Get best of the week

Semantics and activity

The text “ Semantic digital systems ” noted that modern semantic technologies do not yet have full-fledged tools for describing activities. Semantic data is traditionally represented in the form of an object graph, the nodes of which are entities or values, and edges - properties (relations and attributes). Such graphs fix the static state of the subject area. And such a simultaneously captured data set should rather be considered as a graph of knowledge about the subject area, and not as its model, especially if the subject area is an activity, and not a set of unchanging facts.

When considering a subject area as an activity, we should first of all be interested not in entities and their properties, but in acts, that is, events of changing properties. And of course, a lot of relevant, that is, necessary for modeling a domain, statements should not be limited to a description of the current state of activity, but should be a sequence of statements fixing events of changing entities. In fact, the point is that the semantic model of the subject area should not be represented by a static object graph, but by a temporal acyclic directed graph.

To make it clearer what we are talking about, let's look at a few simple examples.

From the relationship “Sasha and Masha’s spouse” with the fixed attributes of individuals “Sasha man”, “Masha woman”, the relations “Sasha husband Masha”, “Masha’s wife Sasha”, as well as relationship relations between spouses and their parents: son-in-law, daughter-in-law , mother-in-law, father-in-law, etc. However, it should be noted that all this set of facts was generated by one event “Sasha and Masha got married” (received a marriage certificate), which should be saved as an affirmation, and all relations generated by this event can be automatically generated in response to search queries using axioms. But the most interesting and important thing in this example is that the event “married” can be recorded event “Sasha and Masha divorced”, which should cancel all relations generated by the first event.

The above considerations suggest that we strictly distinguish between acts of activity that define semantics and the relationships that they generate, which, without adding content, are used only in communication, that is, to simplify communication. An ordered sequence of events like “Abraham begat Isaac; Isaac begat Jacob; Jacob gave birth to Judah ... ”is an exhaustive basis for understanding who is here who's son, grandfather or great-grandfather. That is, the ontology can be divided into (1) the substantial (event, active) level at which events of changing the status or properties of individuals are recorded, and (2) purely linguistic, in which the rules for naming relations between individuals are determined. Moreover, the second level of the ontology is necessary only for serving search queries, because it is clear that the user will be interested in “how many nephews did Magog have?”,but he won’t formulate his question in such a way “how many children were born by those born from the same one from whom Magog was born?”.

The above is true for all relationships - they are all derived from acts of activity. The relationship “Ivanov works in Gazprom” and “Gazprom has an employee Ivanov” is generated by the event “Ivanov is hired by Gazprom.” It is the last statement that should be preserved as a statement about individuals, and Ivanov’s presence in the list of Gazprom employees and Ivanov’s place of work are obtained from this statement according to elementary rules (axioms). Well, it is obvious that data on individuals must contain the sequence of all events of hiring and dismissal, that is, they should be a temporal directed acyclic graph.

As an example, demonstrating not just advantages, but the semantic necessity of the event approach, we can cite the fixation of the relationship “piece - whole” (piece of bread - a loaf). An attempt to describe directly through the part-whole relationship will lead to the absurdity “a piece of bread is part of a loaf”. After all, it is clear that when we have a loaf, then there is no such object as a piece, and when we cut a piece of bread, there will be no whole, that is, a loaf. Correctly, this situation can be described only with the verbal statement “a piece was cut off from a loaf”, which is tied to a specific moment in time and from which all conclusions can be drawn about the relationship between the loaf that existed before the act of cutting the piece and the new individual “piece” that appeared only after this event .

So, if we set ourselves the goal of modeling activity, then we must fix acts / events (with a verb connective) as statements, building them into temporal sequences - processes and actions. With this approach, the data will be organized into directed acyclic graphs. In this case, object graphs describing the properties of the entities of the subject area should be obtained as a result of a logical inference from a set of acts, that is, events of changes in properties.

More articles:


All Articles