PMC:7799291 / 20672-46680 JSONTXT

{"target":"http://pubannotation.org/docs/sourcedb/PMC/sourceid/7799291","sourcedb":"PMC","sourceid":"7799291","source_url":"https://www.ncbi.nlm.nih.gov/pmc/7799291","text":"Text mining systems\nNumerous text mining systems for COVID-19 literature have been released in 2020 thus far. We compile a list of 39 systems in Table 2 (we maintain an up-to-date list of systems on the CORD-19 GitHub page https://github.com/allenai/cord19). These text mining systems are collected through a public form on the CORD-19 website, by searching COVID-19 papers and preprints in the CORD-19 corpus and from social media. We omit systems that appear to index documents using off-the-shelf software (e.g. ElasticSearch) without additional data or methodological extensions or without other obvious distinguishing system features.\nTable 2 COVID-19 text mining systems, including both production systems and research prototypes, covering a range of text mining tasks. Note that ‘Data’ and ‘Methods/Models’ only include known data sources and modeling/implementation details discussed in the associated documentation of these systems. Under ‘Affiliation’, we use for industry, for nonprofit and no symbol for academic affiliations; if no affiliation is provided, the work is conducted by independent researchers ‘Search’ - users issue queries to system to find relevant content. ‘Augmented reading’ - system provides interface for reading papers with additional features (e.g. term highlighting). ‘Exploration’ - users use system to explore available content, possibly without specific informational need. Often used to understand the underlying data source. ‘KB construction’ - system constructs a KB using extracted entities and relations to support a system function. ‘Visualization’ - data visualization is component of how user interacts with system. ‘Clinical diagnostic support’ - system assists healthcare providers in disease diagnosis. ‘Question answering’ - system expects a query in question form and directly answers user-written query with an (extracted) answer. ‘Summarization’ - system surfaces automated summaries of paper text. ‘Claim verification’ - system expects a query in claim or assertion form and verifies or refutes it.\nTasks System Affiliation Link Data Methods/Models User interface\nS1 Search Covidex [114] University of Waterloo and NYU https://covidex.ai/ CORD-19, ClinicalTrials.gov through TrialStreamer Retrieves passages using Anserini [109]. Reranking using T5-base model [72] finetuned on biomedical text and trained for ranking on MS MARCO [14]. Supports user-written keyphrase queries. Highlights matched terms in abstract. Toggle for searching different corpora.\nS2 Search COVID papers browser – https://github.com/gsarti/covid-papers-browser CORD-19, SNLI, MultiNLI Matches queries to papers via pretrained sentence embeddings: SentBERT [74] training procedure with SciBERT [6], BioBERT [49], CovidBERT, and ClinicalCovidBERT. Trained on SNLI [11] and MultiNLI [104] datasets. Supports interactive querying via command line.\nS3 Search fatcat [Covid-19] Internet archive https://covid19.fatcat.wiki/ CORD-19, WHO, Wanfang, CNKI, Internet Archive ElasticSearch Supports user-written keyphrase queries. Highlights matched terms in abstract. Toggle for searching different corpora.\nS4 Search KDCovid Google, UMass Amherst, MSR Montreal, UToronto, CMU http://kdcovid.nl/ CORD-19 Sentence-based retrieval using a similarity metric derived from BioSentVec [17]. BeFree [12] for entity linking. Genes are linked to UniProt and diseases to MedGen. Relations between genes and diseases from DisGeNET (v6) [30]. Supports user-written keyphrase queries, returns relevant papers. Biomedical entities in abstracts color-coded by type. Entity hyperlinks to associated webpages. Gene-disease relations presented for each paper.\nS5 Search DOC Search DRE https://covid-search.doctorevidence.com/ CORD-19, PubMed, ClinicalTrials.gov, WHO ICTRP, news articles, etc. – Supports user-written keyphrase and boolean queries comprised of paper metadata, entities and PICO elements\nS6 Search CoronaSearch – https://coronasearch.net/ CORD-19 Embeds documents using Google’s Multilingual Universal Sentence Encoder [110]. Retrieves relevant documents using Facebook’s Faiss library [37] Supports user-written keyphrase queries, with specific emphasis on multi-lingual queries\nS7 Search DISCOVID.AI Karlsruhe Institute of Technology https://discovid.ai/ CORD-19 – Supports user-written keyphrase queries. Results are linked to associated clinical trials.\nS8 Search Vapur [43] Bogaziçi University https://vapur.herokuapp.com/ CORD-19, ChemProt NER with BERN [41], relation extraction model if BioBERT [49] trained on ChemProt [86], Supports user-written keyphrase queries or query on chemical/gene/RNA compound identifiers (e.g. ChEBI chemical identifier, HGNC gene name). Results organized by relationships with other potential drug targets and entities of interest.\nS9 Search Covid-19 Search Microsoft Azure https://covid19search.azurewebsites.net/ CORD-19 – Supports user-written keyphrase or boolean queries. Filter resiults using extracted biomedical entity types (e.g. anatomy, disease, gene, drug, etc). Recommends similar papers.\nS10 Search Research-Covid19.ai Gowi https://research-covid19.ai/ CORD-19 Search using Azure Cognitive Search. Entities extracted and normalized using BERN [41]. Supports user-written keyphrase or boolean queries. Filter results using extracted biomedical entity types (e.g. species, disease, gene, drug).