Large-scale Information Extraction Research Articles

Constructing adverse event timelines for patients receiving CAR-T therapy using large language models.

2555 Background: Chimeric antigen receptor T (CAR-T) therapy is associated with a high risk of severe adverse events often only detailed in clinical notes. Monitoring them demands significant time and effort for manual chart review. Recent developments in large language modeling (LLMs) show promise for large-scale information extraction from clinical text. We performed a pilot study to evaluate the capability of the GPT-4 LLM to extract adverse events documented in the progress reports. Methods: We extracted progress notes within 30 days of any CAR-T administration from the UCSF deidentified clinical data warehouse. GPT-4, accessed through a HIPAA compliant Microsoft Azure Studio API, was used to extract CAR-T adverse events resulting in clinical intervention. A random sample of adverse events from 10% of patient notes were evaluated by a clinical reviewer (JG, PharmD). Topic modeling using BERTopic was used to cluster all adverse events to identify trends over time. Results: We identified 4183 clinical notes written within 30 days of CAR-T administration from 253 patients (39.1% women, 60.9% men). Mean age was 60.6 (SD:17.7). Manual validation of clinical notes from 10% of patients with CAR-T therapies (n=25) demonstrated that GPT4 was able to extract CAR-T related adverse events with 64% accuracy. We used BERTopic to cluster all extracted adverse events into 19 topics. Clusters with key terms “hyponatremia, leukocytosis, encephalopathy, toxicities, and neurologic” occurred most often (n=277), and primarily documented 12.9 days after CAR-T administration (Table). Conclusions: Although limited by use of de-identified data and absence of prompt engineering, this pilot supports the further investigation of LLMs for extraction of adverse events from unstructured clinical text. [Table: see text]

Mountain Vegetation Classification Method Based on Multi-Channel Semantic Segmentation Model

With the development of satellite remote sensing technology, a substantial quantity of remote sensing data can be obtained every day, but the ability to extract information from these data remains poor, especially regarding intelligent extraction models for vegetation information in mountainous areas. Because the features of remote sensing images (such as spectral, textural and geometric features) change with changes in illumination, viewing angle, scale and spectrum, it is difficult for a remote sensing intelligent interpretation model with a single data source as input to meet the requirements of engineering or large-scale vegetation information extraction and updating. The effective use multi-source, multi-resolution and multi-type data for remote sensing classification is still a challenge. The objective of this study is to develop a highly intelligent and generalizable classification model of mountain vegetation utilizing multi-source remote sensing data to achieve accurate vegetation extraction. Therefore, a multi-channel semantic segmentation model based on deep learning, FCN-ResNet, is proposed to integrate the features and textures of multi-source, multi-resolution and multi-temporal remote sensing data, thereby enhancing the differentiation of different mountain vegetation types by capturing their characteristics and dynamic changes. In addition, several sets of ablation experiments are designed to investigate the effectiveness of the model. The method is validated on Mt. Taibai (part of the Qinling-Daba Mountains), and the pixel accuracy (PA) of vegetation classification reaches 85.8%. The results show that the proposed multi-channel semantic segmentation model can effectively discriminate different vegetation types and has good intelligence and generalization ability in different mountainous areas with similar vegetation distributions. The multi-channel semantic segmentation model can be used for the rapid updating of vegetation type maps in mountainous areas.

Large-scale genealogical information extraction from handwritten Quebec parish records

This paper presents a complete workflow designed for extracting information from Quebec handwritten parish registers. The acts in these documents contain individual and family information highly valuable for genetic, demographic and social studies of the Quebec population. From an image of parish records, our workflow is able to identify the acts and extract personal information. The workflow is divided into successive steps: page classification, text line detection, handwritten text recognition, named entity recognition and act detection and classification. For all these steps, different machine learning models are compared. Once the information is extracted, validation rules designed by experts are then applied to standardize the extracted information and ensure its consistency with the type of act (birth, marriage and death). This validation step is able to reject records that are considered invalid or merged. The full workflow has been used to process over two million pages of Quebec parish registers from the 19–20th centuries. On a sample comprising 65% of registers, 3.2 million acts were recognized. Verification of the birth and death acts from this sample shows that 74% of them are considered complete and valid. These records will be integrated into the BALSAC database and linked together to recreate family and genealogical relations at large scale.

