In a typical PCS, a context-aware application interacts with the physical environment and the user’s system in order to provide appropriate services. This interaction may be a response to the user’s request for a specific service or to the current context information with the aim of providing services that are relevant to the user. In such an environment, semantic similarity measures have been applied at several levels  : the comparison of an application’s components with respect to their appropriateness in a current context; the recommendation of services and collaborative filtering when comparing the preferences of multiple users with the ranking of services according to their relevance during the recommendation process; service discovery by the matching the description of a request with available services; lastly, the comparison of the current context with already known contexts or the detection of current situations.

Semantic similarity measures and context 

The definition of context according to Petit (2005) along with the majority of researchers is based on the four following axes:
1. There is no context without context: the concept of context must be defined in terms of a purpose. For example, the aim may be to adapt the interactive capabilities of a system dynamically;
2. Context is an information space that serves the interpretation: context capture is not an end in itself, but captured data must serve an objective;
3. Context is an information space shared by several actors: the user and the system;
4. Context is an infinite and dynamic space of information: context is not permanently fixed, but is constructed over time.

The following definitions of context should be in accordance with the aforementioned axes. First, Brezillon et al. (1999) defined two concepts relating to context: 1) the set of contextual knowledge (e.g. time, location) to be used in a decision problem, which is latent and cannot be used without an emergent objective; 2) the context as the product of the emergent objective or intention that uses a large part of contextual knowledge.

En 1994, Schilit and Adams categorized context according to six areas. The first three relate to the human factor: user information (e.g. clothes, biophysical conditions), social environment (e.g. proximity to other people), and user tasks (e.g. active user tasks). The other three areas concern the physical environment: location, infrastructure (e.g. resources, communication), and physical conditions (e.g. noise, brightness, weather conditions).

The definition of Dey et al. (2001) is the most cited: “context is any information that can be used to characterize the situation of an entity. An entity is a person, or object that is considered relevant to the interaction between a user and an application, including the user and the application themselves” (p. 5). This definition is evidently similar to Schilit’s because context is defined as a set of information collected from the user environment (person), physical environment (physical object), or system environment, with the objective of collection being the characterization of these environments.

Given the preceding definitions, we may say that context is definitely a set of information characterizing an environment, whether the user, physical, or system environment, and that the collection of this information must serve for an objective.

Impact of context 

Keßler (2007) defines context relative to the similarity measure in the following terms: “A similarity measurement’s context is any information that helps to specify the similarity of two entities more precisely concerning the current situation. This information must be represented in the same way as the knowledge base under consideration, and it must be capturable at maintainable cost” (p. 4). This definition gives rise to the following questions regarding the choice of contextual information to be included in the similarity measure between two concepts:
1. Impact: does the chosen contextual information improve the accuracy of the semantic similarity?
2. Representation: can this contextual information be represented in the knowledge base?
3. Acquisition: can this contextual information be acquired at a reasonable cost?

Semantic similarity between contexts 

In a PCS, the services provided to a user relate to the user context (environmental, systembased). The identification of context is thus an essential task. The question that arises is therefore, “What services must an intelligent device in a PCS provide to a user when the current context is identified?” The identification of the current context is defined by the contextual information related to the triggering of a service as well as a situation or “current context” in the set of current contextual information, similar to a known situation or context (Benazzouz 2012), with each identified situation being linked to one or more of the services to be provided. This identification forms the basis of the rule-based adaptation mechanism, which is a set of conditional rules with the form: if (contextual information I) then (service S).

A situation is “a snapshot of the environment at a given point in time” (Ramparany et al. 2011). Identifying a situation is based on data mining techniques. Once identified, semantic similarity measures are applied in order to compare it with situations with known services. In Dietze et al. (2008), semantic similarity is measured against the Euclidean distance between the contextual data vectored in mobile situation spaces. Gicquel (2012) modeled the spatiotemporal context of a museum visitor in an ontological form, with the semantic similarity measures being used to recommend artwork similar to the interests of the user by comparing the properties of two concepts in the knowledge base. The similarity measure is a modified version of the similarity proposed by Pirró and Euzenat (2010), which combines the similarity calculation based on Tversky’s model with that of informational content.