\nS11 Search, augmented reading DeScign COVID-19 Search DeScign http://covid.descign.com/ CORD-19 – Supports user-written keyphrase queries. Filter results using entities (e.g. viral anatomy, chemicals, diseases, biomolecules, etc.). Supports reading of extracted paper full text with highlighted entities.\nS12 Search, augmented reading COVID-19 Intelligent Insight Sinequa https://covidsearch.sinequa.com/ CORD-19, Elsevier, Clinical trial info from WHO’s ICTRP database, arXiv, bioRxiv, medRxiv, COVID-19 papers from BMJ, Web text from WHO and CDC – Supports user-written keyphrase queries. Filter results using facets (e.g. indication, human phenotype). Supports reading of paper PDFs with highlighted entities.\nS13 Search, exploration Covid AI-powered Search Curiosity https://covid.curiosity.ai/ CORD-19, SciHub, UMLS, OntoBee, MAG, etc. – Supports user-written keyphrase queries. Filter results by paper topics and extracted disease entities linked to KGs (e.g. UMLS, OntoBee, etc.). Provides access CORD-19 papers via other sources (e.g. SciHub).\nS14 Search, exploration COVID-19 Navigator IBM Watson https://covid-19-navigator.mybluemix.net/ CORD-19, Medline, PubMed Open Access, ClinicalTrials.gov, patents from the US Patent Office, UMLS – Supports boolean queries using UMLS concepts and semantic types.\nS15 Search, exploration SPIKE-CORD [88] Allen Institute for AI https://spike.covid-19.apps.allenai.org/search/covid19 CORD-19 Entities and syntax extracted using ScispaCy [59]. Data indexed using Odinson [93]. Support custom query syntax. Specialized query language supports regex operators (e.g. wildcards, number of matches), matching on entity types, and syntactic patterns (e.g. similar verbs).\nS16 Search, exploration EVIDENCEMINER [101] UIUC https://evidenceminer.com/ CORD-19, PubMed, UMLS Retrieves sentences with similar biomedical entities to the query using distantly supervised NER and OpenIE, details in [100]. Supports user-written keyphrase queries that can take the form of a claim, results ranked by the level of evidence they provide toward the query. Entities are highlighted in results. Filter results by entity type.\nS17 Search, exploration, KB construction, visualization Carnap Funktor LLC https://carnap.ai/ CORD-19 – Supports entity-based queries with medical terms, returns relationships from an underlying KB, supported by papers. Filter results by entity, relation, domain study, or the strength of relationships.\nS18 Search, exploration, KB construction, visualization, clinical diagnostic support Kahun Kahun https://coronavirus.kahun.com/ CORD-19, SNOMED, LOINC, and other clinical ontologies – Supports entity-based queries with clinical entities, returns a graph of clinical relationships related to the query entity and COVID-19, supported by papers.\nS19 Search, visualization COVID-SEE [95] University of Melbourne, IBM Research https://covid-see.com/search CORD-19, EBM-NLP Retrieves documents using Covidex API [114], entity tagging using MetaMap [23], PICO element extraction using a BiLSTMCRF model [46] model trained on EBM-NLP [62]. Supports user-written keyphrase queries. Browse search results using a Sankey diagram of PICO extractions.\nS20 Search, exploration, visualization Covidexplorer IIT Gandhinagar’s Lingo Group http://covidexplorer.in/ CORD-19, Twitter Biomedical entities (e.g. proteins, diseases, cell types) extracted using SciBERT [6]. Supports user-written keyphrase queries. Papers are tagged with extracted biomedical entities. Filter results using year or entities. Entity pages show frequent co-mentioned entities and timelines of paper mentions. Visualizes COVID-19 Twitter mention trends.\nS21 Search, exploration, visualization CovidScholar UC Berkeley https://www.covidscholar.com/ CORD-19, Elsevier, LitCovid, Lens, Dimensions, human submissions Adapts the MATSCHOLAR [102] system for identifying relevant papers given entity-centric queries. Supports user-written keyphrase queries comprised of entities. Filter results on paper facets. Search similar papers. Visualizes word embeddings. Integrates user-submitted data corrections.\nS22 Search, claim verification SciFact: CORD-19 Claim Verification [97] Allen Institute for AI https://scifact.apps.allenai.org/ CORD-19, S2ORC, SciFact Retrieves documents using Covidex API [114]. Uses RoBERTa [53] for evidence selection, and to classify claim-evidence pairs as Supported/Refuted. Supports user-written queries that take the form of a scientific claim, returns papers supporting or refuting the claim, along with confidence scores.\nS23 Search, QA CO-Search [29] Salesforce https://sfr-med.com/search CORD-19, ChemProt Retrieves documents using an ensemble of SiameseBERT [75], TF-IDF and BM25. Reranking model composed of multi-hop question answering module and multi-paragraph abstractive summarizer. Supports user-written keyphrase queries or natural questions, returns a ranked list of matching articles with highlighted answer spans.\nS24 Search, QA AWS CORD-19 Search [7] Amazon Web Services (AWS) https://cord19.aws/ CORD-19, Amazon Comprehend Medical Multilabel topic classification of papers using Amazon Comprehend Medical [8]. Search using Amazon Kendra. Research topics learned using LDA. Supports user-written natural questions, returns a ranked list of matching articles with highlighted answer spans. Filter results by topic. Recommends similar papers.