Knowledge Graph Embedding for Link Prediction

Knowledge Graphs (KGs) have found many applications in industrial and in academic settings, which in turn, have motivated considerable research efforts towards large-scale information extraction from a variety of sources. Despite such efforts, it is well known that even the largest KGs suffer from incompleteness; Link Prediction (LP) techniques address this issue by identifying missing facts among entities already in the KG. Among the recent LP techniques, those based on KG embeddings have achieved very promising performance in some benchmarks. Despite the fast-growing literature on the subject, insufficient attention has been paid to the effect of the design choices in those methods. Moreover, the standard practice in this area is to report accuracy by aggregating over a large number of test facts in which some entities are vastly more represented than others; this allows LP methods to exhibit good results by just attending to structural properties that include such entities, while ignoring the remaining majority of the KG. This analysis provides a comprehensive comparison of embedding-based LP methods, extending the dimensions of analysis beyond what is commonly available in the literature. We experimentally compare the effectiveness and efficiency of 18 state-of-the-art methods, consider a rule-based baseline, and report detailed analysis over the most popular benchmarks in the literature.

Optical Tissue Clearing: Illuminating Brain Function and Dysfunction.

Tissue optical clearing technology has been developing rapidly in the past decade due to advances in microscopy equipment and various labeling techniques. Consistent modification of primary methods for optical tissue transparency has allowed observation of the whole mouse body at single-cell resolution or thick tissue slices at the nanoscale level, with the final aim to make intact primate and human brains or thick human brain tissues optically transparent. Optical clearance combined with flexible large-volume tissue labeling technology can not only preserve the anatomical structure but also visualize multiple molecular information from intact samples in situ. It also provides a new strategy for studying complex tissues, which is of great significance for deciphering the functional structure of healthy brains and the mechanisms of neurological pathologies. In this review, we briefly introduce the existing optical clearing technology and discuss its application in deciphering connection and structure, brain development, and brain diseases. Besides, we discuss the standard computational analysis tools for large-scale imaging dataset processing and information extraction. In general, we hope that this review will provide a valuable reference for researchers who intend to use optical clearing technology in studying the brain.

QA4IE: A Question Answering Based System for Document-Level General Information Extraction

Information Extraction (IE) is the task of distilling structured information from unstructured texts by identifying references to named entities as well as relationships between such entities. Existing IE solutions, including Relation Extraction and Open IE, can hardly take cross-sentence information like coreferences into account and are severely restricted by limited relation types as well as informal relation specifications (e.g., free-text based relation triples). In order to overcome the weaknesses, we propose a novel IE framework named QA4IE, which leverages the flexible question answering approaches to produce high-quality relation triples across sentences. Based on this framework, we develop a real-time IE system, which can perform general IE throughout the entire document. For training and evaluating our system, we build a large-scale IE benchmark using distant supervision under human evaluation. We deploy both component analyses and pipeline experiments to evaluate our system. The results show that our system can generalize on unseen entities and relations, as well as achieve significant improvements over existing IE systems.

Automated Extraction of Human Settlement Patterns From Historical Topographic Map Series Using Weakly Supervised Convolutional Neural Networks

Information extraction from historical maps represents a persistent challenge due to inferior graphical quality and the large data volume of digital map archives, which can hold thousands of digitized map sheets. Traditional map processing techniques typically rely on manually collected templates of the symbol of interest, and thus are not suitable for large-scale information extraction. In order to digitally preserve such large amounts of valuable retrospective geographic information, high levels of automation are required. Herein, we propose an automated machine-learning based framework to extract human settlement symbols, such as buildings and urban areas from historical topographic maps in the absence of training data, employing contemporary geospatial data as ancillary data to guide the collection of training samples. These samples are then used to train a convolutional neural network for semantic image segmentation, allowing for the extraction of human settlement patterns in an analysis-ready geospatial vector data format. We test our method on United States Geological Survey historical topographic maps published between 1893 and 1954. The results are promising, indicating high degrees of completeness in the extracted settlement features (i.e., recall of up to 0.96, F-measure of up to 0.79) and will guide the next steps to provide a fully automated operational approach for large-scale geographic feature extraction from a variety of historical map series. Moreover, the proposed framework provides a robust approach for the recognition of objects which are small in size, generalizable to many kinds of visual documents.

Named Entity Recognition and Normalization Applied to Large-Scale Information Extraction from the Materials Science Literature.

