A Vocabulary learning system that verifies and optimizes vocabulary acceleration effects based on etymological clusters and learning logs
The vocabulary learning system addresses automation, structured transfer, and feedback challenges by using etymology clusters and logs to optimize vocabulary acquisition through automated keyword selection, usage recording, and data-driven feedback.
Patent Information
- Application Number
- KR1020260056642
- Authority / Receiving Office
- KR · KR
- Patent Type
- Patents
- Current Assignee / Owner
- Filing Date
- 2026-03-30
- Publication Date
- 2026-11-06
- Estimated Expiration
- 2046-03-30
AI Technical Summary
Existing book-based English learning systems face challenges in automating word selection, lack of structured etymology-based transfer, difficulty in quantifying learning outcomes, and inadequate feedback mechanisms for improving vocabulary acquisition.
A vocabulary learning system that uses etymology clusters and learning logs to automate keyword selection, record and analyze student usage, calculate effective exposure values, and provide data-driven feedback for targeted learning reinforcement.
Automates word selection, records actual usage and transfer, quantifies learning efficiency, and provides data-driven feedback to enhance vocabulary acquisition consistency and speed.
Smart Images

Figure 112026038250268-PAT00028_ABST
Abstract
Description
Technology Field
[0001] The present invention relates to a vocabulary learning system that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs. The present invention is a data processing-based computer implementation system, rather than a simple educational method. Background Technology
[0002] Book-based English learning is widely used in educational settings due to the advantage of enabling vocabulary acquisition within natural contexts. However, in actual operation, teachers and administrators consistently face the following problems.
[0003] 1. Problem with word selection
[0004] Textbooks contain a mix of words with high learning value, as well as those with low frequency, excessive difficulty, or little contextual importance. As a result, teachers must manually select "which words to teach" for every lesson; this increases time and costs, and leads to a lack of reproducibility and consistency as selection criteria vary from teacher to teacher.
[0005] 2. The Problem of Learning Extension (Transfer)
[0006] There is a lack of a structure that systematically guides and records not only form-based transfer extending to derivatives based on the same etymology, but also context-based transfer in which learners utilize target vocabulary in new contexts, syntactic structures, and discourse functions.
[0007] 3. Problems with Measuring Learning Outcomes
[0008] In actual classes and camps, how much a student has encountered, used, and expanded upon words has a decisive impact on achievement; however, since this learning process largely relies on observation or subjective records, it is difficult to quantitatively accumulate or analyze.
[0009] 4. Issues with class improvement (feedback)
[0010] Due to a lack of data on which word groups are acquired slowly and which activities induce more Use / Transfer, it is difficult for operators to systematically determine "what needs to be reinforced," and consequently, class improvement becomes dependent on experience.
[0011] Therefore, in book-based vocabulary learning, there is an urgent need for technology that simultaneously satisfies i) automation of core vocabulary selection based on book text and ensuring explainability of the selection basis, ii) expansion of word groups and induction of transfer learning based on etymology clusters (UnitID), iii) collection of exposure / usage / transfer event logs occurring during the learning process (including automatic or operator input-based) and automatic judgment / quantitative analysis, and iv) provision of learning prescriptions based on the above data and feedback-based optimization loops.
[0012] The technology forming the background of the present invention is disclosed in Korean Registered Patent Publication No. 10-2219886. The problem to be solved
[0013] The present invention aims to solve the problems of the aforementioned prior art and to resolve the following problems that occur during the actual operation of book-based vocabulary learning.
[0014] (1) We want to solve the problem of teachers having to manually select words every time.
[0015] In other words, the present invention aims to reduce the teacher's preparation burden and ensure consistency between lessons by automatically selecting "words of high teaching value (i.e., final keywords)" from book texts through a keyword determination unit and providing the basis for such selection in the form of score decomposition and basis logs.
[0016] (2) It aims to solve the problem of being unable to record, based on evidence, whether the student actually "used / extended use".
[0017] In other words, the present invention collects or receives student speech and writing outputs through a log management unit, and determines and records event logs related to target vocabulary use and transfer in the outputs in an automatic or semi-automatic manner from the perspective of UnitID-based form expansion (use of derived words / variants) and / or context-based expansion (utilization in new semantic categories / discourse functions / syntactic structures), thereby accumulating the learner's actual acquisition process as evidence-based data.
[0018] (3) It aims to solve the problem of being unable to objectively calculate "learning efficiency (acceleration)" rather than simple exposure.
[0019] In other words, the present invention aims to quantitatively demonstrate how much faster a learner has acquired knowledge compared to conventional learning by calculating an Effective Exposure Value (EffExp) that integrates Exposure, Use, and Transfer events through an Acceleration Effect Analysis Unit, and by calculating and providing an Acceleration Index (AI) relative to a Reference Number of Exposures (E*). For reference, in this invention, AI refers to the Acceleration Index, not Artificial Intelligence, and may be referred to by other terms such as AI Acceleration Index below.
[0020] (4) It aims to solve the problem of being unable to automatically reinforce word groups with slow learning achievement.
[0021] That is, the present invention aims to enable automatic learning prescription during operation by automatically identifying keywords or UnitID word groups with low AI indicators through a feedback loop management unit and performing keyword re-recommendation or activity package reinforcement for the corresponding area.