\nS25 Search, QA covidAsk [48] DMIS Lab of Korea University https://covidask.korea.ac.kr/ CORD-19, Natural Questions, SQUAD BEST [50] for keyword matching. DenSPI [79] for longer questions. BERN [41] for named entity extraction. BioSyn [85] for entity linking to CTD or NCBI. Trained on Natural Questions [45] and SQuAD [] datasets. Supports user-written natural questions, returns a ranked list of matching articles with highlighted answer spans. Entities in document text are also linked to external databases.\nS26 Search, QA AUEB Covid-19 Search Engine [56] AUEB’s NLP Group http://cslab241.cs.aueb.gr:5000/ CORD-19, BioASQ Uses the QA model from [56] trained on BioASQ [73] data. Supports user-written or templated questions, returns a ranked list of matching articles with highlighted answer spans. Can restrict to sections for search, e.g. Introduction and Methods.\nS27 Search, QA CovidSearch IEETA, University of Aveiro http://covidsearch.web.ua.pt/ CORD-19 Uses the QA model from [1]. Supports user-written natural questions, returns a ranked list of matching articles with highlighted answer spans.\nS28 Search, QA COVID-19 Research Explorer Google Research https://covid19-research-explorer.appspot.com/ CORD-19 – Supports user-written natural questions, returns a ranked list of matching articles with highlighted answer spans. Can ask follow-up questions.\nS29 Search, QA, summarization CAiRE-Covid [83] The Centre for Artificial Intelligence Research (CAiRE), HKUST https://caire.ust.hk/covid/ CORD-19, Biomedical reviews [111] Keyword-based retrieval of paragraphs using Anserini []. Reranking and answer selection using ensemble of BioBERT QA model [49] and generalized MRQA model [82]. Summarize answers across multiple documents abstractively with BART [51] and UniLM [26], and extractively with nearest neighbor ALBERT [47] sentence embeddings. Supports user-written natural questions, returns a ranked list of matching articles with highlighted answer spans. Provides extractive and abstractive summary over all answer spans.\nS30 Search, summarization CORD-19 Search Vespa https://cord19.vespa.ai/ CORD-19 Generates summaries of papers using T5 [72]. Recommends similar papers using SPECTER paper embeddings [18]. Supports user-written keyphrase queries. Recommends similar papers.\nS31 Exploration tmCovid Emory University http://tmcovid.com/ Pubmed abstracts, PMC full text, PubTator annotations – Explore papers by entity occurrence frequencies.\nS32 Exploration COVIDExplorer Penn State’s Coronavirus-AI Project https://coronavirus-ai.psu.edu/database CORD-19 Unsupervised clustering of documents with maximum modularity clustering [25]. Query matching is based on bag of words similarity between query and document clusters. Filter papers using many interactive filters or extracted topics and keywords.\nS33 Exploration CORD-19 Topic Browser MITRE https://topicbrowser.c19hcc.org/ CORD-19 Topics are extracted using MITRE’s Topic Modeling Neural Toolkit (TMNT) (https://github.com/mitre/tmnt) Explore papers using extracted topics. User can select different granularities of topics.\nS34 Exploration Topic Forest – http://topicforest.com/biomed/coronavirus Exploration Topics are extracted in an unsupervised manner using variant of SGRank [21]. Explore papers through extracted hierarchy of topics and keywords.\nS35 Exploration, visualization COVID-19 Explorer Department of Knowledge Technologies, Joz^ef Stefan Institute http://covid19explorer.ijs.si/ CORD-19 Keyphrases are computed with RaKUn [80]. Documents are ranked by keyphrase similarity to query. Supports boolean queries using extracted keyphrases. Visually explore embedded keyphrases.\nS36 Exploration, visualization SemViz [92] Laboratory for Linguistics and Computation, Brandeis University https://www.semviz.org/ CORD-19, Blender Lab COVID-19 KG [99], Protein-protein-causal-assertions dataset Applies semantic visualization techniques to several COVID-19 graph datasets as described in [92]. Visualizes chemical-gene, pathway-disease and protein–protein interaction KBs with evidence of mentions in papers.\nS37 Exploration, visualization VIDAR-19 [106] Yotta Conseil (independent) https://vidar-19.yotta-conseil.fr/ CORD-19, ICD-11 Risk factors are extracted using keyword matching and regular expressions. Visualizes risk factors within a disease hierarchy.\nS38 Exploration, visualization, KB construction SciSight [33] Allen Institute for AI https://scisight.apps.allenai.org/ CORD-19, MAG Visualizes author, citation, and entity graphs using methods described in [33] Explore papers based on affiliation and author networks, or by extracted entities and entity co-occurrences.\nS39 KB construction AIM COVID-19 Database AIM (by APEL) https://covid19-help.org/database CORD-19, PubMed – Data presented in tabular format tracks state of treatment and vaccine development. Displays extracted entities (drugs, phase of research, class of molecule) supported by evidence from mentioning papers. All of the included systems facilitate search or exploration over the COVID-19 literature, though some feature more specific text understanding tasks such as summarization, QA and claim verification. To facilitate a comparison between systems, we provide the following in Table 2: (i) data used, (ii) models/methods used or implemented by each system and (iii) supported user interface features. In some cases, information is not provided or could not be found about the data or models/methods used; we have indicated this using ‘–’.