Types Of Learning Activities Research Articles

Adversarial active learning for the identification of medical concepts and annotation inconsistency.

Named entity recognition (NER) is a principal task in the biomedical field and deep learning-based algorithms have been widely applied to biomedical NER. However, all of these methods that are applied to biomedical corpora use only annotated samples to maximize their performances. Thus, (1) large numbers of unannotated samples are relinquished and their values are overlooked. (2) Compared with other types of active learning (AL) algorithms, generative adversarial learning (GAN)-based AL methods have developed slowly. Furthermore, current diversity-based AL methods only compute similarities between a pair of sentences and cannot evaluate distribution similarities between groups of sentences. Annotation inconsistency is one of the significant challenges in the biomedical annotation field. Most existing methods for addressing this challenge are statistics-based or rule-based methods. (3) They require sufficient expert knowledge and complex designs. To address challenges (1), (2), and (3) simultaneously, we propose innovative algorithms. GAN is introduced in this paper, and we propose the GAN-bidirectional long short-term memory-conditional random field (GAN-BiLSTM-CRF) and the GAN-bidirectional encoder representations from transformers-conditional random field (GAN-BERT-CRF) models, which can be considered an NER model, an AL model, and a model identifying error labels. BiLSTM-CRF or BERT-CRF is defined as the generator and a convolutional neural network (CNN)-based network is considered the discriminator. (1) The generator employs unannotated samples in addition to annotated samples to maximize NER performance. (2) The outputs of the CRF layer and the discriminator are used to select unlabeled samples for the AL task. (3) The discriminator discriminates the distribution of error labels from that of correct labels, identify error labels, and address the annotation inconsistency challenge. The corpus from the 2010 i2b2/VA NLP challenge and the Chinese CCKS-2017 Task 2 dataset are adopted for experiments. Compared to the baseline BiLSTM-CRF and BERT-CRF, the GAN-BiLSTM-CRF and GAN-BERT-CRF models achieved significant improvements on the precision, recall, and F1 scores in terms of NER performance. Learning curves in AL experiments show the comparative results of the proposed models. Furthermore, the trained discriminator can identify samples with incorrect medical labels in both simulation and real-word experimental environments. The idea of introducing GAN contributes significant results in terms of NER, active learning, and the ability to identify incorrect annotated samples. The benefits of GAN will be further studied.

PENINGKATAN HASIL BELAJAR IPA MATERI TEMPAT HIDUP MAKHLUK HIDUP MELALUI PEMBELAJARAN AKTIF TIPE CARD SORT DI KELAS II SDN KETAWANG KARAY I KEC. GANDING KAB. SUMENEP SEMESTER 1 TAHUN PELAJARAN 2018/2019

The background of the research was that the learning outcomes of science students in class II were still low, namely under KKM 70, out of 14 students only 5 students (36%) completed while 9 students (64%) had not yet finished. This research was conducted to improve learning outcomes of students in science learning material where living things in class II Ketawang Karay I SDN Ganding Subdistrict Sumenep Regency 1st Semester 2018/2019 through the Card Sort Type Active Learning Strategy. This research was conducted on second grade students of Ketawang Karay I Elementary School in Ganding Subdistrict, Sumenep Regency during science learning with living things using living things using the Card Sort Type Active Learning Strategy. The action plan in this study will be carried out in 2 cycles and is planned to be carried out with steps of planning, action, observation, and reflection. The results showed that in the pre-cycle, the condition of student learning outcomes had not been satisfactory, there were only 36% or 5 new students completed in learning and the remaining 9 people or about 64% of students who had not completed the classical average of 47. So that repairs are continued by explaining and using the learning steps with an active learning strategy type card sort. In the first cycle there was an increase in student learning outcomes that was quite good, where there were 9 students or about 64% who completed the learning and the remaining 5 students or about 36% of students who have not completed the classical average of 73. Then in the second cycle, the learning outcomes of students experienced a very significant increase in which 14 students all completed or 100% in learning with a classical average of 87. This shows that using the Card Sort type active learning strategies can improve learning outcomes students in science learning in class II Ketawang Karay I Elementary School Ganding District Sumenep Regency.

