JP2007310425A - Learning process optimization system - Google Patents

Abstract

<P>PROBLEM TO BE SOLVED: To integrate and reuse intellectual resources beyond time and space and realize optimum distribution of intellectual resources by quantifying the effect of education or learning in an educational scene and most adequately providing an opportunity of re-education. <P>SOLUTION: The effect of education to such a present condition is measured, and academic ability is analyzed as objective and quantitative data for improvement in academic ability. For this purpose, variations between a standard academic ability model, which an instructor sets as an expected value, and the comprehension degrees of learners are computed, and teaching materials, which are most suitable for academic ability-poor parts, are automatically computed at meta-level and then provided. From a relative comparison in academic ability of a learner with other learners from a comprehension degree or a result of grading to an exercise, or from an absolute comparison in progress of a learner from a learning history of the learner himself or herself, or by execution of questionnaires on learning environments, academic ability-affecting factors are quantitatively analyzed in a large scale and can be used as data which propose an optimum solution. <P>COPYRIGHT: (C)2008,JPO&INPIT

Description

  It is related to knowledge base using multimedia database.

    This is a method of measuring the distance between information using a vector space model, and is based on a mathematical semantic model that measures the closeness of information.

Integrate and reuse intellectual resources across time and space to achieve optimal allocation of intellectual resources. The effectiveness of education in the field of education is not optimally demonstrated to learners who receive it. Measuring the effectiveness of training conducted at companies, etc., and providing retraining opportunities to learners who did not achieve the benefits, from a management perspective, if the cost-effectiveness cannot be measured, It becomes difficult to provide. In particular, exams conducted at schools and the like are conducted not for the purpose of improving academic ability but for examining academic ability and comparing with other learners. As a result, with the exception of some limited and excellent learners, the analysis of their own academic ability cannot be carried out constructively, the optimal effort of course is not made, and data directly linked to the improvement of academic ability can be used. Not. As a result, there seem to be many students who have not improved their academic ability. Quantify the effectiveness of education or learning at each site to provide the best opportunity for re-education. In particular, the academic achievement tests that have been conducted so far are not intended to improve academic ability, but rather to examine academic ability and compare it with other learners. It has been implemented. As a result, with the exception of some limited and excellent learners, the analysis of their own academic ability cannot be carried out constructively, the optimal effort of course is not made, and data directly linked to the improvement of academic ability can be used. Not. As a result, there seem to be many students who have not improved their academic ability.

    Analyzing academic ability as objective quantitative data for improving academic ability, calculating the variation with the standard academic ability model as the expected value set by the instructor, and optimal for the lack of academic ability Automatic learning materials are automatically calculated and provided at the meta level. Also, based on comprehension and exercise scoring results, scholastic ability is compared with other learners, absolute comparison of progress from learner's own learning history, etc., or questionnaire with learning environment By doing so, factor analysis that affects academic ability can be quantitatively realized on a large scale and used as data for proposing optimal solutions.

(1) Measuring the learner's phase in the knowledge space The invisible knowledge can be analyzed by structurally organizing, subdividing, and writing it. For a specific knowledge domain, it is possible to recognize the top / bottom / hierarchical relationship of the concept and to quantitatively express the temporal order relationship. Each organized knowledge can generate a vector as an element, and can measure the learner's phase on the vector space.

(2) Generally, there are a plurality of processes for solving a certain problem of measuring factors of superiority or inferiority in academic ability. Select the knowledge necessary for problem solving and combine them systematically to solve the problem. Combining knowledge for problem solving makes it very difficult to know how people are processing. However, it is possible to know the selected knowledge and its logical combination.
For example, if you consider a simple logical expression (A∧B) VC⇒D, if you combine A and B or know C, you can draw the conclusion that D. In other words, if you don't know A and B or C at least, you can't get a conclusion of D. Conversely, we can at least analyze that the person who obtained D knows A and B or C. I think that factors can be analyzed by expressing the standard process of problem solving with logical expressions and evaluating the knowledge groups contained in them.
Furthermore, suppose C finds multiple knowledge relationships.
C = {c ₁ , c ₂ , c ₃ }
If the knowledge to be expressed is expressed, it can be further hierarchized.
c ₁ = {c ₁₀ , c ₁₁ , c ₁₂ , c ₁₃ , c ₁₄ }
By evaluating and analyzing these hierarchical knowledge groups, it is possible to comprehensively evaluate the overall picture of knowledge, analyze factors that are superior or inferior in academic ability, and logically determine whether or not there is knowledge. Measure causality with the ability to recognize relationships.