The number of published materials science articles has increased manyfold over the past few decades. Now, a major bottleneck in the materials discovery pipeline arises in connecting new results with the previously established literature. A potential solution to this problem is to map the unstructured raw text of published articles onto structured database entries that allow for programmatic querying. To this end, we apply text mining with named entity recognition (NER) for large-scale information extraction from the published materials science literature. The NER model is trained to extract summary-level information from materials science documents, including inorganic material mentions, sample descriptors, phase labels, material properties and applications, as well as any synthesis and characterization methods used. Our classifier achieves an accuracy (f1) of 87%, and is applied to information extraction from 3.27 million materials science abstracts. We extract more than 80 million materials-science-related named entities, and the content of each abstract is represented as a database entry in a structured format. We demonstrate that simple database queries can be used to answer complex "meta-questions" of the published literature that would have previously required laborious, manual literature searches to answer. All of our data and functionality has been made freely available on our Github( https://github.com/materialsintelligence/matscholar )and website ( http://matscholar.com ), and we expect these results to accelerate the pace of future materials science discovery.

Map Archive Mining: Visual-Analytical Approaches to Explore Large Historical Map Collections.

Historical maps are unique sources of retrospective geographical information. Recently, several map archives containing map series covering large spatial and temporal extents have been systematically scanned and made available to the public. The geographical information contained in such data archives makes it possible to extend geospatial analysis retrospectively beyond the era of digital cartography. However, given the large data volumes of such archives (e.g., more than 200,000 map sheets in the United States Geological Survey topographic map archive) and the low graphical quality of older, manually-produced map sheets, the process to extract geographical information from these map archives needs to be automated to the highest degree possible. To understand the potential challenges (e.g., salient map characteristics and data quality variations) in automating large-scale information extraction tasks for map archives, it is useful to efficiently assess spatio-temporal coverage, approximate map content, and spatial accuracy of georeferenced map sheets at different map scales. Such preliminary analytical steps are often neglected or ignored in the map processing literature but represent critical phases that lay the foundation for any subsequent computational processes including recognition. Exemplified for the United States Geological Survey topographic map and the Sanborn fire insurance map archives, we demonstrate how such preliminary analyses can be systematically conducted using traditional analytical and cartographic techniques, as well as visual-analytical data mining tools originating from machine learning and data science.

Unlocking echocardiogram measurements for heart disease research through natural language processing

BackgroundIn order to investigate the mechanisms of cardiovascular disease in HIV infected and uninfected patients, an analysis of echocardiogram reports is required for a large longitudinal multi-center study.ImplementationA natural language processing system using a dictionary lookup, rules, and patterns was developed to extract heart function measurements that are typically recorded in echocardiogram reports as measurement-value pairs. Curated semantic bootstrapping was used to create a custom dictionary that extends existing terminologies based on terms that actually appear in the medical record. A novel disambiguation method based on semantic constraints was created to identify and discard erroneous alternative definitions of the measurement terms. The system was built utilizing a scalable framework, making it available for processing large datasets.ResultsThe system was developed for and validated on notes from three sources: general clinic notes, echocardiogram reports, and radiology reports. The system achieved F-scores of 0.872, 0.844, and 0.877 with precision of 0.936, 0.982, and 0.969 for each dataset respectively averaged across all extracted values. Left ventricular ejection fraction (LVEF) is the most frequently extracted measurement. The precision of extraction of the LVEF measure ranged from 0.968 to 1.0 across different document types.ConclusionsThis system illustrates the feasibility and effectiveness of a large-scale information extraction on clinical data. New clinical questions can be addressed in the domain of heart failure using retrospective clinical data analysis because key heart function measurements can be successfully extracted using natural language processing.

Large-Scale Information Extraction from Textual Definitions through Deep Syntactic and Semantic Analysis

We present DefIE, an approach to large-scale Information Extraction (IE) based on a syntactic-semantic analysis of textual definitions. Given a large corpus of definitions we leverage syntactic dependencies to reduce data sparsity, then disambiguate the arguments and content words of the relation strings, and finally exploit the resulting information to organize the acquired relations hierarchically. The output of DefIE is a high-quality knowledge base consisting of several million automatically acquired semantic relations.

Bibliome Mining Platform and Application for Building Metabolic Interaction Network

Bibliome mining is a powerful technique for large-scale information extraction from textual data and connecting between biological entities as well as functional hypotheses. Currently, most bibliome mining is used for some specific studies involving genes and proteins; however, much less efforts have been focused on metabolites. In addition to application works, the focus has been on proving the concepts and algorithms, but very few reports on development of applicable text mining platform. In this study, we aimed to develop a bibliome mining platform that could be used to perform basic text mining tasks and further be used for building metabolic interaction networks. We developed a platform with the evaluated tools and subsequently tested its functions for extraction of interactions between biomolecules (e.g. genes, enzymes, proteins and metabolites) in yeast Saccharomyces cerevisiae. The results were then manually curated afterwards using KEGG LIGAND and public yeast database. Of the collected 11 text mining tools, we selected 3 suitable tools, namely ABNER, OpenNLP and LingPipe for further implementing the bibliome mining platform. In summary, a prototype of a bibliome mining platform was successfully developed and may further be used for building metabolic interaction network of S. cerevisiae. This study can be used as a basic framework for further improvement as well as extension of relevant text mining tasks and broad applications.