Short-Term and Long-Term Effects of POGIL in a Large-Enrollment General Chemistry Course

Process oriented guided inquiry learning (POGIL) is a specific type of active learning centered on a learning cycle where students first explore a concept through scientific models, followed by a concept invention, and finally a concept application phase. In spite of POGIL’s research-based design and the many studies showing it increases learning outcomes, there is still a critical gap in the knowledge behind the effect of POGIL in attitudinal factors and the mechanisms behind it. The current study seeks to build an understanding of the mechanistic ways in which POGIL works and its effect on students’ attitudes and learning. The sample consisted of students who enrolled in General Chemistry I in the Fall semester across 7 sections (classrooms) at an R1 (large research) university in the eastern part of the US. Four sections used POGIL instruction (N = 809) while the other three used Traditional teaching methods (N = 543). Statistical models using multilevel statistics show students in the POGIL condition had higher chemistry identity, competency beliefs, and chemistry grades. Furthermore, performance in General Chemistry I appeared to be a core mediator of all the observed differences in General Chemistry II where students in the POGIL condition still performed better and had higher chemistry-related attitudes.

Do Early Simulations Work? Simulations in Gateway Political Science Courses at Community Colleges

Simulation methodology has moved toward using different types of real-world scenario-based learning constructs to improve applied understanding of political science concepts and theories. This type of active learning has become popular in the upper divisions of undergraduate and graduate political science. This reflection addresses the variance in approach to conducting simulations in federal and state general education courses at an open-enrollment community college. Defining the differences conducting these active learning assessments between two and four-year institutions enhances pedagogy at the two-year level and betters connection between these institutions and four-year counterparts. Students are assessed before the simulation and after to gauge progress and competency. At the conclusion, we offer views on how to design and implement simulations among a student population that is diverse academically and demographically.

Agile in Higher Education - A collaborative research project report

This case study shows how complex future-oriented skills can be developed in higher education. Demonstrating a student collaboration project process, I look at adopting an agile approach in teaching and explain how a mix of methods and tools used to facilitate this type of active learning. Students were encouraged to explore different learning strategies, benefiting from the various forms of cooperation pair- and teamwork. Subsequently they individually published their own studies. Students engaged in work during the course and enjoyed experiencing different learning strategies.

Comparison of active learning models in schools in increasing the cognitive value of students

The term learning model is distinguished from the term learning strategy, method learning or learning procedure. The term learning model has broader meaning than a strategy, approach, method or procedure. Learning models are patterns of interaction between students, teachers, and material learning that includes strategies, approaches, methods and techniques learning. Elaborate that learning strategies are tactics or tips planned by the teacher related to the whole preparation of learning can be achieved effectively. While the approach is the way taken by the teacher in the implementation of learning so that the active ideas presented can be adapted to then be understood by students. There are two types of active learning, namely: a methodological approach and material approach. Methodological approach concerns the way students Adapting the active ideas presented into their cognitive structures that are in line in the way taken by the teacher in presenting learning materials the. Examples of methodological approaches are intuitive, analytic, synthetic, spiral, inductive, deductive, thematic, realistic and heuristic. The result of this research is the material approach able to present an active concept through other active concepts already owned by students. The example presents the concept of addition with using the number line or set approach.

