Post-industrial society operates with huge volumes of information both in everyday and professional activities. A large amount of information causes difficulties in making decisions within the framework of rigid time constraints.
A variety of software automation of human activities are used to solve this problem. However, it is necessary to adapt them to the specifics of a particular problem area (PrA) and its contexts for the effective operation of these tools [1,2,7,10,18,19,20].
Thus, "trained" automation software solves the tasks more efficiently, but they require considerable resources (human and temporary) for training.
In this paper, an attempt is made to construct a KB. The content of the KB is an applied ontology. The basic requirements for KB are [26]:
• adaptation to the specifics of PrA based on contexts;
• reliability and speed of ontology storage;
• the presence of a mechanism of logical inference;
• availability of tools to simplify work with the KB for unprepared users;
• availability of mechanisms for importing data from external information resources.
As you can see from Figure 1, the KB consists of the following subsystems:
(i) Ontology store:
• Neo4j [12];
• content management module;
• ontology import/export module. (ii) Inference subsystem:
• inference module. (iii) A subsystem for interaction with users:
• screen forms generation module. (iv) A subsystem for importing data from wiki-resources:
• a module for importing data from wiki-resources. (v) A subsystem for importing data from relational databases:
• a module for importing data from relational databases.
Figure 1. Knowledge base architecture.
Ontology is a model of the representation of the PrA in the form of a semantic graph [9]. Graph-oriented database management system (Graph DBMS) Neo4j is the basis of the ontology store for KB. Neo4j is currently one of the most popular graph databases and has the following advantages:
(i) Having a free community version. (ii) Native format for data storage. (iii) One copy of Neo4j can work with graphs containing billions of nodes and relationships. (iv) The presence of a graph-oriented query language Cypher. (v) Availability of transaction support. Neo4j was chosen to store the description of the PrA in the form of an applied ontology, since the ontology is actually a graph. In this case, it is only necessary to limit the set of nodes and graph relations into which ontologies on RDF and OWL will be translated.
The context of an KB is some state of content of KB, obtained during versioning or building a content of KB using different "points of view" [6,8]. Figure 2 shows an example of the translation of the owl representation of ontology of family relations into the entities of the KB. Formally, the content of the KB can be represented by the following equation:
where t is a number of the KB contexts,
1 , I T i 2 , . . . , I T i n } is a set of KB objects within the i-th context, P T i = {P T i 1 , P T i 2 , . . . , P T i n } is a set of KB classes properties within the i-th context, S T i = {S T i 1 , S T i 2 , . . . , S T i n } is a set of KB objects states within the i-th context,
} is a set of the logical rules fixed in the KB within the i-th context, R T i is a set of KB relations within the i-th context defined as:
where R T i C is a set of relations defining hierarchy of KB classes within the i-th context, R T i I is a set of relations defining the "class-object" KB tie within the i-th context, R T i P is a set of relations defining the "class-class property" KB tie within the i-th context, R T i S is a set of relations defining the "object-object state" KB tie within the i-th context, R T i F is a set of relations generated on the basis of logical KB rules in the context of i-th context. Principles similar to the paradigm of object-oriented programming are at the basis of the content of the KB:
• KB classes are concepts of the PrA;
• classes can have properties, the child-class inherits properties of the parent class;
• objects of KB describe instances of the concepts of the PrO;
• specific values for the properties of objects inherited from the parent class are determined by the states; • logical rules are used to implement the functions of inference by the content of KB.
The inference is the process of reasoning from the premises to the conclusion. Reasoners are used to implement the function of inference. Reasoners form logical consequences on the basis of many statements, facts and axioms. The most popular at the moment reasoners are [5,17]:
These reasoners are actively used in the development of intelligent software. However, Neo4j does not assume the possibility of using similar default reasoners. Thus, there is a need to develop a mechanism for inference based on the content of a KB [3,4].
Currently the Semantic Web Rule Language (SWRL) is used to record logical rules [24]. These SWRL rules describe the conditions under which object a has "nephew-uncle" relation with object c. Formally the logical rule of the KB is:
where T i is the i-th context of the the KB, A T ree is the tree-like representation of a logical rule
The tree-view A T ree of a logical rule F T i is:
where Ant = Ant 1 ΘAnt 2 Θ . . . ΘAnt n is the antecedent (condition) of the logical rule F T i ; Θ ∈ {AN D, OR} is a set of permissible logical operations between antecedent atoms; Cons is the consequent (consequence) of a logical rule F T i . Figure 3 shows an example of a tree-like representation of two logical rules for the ontology of family relations. That rules describes the father-child relationships.
The tree-like logical rule is translated into the following SWRL: Thus, the rules are translated into their tree-view when imported into the KB of logical rules in the SWRL language.
