Context Awareness and Context Recognition in Modern Decision-Making - 2

 

Section Two: Context

In this section we look at the concept of context as it applies to both human and computer cognition. After laying out some perspectives of context and its role in decision-making, we identify and describe three major interpretations of context: as a schema, as a frame, and as a model.

Perspectives on Context

Context is “a complex description of shared knowledge about physical, social, historical, or other circumstances within which an action or an event occurs… (that) does not intervene explicitly in a problem solving but constrains it” (Brézillon, 2004).

Dey and Aboud (1999) define context as “any information that can be used to characterize the situation of an entity. An entity is a person, place, or object that is considered relevant to the interaction between a user and an application, including the user and applications themselves” (See also Aboud, Dey, et al., 2001; See Zainol and Nakata (2010, pp. 126-127) for additional definitions along the same lines).

Winograd (2001, p. 5) argues that context is defined by use rather than by features. “Context is an operational term: something is context because of the way it is used in interpretation, not due to its inherent properties.” (Winograd, 2001) He offers a communication and application programming architecture using a ‘blackboard’ metaphor that supports context-aware computing.

Sato (2003, p. 1324) argues that we should represent context through “a pattern of behavior or relations among variables that are outside of the subjects of design manipulation and potentially affect user behavior and system performance.” He describes a three-part strategy for context-sensitivity: sensing contextual changes, re-configurable architecture, and creating and managing contexts (p. 1327).

Dourish (2004) describes an incompatibility between two views of context.

• One comes from positivist theory—context can be described independently of the actions done; the definition proposed by Dey matches this view.

• Another view can be sustained by phenomenological theory—context emerges from the activity and cannot be described independently.

Guarino & Guizzardi (2015, 2016) offer an account of context as a ‘scene’ such that “events emerge from scenes as a result of a cognitive process that focuses on relationships: relationships are therefore the focus of events” (2016, p.2) and where ‘scenes’ are whatever occurs in a certain region of spacetime.

Types of Context

Some types of context, as described in the literature:

• Activity, Identity, Location, and Time (AILT), (Dey and Aboud,1999)

• Context as related to location, nearby person, hosts or objects, as well as changes of them over time (Schilit, et al. , 1995)

• Context as tailored to the individual’s circumstances (Brown et al., 1997). For example, this “includes the capabilities of the mobile devices, the characteristics of the network connectivity and user specific information such as emotional state, attention focus, and orientation.

• Context as behavioural. The Context-based Reasoning (CxBR) modelling paradigm asserts that context “contains the functionality to allow the agent to successfully ‘navigate’ through the current situation."

• Context across levels of personal, project, group and organisation. It consists of people and their expertise, information sources, informational documents and the evaluation of their relevance, and relevant pragmatic documents (Snowden and Grasso, 2000).

• Context as a spectrum of data described from a user perspective: computing context, user context, physical context, time context, and social context (Gu, 2009).

• Context as related to capability and affordances. No reference for this was found, but the representation is a natural progression from the previous perspectives.

Context-Aware Decision Support

Computational context-aware decision support (CaDS) systems are evolving from ontology-based expert systems to attribute-based neural network systems. A (CaDS) system “consists of a situation model for shared situation awareness” (Feng, et al., 2009, p. 455).

Such systems are intended to address information overload, for example, in a Tactical Information Prioritization System (TIPS) (Marmelstein et al., 2008, p. 259). The aim of such research “is to enhance the decision-maker's perception, comprehension, and projection of the underlying knowledge space”(Hanratty, et al., 2009, p.1).

Dourish and Bellotti (1992, p. 107) state that awareness is an understanding of the activities of others, which provides a context for your own activity. “Awareness supposes that one is able to transform pieces of contextual knowledge into a proceduralized context at the current focus of attention” (Mäkelä et al., 2018, p. 7253).

To date, context-aware decision support systems have been designed along the lines of expert systems, employing "ontology-based decision support” and consisting of “sensor agents to detect raw-level data, a context management agent for handling context data, an information service agent, an operational decision support agent, and user agents for maintaining user information” (Song et al., 2010, p.1).

Contemporary approaches are studying the use of deep learning. Early work found “the intuition of equating the template attribute weights to neural network weights resulted in a good method to learn the weights directly from observation of prior agent behaviour” (Gonzalez, 2004, p. 169) supporting Context-based Reasoning (CxBR) as “a human behaviour modelling technique that uses this approach to model human behavior.”