Penerapan Model Pembelajaran Quiz Team Untuk Meningkatkan Hasil Belajar Ekonomi Peserta Didik Kelas XI IPS 2 SMAN 8 Kediri Di Semester Ganjil Tahun Pelajaran 2018/2019

Economic Learning in SMAN 8 Kediri specifically in class XI IPS 2, so far it is still being agreed by the teacher. The student's questioning activity is still low. Mostly, students do not understand the concepts and the interrelationships between economic concepts. The researcher chose the Quiz Team learning method. Researchers chose the Quiz learning model because learning with the opinion of Silberman. The Quiz Team is a technique to increase participants' responsibility for what they do in a fun and non-scary way. This active learning type of Quiz Team is a technique to increase the responsibility of students towards what they improve through fun and non-frightening ways. In this study, researchers used the action research design model Kemmis and Mc. In the research design there are four stages of research namely planning (planning), action (implementation), observation (observing) and reflection (reflection). In cycle 1, researchers divided 3 large groups, A, B, and C, each group consisting of 11 students. The researcher chooses three topics which will be presented in three segments. The results of the study increased the number of students who achieved minimal completeness in each cycle. In cycle 2, researchers divided the group into smaller ones, in cycle II there were 6 groups, A, B, C, D, E and F. In cycle II, researchers divided into 6 national income calculation concepts namely the concept: Gross Domestic Product (GDP), Gross National Product (GNP), Net National Product (NNP), Net National Income (NNI), Personal Income (PI), Disposible Income (DI). Indicators Success in improving student learning outcomes in this study was seen from 80% of students completing KKM (with a value of 77). The results of the study increased the number of students who achieved minimal completeness in each cycle. At the 42% level (14) students complete KKM. In cycle 1 Approximately 61% (20) of students have completed KKM. In cycle 2, around 82% or 26 students completed KKM. The data has increased from pre-cycle to cycle 2.
 Keywords: Learning Outcomes, Class XI IPS 2, National Revenue, Quiz Team.

UPAYA MENINGKATKAN HASIL BELAJAR IPA MELALUI METODE PEMBELAJARAN ACTIVE LEARNING TIPE QUIZ TEAM PADA SISWA KELAS V SDN 12 SUNGAI LIMAU

ABSTRACT The success of learning objectives is determined by many factors including the teacher's factor in carrying out the teaching and learning process, because the teacher can directly influence, foster and improve students' intelligence and skills. To overcome the above problems and to achieve maximum educational goals, the role of the teacher is very important and it is expected that the teacher has a good teaching method and is able to choose the right learning method and in accordance with the concepts of the subjects to be delivered. The problems to be examined in this study are: Is using the Learning Method Active Learning Type Quiz Team able to improve the science learning outcomes of Class V students at SDN 12 Sungai Limau? The purpose of this action research is to improve the learning outcomes of science through the use of the Active Learning Type Quiz Team Learning Method for students VI.C SDN 12 Sungai Limau. This study used classroom action research as much as two rounds. Each round consists of four stages, namely: planning, implementation, observation, reflection. The target of this study was the fifth grade students of SDN 12 Sungai Limau. The data obtained are quantitative and qualitative data. From the results of the analysis it can be seen that the students' understanding of the material presented by the teacher is more stable (learning completeness increases from cycle I, and II, ie 61.76%, and 88.24% respectively). students have achieved classically. The conclusion of this study is that the Learning Method of Active Learning Type Quiz Team has a positive influence, which can improve student learning outcomes as indicated by the average student answer stating that students are interested and interested in the Active Learning Type Quiz Learning Method Team so that they become motivated to learn

Biomedical Image Segmentation via Representative Annotation

Deep learning has been applied successfully to many biomedical image segmentation tasks. However, due to the diversity and complexity of biomedical image data, manual annotation for training common deep learning models is very timeconsuming and labor-intensive, especially because normally only biomedical experts can annotate image data well. Human experts are often involved in a long and iterative process of annotation, as in active learning type annotation schemes. In this paper, we propose representative annotation (RA), a new deep learning framework for reducing annotation effort in biomedical image segmentation. RA uses unsupervised networks for feature extraction and selects representative image patches for annotation in the latent space of learned feature descriptors, which implicitly characterizes the underlying data while minimizing redundancy. A fully convolutional network (FCN) is then trained using the annotated selected image patches for image segmentation. Our RA scheme offers three compelling advantages: (1) It leverages the ability of deep neural networks to learn better representations of image data; (2) it performs one-shot selection for manual annotation and frees annotators from the iterative process of common active learning based annotation schemes; (3) it can be deployed to 3D images with simple extensions. We evaluate our RA approach using three datasets (two 2D and one 3D) and show our framework yields competitive segmentation results comparing with state-of-the-art methods.

