The methods and means of using ontologies within systems for the categorization of textual content were created. Also, a method for optimizing the definition of which rubrics best relate to a certain text content was developed. The intellectual system that will use the methods developed earlier, as well as other research results was implemented. The results will allow users to easily filter their text content. The system developed has intuitive user interface.
The task of content analysis is becoming more and more relevant in connection with
the rapid growth of the popularity of the World Wide Web, as well as the exponential
increase in the amount of content inside the network [
The aim of work is designing and developing a system of text categorization. The following structure of research was created to achieve the goal of work:
Analysis of machine learning algorithms.
Creating an algorithm for determining relevant categories in text content. Object of research is the process of creation of intellectual systems of text categorization, the main purpose of which is convenient and qualitative classification of the text content. The subject of the study is means, methods and ways of developing intelligent systems of text content text categorization using an ontological approach. The main requirement of the system is to eliminate the need for manual categorization of the text. The second requirement is to create a user-friendly system for text categorization. The final result is a system that will allow users to quickly and accurately categorize text content. Expected results of the development of such a system are: 1. 2. 3. 4.
Development of methods and tools for using ontologies in the text categorization systems.
Development of methods for optimizing the definition and improvement of the relevance of the categorization to the text.
Development of a system that would use both existing and innovative methods of categorization.
Development of user-friendly and clear user interface of the system. 2
In order to understand the design and development of the intellectual system, the
definition of this term should be specified. So, under the term «intellectual system»
means a technical or software system that is capable of solving tasks that are
considered to be traditional. These tasks belong to a specific subject area, knowledge of
which is stored in the memory of such system. An intelligent system usually includes
three basis blocks: knowledge base; mechanism of decision making; intelligent
interface. Intellectual system can be understood as intelligent information systems with
intellectual support that solves the problem without the participation of the decision
maker, in contrast to the intellectualized system in which the operator is present [
If we analyze the main methods of content categorization, namely text, one can
conclude that one of the most successful methods of categorization is methods used
by ontologies. It is also worth noting that these systems, for the most part, do not use
the full spectrum of opportunities and benefits of ontologies, which provides great
prospects for future developments in this direction. If to compare ontologies with
other methods of constructing knowledge bases, you can easily see the benefits of the
first one. Ontology is a standard of knowledge engineering, which has proven itself as
one of the best methods for representing objective knowledge [
However, in the field of ontologies there is a set of unresolved problems, the
solution of which will allow the development of fast and efficient systems for working out
the text, namely its categorization. The list of such tasks includes [
Under the formal model of ontology, O understand the three of the following form:
O = <C, R, F>, (1)
where C is a finite set of concepts (concepts, terms) of the domain (PO), which is
given by the ontology of O; R: C → C - a finite set of relations between concepts
(concepts, terms) of a given software; F - finite set of functions interpretation
(axiomatization, restrictions) defined on concepts or ways ontology O. It is worth noting
that the set C is finite and non-empty, and F and R must be finite [
O = <C, R, F, W, L>, (2)
where W is the importance of the concepts C, and L is the importance of the
relationship R. Such an expansion improves the system's results at times, since categories the
importance of meeting a particular concept differ for different categories. The same is
with the concepts: the connection between them may have different meanings [
In most cases, the graph is often used to provide ontologies (often a conceptual
graph). In the graph, vertices are the concept of software, and arcs are the relation
between different concepts. Depending on whether the axioms of the concepts are
defined, the vertices are divided into interpreted and not interpolated. The arches
(relation) can be vertical or horizontal. With the help of vertical arcs, the taxonomy of
the concepts of software is given. The requirement of horizontal arcs is to determine
the set of values and the area of definition of relations. In general, the structure of the
ontology can be defined by four categories of elements: concept; relation; axioms;
attributes. Concepts (classes) is general categories are organized according to the
hierarchy. The class can be considered as the union of all representatives of a certain
entity. That is, each class describes a group of entities united by common properties.
