Lewis R. L., Tollinger I., Chu A., Howes A. and Vera A. Generating Automated Predictions
of Behavior Strategically Adapted To Specific Performance Objectives ACM/SIGCHI Conference on
Human Factors in Computing Systems, pp. 621-630
[13] Howes A., Vera A., Lewis R.L. and Mccurdy (2004), M. Cognitive Constraint Modeling: A Formal
Approach To Reasoning About Behavior, Annual meeting of the Cognitive Science Society Lawrence
Erlbaum Associates.
[14] John B. E. and Kieras D.(1996) The GOMS Family of User Interface Analysis Techniques:
Comparison And Contrast. ACM Transactions On Computer Human Interaction 3, pp. 320-351.
[15] Motomura Y., Yoshida K., Fujimoto K.(200), Generative user models for Adaptive Information
Retrieval, Proc. IEEE Conf. on Systems.
[16] Norcio F.(1989), Adaptive Interfaces: Modelling Tasks and Users, IEEE Trans. Systems, Man,
Cybernetics, 19(2) pp. 399-408.
[17] Allen, R. B. (1997), Mental Models and User Models In M. G. Helander, T. K. Landauer, & P.
[18] V. Prabhu (Eds.), Handbook of Human-Computer Interaction, Volume 1, Elsevier Science B.V.,
Amsterdam, pp. 49-63.
[19] Hollan, J. D. (1990), User Models and User Interfaces: A Case for Domain Models, Task Models, and
Tailorability, In Proceedings of AAAI-90, Eighth National Conference on Artificial Intelligence,
AAAI Press/The MIT Press, Cambridge, MA, pp. 1137.
[20] Billings, C. E. (1991) Human-Centered Aircraft Automation: A Concept and Guidelines, NASA
Technical Memorandum, Report No. 103885, NASA Ames Research Center.

A SURVEY ON USER MODELING IN HCI

More Related Content

What's hot (17)

Similar to A SURVEY ON USER MODELING IN HCI (20)

Recently uploaded (20)

A SURVEY ON USER MODELING IN HCI