PENINGKATAN MINAT BELAJAR SISWA DALAM MATA PELAJARAN MATEMATIKA MELALUI STRATEGI PEMBELAJARAN AKTIF TIPE INDEX CARD MATCHDI KELAS IV SDN 06 LUBUK JANTAN

The interest in learning is one of the important factors for students' success in learning. However, interest in learning fourth grade students of SDN 06 Lubuk Jantan in mathematics subjects was found to be still low. The purpose of this study was to describe and obtain information about increasing students' interest in learning in mathematics through the Index Card Type Active Learning Strategy Match in Class IV SDN 06 Lubuk Jantan. This research is a classroom action research. The research procedures in this study include planning, action, observation and reflection. This study consisted of two cycles with four meetings. The research subjects consisted of 23 fourth grade students of SDN 06 Lubuk Jantan. The research data was collected using a test sheet. Data were analyzed using percentages. Based on the results of research and discussion that have been put forward, it can be concluded that the Active Learning Strategy Type Index Card Match can increase students' interest in learning in mathematics subjects in Class IV SDN 06 Lubuk Jantan. The interest in student learning in the first cycle was 69.67 increasing to 84.36.

MENINGKATKAN HASIL BELAJAR PESERTA DIDIK MENGGUNAKAN STRATEGI PEMBELAJARAN AKTIF TIPE GUIDED NOTE TALKING PADA MATERI PERUSAHAAN DAGANG DI KELAS XII IPS 1 SMA NEGERI 2 PEKANBARU

Teaching the learning materials of trading companies the author found that only a few students were active in expressing their opinions, as well as asking questions. Students are even passive, reluctant to ask the teacher if they do not understand the material being taught, unfocused and sleepy when the learning takes place so that the learning outcomes obtained by students on the subject of trading companies are on average 78. Students who complete only about 10 students and 24 students did not complete, while the minimum completeness criteria (KKM) were determined 82. To overcome the problem of lack of attention and activeness of students, the authors tried to apply Using the Guided Note Talking Type Active Learning Strategy to the Material of Trading Companies. This type of research is classroom action research. As for the object of this research are students of class XII IPS 1 Pekanbaru 2 High School. The method of data collection in this study used tests, observation and documentation. The research instrument is a formative test. The results showed an increase in accounting learning outcomes of students of class XII IPS 1 Pekanbaru 2 High School, amounting to 34 students. The application of active Guided Note Talking learning strategies can improve student learning activities. Where the teaching and learning process can be carried out pleasantly so that students can be motivated to learn. The application of active Guided Note Talking learning strategies can improve student learning outcomes with a percentage of the first cycle of 50%, and a second cycle of 85.29%, and the average learning value of students in cycle I 84.38 and cycle II 90.5.Keywords: Learning Outcomes, Economics, Accounting, Talking Guided Note Type Active Learning Strategies

PENGARUH PENERAPAN MODEL PEMBELAJARAN ACTIVE LEARNING TIPE QUIZ TEAM TERHADAP KEMAMPUAN BERPIKIR KRITIS DENGAN VARIABEL MODERATOR MOTIVASI BELAJAR PESERTA DIDIK