\nThe majority of systems we document here make use of public corpora and data resources, which are easily accessible from their source. Corpora like CORD-19 and LitCovid and other commonly used data resources like ClinicalTrials.gov, UMLS and biomedical ontologies adhere to FAIR principles of Findability, Accessibility, Interoperability and Reusability [103], though some systems [e.g. CovidScholar (Row S21), DOC Search (Row S5), COVID-19 Intelligent Insight (Row S12)] leverage proprietary corpora or private annotations in addition to public datasets. Additionally, though many of these systems have transparent methods or provide source code for reproducibility, a number of systems do not, as noted by missing model descriptions in Table 2.\nThe rest of this section is organized as follows: we define the text mining tasks used to categorize and assess the surveyed systems and use these tasks to anchor discussion and comparison of systems described in Table 2. For each task, we (i) summarize features and methodology used by the associated systems and (ii) highlight specific systems that have taken additional steps to tailor their interface for real-world use by biomedical and clinical researchers and practitioners. Such additional steps include joining literature data with biomedical KBs used in clinical settings or adding annotations created by medical experts specifically for COVID-19-related tasks. For each mentioned system, we provide a link to its corresponding row in Table 2.\nSearch systems provide search experiences in which users issue queries expressing informational needs that the system satisfies with a returned collection of relevant documents. Queries can be collections of keyphrases, similar to those supported by traditional search engines like Google or PubMed. Indexing and retrieval can be implemented using open-source tools like Anserini [109] or commercial software like Amazon Kendra (https://aws.amazon.com/kendra/) or Azure Cognitive Search (https://azure.microsoft.com/en-us/services/search/). Systems like COVID papers browser (Row S2), CoronaSearch (Row S6) and CovidScholar (Row S21) compute embeddings for queries and paper text spans (i.e. sentences or entities) and retrieve documents containing nearest-neighbor spans as results. Some systems constrain the query vocabulary to entities in a known KB (e.g. COVID-19 Navigator (Row S14) allows query terms in the form of UMLS concepts). SPIKE-CORD [88] (Row S15) supports specification of regular expression-like patterns to afford users greater control over search results.\nAmong these search systems, Covidex [114] (Row S1), fatcat (Row S3), DOC Search (Row S5), COVID-19 Intelligent Insight (Row S12), Covid AI-powered Search (Row S13), COVID-19 Navigator (Row S14) and CovidScholar (Row S21) integrate data from many sources, going beyond documents in CORD-19 or LitCovid to other databases such as ClinicalTrials.gov, Lens, Dimensions, documents from the WHO or CDC websites and more. Several systems also leverage external KBs for entity linking, such as Vapur (Row S8), which links to ChemProt [86], COVID-19 Navigator (Row S14) and EVIDENCEMINER (Row S16), which link to UMLS, or AWS CORD-19 Search [7] (Row S24), which uses external knowledge from the Comprehend Medical KB [8, 105]. DOC Search (Row S5) and COVID-SEE [95] (Row S19) are interesting systems that incorporate extracted PICO elements and relationships in visualization and exploration, which can be especially helpful when viewing results from clinical trial papers.\nExploration-focused systems assist users with discovery and understanding of documents in a corpus. Such systems may not aim to satisfy a specific informational need but are rather used to help users understand the underlying data source; as such, their interfaces facilitate unfocused data exploration and repeated interactions. Instead of supporting arbitrary user-written queries, these systems may provide a predefined set of topics or keyphrases with which to filter the documents. Keywords or keyphrases can be extracted from documents using supervised biomedical entity extraction (e.g. ScispaCy [59] and BERN [41]) or unsupervised keyphrase extraction (e.g. SGRank [21]). Paper topics can similarly be assigned via supervised document classification, as in AWS CORD-19 Search [7] (Row S24), which classifies papers using entities in the Comprehend Medical KB or in an unsupervised manner by clustering extracted keyphrases, as in COVIDExplorer (Row S32). TopicForest (Row S34) is interesting because it makes use of a learned topic hierarchy that organizes extracted keyphrases for users, although the user interface is under-developed.\nAmong the systems that leverage KBs, those that use curated domain-specific KBs tend to provide a better user experience, since the entities and relations in these KBs have been vetted by domain experts. IBM Watson’s COVID-19 Navigator (Row S14, https://covid-19-navigator.mybluemix.net/), perhaps the best example of this, allows users to perform boolean queries using UMLS concepts and semantic types [9].\nQA systems accept queries in the form of questions and provide extracted answer spans from documents. Most QA systems over COVID-19 literature provide both search and QA functionalities, retrieving relevant documents and surfacing answering spans. Several provide additional features such as generating summaries across answers, as in CAiRE-Covid [83] (Row S29), or the ability to ask follow-up questions, as in Google’s COVID-19 Research Explorer (Row S28). Due to a lack of abundant training data specific to COVID-19, most existing QA systems needed to bootstrap their own QA training data or are trained on non-scientific domain datasets like SQuAD [73] or smaller biomedical domain QA datasets like BioASQ [91], which may result in less performant systems. Efforts like [87] and EPIC-QA (Section 5.3) aim to change this by creating public COVID-19 QA datasets for finetuning these QA systems.