Svetla Boytcheva

Focusing on scenario recognition in information extraction

This paper reports a research effort in Information Extraction, especially in template pattern matching. Our approach uses reach domain knowledge in the football (soccer) area and logical form representation for necessary inferences of facts and templates filling. Our system FRET1 (Football Reports Extraction Templates) is compatible to the language-engineering environment GATE and handles its internal representations and some intermediate analysis results.

Text Mining and Big Data Analytics for Retrospective Analysis of Clinical Texts from Outpatient Care

Abstract This paper presents the results of an on-going research project for knowledge extraction from large corpora of clinical narratives in Bulgarian language, approximately 100 million of outpatient care notes. Entities with numerical values are mined in the free text and the extracted information is stored in a structured format. The Algorithms for retrospective analyses and big data analytics are applied for studying the treatment and evaluating the diabetes compensation and control of arterial blood pressure.

Applying Language Technologies on Healthcare Patient Records for Better Treatment of Bulgarian Diabetic Patients

This paper presents a research project integrating language technologies and a business intelligence tool that help to discover new knowledge in a very large repository of patient records in Bulgarian language. The ultimate project objective is to accelerate the construction of the Register of diabetic patients in Bulgaria. All the information needed for the Register is available in the outpatient records, collected by the Bulgarian National Health Insurance Fund. We extract automatically from the records’ free text essential entities related to the drug treatment such as drug names, dosages, modes of admission, frequency and treatment duration with precision 95.2%; we classify the records according to the hypothesis “having diabetes” with precision 91.5% and deliver these findings to decision makers in order to improve the public health policy and the management of Bulgarian healthcare system. The experiments are run on the records of about 436,000 diabetic patients.

Towards Extraction of Conceptual Structures from Electronic Health Records

This paper presents the general framework and the current results of a project that aims to develop a system for knowledge discovery and extraction from the texts of Electronic Health Records in Bulgarian language. The proposed hybrid approach integrates language technologies and conceptual processing. The system generates conceptual graphs encoding the patient case history, which contains templates for the patients diseases, symptoms and treatments. We describe simple inference in the generated graphs resource bank. Some experiments and their evaluation are presented in the article.

Integration of Resources and Components in a Knowledge-Based Web-Environment for Terminology Learning

This paper presents the design and currently elaborated components in the knowledge-basedl earning environment called STyLE. It supports learning of English terminology in the domain of finances with a target user group of non-native English speakers. The components elaboratedso far allow for the discussion of the Web-based learning environment, the approach to the building of a learner model, and the adaptive strategies for instructional and content planning depending on the learning situation. The paper emphasises on the specific aspects of learning terminology in a secondlan guage andc hecking the correctness of learners performance within the application of STyLE.

Mining comorbidity patterns using retrospective analysis of big collection of outpatient records

BackgroundStudying comorbidities of disorders is important for detection and prevention. For discovering frequent patterns of diseases we can use retrospective analysis of population data, by filtering events with common properties and similar significance. Most frequent pattern mining methods do not consider contextual information about extracted patterns. Further data mining developments might enable more efficient applications in specific tasks like comorbidities identification.MethodsWe propose a cascade data mining approach for frequent pattern mining enriched with context information, including a new algorithm MIxCO for maximal frequent patterns mining. Text mining tools extract entities from free text and deliver additional context attributes beyond the structured information about the patients.ResultsThe proposed approach was tested using pseudonymised reimbursement requests (outpatient records) submitted to the Bulgarian National Health Insurance Fund in 2010–2016 for more than 5 million citizens yearly. Experiments were run on 3 data collections. Some known comorbidities of Schizophrenia, Hyperprolactinemia and Diabetes Mellitus Type 2 are confirmed; novel hypotheses about stable comorbidities are generated. The evaluation shows that MIxCO is efficient for big dense datasets.ConclusionExplicating maximal frequent itemsets enables to build hypotheses concerning the relationships between the exogeneous and endogeneous factors triggering the formation of these sets. MixCO will help to identify risk groups of patients with a predisposition to develop socially-significant disorders like diabetes. This will turn static archives like the Diabetes Register in Bulgaria to a powerful alerting and predictive framework.

Conceptual Graphs self-tutoring system

The present paper describes a self-tutoring system CG-EST (Conceptual Graphs Elearning Self-Tutoring system). The purpose of this system is to generate questions to the user from given text lessons, to process user answers and to check their correctness. CG-EST includes modules for: parsing and producing logical forms (LF); generating conceptual graphs (CG) based on LF; generating questions based on LF; defining the scope of correct answer using CG operations; matching the user’s and correct answers. This approach assists the automatic eLearning systems being developed for various domains with some requirements: all sentences in the lessons have to use Controlled English (CE); to each lesson a simple concept type hierarchy and a synonyms set have to be attached. However these restrictions are useful for semantic interpretation of Natural Language (NL) texts. Thus CG-EST presents a solution for self-service eLearning that is ideal for training in a specific field.

Mining Association Rules from Clinical Narratives.

Shallow text analysis (Text Mining) uses mainly Information Extraction techniques. The low resource languages do not allow application of such traditional techniques with sufficient accuracy and recall on big data. In contrast, Data Mining approaches provide an opportunity to make deep analysis and to discover new knowledge. Frequent pattern mining approaches are used mainly for structured information in databases and are a quite challenging task in text mining. Unfortunately, most frequent pattern mining approaches do not use contextual information for extracted patterns: general patterns are extracted regardless of the context. We propose a method that processes raw informal texts (from health discussion forums) and formal texts (outpatient records) in Bulgarian language. In addition we use some context information and small terminological lexicons to generalize extracted frequent patterns. This allows to map informal expression of medical terminology to the formal one and to generate automatically resources.

Identification of Risk Factors in Clinical Texts through Association Rules.

We describe a method which extracts Association Rules from texts in order to recognise verbalisations of risk factors. Usually some basic vocabulary about risk factors is known but medical conditions are expressed in clinical narratives with much higher variety. We propose an approach for data-driven learning of specialised medical vocabulary which, once collected, enables early alerting of potentially affected patients. The method is illustrated by experimens with clinical records of patients with Chronic Obstructive Pulmonary Disease (COPD) and comorbidity of CORD, Diabetes Melitus and Schizophrenia. Our input data come from the Bulgarian Diabetic Register, which is built using a pseudonymised collection of outpatient records for about 500,000 diabetic patients. The generated Association Rules for CORD are analysed in the context of demographic, gender, and age information. Valuable anounts of meaningful words, signalling risk factors, are discovered with high precision and confidence.

Mining Clinical Events to Reveal Patterns and Sequences

This paper presents results of ongoing project for discovering complex temporal relations between disorders and their treatment. We propose a cascade data mining approach for frequent pattern and sequence mining. The main difference from the classical methods is that instead of applying separately each method we reuse and extend the result prefix tree from the previous step thus reducing the search space for the next task. Also we apply separately search for diagnosis and treatment and combine the results in more complex relations. Another constraint is that items in sequences are distinct and we have also parallel episodes and different time constraints. All experiments are provided on structured data extracted by text mining techniques from approx. 8 million outpatient records in Bulgarian language. Experiments are applied for 3 collections of data for windows with size 1 and 3 months, and without limitations. We describe in more details findings for Schizophrenia, Diabetes Mellitus Type 2 and Hyperprolactinemia association.