This research was motivated by observations and interviews stating that the critical thinking skills by taking into account the motivation of learners is still low. The purpose of this study was to determine whether there is influence from learning model Active Learning Type Quiz Team about the critical thinking skills of learners, whether there is influence learning motivation of students towards critical thinking skills of learners, and whether there is an interaction effect between teaching model active learning type of quiz team to learning motivation towards critical thinking skills of learners. The method used is experimental method with factorial design. This caused the hypothesis testing is done by using two way analisys of variance in order to avoid noise or a possibility that states there is an effect due to mixing of the data analysis. Meanwhile, the population in this study were all students of class X as many as 309 students and the sample is class X-2 as the experimental class and X-3 as the control class, each of which consists of 39 students. Based on the results of the study showed: There is the influence of learning model Active Learning type Quiz Team of the critical thinking skills of learners with sig. 0.000 <0.05. There is the influence of learning motivation towards the critical thinking skills of learners with sig. 0.003 <0.05. There is an interaction effect between the active learning teaching model type quiz team with the learning motivation towards the critical thinking skills of learners with sig. 0.000 <0.05. To reach the maximum critical thinking skills, it is recommended that the teachers continue to practice, formulate and present problems that exist into the learning process as a stimulus for students to get them used to find the solution of a problem that occurs in daily life.

Peningkatan Kemampuan Membaca Bahasa Jerman Melalui Metode Active Learning Tipe Reading Guide

The purpose of this study was to obtain data on planning, process, and results of the reading guide type active learning method in students' German reading skills. The data of this study are qualitative data and quantitative data. Qualitative data is obtained through observations and quantitative data obtained through the ability to read the cycle I and cycle II. Data were analyzed using percentage techniques. The results of this study indicate that the reading ability with the percentage value obtained by students in the first cycle reached 66.78% and the second cycle reached 82.14%. These results indicate that the active learning type of reading guide can improve students' reading skills

UPAYA MENINGKATKAN HASIL BELAJAR SISWA KOMPETENSI MINYAK BUMI DAN GAS ALAM MELALUI PENERAPAN PEMBELAJARAN COLLEGE BALL

learning outcomes of class X.1 students at SMAN 1 Ciwaringin have not reached the standard of competence (KKM) set at school yet. The problem of chemistry learning is that the teacher in the learning process still uses conventional methods with teacher-centered lectures, not using student-centered learning models yet, so that students' thinking skills are less optimized to be more critical and more effective through application of college ball type active learning. This Classroom Action Research design consists of two cycles, each cycle consisting of four stages, namely planning, observing and reflecting. Research subjects are teachers and students of class X.1. Data collection techniques use observation and test results. The results of the study showed that before the study of chemistry learning outcomes, students obtained an average value of 68.10 with classical completeness of 37.43%. After conducting research in the first cycle obtained an average value of 78.20 with classical completeness of 83.40. In the second cycle the average value and completeness of classical learning has reached the indicator of minimum completeness success, namely the value of completeness learning ? 76 with classical completeness 83%. The conclusion of this study is the application of college ball type active learning can improve student activity, teacher performance and student learning outcomes in teaching and learning process.
 
 Keyword : Learning outcomes, collage ball learning, hasil belajar

PENGARUH PENERAPAN MODEL PEMBELAJARAN ACTIVE LEARNING TIPE “GIVING QUESTION AND GETTING ANSWERS“ TERHADAP HASIL BELAJAR DASAR-DASAR KELISTRIKAN SISWA DI KELAS X TEKNIK AUDIO VIDEO SMK NEGERI 1 BUKITTINGGI