\nSummarization systems aim to provide a condensed version of a longer piece of text. The motivation is to allow readers to derive the main points of a document without expending as much effort in reading or to provide a quick overview of a document for the reader to decide whether or not to invest more time. Two systems in Table 2 incorporate summarization components: Vespa CORD-19 Search (Row S30), which generates paper-level summaries, and CAiRE-Covid [83] (Row S29), a QA system that generates multi-document summaries across answering spans. The CAiRE-Covid system generates both extractive and abstractive summaries by aggregating information across answering spans for an input query, providing a quick, high-level overview of current research.\nKB construction describes systems that create KBs by extracting entities and relations from text. The KB can be used to support other goals like search or exploration, or may be the primary goal, as in the AIM COVID-19 database (Row S39), which links papers to their corresponding clinical trials and trial results. The AIM database allows users to track the state of treatment and vaccine development for COVID-19.\nVisualization provides a visual way to interact with and understand data. Visualizations are usually coupled with extracted KBs or citation networks and provide an alternate way to explore a corpus of scientific papers. Examples include SemViz [92] (Row S36), which focuses on exploration of the CORD-19 corpus, the Blender Lab COVID-KG and protein–protein interaction datasets and SciSight [33] (Row S38), which allows users to browse the documents in CORD-19 by author, institutional affiliation, extracted entities and network relationships.\nAugmented reading systems attempt to improve upon the standard reading experience of papers by providing features such as entity highlighting or within-document and between-document links, e.g. COVID-19 Intelligent Insight (Row S12) highlights extracted entities directly on a paper PDF.\nOther tasks may be more specialized. For claim verification, a system identifies papers containing evidence that supports or refutes a claim provided in a query. SciFact [97] (Row S22) is an example of such a system. For clinical diagnostic support, a system aims to assist healthcare providers in clinical practice, e.g. the Kahun system (Row S18) allows providers to enter patient signs and symptoms, laboratory values and medical history, and provides likely diagnostic outcomes based on known associations derived from literature and other sources.\nSeveral of the systems we catalog use KBs or provide tight integration with controlled vocabularies (e.g. UMLS, ICD-10) or ontologies (e.g. Gene Ontology). These systems are well positioned to integrate with other data sources that use the same shared vocabularies and to leverage the automated reasoning or inference capabilities of structured KBs. We also observe that very few text mining systems in production have a clinical focus. Those that are better integrated with clinical trial data (e.g. Covidex (Row S1), DOC Search (Row S5), COVID-19 Intelligent Insight (Row S12) and AIM COVID-19 Database (Row S39)) may provide better insights for clinicians and clinical researchers. Going forward, we expect more opportunities for integrating these systems into clinical environments, where novel diagnostic and treatment strategies identified in the literature can be quickly adapted into practice.","divisions":[{"label":"title","span":{"begin":0,"end":19}},{"label":"p","span":{"begin":20,"end":639}},{"label":"table-wrap","span":{"begin":640,"end":16022}},{"label":"label","span":{"begin":640,"end":647}},{"label":"caption","span":{"begin":649,"end":2056}},{"label":"p","span":{"begin":649,"end":2056}},{"label":"table","span":{"begin":2057,"end":16022}},{"label":"tr","span":{"begin":2057,"end":2133}},{"label":"th","span":{"begin":2057,"end":2063}},{"label":"th","span":{"begin":2064,"end":2071}},{"label":"th","span":{"begin":2073,"end":2085}},{"label":"th","span":{"begin":2087,"end":2092}},{"label":"th","span":{"begin":2094,"end":2099}},{"label":"th","span":{"begin":2101,"end":2116}},{"label":"th","span":{"begin":2118,"end":2133}},{"label":"tr","span":{"begin":2134,"end":2531}},{"label":"td","span":{"begin":2134,"end":2136}},{"label":"td","span":{"begin":2138,"end":2144}},{"label":"td","span":{"begin":2146,"end":2159}},{"label":"td","span":{"begin":2161,"end":2191}},{"label":"td","span":{"begin":2193,"end":2212}},{"label":"td","span":{"begin":2214,"end":2263}},{"label":"td","span":{"begin":2265,"end":2411}},{"label":"td","span":{"begin":2413,"end":2531}},{"label":"tr","span":{"begin":2532,"end":2901}},{"label":"td","span":{"begin":2532,"end":2534}},{"label":"td","span":{"begin":2536,"end":2542}},{"label":"td","span":{"begin":2544,"end":2564}},{"label":"td","span":{"begin":2566,"end":2567}},{"label":"td","span":{"begin":2569,"end":2615}},{"label":"td","span":{"begin":2617,"end":2640}},{"label":"td","span":{"begin":2642,"end":2852}},{"label":"td","span":{"begin":2854,"end":2901}},{"label":"tr","span":{"begin":2902,"end":3161}},{"label":"td","span":{"begin":2902,"end":2904}},{"label":"td","span":{"begin":2906,"end":2912}},{"label":"td","span":{"begin":2914,"end":2931}},{"label":"td","span":{"begin":2933,"end":2949}},{"