基于词向量相似度的食品安全问答系统设计与实现

针对目前食品安全问答系统准确率不高且无法满足智能化问答要求等问题,基于词向量相似度设计食品安全问答系统。采用深度学习方法构建食品安全领域知识库及词向量模型,结合近义词库提出问句相似度计算方法,将问句与知识库内所有问句进行匹配,返回相似度最高问句对应的答案。实验结果表明,该系统问答准确率达到80%,能满足食品行业用户的日常问答需求。     

基于人工智能技术的医学知识问答系统研究

将人工智能技术应用于医学知识问答系统构建,帮助患者更准确性、更有针对性地认识疾病。建立标准化医学知识问答数据库,并通过医院门诊收集患者问题,利用Siamese LSTM获取问题间的相似度,将非标准化问题纳入问答系统,通过小程序完成问答系统与用户之间的交互,并选取100位门诊患者进行程序回答测试。程序实际测试的回答有效率为82%,无效率为18%,对于内容实质相似的问题回答准确。目前为止,程序运行的准确性处于较高水平,有效率达到82%,可以很好地解答实质相似、问法不同的患者提问。今后,有望通过进一步的数据扩充及程序训练,给患者提供正确的相关医学知识,进而提高就医效率,减缓临床压力,缓和医患关系,提高患者就医满意度。     

Combined online and in‐class pretesting improves exam performance in general psychology

This study examined the effect of distributed questioning on learning and retention in a college lecture course. A total of 48 question pairs were presented over four exams. The 16 question pairs associated with each of the three blocks of the course appeared on the block exams, and all 48 appeared on the final exam. The two questions in each pair were related to each other, so that knowing the answer to one question usually implied knowing the answer to the other. One question in each pair was included in an experimental condition, in which questions were presented online, in class, or both online and in class, before appearing in exams. These conditions were counter‐balanced across the sample. The control questions appeared only in exams. Providing a question online in advance of class, as well as in class, had a significant long‐term effect on the probability of knowing the answers to both experimental and control questions when they appeared in exams. These results demonstrate that coordinated online and in‐class instruction can significantly improve exam performance. These results are consistent with the hypothesis that distributed instruction creates more robust memory traces, rather than the hypothesis that it creates additional memory traces.     

A Method of Providing a More Valid Distribution of School Marks

The cognitive thought processes involved in students’ answers to different kinds of teachers’ questions were investigated using data obtained from a previous study. The dimensions examined were (a) the degree of correspondence between the cognitive level of teachers’ questions and the cognitive level of students’ answers, and (b) the relation of that correspondence to the type of cognitive coding system used, grade level, and clarity of the questions and answers. It was found that the chances are about even that there will be a correspondence between the cognitive level of the question asked and the cognitive level of the response that was elicited. The coding system used, grade level of the students, and clarity of the questions each moderated this effect.     

Responding to children’s answers: questions embedded in the social context of early childhood education

This article presents analysis of question–answer sequences during problem inquiry between a teacher and two children in an early childhood crèche in New Zealand. Conversation analysis is used to reveal which questions the teacher asks, how children answer the questions, and the teacher’s responses to the child’s answers. Although adults’ ‘effective’ questions were identified and promoted in the REPEY study much less attention has been given to how adults respond to children’s answers. It is imperative to investigate the sequences of talk which follow a question in order to establish how teaching and learning is co-constructed in context, one utterance at a time and as a joint project between teacher and child. The findings suggest that task problems and emotional problems are treated in a similar way during problem inquiry, highlighting the complexity of interactions when teachers are providing both emotional care and educational support for young children.     

Multiple‐Choice Tests and Student Understanding: What Is the Connection?

Instructors can use both “multiple‐choice” (MC) and “constructed response” (CR) questions (such as short answer, essay, or problem‐solving questions) to evaluate student understanding of course materials and principles. This article begins by discussing the advantages and concerns of using these alternate test formats and reviews the studies conducted to test the hypothesis (or perhaps better described as the hope) that MC tests, by themselves, perform an adequate job of evaluating student understanding of course materials. Despite research from educational psychology demonstrating the potential for MC tests to measure the same levels of student mastery as CR tests, recent studies in specific educational domains find imperfect relationships between these two performance measures. We suggest that a significant confound in prior experiments has been the treatment of MC questions as homogeneous entities when in fact MC questions may test widely varying levels of student understanding. The primary contribution of the article is a modified research model for CR/MC research based on knowledge‐level analyses of MC test banks and CR question sets from basic computer language programming. The analyses are based on an operationalization of Bloom's Taxonomy of Learning Goals for the domain, which is used to develop a skills‐focused taxonomy of MC questions. However, we propose that their analyses readily generalize to similar teaching domains of interest to decision sciences educators such as modeling and simulation programming.     

