Concept, algorithm and scheme of information space formation is considered in this paper. This algorithm involves the use of an information system, which is actually a software basis for maintaining the information space. Identification of the input object in the information space allows us to identify it univocally according to the relevant features. For this, an identification method based on consecutive analysis of the object's characteristics is used. Classification of the information objects in view to add them to the information space on basis of multidimensional discriminant analysis is considered. Efficiency analysis of the input objects search in information space is perform.
Unified information space can be considered as the information model of a complex subject area. It
includes the information objects, the links between them, the space environment and the processes for
the creation the unified information space [
The contradiction arises between the heterogeneous nature of sensors for collecting features of
objects and the requirement for unified data presentation [
An information object is a formal description of some object by its main parameters. It describes as
a tuple of parameters of the object, and all values of parameters are determined by characteristics of
real object. The information about objects in the unified information space changes in dynamic way [
The unified information space allows us to give the access to the common information without the limits for place and time. The computer system is form as the basis for unified information space and it performs the following main functions: transformation of information about objects and formation of unified in-formation space; to provide users with information about objects.
The creation of an information space is intended to provide access to some common information without the space and time limits. The basis of the information space is a set of computer systems, local networks, open networks (Internet), software (operating system, applications, databases, mail services, etc.). Also, the creation the information space is based on the mechanisms that combine the different information systems to each other.
The main purpose of the information space is to unambiguously compare each input object to the
respective information object in information space [
Identification of the information object is performed by the external or internal parameters, taking
into account the interaction of the information object in the information space. To do this, each
information object is assigned a set of parameters that, to some extent, characterize the object, i.e. form
the image of the information object. Similarly, an information request to input object is formulated as
brief description [
Due to this, the procedure of the input object identification is just a simple comparison of its
parameters with a certain image of the query [
We present an algorithm for the information space design that based on the information system (Fig. 1):
1. Information system is a distributed system that performs the conversion of information from various sources and usually it comes in different formats, into a single set of information object’s parameters, by which users of information space can uniquely identify the input object.
2. The user's entry point into the information system is created. There may be many such entry points. 3. A single hardware space formed for the user regardless of the certain entry point. That is, to process the information request of any user, the same hardware is used.
4. The software tools are available to the user from anywhere.
5. Each user has an account, which contains the software tools that are often used by him, as well as his actions, including in relation to specific services.
6. User accounts are stored remotely and user may access them from the different login points. 7. The user enter into the information space via entry point, and there is performed the adjusting of his environment, i.e. the services that he often used are fixed, and his actions in for the specific services are analyzed.
8. As a result, the information system is present to the user as a single space, regardless of the current entry point.
Figure 2 shows the general scheme of the information space formation.
The data collection in the information space and its formation is presented in the form of a graph, which reflects an integrated model of the information space, where the nodes of the graph correspond to the information objects, and arcs are the relationship between these objects.
, , , , (1) where is the set of information objects; is the set of parameters of information objects; is the set of connections types; is mapping that specifies a certain relationship from set of relationships between the information objects.
We define the set of connections types as the union of the set of connections between information objects and the set of connections between the parameters of information objects: , (2) where is the set of connections between information objects; the parameters of information objects.
In turn, the single information space is a union of information spaces: is the set of connections between where is the number of information spaces.
Moreover, each single information space is presented as a graph: , , , , , , where is the set of information spaces; is the set of connections types, which is similar to expression (2) are the union of a set of connections between information objects of information spaces set of relationships between the parameters of information objects of information spaces and the
is mapping that specifies a certain relationship of set of types between set of the information ⋃ spaces .
For each input object 1, information object
1, is determined. In case when , ⋂ ⋂
, that enters the inputs of the information space and , which already are in the information space, the set of parameters ,
does not completely match the set of parameters and , is considered identified and is not added to the information space . such an input object
is considered new and is added to the information space .