label":"td","span":{"begin":2951,"end":2979}},{"label":"td","span":{"begin":2981,"end":3026}},{"label":"td","span":{"begin":3028,"end":3041}},{"label":"td","span":{"begin":3043,"end":3161}},{"label":"tr","span":{"begin":3162,"end":3702}},{"label":"td","span":{"begin":3162,"end":3164}},{"label":"td","span":{"begin":3166,"end":3172}},{"label":"td","span":{"begin":3174,"end":3181}},{"label":"td","span":{"begin":3183,"end":3233}},{"label":"td","span":{"begin":3235,"end":3253}},{"label":"td","span":{"begin":3255,"end":3262}},{"label":"td","span":{"begin":3264,"end":3490}},{"label":"td","span":{"begin":3492,"end":3702}},{"label":"tr","span":{"begin":3703,"end":3953}},{"label":"td","span":{"begin":3703,"end":3705}},{"label":"td","span":{"begin":3707,"end":3713}},{"label":"td","span":{"begin":3715,"end":3725}},{"label":"td","span":{"begin":3727,"end":3730}},{"label":"td","span":{"begin":3732,"end":3772}},{"label":"td","span":{"begin":3774,"end":3841}},{"label":"td","span":{"begin":3843,"end":3844}},{"label":"td","span":{"begin":3846,"end":3953}},{"label":"tr","span":{"begin":3954,"end":4252}},{"label":"td","span":{"begin":3954,"end":3956}},{"label":"td","span":{"begin":3958,"end":3964}},{"label":"td","span":{"begin":3966,"end":3978}},{"label":"td","span":{"begin":3980,"end":3981}},{"label":"td","span":{"begin":3983,"end":4008}},{"label":"td","span":{"begin":4010,"end":4017}},{"label":"td","span":{"begin":4019,"end":4162}},{"label":"td","span":{"begin":4164,"end":4252}},{"label":"tr","span":{"begin":4253,"end":4437}},{"label":"td","span":{"begin":4253,"end":4255}},{"label":"td","span":{"begin":4257,"end":4263}},{"label":"td","span":{"begin":4265,"end":4276}},{"label":"td","span":{"begin":4278,"end":4311}},{"label":"td","span":{"begin":4313,"end":4333}},{"label":"td","span":{"begin":4335,"end":4342}},{"label":"td","span":{"begin":4344,"end":4345}},{"label":"td","span":{"begin":4347,"end":4437}},{"label":"tr","span":{"begin":4438,"end":4856}},{"label":"td","span":{"begin":4438,"end":4440}},{"label":"td","span":{"begin":4442,"end":4448}},{"label":"td","span":{"begin":4450,"end":4460}},{"label":"td","span":{"begin":4462,"end":4481}},{"label":"td","span":{"begin":4483,"end":4511}},{"label":"td","span":{"begin":4513,"end":4530}},{"label":"td","span":{"begin":4532,"end":4619}},{"label":"td","span":{"begin":4621,"end":4856}},{"label":"tr","span":{"begin":4857,"end":5133}},{"label":"td","span":{"begin":4857,"end":4859}},{"label":"td","span":{"begin":4861,"end":4867}},{"label":"td","span":{"begin":4869,"end":4884}},{"label":"td","span":{"begin":4886,"end":4901}},{"label":"td","span":{"begin":4903,"end":4943}},{"label":"td","span":{"begin":4945,"end":4952}},{"label":"td","span":{"begin":4954,"end":4955}},{"label":"td","span":{"begin":4957,"end":5133}},{"label":"tr","span":{"begin":5134,"end":5445}},{"label":"td","span":{"begin":5134,"end":5137}},{"label":"td","span":{"begin":5139,"end":5145}},{"label":"td","span":{"begin":5147,"end":5166}},{"label":"td","span":{"begin":5168,"end":5172}},{"label":"td","span":{"begin":5174,"end":5202}},{"label":"td","span":{"begin":5204,"end":5211}},{"label":"td","span":{"begin":5213,"end":5300}},{"label":"td","span":{"begin":5302,"end":5445}},{"label":"tr","span":{"begin":5446,"end":5757}},{"label":"td","span":{"begin":5446,"end":5449}},{"label":"td","span":{"begin":5451,"end":5476}},{"label":"td","span":{"begin":5478,"end":5501}},{"label":"td","span":{"begin":5503,"end":5510}},{"label":"td","span":{"begin":5512,"end":5537}},{"label":"td","span":{"begin":5539,"end":5546}},{"label":"td","span":{"begin":5548,"end":5549}},{"label":"td","span":{"begin":5551,"end":5757}},{"label":"tr","span":{"begin":5758,"end":6172}},{"label":"td","span":{"begin":5758,"end":5761}},{"label":"td","span":{"begin":5763,"end":5788}},{"label":"td","span":{"begin":5790,"end":5818}},{"label":"td","span":{"begin":5820,"end":5827}},{"label":"td","span":{"begin":5829,"end":5861}},{"label":"td","span":{"begin":5863,"end":6005}},{"label":"td","span":{"begin":6007,"end":6008}},{"label":"td","span":{"begin":6010,"end":6172}},{"label":"tr","span":{"begin":6173,"end":6518}},{"label":"td","span":{"begin":6173,"end":6176}},{"label":"td","span":{"begin":6178,"end":6197}},{"label":"td","span":{"begin":6199,"end":6222}},{"label":"td","span":{"begin":6224,"end":6233}},{"label":"td","span":{"begin":6235,"end":6262}},{"label":"td","span":{"begin":6264,"end":6305}},{"label":"td","span":{"begin":6307,"end":6308}},{"label":"td","span":{"begin":6310,"end":6518}},{"label":"tr","span":{"begin":6519,"end":6786}},{"label":"td","span":{"begin":6519,"end":6522}},{"label":"td","span":{"begin":6524,"end":6543}},{"label":"td","span":{"begin":6545,"end":6563}},{"label":"td","span":{"begin":6565,"end":6575}},{"label":"td","span":{"begin":6577,"end":6618}},{"label":"td","span":{"begin":6620,"end":6717}},{"label":"td","span":{"begin":6719,"end":6720}},{"label":"td","span":{"begin":6722,"end":6786}},{"label":"tr","span":{"begin":6787,"end":7192}},{"label":"td","span":{"begin":6787,"end":6790}},{"label":"td","span":{"begin":6792,"end":6811}},{"label":"td","span":{"begin":6813,"end":6828}},{"label":"td","span":{"begin":6830,"end":6852}},{"label":"td","span":{"begin":6854,"end":6908}},{"label":"td","span":{"begin":6910,"end":6917}},{"label":"td","span":{"begin":6919,"end":7031}},{"label":"td","span":{"begin":7033,"end":7192}},{"label":"tr","span":{"begin":7193,"end":7638}},{"label":"td","span":{"begin":7193,"end":7196}},{"label":"td","span":{"begin":7198,"end":7217}},{"label":"td","span":{"begin":7219,"end":7238}},