Defining which class belongs to one class is one of the most common tasks in systems
that use an ontological approach. Such a task is called categorization. Aattributes are
an ontology element that represents a certain class. It is a specific element that
belongs to one category. In elements of ontology there is a specific hierarchy. At the
lowest level, there are specific representatives (attributes). There are categories above
the instances. Above them are the relation between categories. In the top of hierarchy
there are axioms and rules that combine all these steps. Below is a schematic
representation of this hierarchy (Fig. 1). In order to build ontological model, first of all, it
is necessary to define the hierarchy of concepts (set C). Also, during the construction
of ontology, as an infological model, experts in subject areas should participate. For
qualitative ontology construction it is necessary that these specialists skillfully use
abstraction and combination. Also, when constructing an ontology, it is necessary to
construct atomic concepts from a set of differential attributes. For convenience, when
constructing ontologies, classification is often used. The classification is the method
of streamlining knowledge. Using classification approach, we divide objects into
groups based on their structure and behavior. In object-oriented analysis, by means of
determining the general properties of objects, the simplicity of the architecture of the
model system is achieved. It is because of the simplicity of the infological model of
the system that key mechanisms and abstractions are easy to find. In modern studies,
it is considered that there is no ideal hierarchy of classes, as well as the correct
classification of objects. Since there are no strict methods and rules for classification of
objects and their classes. This is due to the fact that is a compromise solution to
choose classes with which the system will operate [
It can be assumed that there is no system for text content categorization that does not use parsing or keyword typing. The key is a word, or a stable expression of the natural language, which expresses some aspect of the content of the document; a word that contains an important semantic load. Such a word can act as a key when searching for information. Parsing is a process in which the input sequence of characters is analyzed. The purpose of parsing is to break up the grammatical structure according to the prescribed formal grammar. In this analysis, the content becomes a kind of data structure. Often, as a data structure, a tree is used that repeats the syntactic structure of the input data. This is due to the fact that such a structure is very well suited for further processing. Stemming is algorithms work by cutting off the word to the base. This is achieved by rejecting the suffix, ending, and other auxiliary parts of the word. Although the results of the sedation, often reminiscent of the root of the word, the emulation algorithms are based on other principles than algorithms for determining the root of the word. Therefore, the word after stamping may be different from the morphological root of the word. In the tasks of information retrieval and linguistic morphology, the process of sedation is often used.
The stemming algorithms contain the 2 most common problems [
Over-stemming is when two words with different stems are stemmed to the same root. This is also known as a false positive.
Under-stemming is when two words that should be stemmed to the same root are not. This is also known as a false negative.
Stemming algorithms attempt to minimize each type of error, although reducing one
type can lead to increasing the other [
After analyzing modern works [
After analyzing the scientific studies of foreign and our scientists, one can conclude
that in the area of processing of information resources, such aspects as assessing the
quality of ontology, extracting knowledge from heterogeneous sources and
developing methods of integration between ontologies are the main ones in the direction of
research ontologies. Modern areas of researches that are related to the teaching of
ontologies, as well as their practical use in intelligent information systems, are:
1. Learning ontologies based on the analysis of texts in the natural language [
DAML + OIL) [
A thorough search of sites with the possibility of the text categorization was carried
out for the analysis of finished software. After searching for such sites, one can
conclude that the text classification market has few rivals for the system under
development [
Bureau (IAB).
The results of the categorizing were also better than in the previous systems and were fairly accurate. Also, this system has a much friendlier user interface (Fig. 4-5).
The main purpose of the system is the text content categorization. The main objectives of this system are the user-friendly interaction with the system, and qualitative categorization/clasification of the content. In order to achieve each of these objectives, we will break them down. This partition will look like this: 1.
o Authorization o Sending request for categorizing. o Reviewing categorized/classified articles.
o Filtering articles by sections.
Categorization content.
o Processing a categorization request. o Searching key essentials. o Analyzing found key entities. o Defining categories for articles.
o Optimizing weights for categories. In the text content categorization system an object that changes its state over time is the article itself. Fig. 9 shows the state diagram of the article. A sequence diagram shows a set of messages arranged in time sequence. Also, this diagram depicts which messages are transmitted between objects during an action. In UML, when constructing such a diagram, processes and objects are represented in the form of vertical lines, and messages between them are represented as horizontal lines. Messages must be ordered at the time of departure. Fig. 10 shows the sequence diagram for the text content categorization system. Since the diagram as a result of such association is rather voluminous, in order not to confuse people who will work with this diagram, the actions are numbered. The numbering begins with 1 and continues on the message movement in the system. Collaboration diagram is one of the most comprehensive. This diagram is most commonly used by system designers since they can see the overall picture of the system they are developing. Fig. 11 shows a collaboration diagram.