Extraction of Hongze Lake Reclamation Area Based on RADARSAT SAR and LANDSAT ETM+

Abstract Due to some limitations of the visible and near infrared bands in remote sensing, radar remote sensing technology, which is capable in all-time, all-weather, and penetrating to some degree and so forth, has become a new technology and new method in rapid, large-scale information extraction. In this paper, the feasibility of using radar image as data source in terms of information extraction of Hongze Lake reclamation area is discussed. Firstly, based on RADARSAT-1 data, radar image was calibrated, geometrically corrected and filtered. Secondly, classification was carried out on SAR data and ETM+ data by using thresholds with the help of groundtruth data. Then, the ETM+ data was classified into the categories of aquaculture, open water, rice paddy and wetland vegetation by using wetness and spectral characteristics. Finally, the categories extracted from two different data sources were combined together. The results show that the methods used in this research are practical and efficient. The overall accuracy of classification is 93% with high Kappa Coefficient of 0.87.

A simple and robust method for extracting terminology

In this paper we will present a simple, yet effective, method for extracting terminology from technical text. The method is based on the observation that for technical domains it is much simpler to describe what a valid terminological unit cannot be than what it can possibly be. Our method relies on a set of filters that exclude multi-word units according to simple rules regarding their context and internal lexical structure, and it does not require any special pre-processing such as POS tagging. Rules were hand-coded in a simple incremental process and may be ported to several languages with little effort. Additionally, the method is able to process more than two million words per minute on a standard computer. Although the method was originally intended for semiautomatic terminological extraction, we believe that it can also be applied in fully automated procedures, making it appropriate for large-scale information extraction. We will start by explaining our main motivation for building this method and we will describe its role in a larger framework, the Corpógrafo. We will then present the process of building the current method, from the first very simple approaches to the current version, pointing out the problems encountered at each step. We will then present results of applying the current version of the extraction method to specific domain corpora in English. Finally, we will present future plans and explain how we are currently in the process of building a small semantic lexicon for helping future large-scale information extraction procedures.

Prediction and precise diagnosis of diseases by data pattern analysis in multiparameter flow cytometry: Melanoma, juvenile asthma, and human immunodeficiency virus infection

Publisher Summary The CLASSIF1 algorithm assures the fast, exhaustive, and unbiased extraction of discriminant biomolecular data pattern from any kind of flow cytometric or other multiparameter data on typical personal computers, that is, out of hundreds or thousands of data columns. CLASSIF1 data classifications require specific and precise measurements according to standardized methods. The intellectual analysis of the selected discriminant data patterns of the classifiers favors scientific hypothesis formation through intuitive result presentation. Thus, large scale information extraction from multiparameter data is available, as an important precondition for individualized disease course predictions in patients. Data analysis may simultaneously concern flow cytometric, humoral biochemistry, or clinical patient data to determine the most discriminant parameter pattern from the totality of the available information. The predictive capacity of the melanoma classifier is similar to the one for survival prediction in colorectal carcinoma patients that is, lower than for the estimation of sepsis outcome in intensive care medicine and for the preoperative prediction of postcardiotomy syndrome in children with open heart surgery.

A generic information extraction architecture for financial applications

The advent of computing has exacerbated the problem of overwhelming information. To manage the deluge of information, information extraction systems can be used to automatically extract relevant information from free-form text for update to databases or for report generation. One of the major challenges to the information extraction is the representation of domain knowledge in the task, that is how to represent the meaning of the input text, the knowledge of the field of application, and the knowledge about the target information to be extracted. We have chosen a directed graph structure, a domain ontology and a frame representation, respectively. We have further developed a generic information extraction (GIE) architecture that combines these knowledge structures for the task of processing. The GIE system is able to extract information from free-form text, further infer and derive new information. It analyzes the input text into a graph structure and subsequently unifies the graph and the ontology for extraction of relevant information, driven by the frame structure during a template filling process. The GIE system has been adopted for use in the message formatting expert system, a large-scale information extraction system for a specific financial application within a major bank in Singapore.