The presence of a tree-like representation of a logical rule allows to form both a SWRLrepresentation of a logical rule and a Cypher-representation based on it.
Relations of a special type are formed by using Cypher to represent the logical rule between entities of the KB. Figure 4 shows the content of KB after executing the Cypher queries that were built for the logical rule shown in Figure 3. These relations correspond to the antecedent atoms of the logical rule. Formed relationships provide the inference from the contents of the KB.
The dynamic graphical user interface (GUI) mechanism is used to simplify the work with KB of untrained users and control of user input [11,13,21].
You need to map the KB entities to the GUI elements to build a GUI based on the contents of the KB. Formally, the GUI model can be represented as follows:
where L = {L 1 , L 2 , . . . , L n } is a set of graphical GUI components (for example, ListBox, TextBox, ComboBox, etc.), C = {C 1 , C 2 , . . . , C n } is a set of KB classes, I = {I 1 , I 2 , . . . , I n } is a set of KB objects, P = {P 1 , P 2 , . . . , P n } is a set of properties of KB classes, S = {S 1 , S 2 , . . . , S n } is a set of states of KB objects. Thus, the contents of the KB are mapped to set of GUI components. This makes it easier to work with KB for a user who does not have skills in ontological analysis and knowledge engineering. It also allows you to monitor the logical integrity of the user input, which leads to a reduction in the number of potential input errors.
At present, wiki-technologies are used to organize corporate KB. It is necessary to solve the task of knowledge extracting from wiki-resources [14,15,16,23,27]. Table 1 contains the result of mapping the KB entities to the wiki-resource entities [22]. Thus, it becomes possible to import the structure of external wiki resources for initial filling of the KB contents.
Table 1. The correspondence between the wiki-resource entities and the entities of KB.
The entities of wiki-resource Class Category Subclass Subcategory Object Page Class properties
The infobox elements (properties) Object states
The infobox elements (values) Relations Hyperlinks Also, a content of KB can be built on the basis of an analysis of the content of wiki-resources pages. In this work the Syntaxnet [22] framework to construct a syntactic tree Synt of content of wiki-resources pages is used. Further, using a set of rules Rule Synt , a syntax tree Synt is translated into entities of KB.
Formally the functions of translating a syntactic tree into entities of KB:
where N Synt is a set of nodes of the syntactic tree Synt, Rule Struct Synt is a set of rules to translating nodes of syntactic into structure entities of the KB, Rule Content Synt is a set of rules to translating nodes of syntactic into content entities of the KB, {C O , P O , R O P } T i is a set of structure entities of the KB within the context T i (eq. 1), {I O , S O , R O I , R O S } T i is a set of content entities of the KB within the context T i (eq. 1). Formally the rules to translating nodes of syntactic into entities of the KB:
where N Synt i is the i-th node of syntactic tree. Thus, it becomes possible to extract knowledge from the structure of wiki-resource and contents of wiki-resource pages and present the extracted knowledge as a content of KB.
Relational databases are widely used for data storing and contains subject area description in the form of interconnected tables. Nowadays, researchers of various scientific groups are involved in solving the problem of extracting knowledge from relational databases.
The relational data model can be represented as the following expression:
where E = {E 1 , E 2 , . . . , E n is a set of database tables (entities), R = {R 1 , R 2 , . . . , R i , . . . R n is a set of relationships between database tables:
where E j , E k are database entities; F (x) is the relationship between entity E j and entity E k , G (x) is the relationship between entity E k and entity E j . Scope of functions F (x) and G (x) are U -single relationship and N -multiply relationship.
For mapping of relational database structure with KB structure special functions are used: Thus, it becomes possible to extract knowledge from the contents of relational databases and present the extracted knowledge as a content of KB.
Thus, the use of KB stored in the Graph DBMS in the decision support process presupposes the existence of a certain set of mechanisms:
• organization of inference on the content of KB by translating SWRL-rules into Cypherstructures; • building a graphical user interface based on the contents of KB;
• automated import of knowledge from structure and content of wiki-resources;
• automated import of knowledge from relational databases. These mechanisms allow to automate the learning process of KB and simplify the work of specialists with KB. The application of a contextual approach to the storage of knowledge raises the effectiveness of the use of subject ontologies, allowing to adapt the KB to the characteristics of the PrA and to the requirements of specialists. This approach provides them with a tool that is convenient in a software dynamically changeable depending on the contents of the KB.
[1] Berant J, Chou A, Frostig R and Liang P 2013 Semantic parsing on freebase from question-answer pairs Proceedings of the Conference on Empirical Methods in Natural Language Processing (EMNLP) 1533-1544
This work was financially supported by the Russian Foundation for Basic Research (Grant No. 16-47-732054).