Effects of different cue types on the accuracy of primary school teachers' judgments of students' mathematical understanding

To gain insight into how teachers' judgment accuracy can be improved, we investigated effects of cue-type availability. While thinking aloud, 21 teachers judged their fourth grade students' (n = 176) decimal magnitude understanding. Sensitivity (correctly judging what students did understand) did not improve from availability of both answer cues (students' answers to prior practice problems) and student cues (knowledge of students triggered by knowing their names), and was lower when only answer cues were available, compared to only student cues. Specificity (correctly judging what students did not understand) was higher when only answer cues were available, compared to only student cues or both student and answer cues.     

Pace: An Alternative Measure of Student Question Response Time

In some professions, speed and accuracy are as important as acquired requisite knowledge and skills. The availability of computer-based testing now facilitates examination of these two important aspects of student performance. We found that student response times in a conventional non-speeded multiple-choice test, at both the global and individual question levels, closely approximated lognormal distributions. We propose a new measure, pace, which is derived from the survival function of these distributions for analysis of individual person response times. These pace estimates could be used both to rank and compare students; pace also performed maximally compared to other parameterizations in generalizability and dependability studies. While pace was very weakly related to person ability, there was no detectable relationship to question parameters of shift, natural logarithmic mean, or natural logarithmic standard deviation. That is, pace was a person-dependent, question-independent measure. Pace measurements were also successfully used as covariates in models for estimation of person response time to specified questions and person accuracy in response to specified questions. Thus, the analysis of pace can contribute significantly to comprehensive evaluation of student performance in both the speed and ability domains and is a requisite to best practice in testing and assessment.     

Guess Where: The Position of Correct Answers in Multiple-Choice Test Items as a Psychometric Variable

In this article, the authors show that test makers and test takers have a strong and systematic tendency for hiding correct answers—or, respectively, for seeking them—in middle positions. In single, isolated questions, both prefer middle positions to extreme ones in a ratio of up to 3 or 4 to 1. Because test makers routinely, deliberately, and excessively balance the answer key of operational tests, middle bias almost, though not quite, disappears in those keys. Examinees taking real tests also produce answer sequences that are more balanced than their single question tendencies but less balanced than the correct key. In a typical four-choice test, about 55% of erroneous answers are in the two central positions. The authors show that this bias is large enough to have real psychometric consequences, as questions with middle correct answers are easier and less discriminating than questions with extreme correct answers, a fact of which some implications are explored.     

试题命制的科学性：结构要素及其技术规范

试题命制是考试的基础要素和关键环节,体现着考试作为测量手段的科学理性。试题的结构要素分别是刺激情境(线索材料)、设问(作答指令)、分值赋予、答案及评分标准等。试题命制的科学性指试题整体及其结构要素所涉及的情境素材、概念原理和推理论证等准确可靠,没有错误或者歧义。科学性维度的技术规范要求命题时首先明确测试的"构念",即测试要测的到底是被试哪方面的特质,"构念"相对抽象,需要在内容领域和认知能力等方面对"构念"进行分解使之具体化;基于测试"构念",考核重要的而非无关紧要的学习内容;其次要求试题刺激情境自身的合理性与科学性,要求试题的作答指令清晰明确无歧义,要求分值赋予合理,要求参考答案与评分标准精确并与情境素材保持逻辑一致。试题命制还需要关注试题的局部独立性。科学性是任何考试都必须遵循的基础性原则,试题命制自始至终都必须规避因科学性错误带来的测试偏差和风险。     

Individual differences in memory-monitoring accuracy

The reliability of memory-monitoring accuracy was examined in two studies. In Study 1, college students recalled answers to general knowledge questions and rated confidence in the answers. Odd-even reliability of monitoring accuracy was not significantly different from zero; students with at least 20 correct and 20 incorrect answers had higher reliability when the odd and even subsets of questions were matched on recall accuracy. The best reliability was observed when the distribution of confidence ratings was equated across odd and even question subsets, and students had at least 20 confidence judgments at each level. In Study 2, a 100-item mutiple-choice general knowledge test led to odd-even reliability for confidence-recognition accuracy that was fairly high, relative to previous research, but well below desired levels. For FOK-recognition, odd-even reliability was significantly different from zero, but low. These data indicate that current methods allow an individuals' monitoring accuracy to be assessed with low to moderate reliability. Suggestions for further research are offered.     

Adaptively selecting biology questions generated from a semantic network

The paper describes a biology tutoring system with adaptive question selection. Questions were selected for presentation to the student based on their utilities, which were estimated from the chance that the student’s competence would increase if the questions were asked. Competence was represented by the probability of mastery of a set of biology knowledge components. Tasks were represented and selected based on which knowledge components they addressed. Unlike earlier work, where the knowledge components and their relationships to the questions were defined by domain experts, this project demonstrated that the knowledge components, questions and their relationships could all be generated from a semantic network. An experiment found that students using our adaptive question selection had reliably larger learning gains than students who received questions in a mal-adaptive order.     