The latest diagram was a deployment diagram. Just like the previous diagram, this diagram is more technical. As a component diagrams, it is very popular among software developers, especially among DevOps. Because it is their competence to start and configure the system's nodes. This diagram can also be used in software development, in order to outline which nodes are needed by the system, which can facilitate the development, since no extra work will be done. Fig. 12 shows the deployment diagram of the intellectual system of categorization of the text content.
As described in the previous section, there are already systems that would implement the possibility of categorizing text content. There are different types of systems, both public and closed systems, access to which must be purchased. However, none of the systems can save results, although they can be used to improve the results. That is why it should be used to improve the competitiveness of the system. The first and most obvious problem is the user interface. Many existing systems have a big problem in this direction. And this should be used in the development of this system. The user interface is a very important part of the system, since it makes no sense to design a system that nobody will use. The next problem is the ability for users to save their articles. This can improve the system time, since it will not be necessary to rearrange articles. And if the user loses the result of the previous categorizing, he will always be able to view the previous results. The third problem also applies to the preservation of information. We can use the history of categorizations to improve its quality. Knowing to which categories are the user's articles belongs, we can improve the accuracy of classification. That is, if a user often writes on a single topic, then the possibility that his new article belongs to this topic is much bigger than that the article belongs to a completely new topic for the user. To solve this problem, you can connect a machine learning algorithm that could analyze the user's history. 7
For full operation of the system, a permanent connection with the server part of the system is required. Also, on the server should be a mandatory access to the Internet, because the server uses external resources. The system is based on the MERN stack using the JavaScript programming language. The user interface is a web page where the user can log in and take advantage of the system.
The purpose of the system is to automate the classification of the text content. The classification problem is quite acute in modern realities, since manual classification is a time-consuming process, and correctly selected columns improve both SEO and user-friendliness of the system. The system being developed uses the Dandelion API. Dandelion API - an open API designed for text analysis. In this system, Dandelion API is used to get the essence of the text. The entity that returns this API is part of the DBPedia ontology. Dandelion API is a multilingual API. It supports a stable analysis of 7 languages and more than 50 languages are in active development. DBPedia is a crowd-sourced community effort to extract structured content from the information created in various Wikimedia projects. Most systems of text classification have a limited number of available classifications, but the system developed thanks to DBPedia will have almost the maximum available at the moment the list of classifies.
The main part of the whole process of analyzing the text on the server is to process the essentials derived from DandelionAPI. For large text, DandelionAPI returns many entities among which the most important ones to be identified, as well as determine which sections they belong to. The solution to the task involves the following steps:
Definition of the language of the text. 2. Splitting the text into pieces. 3. Configure Dandelion API request. 4. Sending Dandelion API request. 5. Getting results from the Dandelion API. 6. Rejecting entities with low confidence. 7. Finding and removing entities that are alternate names of other entities. 8. Obtaining categories from entities. 9. Counting the number of times each of the categories met. 10. Finding the maximum weight for each category. 11. Finding the weight of each category. 12. Correction of weight of each category according to the user's history. 13. Sifting categories with low weights. 14. Saving categories to user history.
15. Saving categories as clasifications for text content.
The problem text classification is relevant because none of the top sites for blogs / articles does not offer automatic categorization of text, which means that users or moderators of the data need to manually categorize content on each site. Content classification is an important function of such a resource, as it will simplify the navigation, search and filtering content to users. The advantages of the system are:
No need for support.
Automation of categorizing process. No constraints on the subject of convent.
Ability to improve the system.
