Search

Abstract Background Ischemic stroke (IS) is a major health risk without generally usable effective measures of primary prevention. Early warning signals that are easy to detect and widely available can save lives. Estonia has one nation-wide Electronic Health Record (EHR) database for the storage of medical information of patients from hospitals and primary care providers. Methods We extracted structured and unstructured data from the EHRs of participants of the Estonian Biobank (EstBB) and evaluated different formats of input data to understand how this continuously growing dataset should be prepared for best prediction. The utility of the EHR database for finding blood- and urine-based biomarkers for IS was demonstrated by applying different analytical and machine learning (ML) methods. Results Several early trends in common clinical laboratory parameter changes (set of red blood indices, lymphocyte/neutrophil ratio, etc.) were established for IS prediction. The developed ML models predicted the future occurrence of IS with very high accuracy and Random Forests was proved as the most applicable method to EHR data. Conclusions We conclude that the EHR database and the risk factors uncovered are valuable resources in screening the population for risk of IS as well as constructing disease risk scores and refining prediction models for IS by ML.

This paper describes the identification, adaptation and creation of tools that are needed for creating a quote extractor for Estonian media texts that would be able to properly extract both direct and indirect quotes and attribute them to the correct person identified by full name and profession. This includes named entity recognition and resolution as well as grammar-based extraction of direct and indirect quotes. To get a further understanding of indirect speech in Estonian media, we also performed a corpus linguistic analysis of the quotes extracted with our tools from one week of Estonian news. *** Tsitaatide eraldamine eestikeelsetest meediatekstidest: analüüs ja töövahendid Artikkel annab ülevaate eesti keele tsitaadituvastaja loomise esimesest etapist. Tsitaadituvastaja eesmärk on eraldada nii otseses kui kaudses kõnes väljendatud tsitaate koos tsiteeritud isiku täisnime ning võimalusel ka ametiga. Artiklis selgitasime, milliseid komponente tsitaadituvastaja jaoks oleks vaja ning vastavalt sellele testisime ja kohandasime olemasolevaid ning lõime veel puuduvaid töövahendeid. Samuti identifitseerisime tsitaadituvastaja arenduseks vajalikud parandused ja lisatööriistad ning analüüsisime uudistes otsese ja kaudse kõne edastamiseks kasutatavaid saatelauseid. Isikunimede leidmiseks kasutasime EstNLTK teegi standardset CRF-põhist nimeüksuste märgendajat. Lisasime sellele ühestaja, mis leiab tekstist tsiteeritud isiku täisnime, juhul kui saatelauses on kasutatud ainult ees- või perekonnanime. Sel moel suutsime ära lahendada 61,0% ühesõnalistest isikunimedest. Elukutsete leidmiseks lõime 5659 sõna suuruse eestikeelse ametite leksikoni ning märgendasime selle põhjal nimeüksuste tuvastaja treeningkorpuses ka elukutsed. Seejärel treenisime nimeüksuste tuvastaja ümber tuvastama ka elukutseid. Tulemusi hinnates leidsime, et kõige parem on kasutada leksikonipõhist lähenemist koos ümbertreenitud CRF-märgendajaga, mis andis elukutsete tuvastamise F1-skooriks 86,1%. Otsekõne eraldamiseks kasutasime regulaaravaldistepõhist lähenemist. Kuna otsekõne on jutumärkidega selgelt markeeritud, saime sel moel 95,0%-se F1-skoori. Otsekõne saatelausetest saime sisendit kaudse kõne eraldamiseks: lõime verbide, nimisõnade ning määruste leksikoni, mis viitavad, et lause edastab vahendatud mõtteid. Grammatikapõhise lähenemisega saime F1-skooriks 84,7%, seejuures oli täpsus 93,5%. Uurides tuvastamata jäänud kaudse kõne lauseid, leidsime veel mitut tüüpi konstruktsioone, mida saagise parandamiseks grammatikapõhises lähenemises käsitleda võiks. Lõpetuseks analüüsisime vastloodud töövahendite abil ühe nädala Eesti meediatekstidest eraldatud tsitaate ja nende saatelauseid, käsitledes nii leksikaalseid, morfoloogilisi, süntaktilisi kui ka semantilisi jooni. Tulevikus on plaanis peale eraldiseisvate töövahendite parandamise luua ka vabalt kasutatav terviklahendus eestikeelsete tsitaatide ja tsiteeritute tuvastamiseks.

"Syntactic analysis of Estonian netspeak using Constraint Grammar" The paper provides an overview of an attempt to adapt the Estonian Constraint Grammar rule set for netspeak. The rule set has been developed by Kaili Müürisep and Tiina Puolakainen for shallow and dependency parsing of Estonian literary language, and it has previously been adapted for shallow parsing of spoken Estonian by Kaili Müürisep and Heli Uibo. First, in order to adapt the rules, a chatroom corpus was parsed with the existing rule set. The corpus was manually revised and based on the errors that were found, changes were made to the rule set. The changes regarded detection of clause boundaries and particle verbs, as well as assignment of syntactic tags and dependency relations. Extensive use of discourse particles and direct addresses, short sentence length, and small percentage of attributes among the syntactic functions used in text appeared to be the most distinctive features of netspeak, as well as the large amount of elliptical sentences from which, in addition to other syntactic functions, a predicate can be left out. As a result of adapting the rule set, the results of both shallow and dependency parsing improved. The most error-prone syntactic functions were subjects, predicatives, and adverbials. In dependency parsing, the largest number of errors was made in determining the governors of adverbials.