“How do You Know this Answer?” – Children’s Use of Text Data and General Knowledge in Story Comprehension

This study investigates how, and to what extent, young readers (7–8 year-olds) use text information or their prior knowledge when answering comprehension questions about narrative texts. The children were asked to explain how they found out their responses by answering the following question: “how do you know this answer?” Their answers and justifications were analysed both qualitatively and quantitatively. The text proved to be the main source of information for these readers. However, sometimes the children seemed to ignore the text and over rely on their prior knowledge to answer the questions. The procedure of asking children to justify their answers was shown to be a good way of specifying more precisely some of their problems in text comprehension. It also seemed to encourage them to look back at the text and review their responses and, as such, it could be considered a useful tool to improve children’s reading comprehension.     

Coordinating Procedural and Conceptual Knowledge to Make Sense of Word Equations: Understanding the complexity of a ‘simple’ completion task at the learner’s resolution

The present paper discusses the conceptual demands of an apparently straightforward task set to secondary‐level students—completing chemical word equations with a single omitted term. Chemical equations are of considerable importance in chemistry, and school students are expected to learn to be able to write and interpret them. However, it is recognised that many students find them challenging. The present paper explores students’ accounts of their attempts to identify the missing terms, to illuminate why working with chemical word equations is so challenging from the learner’s perspective. Three hundred secondary‐age students responded to a five‐item exercise based on chemicals and types of reactions commonly met at school level. For each item they were asked to identify the missing term in a word equation, and to explain their answers. This provided a database containing more than 1,000 student accounts of their rationales. Analysis of the data led to the identification of seven main classes of strategy used to answer the questions. Most approaches required the coordination of chemical knowledge at several different levels for a successful outcome; and there was much evidence both for correct answers based on flawed chemical thinking, and appropriate chemical thinking being insufficient to lead to the correct answer. It is suggested that the model reported here should be tested by more in‐depth methods, but could help chemistry teachers appreciate learners’ difficulties and offer them explicit support in selection and application of strategies when working with chemical equations.     

Extended matching sets questions for online numeracy assessments: a case study

Extended matching sets questions (EMSQs) are a form of multiple‐choice question (MCQ) consisting of a stem (the question or scenario) with an extended number of possible answers. Although there is no consensus on their absolute format, for the purpose of this paper a multiple‐choice question with ten or more alternative answers is considered to be an EMSQ. Faced with the limitations imposed by virtual learning environment software, I have conducted a case study into the use of the EMSQ format in online assessment of numerical and statistical ability which shows that properly constructed questions of this type can play a valuable role in assessment of numeracy. The extended format was found to work well on screen and resulted in an increase in both student marks and student satisfaction when compared with other answer input formats. This case study indicates that the EMSQ format has much more widespread applicability for online assessment that its traditional uses.     

The Effect of Worked Examples When Learning to Write Essays in English Literature

Worked examples, commonly used in technical domains, are rarely used in language areas such as English literature. In 3 experiments, Korean university students for whom English was a foreign language received worked examples intended to facilitate problem solving in the ill-structured domain of English literature. During the learning phase, half of the students were presented conventional essay questions that they were asked to answer. The other half of the students were presented the same questions along with model answers that they were asked to study, followed by similar questions that they had to answer themselves. All students then were asked to answer retention, near and far transfer tests. Relatively more knowledgeable students were assigned to Experiment 1 than to Experiment 2, who, in turn, were more knowledgeable than were the students in Experiment 3. Results indicated that the effectiveness of worked examples increased with decreasing student knowledge.     

Repeated reading to enhance fluency: Old approaches and new directions

As phoneme awareness deficits and resulting decoding weaknesses are increasingly addressed, there is heightened awareness of the role of fluency in reading. This paper reviews the history of fluency training, discusses the theoretical bases of such training, and summarizes the current knowledge about the efficacy of training procedures. We focus on Repeated Reading (RR), the most familiar and researched approach to fluency training. Outcome data on Repeated Reading, presented in the form of questions, is meant to answer practitioners’ questions about implementation and efficacy and to provide a starting point for researchers interested in the topic. Although some answers are straightforward, others indicate the subtleties involved in answering the broad question, “Does Repeated Reading work?” In addition to a list of practical suggestions based on Repeated Readings findings, three new approaches to fluency training are introduced.     

What are Middle School Students Talking About During Clicker Questions? Characterizing Small-Group Conversations Mediated by Classroom Response Systems

There is a growing interest in using classroom response systems or clickers in science classrooms at both the university and K-12 levels. Typically, when instructors use this technology, students are asked to answer and discuss clicker questions with their peers. The existing literature on using clickers at the K-12 level has largely focused on the efficacy of clicker implementation, with few studies investigating collaboration and discourse among students. To expand on this work, we investigated the question: Does clicker use promote productive peer discussion among middle school science students? Specifically, we collected data from middle school students in a physical science course. Students were asked to answer a clicker question individually, discuss the question with their peers, answer the same question again, and then subsequently answer a new matched-pair question individually. We audio recorded the peer conversations to characterize the nature of the student discourse. To analyze these conversations, we used a grounded analysis approach and drew on literature about collaborative knowledge co-construction. The analysis of the conversations revealed that middle school students talked about science content and collaboratively discussed ideas. Furthermore, the majority of conversations, both ones that positively and negatively impacted student performance, contained evidence of collaborative knowledge co-construction.