2. Input that is the result of a request to the Dandelion API.
3. Output data, which the system generates.
Inputs made by the user are user data, as well as text content for classification. The input, that is the result of a request to the Dandelion API - is the data received by the system after the analysis of the user's text from the Dandelion API. Output is classified articles that are received upon a user request. Integrated data is data that affect the operation of the algorithm. In this system it is the history of user’s categories. 8
The «Text Categorizer» is a website. To access to the website Internet is needed. The website is available on both personal computers and mobile devices. The target audience of this application is people who generate or consume text content. All articles and their classifications are publicly available on the site. To work with the system as the user, it is only necessary to fill in all the information about the article (name, text), after which the system will automatically select the category’s. The user does not need any actions to do that. Due to the high automation of the process, there is only one type of user in the system: Author. To work on the Author system you need to be authenticated. After passing authentication, the author will be able to: view the classified articles; filter articles by categories; categorize his own articles.
To access the application, the user must first go to the site address, then authenticate to the system. After passing the authentication process, the user will go directly on the main page of the application (Fig. 13). On the main page, the user can see the cards of the classified articles. Each card contains the main information about the article: title, author, date of creation, category. At the top of all pages of the system, the user can see the header. The left-handed menu consists of the following buttons:
On the right side, the user can see under which username he or she is locked, as well as the exit button. In the bottom right corner of each card there is a button «View Full», which allows you to go to the page detailed review article (Fig. 14). On the page's detailed article view, you can see the full article. In the upper right corner of the page is a button «Delete», which function is to delete the article from the system. Also, the user has the option to click on one of the categories of the article and return to the main page where the category to which the user has clicked will be included in the article search filter.
There is an option to add a category to the filter list by using several methods:
Clicking on the category on the article card on the main page;
Entering a category in the search box input.
When entering the category in the search box, the user can enter any value. However, one can select only one of the existing categories (Fig. 15). During updating of the list of categories, on the main page are filtered and are appeared only selected classifications. (Fig. 16). To add an article to the system, go to the «Create» page (Fig. 4.5). This page has an article entry form. The first input field is the title of the article. Then the user has a choice:
To download the text document by clicking the «Select file» button.
To enter the article text manually in the «Article Body» entry field.
Next, the user must click on the «Submit» button, after which the article will automatically be classified down and saved in the system. The user will be redirected to the main page of the site. 9
In order to check the work of the website one must consider the main types of work with the system. An important part of the developed system is the possibility of multilanguage use. Therefore, it is worth checking out the work with the major world languages: English; German; Italian; French You also need to check whether it is possible to download articles as a text document. Each author will start work the same: authorization, transition to the main page, the click of a «Create» button, after which will be done the main work with the system. The system`s work was first reviewed in English (Fig. 18-19) and in German (Fig. 20-21).
Fig. 21. View the result of the categorization of the German text
As a result of a practical implementation, a resource has been created. The main proposes is solving the problem of automating the categorization of text content. A description of the created software product was made. Using the ER-diagrams for the Chen note, the database was described. In this section, there were also illustrated user roles in the system and their main types of interaction with the system. Test case of the main functional of the system was carried out and analyzed.
During the implementation of work, an analysis and review of literary sources were conducted, in which key concepts, recent research and ready-made software solutions of the problem were described and analyzed. As a result of this analysis, it was determined that the developed system would be successful among users. It was also determined which deficiencies of competitive systems need to be corrected in the system being developed. System was analyzed after the analysis of literary sources. In this analysis, an objective tree, a UML diagram and a system hierarchy were developed. During this stage of the system development, it was determined which goals are necessary to create the system. The next step at this stage was to determine the ways of moving data in the system. The next stage was the choice of software solutions. The MERN stack was selected as the backbone of the system. This stack was chosen due to the ability to create an isomorphic JavaScript application. After the successful determination of software products, the hardware requirements of the system under development were determined. After the previous stages, the practical implementation of the system was developed. The finished product has been described in detail. Also, during this stage of development has been described user's manual. The system was tested on a test case. The final stage of development was the analysis of the economic feasibility of the system. It was considered economically feasible and competitive.
As a result of the implementation of the graduation paper, we received a software product that provides a convenient text categorization. Anyone can access this web site and to categorize their own text through a website. The system is open to improvements through the expansion of supported languages, as well as improving the speed and quality of the definition of the categories.