The problem in this research is the fact that found in SMK N 1 Bukittinggi, there are many students of class X Audio Video Engineering is result study of learning outcomes below the minimum standard (KKM) in the subject Fundamentals of electricity at school set is set 75. Total students that are under KKM 60.35% and students those are above KKM 39.65%. Many factors that affect it, those are internal factors and external factors, as well as the learning model. This study aims to prove the existence of differences in learning outcomes using learning Active Learning Model Type "Giving Question and Getting Answers " with direct instructional model (MPL). This study uses a Quasi Experimental Design, this study population is a class X student of SMK N 1 Bukittinggi TAV academic year 2013/2014. Sample technique done by randomly. Experiment class is a class that is treated by using the Active Learning Model Application Effects of Learning Type "Giving Question and Getting Answers " and that becames the control group is a class that uses MPL. Data of achievement test in the form of 20 questions about the essay. Data obtained from the analysis of the data manually to test for normality, homogeneity, hypothesis testing, research test results obtained from the average value of students learning model that uses Active Learning Type "Giving Questions and Getting Answers" is 79.53 while students are using the MPL amounting to 60.78. The result of the hypothesis by using the formula manually found that tcount 9.886> 2.35 ttable altenative hypothesis (H0) is rejected and (H1) is accepted. That means, significantly the average experimental class learning outcomes greater than the average of the control class learning outcomes. Keywords: Learning Active Learning model type "Question Giving and Getting Answers", MPL, Control and Experiment

PENGARUH PENERAPAN MODEL PEMBELAJARAN ACTIVE LEARNING TIPE“DYNAMIC TRUE OR FALSE“ TERHADAP HASIL BELAJAR SISWA KELAS X DALAM MATA PELAJARAN MENERAPKAN DASAR-DASAR ELEKTRONIKA PROGRAM KEAHLIAN TEKNIK AUDIO VIDEO DI SMKN 1 BUKITTINGGI

This research is motivated by the problem of low learning outcomes in subjects apply the basics of electronics class X Audio Video Engineering Department (TAV) Vocational School (SMK) Bukittinggi School 1, an experimental study using two class is a class that acts as a control and experiment. This study aimed to see whether there is influence learning outcomes after implementing the learning model of Active Learning type Dynamic true or False. The problem in this study is the fact that the field is found in SMKN 1 Bukittinggi, there are many students of class X Audio Video Engineering who scored below the standart of learning outcome minimum completeness criteria in subjects applying the basic fundamentals of electronics set a school that is 75.The population of this research is class X TAV SMKN 1 Bukittinggi TA 2013/2014. The data collected from the test results of a study about the essay as many as 20 items. Data were analyzed using Microsoft Excel to test normality, test homogenity, and test hypothesis. Normality test all values obtained χ2count is smaller than the value χ2 table. Homogenitas test obtained χ2 count< χ2table , atau -1,03 < 14,067. Hypothesis test tcount> ttable yaitu (5,613 >1.697). From the test results of research obtained an average value of student which uses the learning model of Active Learning Type Dynamic true or False is 79,53 while the control group of student who use learning directly lower the 60,78. The results of the hypothesis by using Microsoft Excel in t count found that -1,03 < t table 14,067, so the alternative hypothesis (H1) is accepted or rejected the null hypothesis (H0). This means that on average significantly experimental class learning outcomes greater than the average of the control class learning outcomes. So it can be concluded that the application of Dynamic True or False effect on learning outcomes.Keywords : Active Learning, Dynamic True or False

Physical models can provide superior learning opportunities beyond the benefits of active engagements.

The essence of molecular biology education lies in understanding of gene expression, with subtopics including the central dogma processes, such as transcription and translation. While these concepts are core to the discipline, they are also notoriously difficult for students to learn, probably because they cannot be directly observed. While nearly all active learning strategies have been shown to improve learning compared with passive lectures, little has been done to compare different types of active learning. We hypothesized that physical models of central dogma processes would be especially helpful for learning, because they provide a resource that students can see, touch, and manipulate while trying to build their knowledge. For students enrolled in an entirely active‐learning‐based Cell & Molecular Biology course, we examined whether model‐based activities were more effective than non‐model based activities. To test their understanding at the beginning and end of the semester, we employed the multiple‐select Central Dogma Concept Inventory (CDCI). Each student acted as their own control, as all students engaged in all lessons yet some questions related to model‐based activities and some related to clicker questions, group problem‐solving, and other non‐model‐based activities. While all students demonstrated learning gains on both types of question, they showed much higher learning gains on model‐based questions. Examining their selected answers in detail showed that while higher performing students were prompted to refine their already‐good mental models to be even better, lower performing students were able to construct new knowledge that was much more consistent with an expert's understanding. © 2018 The Authors. Biochemistry and Molecular Biology Education published by Wiley Periodicals, Inc. on behalf of International Union of Biochemistry and Molecular Biology., 46(5):435–444, 2018.