(3) Measuring the causal relevance of knowledge subspace The difference between experts and non-experts in a certain knowledge area is considered to be in the ability to recognize that the relevance of knowledge can be identified and linked. When logical association is possible, it recognizes whether an event has a one-to-one relationship with knowledge, a one-to-many relationship, or even a many-to-many relationship. it can. Fragmented knowledge does not allow a logically wide and deep understanding of the problem. It can be said that the degree of recognition in the knowledge domain can be analyzed by the degree of ability to recognize this knowledge relevance. Therefore, the analysis of superiority or inferiority of academic ability in the knowledge domain seems to depend not only on whether or not there is knowledge but also on recognizing the relevance of knowledge. If these abilities can be analyzed, the lack of logical connection and relevance will be the factor that determines the superiority or inferiority of academic ability. If this lack can be analyzed, the right material can be selected.

Knowledge quantification techniques Knowledge areas can be subdivided just as academic fields are subdivided and academic societies are subdivided. Considering that a certain academic area is synonymous with the knowledge area, the knowledge is divided for the purpose of organizing and subdividing the knowledge and expressing the connection and the relation quantitatively. For example, a certain knowledge area is classified as follows.
By such graphing, the decomposed knowledge is symbolized and a vector is generated as follows.
M = {m ₁ , m ₁₁ , m ₁₂ , m ₁₁₁ , m ₁₁₂ , m ₁₁₃ }
From this vector, the hierarchical relationship, the adjacency relationship, or the order relationship can be quantified.

Procedure 1 Subdivide the attributes of knowledge for quantifying the hierarchical relationship, and clarify the hierarchical structure that inherits the attributes.
Knowledge is an abstract subject. However, knowledge can be shared as explicit knowledge by expressing it visually. By disassembling the vague knowledge as a whole and expressing it visually, it can be symbolized into the smallest recognizable unit. While maintaining the bidirectionality and relevance of knowledge specialization and generalization, attributes can be formalized according to the level of abstraction of knowledge, and the inheritance relationship of attributes can be expressed quantitatively.

Procedure 2 Arbitrary quantification of correlations Adjacent relationships of hierarchical knowledge attributes are quantified.
Quantify any adjacency. (Symmetric matrix) The purpose is to express knowledge relevance and connectivity quantitatively.

Procedure 3 Quantify the order relationship Quantify the hierarchical knowledge attribute adjacency and correlation such as temporal order (adjacency matrix).
Knowledge that combines systematic information such as specialized knowledge and subject knowledge has a learning order. By quantifying such temporal order, the learning process can be analyzed. Conversely, this allows you to analyze the learner's context and suggest the optimal learning path.
The chronological order of knowledge is quantified in a hierarchy based on global classification. Classification of knowledge is very difficult. If too much part is involved, there is a possibility that the whole cannot be understood. Even if partial optimization is performed, there is no guarantee of total optimization, and it becomes more difficult to quantitatively express the relevance of knowledge. Therefore, I think that the classification should be based on the basic principles that govern the whole.
M = {m ₁ , m ₂ , m ₃ , m ₄ , m ₅ }
The order relation of the above vector elements is represented by a graph.
An adjacency matrix that represents a temporal order will not be a symmetric matrix. To indicate the direction, 1 is set so that the directionality can be expressed quantitatively. 1 stands for the number of arrows. As with the adjacency matrix that becomes a symmetric matrix, the diagonal elements have no substantial meaning and are therefore 0.

Measure the learner's situation in the knowledge vector space A specific knowledge area K is arranged and subdivided to obtain i knowledge areas k ₁ , k ₂ , k ₃ ,. . . , Consider a call k _n is added as follows: as a premise.