{"label":"td","span":{"begin":7240,"end":7244}},{"label":"td","span":{"begin":7246,"end":7272}},{"label":"td","span":{"begin":7274,"end":7295}},{"label":"td","span":{"begin":7297,"end":7423}},{"label":"td","span":{"begin":7425,"end":7638}},{"label":"tr","span":{"begin":7639,"end":7949}},{"label":"td","span":{"begin":7639,"end":7642}},{"label":"td","span":{"begin":7644,"end":7695}},{"label":"td","span":{"begin":7697,"end":7703}},{"label":"td","span":{"begin":7705,"end":7716}},{"label":"td","span":{"begin":7718,"end":7736}},{"label":"td","span":{"begin":7738,"end":7745}},{"label":"td","span":{"begin":7747,"end":7748}},{"label":"td","span":{"begin":7750,"end":7949}},{"label":"tr","span":{"begin":7950,"end":8299}},{"label":"td","span":{"begin":7950,"end":7953}},{"label":"td","span":{"begin":7955,"end":8035}},{"label":"td","span":{"begin":8037,"end":8042}},{"label":"td","span":{"begin":8044,"end":8049}},{"label":"td","span":{"begin":8051,"end":8081}},{"label":"td","span":{"begin":8083,"end":8136}},{"label":"td","span":{"begin":8138,"end":8139}},{"label":"td","span":{"begin":8141,"end":8299}},{"label":"tr","span":{"begin":8300,"end":8702}},{"label":"td","span":{"begin":8300,"end":8303}},{"label":"td","span":{"begin":8305,"end":8326}},{"label":"td","span":{"begin":8328,"end":8342}},{"label":"td","span":{"begin":8344,"end":8381}},{"label":"td","span":{"begin":8383,"end":8411}},{"label":"td","span":{"begin":8413,"end":8429}},{"label":"td","span":{"begin":8431,"end":8594}},{"label":"td","span":{"begin":8596,"end":8702}},{"label":"tr","span":{"begin":8703,"end":9181}},{"label":"td","span":{"begin":8703,"end":8706}},{"label":"td","span":{"begin":8708,"end":8742}},{"label":"td","span":{"begin":8744,"end":8757}},{"label":"td","span":{"begin":8759,"end":8788}},{"label":"td","span":{"begin":8790,"end":8814}},{"label":"td","span":{"begin":8816,"end":8832}},{"label":"td","span":{"begin":8834,"end":8920}},{"label":"td","span":{"begin":8922,"end":9181}},{"label":"tr","span":{"begin":9182,"end":9634}},{"label":"td","span":{"begin":9182,"end":9185}},{"label":"td","span":{"begin":9187,"end":9221}},{"label":"td","span":{"begin":9223,"end":9235}},{"label":"td","span":{"begin":9237,"end":9248}},{"label":"td","span":{"begin":9250,"end":9279}},{"label":"td","span":{"begin":9281,"end":9345}},{"label":"td","span":{"begin":9347,"end":9443}},{"label":"td","span":{"begin":9445,"end":9634}},{"label":"tr","span":{"begin":9635,"end":10091}},{"label":"td","span":{"begin":9635,"end":9638}},{"label":"td","span":{"begin":9640,"end":9666}},{"label":"td","span":{"begin":9668,"end":9708}},{"label":"td","span":{"begin":9710,"end":9732}},{"label":"td","span":{"begin":9734,"end":9767}},{"label":"td","span":{"begin":9769,"end":9792}},{"label":"td","span":{"begin":9794,"end":9939}},{"label":"td","span":{"begin":9941,"end":10091}},{"label":"tr","span":{"begin":10092,"end":10504}},{"label":"td","span":{"begin":10092,"end":10095}},{"label":"td","span":{"begin":10097,"end":10107}},{"label":"td","span":{"begin":10109,"end":10123}},{"label":"td","span":{"begin":10125,"end":10135}},{"label":"td","span":{"begin":10137,"end":10163}},{"label":"td","span":{"begin":10165,"end":10182}},{"label":"td","span":{"begin":10184,"end":10367}},{"label":"td","span":{"begin":10369,"end":10504}},{"label":"tr","span":{"begin":10505,"end":10939}},{"label":"td","span":{"begin":10505,"end":10508}},{"label":"td","span":{"begin":10510,"end":10520}},{"label":"td","span":{"begin":10522,"end":10544}},{"label":"td","span":{"begin":10546,"end":10571}},{"label":"td","span":{"begin":10573,"end":10592}},{"label":"td","span":{"begin":10594,"end":10628}},{"label":"td","span":{"begin":10630,"end":10771}},{"label":"td","span":{"begin":10773,"end":10939}},{"label":"tr","span":{"begin":10940,"end":11457}},{"label":"td","span":{"begin":10940,"end":10943}},{"label":"td","span":{"begin":10945,"end":10955}},{"label":"td","span":{"begin":10957,"end":10970}},{"label":"td","span":{"begin":10972,"end":11000}},{"label":"td","span":{"begin":11002,"end":11031}},{"label":"td","span":{"begin":11033,"end":11066}},{"label":"td","span":{"begin":11068,"end":11276}},{"label":"td","span":{"begin":11278,"end":11457}},{"label":"tr","span":{"begin":11458,"end":11823}},{"label":"td","span":{"begin":11458,"end":11461}},{"label":"td","span":{"begin":11463,"end":11473}},{"label":"td","span":{"begin":11475,"end":11507}},{"label":"td","span":{"begin":11509,"end":11525}},{"label":"td","span":{"begin":11527,"end":11559}},{"label":"td","span":{"begin":11561,"end":11576}},{"label":"td","span":{"begin":11578,"end":11634}},{"label":"td","span":{"begin":11636,"end":11823}},{"label":"tr","span":{"begin":11824,"end":12066}},{"label":"td","span":{"begin":11824,"end":11827}},{"label":"td","span":{"begin":11829,"end":11839}},{"label":"td","span":{"begin":11841,"end":11852}},{"label":"td","span":{"begin":11854,"end":11881}},{"label":"td","span":{"begin":11883,"end":11912}},{"label":"td","span":{"begin":11914,"end":11921}},{"label":"td","span":{"begin":11923,"end":11950}},{"label":"td","span":{"begin":11952,"end":12066}},{"label":"tr","span":{"begin":12067,"end":12332}},{"label":"td","span":{"begin":12067,"end":12070}},{"label":"td","span":{"begin":12072,"end":12082}},{"label":"td","span":{"begin":12084,"end":12110}},{"label":"td","span":{"begin":12112,"end":12127}},{"label":"td","span":{"begin":12129,"end":12175}},{"label":"td","span":{"begin":12177,"end":12184}},{"label":"td","span":{"begin":12186,"end":12187}},{"label":"td","span":{"begin":12189,"end":12332}},{"label":"tr","span":{"begin":12333