Context Awareness and Recognition

‘Concept awareness’ denotes the capability or fact of being aware of context; by contrast, ‘context recognition’ describes the process or method of achieving context awareness.

Bricon-Souf and Newman (2007) describe context awareness as including "the ability... to detect, sense, interpret, act and respond to aspects of the environment, such as location, time, temperature or user identity."

We could say it is the ability to examine the environment and react to the dynamical changes such as the location of user, the collection of nearby people, hosts, and accessible devices, and adapt their behavior based on the context of the application and the environment.

We see similar definitions of 'context awareness' applied to both human and computer applications. Dey (1999) for example writes “A system is context-aware if it uses context to provide relevant information and/or services to the user, where relevancy depends on the user's task.”

A variety of context recognition mechanisms may be employed. For example, in a survey of research on context recognition in surgery, Pernek and Ferscha (2017) identify the following:

• Environment tracking - use of “pluggable monitors and devices that can be connected and used to infer surgical workflow” (p. 1722)

• Kinematic tracking - “focused on tracking the location of surgical instruments or quantifying hand movements of surgeons” (p. 1723)

• Video tracking - “recognize actions from intra-body video images” (p. 1723)

• Cognitive state context - through eye-gaze, skin response, heart-rate, force (p. 1724ff)

In computer applications, the predominant mechanisms employed were machine learning or neural network-based pattern recognition algorithms. For example, Radu et al. (2018) study “the benefits of adopting deep learning algorithms for interpreting user activity and context as captured by multi-sensor systems” (p. 157.2). Similarly, Billones et al. (2018) discuss the use of deep learning for vehicular context recognition. Alajaji et al. (2020) “propose DeepContext, a deep learning based network architecture for recognizing a smartphone user's current context.”

Context as schema

The general sense of a schema is as a semantic representation consisting of a form of representation as a generalisation combined with elements or blank spaces that are filled by concrete particulars to constitute an instance of the schema (Bartlett, 1932, Corcoran and Hamid, 2016).

Context may be thought of as “a mental codification of experience that includes a particular organised way of perceiving cognitively and responding to a complex situation or set of stimuli” (Merriam-Webster, 2024). There are senses of ‘schema’ in logic, psychology, computer science. Schemas may be thought variously as:

• consisting of i) set of words and blanks, ii) mechanism for filling blanks, or

• consisting of i) individual concepts, ii) mechanisms to connect concepts (Gal'perin, 1989, p. 65)

• consisting of scripts (Schank and Abelson, 1977)

Schemas are recognized to be constantly changing. Bartlett’s “concept of schema emphasises the dynamic and evolving nature of these cognitive constructs, which continuously adapt as we encounter new information” (Main, 2023).

Schema Development

Depending on the discipline or perspective, schema development may be described as ‘orientation’, ‘view’, or ‘case-based’.

‘Orientation’ is one of the steps in the OODA loop, discussed above. Orientation is depicted as “a schema to elucidate the role of human cognition (perception, emotion, and heuristics) in defense planning in a non-linear world characterized by complexity, novelty, and uncertainty” (Johnson, 2023, p. 43).

In IT and database development for systems such as JBI-IM (discussed above), context is represented as the development of various ‘views’ representing various ways to display underlying data schemas.

Schema Activation

Schema ‘activation’ is the deployment or retrieval of a schema to be applied or descriptive of a particular situation, and is often depicted as a cognitive process. For example: “Activating schemata and training students to use reading strategies are both generally effective in reading comprehension skills” (Cho and Hyun, 2020, p. 49). “Through schema activation, judgments are formed based on internal assumptions (bias) in addition to information actually available in the environment” (Worthy et al., 2024).

Schema Change

Schemas may change either through accommodation or assimilation of new data through either a top-down or bottom-up process.

For example, in the Composition Modeling Framework (CMF) (Staskevich et al., 2007), "when existing schemas change on the basis of new information, we call the process accommodation. In other cases, however, we engage in assimilation, a process in which our existing knowledge influences new conflicting information to better fit with our existing knowledge, thus reduc(ing) the likelihood of schema change” (Worthy et al., 2024).

Context as frame

A ‘frame’ is most generally thought of as an organisation of experience (Goffman, 1974) and in this sense more of a cognitive or psychological construct than semantic. It is an interpretation of reality “that puts the facts or events referred to in a certain perspective” (Morasso, 2012, p. 5).