Procedure 1 Classify knowledge domain K = {k ₁ , k ₂ , k ₃ , k ₄ , k ₅ , k ₆ }

Procedure 2 Quantify the hierarchical relationship Here, the classified knowledge areas are further layered into three layers.
k ₁ = {k ₁₀ , k ₁₁ , k ₁₂ , k ₁₃ , k ₁₄ , k ₁₅ }
Next, the knowledge areas in the first layer are classified in the second layer as follows.
_{k 10 = {k 100, k} 101, k 102, k 103, k 104, k 105}
Finally, the knowledge areas classified in the second layer are classified as follows.
k ₁₀₀ = {k ₁₀₀₀ , k ₁₀₀₁ , k ₁₀₀₂ , k ₁₀₀₃ , k ₁₀₀₄ , k ₁₀₀₅ }

      We can classify any domain of knowledge such as specialized knowledge or subject knowledge using these procedures. Of course, it depends on the specific knowledge domain such as the hierarchical structure and the number of classifications, but the principle of the layering method is the same.

Procedure 3 Evaluation Method The evaluation is performed in order from the vector at the lower level of the hierarchy, and it is carried up to the elements of the vector at the higher level, and finally the knowledge region K is evaluated.

Examination of scholastic ability Tests are conducted, and each question is pre-set as to whether the knowledge is logically related. If you have not taken the test, evaluate it as 0.

The correct answer rate is used as an evaluation value of knowledge. For this purpose, dummy variables are set in order to analyze the qualitative data of questions that are valid and questions that are not. The correct number can be counted by setting the question correctly answered by the learner to 1 and setting the question to 0. Furthermore, the question defines the relevance with the node. Based on the definition, the relevance is measured and used as a node evaluation. The evaluation of the question identifier Qn is defined as follows.
Furthermore, since the knowledge related to the question is defined, the denominator obtains the evaluation value Ei of the knowledge node ki as the related question R using the valid number as the numerator.

A learner's current learning situation and a target learning situation are generated as weight vectors, and an evaluation value is obtained by an inner product.
Similar to the generation method for analyzing the current situation of the learner, the weight vector generation method obtains an evaluation value using the selection learner in the education system as a model learner. The value is normalized by 100 for each knowledge node. It is possible to know the distribution ratio of scores in a specific knowledge area.

Also, the difference between the learner's current situation and the target learning situation is analyzed.
D _i = k _xi -k _yi

From the results obtained up to this point, it is possible to propose an optimal portfolio and to propose optimal media content at the meta level. For example, Di is arranged in descending order, and a portfolio is constructed from the lower three elements. Assume that the lowest order is k5, and the following order relationship is expressed quantitatively.
k ₁ → k ₃ → k ₅ , k ₂ → k ₅

Furthermore, the hierarchical relationship
k ₁ = {k ₁₀ , k ₁₁ , k ₁₂ , k ₁₃ , k ₁₄ , k ₁₅ }
_{k 10 = {k 100, k} 101, k 102, k 103, k 104, k 105}
k ₁₀₀ = {k ₁₀₀₀ , k ₁₀₀₁ , k ₁₀₀₂ , k ₁₀₀₃ , k ₁₀₀₄ , k ₁₀₀₅ }

    The optimal portfolio can be constructed by analyzing each element of the hierarchy for 2 to 5 knowledge areas, rearranging from each evaluation value in descending order, and measuring the factor relationship. In this case, the factor relationship is simply set in ascending order of evaluation values. Problems can be found at the meta level, problems can be limited, and problems can be solved by learning relevant knowledge areas. Regarding the temporal order, the learning procedure can be shown by rearranging in the young order or examining the direction in the order expressed by the matrix and proposing backward the direction.

Temporal Transition Each evaluation value has a t-value representing a time axis, and time series analysis can be performed by arranging k time axes t in order to analyze the transition.