,"end":13015}},{"label":"td","span":{"begin":12333,"end":12336}},{"label":"td","span":{"begin":12338,"end":12363}},{"label":"td","span":{"begin":12365,"end":12381}},{"label":"td","span":{"begin":12383,"end":12445}},{"label":"td","span":{"begin":12447,"end":12474}},{"label":"td","span":{"begin":12476,"end":12509}},{"label":"td","span":{"begin":12511,"end":12832}},{"label":"td","span":{"begin":12834,"end":13015}},{"label":"tr","span":{"begin":13016,"end":13278}},{"label":"td","span":{"begin":13016,"end":13019}},{"label":"td","span":{"begin":13021,"end":13042}},{"label":"td","span":{"begin":13044,"end":13058}},{"label":"td","span":{"begin":13060,"end":13065}},{"label":"td","span":{"begin":13067,"end":13091}},{"label":"td","span":{"begin":13093,"end":13100}},{"label":"td","span":{"begin":13102,"end":13209}},{"label":"td","span":{"begin":13211,"end":13278}},{"label":"tr","span":{"begin":13279,"end":13451}},{"label":"td","span":{"begin":13279,"end":13282}},{"label":"td","span":{"begin":13284,"end":13295}},{"label":"td","span":{"begin":13297,"end":13304}},{"label":"td","span":{"begin":13306,"end":13322}},{"label":"td","span":{"begin":13324,"end":13343}},{"label":"td","span":{"begin":13345,"end":13398}},{"label":"td","span":{"begin":13400,"end":13401}},{"label":"td","span":{"begin":13403,"end":13451}},{"label":"tr","span":{"begin":13452,"end":13817}},{"label":"td","span":{"begin":13452,"end":13455}},{"label":"td","span":{"begin":13457,"end":13468}},{"label":"td","span":{"begin":13470,"end":13483}},{"label":"td","span":{"begin":13485,"end":13520}},{"label":"td","span":{"begin":13522,"end":13561}},{"label":"td","span":{"begin":13563,"end":13570}},{"label":"td","span":{"begin":13572,"end":13737}},{"label":"td","span":{"begin":13739,"end":13817}},{"label":"tr","span":{"begin":13818,"end":14103}},{"label":"td","span":{"begin":13818,"end":13821}},{"label":"td","span":{"begin":13823,"end":13834}},{"label":"td","span":{"begin":13836,"end":13857}},{"label":"td","span":{"begin":13859,"end":13864}},{"label":"td","span":{"begin":13866,"end":13898}},{"label":"td","span":{"begin":13900,"end":13907}},{"label":"td","span":{"begin":13909,"end":14012}},{"label":"td","span":{"begin":14014,"end":14103}},{"label":"tr","span":{"begin":14104,"end":14339}},{"label":"td","span":{"begin":14104,"end":14107}},{"label":"td","span":{"begin":14109,"end":14120}},{"label":"td","span":{"begin":14122,"end":14134}},{"label":"td","span":{"begin":14136,"end":14137}},{"label":"td","span":{"begin":14139,"end":14180}},{"label":"td","span":{"begin":14182,"end":14193}},{"label":"td","span":{"begin":14195,"end":14271}},{"label":"td","span":{"begin":14273,"end":14339}},{"label":"tr","span":{"begin":14340,"end":14683}},{"label":"td","span":{"begin":14340,"end":14343}},{"label":"td","span":{"begin":14345,"end":14371}},{"label":"td","span":{"begin":14373,"end":14390}},{"label":"td","span":{"begin":14392,"end":14453}},{"label":"td","span":{"begin":14455,"end":14485}},{"label":"td","span":{"begin":14487,"end":14494}},{"label":"td","span":{"begin":14496,"end":14591}},{"label":"td","span":{"begin":14593,"end":14683}},{"label":"tr","span":{"begin":14684,"end":15116}},{"label":"td","span":{"begin":14684,"end":14687}},{"label":"td","span":{"begin":14689,"end":14715}},{"label":"td","span":{"begin":14717,"end":14728}},{"label":"td","span":{"begin":14730,"end":14793}},{"label":"td","span":{"begin":14795,"end":14818}},{"label":"td","span":{"begin":14820,"end":14900}},{"label":"td","span":{"begin":14902,"end":15000}},{"label":"td","span":{"begin":15002,"end":15116}},{"label":"tr","span":{"begin":15117,"end":15375}},{"label":"td","span":{"begin":15117,"end":15120}},{"label":"td","span":{"begin":15122,"end":15148}},{"label":"td","span":{"begin":15150,"end":15164}},{"label":"td","span":{"begin":15166,"end":15193}},{"label":"td","span":{"begin":15195,"end":15229}},{"label":"td","span":{"begin":15231,"end":15246}},{"label":"td","span":{"begin":15248,"end":15322}},{"label":"td","span":{"begin":15324,"end":15375}},{"label":"tr","span":{"begin":15376,"end":15703}},{"label":"td","span":{"begin":15376,"end":15379}},{"label":"td","span":{"begin":15381,"end":15424}},{"label":"td","span":{"begin":15426,"end":15439}},{"label":"td","span":{"begin":15441,"end":15463}},{"label":"td","span":{"begin":15465,"end":15499}},{"label":"td","span":{"begin":15501,"end":15513}},{"label":"td","span":{"begin":15515,"end":15593}},{"label":"td","span":{"begin":15595,"end":15703}},{"label":"tr","span":{"begin":15704,"end":16022}},{"label":"td","span":{"begin":15704,"end":15707}},{"label":"td","span":{"begin":15709,"end":15724}},{"label":"td","span":{"begin":15726,"end":15747}},{"label":"td","span":{"begin":15749,"end":15762}},{"label":"td","span":{"begin":15764,"end":15797}},{"label":"td","span":{"begin":15799,"end":15814}},{"label":"td","span":{"begin":15816,"end":15817}},{"label":"td","span":{"begin":15819,"end":16022}},{"label":"p","span":{"begin":16023,"end":16556}},{"label":"p","span":{"begin":16557,"end":17303}},{"label":"p","span":{"begin":17304,"end":18057}},{"label":"p","span":{"begin":18058,"end":19134}},{"label":"p","span":{"begin":19135,"end":20099}},{"label":"p","span":{"begin":20100,"end":21245}},{"label":"p","span":{"begin":21245,"end":21652}},{"label":"p","span":{"begin":21653,"end":22551}},{"label":"p","span":{"begin":22551,"end":23305}},{"label":"p","span":{"begin":23305,"end":23721}},{"label":"p","span":{"begin":23721,"end":24266}},{"label":"p","span":{"begin":24266,"end":24554}},{"label":"p","span":{"begin":24554,"end":25107}}],"tracks":[]}