From a more computational perspective, Minky’s (1974) account is an elaboration of the schema. “Here is the essence of the theory,” writes Minsky. “When one encounters a new situation (or makes a substantial change in one's view of the present problem) one selects from memory a structure called a Frame. This is a remembered framework to be adapted to fit reality by changing details as necessary.”

Similarly, in their consideration of choice theory in uncertain conditions, Tversky and Kahneman argue that “the normative and the descriptive analyses of choice should be viewed as separate enterprises” (1986, p. s275) with framing describing the former (for example, where someone is risk-tolerant or risk-averse).

Lakoff (2010, p. 71) describes frames as “structures (that) are physically realized in neural circuits in the brain. All of our knowledge makes use of frames, and every word is defined through the frames it neurally activates.”

Examples

The concept of a frame is at once less formal and more detailed than the schema, and consists not only of a generalised description of a situation or collection of data, but also objectives, expectations or values. These are illustrated with the following examples:

• “The common good” in diplomacy in military affairs (Karadag, 2017).

• “the frame of arms control” in distribution of verification resources (Avenhaus and Canty, 2011).

• “Luttes de sens, cadrages et grammaire lexicale en contexte révolutionnaire” (Struggles for meaning, framing and lexical grammar in a revolutionary context) (Rey, 2020).

• “Framing war: Public opinion and decision-making in comparative perspective” - “uses the recent war on Iraq as a case study, focusing on the elite and media framing of this event in order to examine the interaction between the political elite and the mass public” (Olmastroni, 2014).

• “the analysis of the strategic culture of Hungary is approached from a perspective that will frame the cultural and ahistorical view of the international policy of the neorealists with its own cultural and historical dimension” (Jeremić, 2021, p. 51).

• “identifying four related concepts that help frame how a COP may improve an organization’s efficiency and effectiveness: effectiveness-based measures, decision rights,schwerpunkt, and neutral integrators” (Pyles et al., 2008, p. 5).

Frame Vs Framework

A frame should be distinguished from the related but distinct concept of the ‘framework’. The latter is not a cognitive or psychological construct, but rather a method or process designed to explain, guide or improve decision-making (for example, Elgoff and Smeets, 2023, p. 502). In this context, a framework is best viewed as a decision-making or design tool (see ‘Decision-Making, above).

Context in metaphor

Metaphor is a powerful instrument for creating and representing frames in cases where literal representation is insufficient.

“The concepts that govern our thought are not just matters of intellect,” writes Lakoff (1980, p. 3). The metaphor ‘argument is war,’ for example, “is one that we live by in this culture; it structures the actions we perform in arguing.” Similarly, Taylor (2008, p. iii) writes, “The conception of literal meaning adopted by both semantic and pragmatic metaphor theorists, which roughly indicates an adherence to a lexical authority and conventionally accepted grammar, is far too limited in scope to account for what is generally taken to include literal meaning in the use of language.”

Metaphor may be thought of “as an eminently cultural linguistic phenomenon”, however, “There are several different ways of thinking about the nature of context in metaphor production that is not necessarily cultural” (Kövecses, 2017, p. 307).

Metaphors both define and are defined by context. “The purpose of metaphorical framing is to convey an abstract or complex idea in easier-to-comprehend terms by mapping characteristics of an abstract or complex source onto characteristics of a simpler or concrete target” (Wikipedia, 2024). It “tends to illuminate certain aspects while obscuring others” (Norscini and Daniela, 2024, p. 14). Thus a complex phenomenon is rendered more concrete.

Context as model

Context as a model is predominantly found in the form of a ‘context model’. “Context models are used to illustrate the operational context of a system - they show what lies outside the system boundaries” (Kurkovsky, 2024; Sommerville, 2015, Chapter 5).

In an ontology, a context model helps define a subject using a semantic analysis of information related to the subject. Wang et al. (2004, pp. 18-19) describe several informal context modelling approaches and present a formal context ontology. A software system context model “explicitly depicts the boundary between the software system and its external environment” (Johnston, 2021). A physical system context model may define an environment for a software simulation, for example, digital twin (Sahlab et al., 2022, p. 463).