Otherwise in case when , matches the set of parameters and , such an input object between the new information object and the existing information objects of the are added. An important goal of the information space design is to transform the
Also, the links information space present unambiguously. input information in the certain way when each information object in the information space will be
The parameters of information objects in the information space should be defined in unified format and their number must be the same. The users of the information space should also perceive information objects unambiguously. Usually, the input data in the information space has heterogeneous nature. The goal of the internal mechanisms of the computer system is to transform the heterogeneous information coming in different formats and from different sources into a single set of parameters of information objects, by which the users of information space can uniquely identify the input object. (3) (4) (5) 3. Classification of information objects in the information space
We use the multidimensional discriminant analysis to classify information objects in order to add them to the information space. To do this, for each information space we built a discriminant function in the following form: calculate the value of the discriminant function .
Then, based on a set of values
for each information space, we calculate the main statistical characteristics of the obtained values of discriminant functions: expected value and dispersion . These characteristics make it possible to determine the affiliation of a new information object ⋯
, ;
where space; is the value of the discriminant function for -th information object in ℎ-th information are the parameters -th information object that are significant (have a value other than NONE) for ℎ-th information space, 1, corresponds to ℎ-th information space, ℎ 1, ; is the free coefficient of discriminant function, which are the coefficients of the discriminant function, which correspond to the significant parameters of each information space, 1, .
Then, for each information object 1, in each information space ℎ 1, , we to a particular information space. discriminant functions are calculated for each information space. New information object case when its characteristics meet the following condition:
For each input object , which should be added as a new information object , the values of and are compared with the obtained statistical characteristics
will be added to the information space in . (6)
The implementation of the above discriminant analysis allow us determine the certain information space to which the new input information object should correspond. space
Let us analyze the effectiveness of the input objects search in the information space. We created an information space of 20,000 information objects for the experiments. The part of the missing parameters in the information objects (NONE) was set at the level 6%. After that, the information space was reconstructed by excluding of the duplicate information objects. We performs 100 experiments with certain probabilities (5, 10, 15, 20, and 25%, respectively), which reflects the fact that the parameter would not be read by sensors (NULL).
Let us consider the case when every of the 20,000 information objects is described by 7 parameters. Table 1 shows a fragment of 10 information objects ( ) of the information space ( ).
IO2 IO3 IO4 IO5 IO6 IO7 IO8 IO9 IO10 Р1 2±0.2 2±0.2 2±0.3 5±0.4 1±0.7 5±0.4 NONE 3±0.4 2±0.3 4±0.7 Р2 2±0.7 6±0.7 5±0.4 6±0.6 6±0.3 4±0.4 4±0.5 2±0.6 3±0.4 4±0.2 Р3 3±0.4
4±0.1 NONE 5±0.9 3±0.7 7±0.6 3±0.8 5±0.6 5±0.3 Parameters Р4 6±0.4 6±0.9 8±0.2 7±0.4 7±0.8 6±0.4 6±0.2 4±0.2 5±0.5 4±0.5 Р5 7±0.6 9±0.8 5±0.9 5±0.6 5±0.8 5±0.9 9±0.2 6±0.5 8±0.6 7±0.5 Р6 10±0.6 9±0.7 10±0.3 8±0.5 8±0.5 8±0.4 6±0.4 6±0.6 8±0.2 6±0.9 Р7 11±0.4
8±0.5
10±0.9 8±0.1 8±0.9 11±0.4
In this case, the interval length for each parameter is 5 units, for example, for parameter P1 [1;6).
The sensors read all parameters values for the input object and there was an unambiguous
New object: 1.6 2.9 4.2 4.7 5.3 8.5 8.4 Search object: New object: 4.9 3.3 4.5 7.9 5.2 6.5 11.7 Search object:
Object absent! information space. IO 00603 1±0.6 2±0.6 4±0.6 4±0.6 5±0.6 8±0.4 8±0.5
The sensors read all parameters values of the input object and did not identify it, it means that no information objects corresponding to the following input object was found in the information space: In this case, the input object considered as a new information object and it can be added to the Object absent! 5. The sensors did not get the value of one or more parameters and after the interaction of the information objects between themselves in the information space it was found that this object did not have this parameter (NONE).