    Optimal efforts to eliminate variations in learner dominance at the level of scholastic ability that can be measured by examinations, etc. are the effects of improving the scholastic ability equivalent to the subjects who have been recognized as superior in the expected level and academic ability sought by the instructor There is. In addition, even for learners with high academic ability levels, by optimizing their efforts, they can eliminate surplus efforts and create opportunities for them to invest time and effort in their applied fields of academic ability such as their hobbies and research topics. be able to. Furthermore, if the academic ability can be increased to the expected level for more learners, the expected level of improvement in mass education can be expected as a breakthrough to certain types of elite education seen in academicism. By setting a standardized requirement level at the meta level on the instructor side, the learner can clarify the learning task, further clarify the lack of academic ability, quantify the learning effect, and And can optimize their efforts. Detailed quantification of the measurement of educational effects that could not be realized with analog technology education forms is more effective for students with lower academic ability levels. It is said that learners with low academic ability levels are often unable to optimize their efforts by clarifying their ability and ability to solve learning tasks based on their own will. In response to such problems, the instructor side can quantify the learning effects based on the learning tasks and the results of the exercises. The effect of learning can be improved.

    In order to create a standard achievement model based on feedback from learners and the importance set by the instructor, as many samples as possible are required. In order to efficiently collect and analyze the information, a system that works with a computer connected to the Internet using a server / client method is required. In addition, intellectual property rights such as copyrights in the content used in the system will make it more difficult to create an optimal learning environment as the usage restrictions increase, so such bottlenecks are removed. It is necessary to conclude a contract so that inconvenience occurs in this system, and create content. In order to continually improve the quality of high-quality systems and the digital content used in such systems, it is necessary to manage talented personnel and maintain high profitability. There is also a need for a business model. Considering that, in order to introduce this system, building a new organizational system, and building a monopoly business model in order to increase the ability to aggregate information in the system and the ability to aggregate revenue to support it, We believe this is the best mode for carrying out the invention.

    It can be applied in industries, organizations and workplaces that need to measure the effectiveness of education or discover opportunities for retraining. It is possible to measure the effects of learners quantitatively and use them as objective data for improvement and re-education in educational settings that tend to be unilateral knowledge transfer activities of educators or instructors. is. The effect of training conducted by companies is a vital element that will determine the future of the organization. For such issues, quantitative analysis of cost-effectiveness is also useful for management decision making. In particular, it is very difficult to measure the learning effects of individual learners when they are used in schools and other educational settings, and education is conducted for the average student. The average student, a virtual learner, cannot take the best approach, but by measuring the learning effects of individual learners as measured values, it is possible to provide optimal learning resources and environments for each. It is possible to improve the learning effect of individual learners and improve the overall learning effect.

Claims (7)

  When conducting a test related to a specific knowledge area, the knowledge area is organized and subdivided into a hierarchical structure to express knowledge in a vector space, thereby measuring the phase of the subject in the knowledge space. A learning process optimization system that can know the current learning status of the subject.   When conducting a test on a specific knowledge area, the knowledge area is categorized and subdivided into a hierarchical structure to represent knowledge on a vector space, thereby quantitatively identifying the knowledge area that is a factor of academic ability A learning process optimization system that can be expressed in terms of the correlation.   When conducting a test on a specific knowledge domain, by expressing the knowledge on a vector space by organizing and subdividing the knowledge domain into a hierarchical structure, by measuring the causal relevance of the knowledge subspace, A learning process optimization system that proposes the order of learning based on the order relation of knowledge domains.   Based on the measurement results obtained in claims 1 to 3, the learning plan for the learning area, such as learning tasks and procedures for the knowledge domain, is analyzed from the learner's current learning situation, so that the decision making of the learner is optimal Learning process optimization system that proposes a unique portfolio.   Organize, subdivide, and stratify the knowledge area of academic ability possessed by subjects who are recognized as having superior academic abilities in tests that are conducted for the purpose of scholastic ability tests and achievements that are trusted by the public, and tests that are aimed at selection Learning process optimization system that can analyze optimal learning model and non-learning learning model by modeling learning and achievement as ideal achievement in the vector space.   As an optimal academic ability model that idealizes the academic ability of subjects recognized as having superior academic ability, measure the difference from the current learning situation of the learner who learns on the system, add the time axis to the evaluation items, A learning process optimization system that proposes the direction of efforts that can make up for the difference as a learning plan.   In the knowledge area expressed in the vector space, the evaluation value in the knowledge area obtained by the subject who is superior in academic ability is generated as a weight vector and set as the target learning situation. A learning process optimization system that measures the distance from the inner product and proposes an optimal learning plan.