Large language models (LLM) also have mechanisms to define context. For example, a ‘context window’ defines the request space for an LLM. A recently released version of Google Gemini defines a 1 million token context window that allows it “to understand up to one hour of video, 11 hours of audio, over 700,000 words (so it could read, digest and answer questions about Tolstoy's War & Peace) or over 30,000 lines of code” (Pichai, 2024).

Today, model context protocols (MCP) are used by generative AI systems such as Claude as a mechanism connecting them to underlying systems and information such as graphs and databases on local filesystems or accessible in the cloud (Anthropic, 2024).

Types of Model

It is beyond the scope of this review to identify and define the full scope of models and model technology; the typologies offered below provide a sense of this scope with respect to context.

Process Models:

• Mathematic - for example, “mathematical models of combat activities and combat means of destructions, and their development paths of the use of troops in the process of preparation” (Mikayilov and Bayramov, 2019, p. 156)

• Deterministic - for example, to define ship speed optimisation from the perspective of cash flow (Beullens et al., 2023).

• Stochastic - for example, decision support in a dynamic environment such as a wildfire (Roozbeh et al., 2021).

• Forces – for example, Porter’s Five Forces (Porter, 1979).

• Neural Network Model - a simplified model of the operations of a human brain involving creation and activation of patterns of connection (IBM, 2021).

Business Models:

• Logic Model - a mechanism for describing inputs and outputs for translation of business data into outcomes or benefits. For example, Paul et al. (2015, p. 17).

• Mental Model - “ an overarching term for any sort of concept, framework, or worldview that you carry around in your mind.” Clear (2024) provides a pretty good list. Boyd (1973) describes forms of mental models as patterns or concepts in his description of OODA (described above).

• Case - a specific business-focused description of a business situation, either for study after the fat or that “puts a proposed investment decision into a strategic context and provides the information necessary to make an informed decision” (TBS, 2009).

Computational Models:

• Data Model - a representation of data structures, may be a schema (see above) or ontology, depicted textually (XML, JSON) or as a diagram (Entity Relationship (ER) or Universal Modelling Language (UML)). See for example ‘context data model’ (Ceri et al., 2007) or ‘contextual design model’ (Holtzblatt and Beyer, 2017).

• Simulation - for example, “The use of modeling and simulation offer a better understanding of the concepts and solutions for commander's decision making” (Cîrciu et al., 2010, p. 93; Connable, et al., 2014).

• Multi-Objective Optimization - for example, a multi-objective model for hub location and cost sharing (Mrabti et al., 2022).

Validation

Models are intended to serve as representations of processes, data or physical environments. As such, unlike schemas or frames, models have a unique requirement of validation. The following terminology is employed:

• Verification: The process of determining that a model implementation and its associated data accurately represent the developer’s conceptual description and specifications.

• Validation: The process of determining the degree to which a model and its associated data provide an accurate representation of the real world from the perspective of the intended uses of the model.

• Accreditation: The official certification that a model, simulation, or federation of models and simulations and its associated data is acceptable for use for a specific purpose. (All quoted from AcqNotes, 2024; DoD 2008; Owen, and Chakrabortty, 2022).

In a wider context, other criteria and terminology may be used to evaluate models, for example, model fit and measurement invariance (Goldammer et al., 2024). Similarly, an ‘inference to the best explanation’ model minimally consists of the following:

• Abduction: The generation of candidate hypotheses and theories.

• Epistemic value: One or more epistemic values that order theories.

• Theory evaluation: An aggregation operation that takes orderings of theories and yields an overall ordering (Quoted from Rast, 2023, p. 3).

Additionally, theory evaluation may consider ‘epistemic virtues’ such as simplicity, paucity, or commensurability.

Summary

In this section we considered the nature and attributes of context as it in forms human and computational cognition, and in particular, expanded upon three major interpretations of context: as a schema, as a frame, and as a model.

It is not clear that any individual interpretation of context offers a comprehensive understanding of decision-making as referenced in section 1. The three interpretations of context are themselves contextual in nature, offering a mixture of mechanism and metaphor in an effort to convey an intuitive understanding of the subject.

In the next section, we will examine the role of data in the decision-making process generally and offer a broader decision-making model that explicitly incorporates contextual factors.

--

Image source: Figma https://www.figma.com/resource-library/context-diagram/

This article is based on work completed for Defence Research and Development Canada, Contract Report DRDC-RDDC-2025-C035

References will be listed after the series is complete.