New object: 2.7 4.7 4.1 8.3 5.5 Null 7.6 Search object: IO 12602 2±0.5 4±0.4 4±0.3 8±0.7 5±0.8 10±0.9 7±0.8 ReCreateObject: 2.7 4.7 4.1 8.3 5.5 Null 7.6 ReCreateObject: 2.7 4.7 4.1 8.3 5.5 Null 7.6 ReCreateObject: 2.7 4.7 4.1 8.3 5.5 None 7.6 Search object: Object absent!
The integrated results of the experiments allow us to draw conclusion about the effectiveness of the input objects search in the information space for 7 parameters of information object and the length of the interval of 5 units. They are presented in Table 2.
Table 2 shows the low level of the efficiency of input objects search in the information space for 7 parameters and for interval length of 5 units.
Next, we reduced the number of parameters that describe the information object, and with the same initial data to repeat the similar experiments with the number of parameters from 4 to 6. The results of the experiments are presented in Table 3.
Table 3 shows the 100% probability for the input object identification in the information space for 4 parameters and the length of the interval of 5 units, i.e. there is no need to reduce the number of parameters. We design a graph on the data from Tables 2 and 3 in order to compare the efficiency of search of input objects in the information space for an interval length of 5 units and the number of parameters from 4 to 7 (Fig. 3). Figure 3: Graph of comparison of efficiency of input objects search in information space for interval length of 5 units and number of parameters from 4 to 7 Table 4 The efficiency of input objects search in information space for 7 parameters and interval length from
Probability that the parameter will not be read by sensors (NULL),%
We may conclude that increasing the length of the interval results in a strong decrease in the efficiency of input objects search in the information space. We design graph on the data in Table 4 in order to compare the efficiency of search of input objects in the information space for 7 parameters and the interval length from 3 to 5 units (Fig. 4). Probability that the parameter will not be read by sensors (NULL),% 3 4 5 Figure 4: Graph of comparison of efficiency of input objects search in information space for 7 parameters and interval length from 3 to 5 units Table 5 Efficiency of input objects search in information space for 7 to 10 parameters and interval length of 3 units
We design a graph on the data in Table 5 in order to compare the efficiency of search of input objects in the information space for 7 to 10 parameters and the interval length of 3 units (Fig. 5). 100 % tce 90 , j tob 80 u inp 70 e tfh 60 o ion 50 t a iifc 40 t ied 30 n f oy 20 t i iab 10 l b roP 0 5 10 15 20 25 Probability that the parameter will not be read by sensors (NULL),% 7 8 9 10 Figure 5: Graph of comparison of efficiency of input objects search in information space for 7 to 10 parameters and interval length of 3 units 5. Comparison of the results obtained by machine learning and experimental data
The adequacy of the suggested method for input objects identification in the information space was checked using machine learning methods.
The multivariate logistic regression using the LogisticRegression class from the Scikit-Learn library was used to identify information objects. By default, it applies a one-on-one strategy for more than two classes.
To switch to multivariate logistic regression, the multi_class hyperparameter is set to “multinomial” and the lbfgs solver is selected.
By default, we apply applied regularization, which can be controlled with the hyperparameter .
X = info_data [“InformationObject”] # prepared parameters vectors for information objects Y = info_data [“IdentifiedObject”] softamx_reg = LogisticRegression (multi_class = “multinomial”, solver = “lbfgs”, C = 10) softmax_reg.fit (X, Y)
This method was applied, for example, for the input object with the parameters [
Thus, we can conclude that the results obtained with the suggested method of input objects identification in the information space are confirmed by the corresponding results obtained with the machine learning method.
We develop an algorithm for the information space formation on the information system, which is actually a software basis for maintaining the information space. Identification of the input object in the information space allow us uniquely identify it by the appropriate parameters. To do this, we suggest the method of identification based on step-by-step analysis of the object parameters.