Best practices in active and student-centered learning in physiology classes.

This review article includes our analysis of the literature and our own experiences in using various types of active learning as best practices for evidence-based teaching in physiology. We have evaluated what physiology students should be expected to learn and what are specific challenges to enhancing their learning of physiology principles. We also consider how the instructor should design his or her teaching to improve buy-in from both students and other faculty members. We include a discussion of how the readers can evaluate their teaching approaches for their successes in enhancing student learning of physiology. Thus we have addressed pedagogical improvements specific to student learning of physiology, with additional suggestions from cognitive psychology approaches that can improve physiology teaching and learning.

Students’ interest towards STEM: a longitudinal study

ABSTRACTBackground: Previous studies have reported the impact of integrated science, technology, engineering and mathematics (STEM)-based programmes on students’ interest towards STEM, however, only few have tracked the longitudinal impact of such programmes across the years.Purpose: This study examined the long-term impact of the Bitara-STEM: Science of Smart Communities Program (Bitara-STEM) on middle school students’ interest towards STEM.Program description: The program exposed students with integrated STEM education through a hands-on project based learning experiences, in an informal learning environment involving the application of the five phases of an engineering design process.Sample: This study involved 121 students who participated in the 2014 cohort of the Bitara-STEM program.Design and methods: Data were collected through a longitudinal survey and interviews. Interviews were conducted with sub sample of students to gauge the reasoning behind the decline, sustain or increment of interest towards STEM.Results: Findings revealed that, 2 years after leaving the programme, the levels of interest towards STEM careers were sustained but interest towards STEM subjects was not. The data from interviews revealed that the possible reasons for the decrease in interest towards STEM subjects could be due to the quality of teaching and learning that they experienced in the classroom.Conclusions: Thus, in order to sustain interest towards STEM subjects, students should be involved with Bitara-STEM type active learning strategies more regularly. Teaching and learning approaches in schools need to reflect the more active learning approaches, involving the integration of STEM subjects, as exemplified in this programme.

Evaluating active learning methods for annotating semantic predications.

ObjectivesThis study evaluated and compared a variety of active learning strategies, including a novel strategy we proposed, as applied to the task of filtering incorrect semantic predications in SemMedDB.Materials and methodsWe evaluated 8 active learning strategies covering 3 types—uncertainty, representative, and combined—on 2 datasets of 6,000 total semantic predications from SemMedDB covering the domains of substance interactions and clinical medicine, respectively. We also designed a novel combined strategy called dynamic β that does not use hand-tuned hyperparameters. Each strategy was assessed by the Area under the Learning Curve (ALC) and the number of training examples required to achieve a target Area Under the ROC curve. We also visualized and compared the query patterns of the query strategies.ResultsAll types of active learning (AL) methods beat the baseline on both datasets. Combined strategies outperformed all other methods in terms of ALC, outperforming the baseline by over 0.05 ALC for both datasets and reducing 58% annotation efforts in the best case. While representative strategies performed well, their performance was matched or outperformed by the combined methods. Our proposed AL method dynamic β shows promising ability to achieve near-optimal performance across 2 datasets.DiscussionOur visual analysis of query patterns indicates that strategies which efficiently obtain a representative subsample perform better on this task.ConclusionActive learning is shown to be effective at reducing annotation costs for filtering incorrect semantic predications from SemMedDB. Our proposed AL method demonstrated promising performance.