An approach, allowing obtaining information from a text that is relevant to users' question, is proposed in this paper. The advantage of the proposed approach is using the same algorithms for analysis of texts from different groups of languages, namely for Latin and Cyrillic alphabets. The most relevant sentences from the text that match to concrete types of questions are found by means of matching questions and attributes of answers in texts. Such attributes are specific for the concrete type of the question. An ontology aimed to consider key information about questions and attributes of answers also is proposed. This ontology allows adding new languages, types of questions, and attributes of answers. Architectural solution of the software system for operations of ontology processing and searching of answers, is represented in the paper. A case study illustrates the process of text processing. As an example of the text, a Bulgarian legend was chosen, because the style of preparing fairytales is one of the informal styles of text representation.
Nowadays values of information in digital world are increased. In order to perform effective search user must proceed a large amount of data. In order to take this process more effective different software systems for searching and processing of information are used. Existing systems have the next difficulties to use: systems are not flexible (based on artificial networks, based on ontologies), expensive, difficult to changes, difficult to be adopted to different languages or environments. From the other side the designing, development, and supporting of such systems requires many efforts to adopt software system for different purposes.
Solution may be in the area – flexible systems, open for other languages, allowing filtering user requests and reducing the number of searched information.
Paper is organized by the following way: Chapter 2 represents literature review and summarizing of drawbacks of existing approach. Chapter 3 represents main steps of the proposed approach. Chapter 4 illustrates a part of designed ontology for different types of questions. As examples, Bulgarian and English languages are considered. Chapter 5 represents the software architecture of the software system for realizing of the proposed approach. Chapter 6 contains a practical example of preparing XML files, as external attributes for two languages. Detailed information for domain analytics about testing the proposed approach is represented in Chapter 7. Chapter 8 illustrates the results of Case Study for asking questions about Bulgarian legend. It is known, that in fairytales and legends the freest style of representing ideas is admitted. That is why, such complex texts are selected for testing. Chapter 8 contains description of main class for corresponding software system realization. Conclusions represents quantity estimation of the proposed approach and ideas for the further research.
Paper [
Corpus acquisition deals with finding a sufficiently large set of texts that could be used to generate a semantic space.
Corpus parsing is used to remove any superfluous information from further consideration. (For example specific XML tags or other, irrelevant data.)
Corpus preprocessing serves to reduce the amount of different words in the corpus, effectively
reducing the context vector dimension:
1. Text cleaning – this includes the deletion of all text characters not belonging to the native
script of the language in question, the removal of numbers and words that contain numbers,
the elimination of punctuation marks and the shifting of all capital letters into lower case [
Choosing an algorithm for the creation of the semantic space and supplying it with the preprocessed corpus text.
The reduction of context vector dimension. Each algorithm has its own post-processing routine
which is encapsulated within the algorithm, as defined in the S-Space package [
Paper [
Approaches devoted to analysis of object state in decision support systems allows to analyze
different states of objects (and as a conclusion – characteristics of entities) with the aim to extract
metainformation about entities and get the answers to questions in the text. Such approaches may be
implemented if there is an information that state of object or domain entities may be changed during
story [
Other implementation of question-oriented analysis of texts is to get questions essential for
specific group of users to predict their relation, emotions, and opinions about texts. Typical question
may be applied to many different texts with the aim to select the best test considering some conditions
of chosen social group. For example searching the most appropriate texts fro children [
After analyzing the existing solutions for processing of texts, requirement specification with the essential software requirements was designed. Points that has the highest level of priorities are considered in this paper.
It is proposed to classify domain entities according to defined characteristics. One entity may correspond to several classes. The list of the proposed characteristics for questions is given below (Table 2 and Table 3). Aim of this classification is to take marks in text according to types of questions.
Component diagram of the proposed software system is represented in the Figure 1. It illustrates the structure of the system and interaction between its components.
XML parser
Metainformation
Text parser Answers
User input processing
Question metadata Matching tool
Metainformation about questions and answers for every language is stored in special XML file. Then, it is necessary to parse texts and questions. Information, extracted from XML parser, is transmitted to text parser and module for the questions processing (Use Input processing module in the Figure 1.).
After parsing texts answers according to the type of question are obtained. After parsing questions keywords are extracted (Table 3). After matching questions and information extracted from texts answers are proposed to user.
In order to represent connection between metainformation in texts and questions part of XML file is proposed. Structure of the file shows that in order to add new features of defining new answers in text it is necessary to modify items of list <MetaList> (See Table 2 and Table 3). In addition, it is easy to add, modify or remove new types of question.
<Bulgarian> <TypeQ> къде? </TypeQ> <MetaList> <Attribute> на </Attribute> <Attribute> в </Attribute> <Attribute> във </Attribute> <Attribute> близо </Attribute> <Attribute> далеко </Attribute> <Attribute> под </Attribute> <Attribute> над </Attribute> </MetaList> <TypeQ> кога? </TypeQ> <MetaList> <Attribute> в нощта </Attribute> <Attribute> нощ </Attribute> <Attribute> деня </Attribute> <Attribute> денят </Attribute> <Attribute> сутрин </Attribute> <Attribute> сутринта </Attribute> <Attribute> днес </Attribute> <Attribute> сега </Attribute> <Attribute> никога </Attribute> </MetaList> </Bulgarian>
In order to use the proposed approach it is necessary to perform the next steps: 1. Domain analytics designs a domain ontology (message 1 in UML Sequence diagram on figure 2). Ontology may looks like a UML class diagram. 2. Using this ontology XML files with questions’ and answers’ attributes are prepared. In order to do this common XML files for specific language may be précised (see examples from chapter 6). Domain analytic gathers specific question attributes for ontology (message 2 on UML Sequence Diagram on figure 2), and then prepares attributes of answers (message 3 on UML Sequence Diagram on figure 2). Then XML files for different languages are prepared for the further using (messages 4.1, 4.2, and 4.3 on UML Sequence Diagram on figure 2). 3. Then domain analytics starts to test prepared XML files for the chosen text. For every question the next activities are performed 3.1 Using XML files attributes of question are extracted (messages 5.1 and 5.2 on UML Sequence Diagram on figure 2). 3.2 Using XML file for concrete language and concrete question and considering text a set of answers are founded (messages 6 on UML Sequence Diagram on figure 2). 4. Answers are returned to domain analytics for analysis (messages 7 on UML Sequence
Diagram on figure 2)
Bulgarian Cред живописните възвишения на Рила, редом до църковните храмове, жилищни комплекси и костници на Рилския манастир, своята снага издига и една сурова на вид сграда от масивни ломени камъни, наречена Хрельовата кула. Нейните дебели зидове не изглеждат никак на място в този храм, посветен на Бога и по-скоро оставят усещането за средновековно укрепление. А всъщност става въпрос именно за това.
Най-старата сграда от целия съвременен комплекс на Рилския манастир е построена от протосеваст Стефан Хрельо Драговол – благородник, който притежавал владения в региона.
От надпис, запазил се в самата кула, разбираме, че отбранителното съоръжение било издигнато от влиятелния властел през 1335 г., заедно с църквата „Св. Богородица Осеновица“ и било отдадено на защитата на богослужебните сгради, които тогава оформяли комплекса посветен на Св. Иван Рилски.
Малко се знае за вида, който е имал тогава Рилския манастир, тъй като пораженията
нанесени от османското завоевание са довели както до загубването на изворови данни, така и
до щети върху самите паметници на българската култура [
Translation into English Among the picturesque elevations of Rila, next to the churches, residential complexes and ossuaries of the Rila Monastery, a harsh-looking building made of massive rubble stones, called the Khrel tower, is also erected. Its thick walls do not look at all out of place in this temple dedicated to God and rather leave the feeling of a medieval fortification. And in fact, this is exactly what it is about.
The oldest building of the entire modern complex of the Rila Monastery was built by Protosevast Stefan Hrelio Dragovol - a nobleman who owned estates in the region.
From an inscription preserved in the tower itself, we understand that the defensive facility was erected by the influential ruler in 1335, together with the church "St. Bogoroditsa Osenovitsa" and was devoted to the protection of the religious buildings that then formed the complex dedicated to St. Ivan Rilski.
Little is known about the appearance that the Rila Monastery had then, since the defeats inflicted by the Ottoman conquest led to both the loss of source data and damage to the monuments of Bulgarian culture. The next activity – is to take question about text. Tables with answers are given below. Possible variants of answers (all sentences with alive entities are considered) Стефан Хрельо Драговол – Stefan Hrelio Dragovol - a nobleman who благородник, който притежавал владения в owned estates in the region. региона.
които тогава оформяли комплекса which then formed the complex dedicated to посветен на Св. Иван Рилски. St. Ivan Rilski.
посветен на Бога и по-скоро оставят усещането за средновековно укрепление dedicated to God and rather leave the impression of a medieval fortification Words – 28 Total value of text - 207
English What conquests destroyed the Rila Monastery?
The development of project is started from the class allowing to save and restore XML file from hard disk. Storing and restoring is made by means of XML serialization. Class dataStore incapsulates the serialization and deserialization operations.
class dataStore {
public string filename { get; set; } public Dictionary<string,List<string>> quest_ans { get; set;} public void SerializeD() { } public void DeSerializeD() { }
The next class should support basic operations with datastore. (Something like CRUD operations when databases are processed.) Class ManageDataStore implements datastore processing operations. class ManageDataStore { dataStore ds { get; set; } public void DataStore_Create() { } public void DataStore_Edit() { } public void DataStore_View() { } public void DataStore_Delete() { } } Class Language is aimed to proceed operation of searching answers in texts. class Language {
dataStore ds { get; set; } public string text { get; set; } public List<string> questions { get; set; } public List<string> answers { get; set; }
Language(string text, List<string> questions, string Lang) { } public void GetInformation() { } public void FindAnswewrs() { } }
The paper proposes the multilingual approach for searching information in texts. Searching procedure matches attributes of the question and the text. The main idea of the proposed approach is to reduce amount of information needed to be processed by user in order to obtain answer to questions.
Table 9 shows the ratio of information needed to be proceed to find answer to question. Time ratio is calculated in the next way: average speed for reading such types of texts is 160 words in minute, It is assumed that reading will take 1 minute for Bulgarian language (Table 4) and 182/160=1.13 minute for English language.
10. References.