[0022] (5) This aims to solve the problem of being unable to verify actual achievement changes through pre / post evaluations.
[0023] In other words, the present invention aims to enhance the reliability and usability of the system by having the evaluation unit calculate score changes for each learner through pre / post assessments and analyze the correlation with log-based acceleration indicators (AI), thereby providing both learning process indicators and achievement change indicators.
[0024] However, the technical problems that the embodiments of the present invention aim to solve are not limited to the technical problems described above, and other technical problems may exist. means of solving the problem
[0025] As a technical means for achieving the above-mentioned technical task, a vocabulary learning device for verifying and optimizing vocabulary acceleration effects based on an etymology cluster and a learning log according to an embodiment of the present invention may include: a keyword determination unit that determines a set of keywords from the book text of a target book; a log management unit that records and manages event logs generated during the learning process of the target book using the set of keywords; and an acceleration effect analysis unit that calculates an effective exposure value and an acceleration indicator regarding a learner using the recorded log.
[0026] Additionally, the vocabulary learning device may further include a candidate keyword generation unit that generates candidate keywords from the book text before determining the keyword set; and an etymology cluster matching unit that calculates a matching reliability based on assigning word group identifiers to the candidate keywords through matching with an etymology cluster database.
[0027] The keyword determination unit above can determine a final keyword set including some of the candidate keywords among the above candidate keywords as the keyword set.
[0028] Additionally, the vocabulary learning device may further include an evaluation unit that performs verification of the reliability of the acceleration indicator based on the results of pre-evaluation and post-evaluation, and a feedback loop management unit that manages a feedback-based optimization loop that performs keyword re-recommendation and parameter update by referring to the acceleration indicator.
[0029] The above-described means for solving the problem are merely exemplary and should not be interpreted as intended to limit the invention. In addition to the exemplary embodiments described above, additional embodiments may exist in the drawings and the detailed description of the invention. Effects of the invention
[0030] The present invention provides a vocabulary learning system, device, and method for verifying and optimizing vocabulary acceleration effects based on etymology clusters and learning logs, thereby providing various effects such as the following.
[0031] (1) "Selection of words to teach" can be automated, and the reason for selection can also be automatically provided.
[0032] Specifically, the present invention receives book text as input, automatically generates candidate keywords, performs UnitID-based etymology matching and multi-score calculation (CIS / EFS / LevelFit / Novelty / Penalty) on the candidate keywords, and then determines the final keywords according to VPS.
[0033] By providing these functions, the present invention can significantly reduce the time teachers spend manually selecting words, and at the same time, allow the reasons why each final keyword was selected (Score Breakdown and Explanation Log) to be stored and provided, thereby improving the consistency and reproducibility of lesson design.
[0034] (2) Instead of learning words individually, UnitID-based word group expansion learning can be automatically implemented using generative AI.
[0035] Specifically, the present invention can match a final keyword to a word family identifier (UnitID) in an etymology cluster (UnitID) database and automatically generate a Word Family Pack including derivatives, inflectional forms, and variants belonging to the corresponding UnitID. Furthermore, the present invention can provide the Word Family Pack and keyword characteristics (level, difficulty, contextual information, etc.) as structured data that can be input into a generative AI model (e.g., LLM), thereby enabling the generative AI model to automatically generate learning activity packages, questions, example sentences, and extended learning tasks including use tasks and transfer tasks based on the input data.
[0036] By providing these functions, the present invention enables learning activities to be automatically configured into an extended learning structure utilizing the entire word group within the same UnitID, rather than being limited to simple word memorization or single word exposure, and helps learners systematically perform vocabulary expansion and transfer learning at the word group level even under the same learning time and text conditions.
[0037] (3) Students' use and transfer behaviors can be logged in an "automation-assisted (semi-automatic)" manner and accumulated as evidence.
[0038] Specifically, in existing book-based classes, there was a problem in quantitatively accumulating learning process data because whether students actually used or expanded words was recorded based on observation or test results.
[0039] In contrast, the present invention can generate automatic determination results (candidates) for Use and Transfer events by collecting or receiving learner output data (output) from voice / video (STT, Speech-To-Text), output images (OCR, Optical Character Recognition), or teacher / staff input memos, and performing UnitID-based matching on the output text. In this case, the present invention allows the determination results to be confirmed through verification / correction by a teacher or operator if necessary, and allows the confirmed events to be stored / accumulated in a database as structured logs.
[0040] By providing these functions, the present invention enables the student's actual vocabulary usage and extended usage processes to be recorded as evidence-based data rather than as simple "memory" or "observation," and subsequently utilized for analysis and feedback design.
[0041] (4) Instead of just "number of exposures," the actual acquisition strength (EffExp) can be calculated and provided.
[0042] Specifically, the present invention can calculate EffExp (Effective Exposure Value) by reflecting the respective learning contributions without treating Exposure, Use, and Transfer equally.
[0043] By providing these functions, the present invention can integrate and calculate how much a learner has performed actions (output / extension) important for acquisition, unlike conventional learning analysis centered on "how many times it has been seen," and provide the relevant information.
[0044] (5) The actual learning speed compared to the standard (E*) can be presented as a numerical value.
[0045] Specifically, the present invention can calculate an AI (acceleration metric) by normalizing EffExp by the reference exposure count (E*), thereby enabling learners to intuitively check numerically how quickly vocabulary acquisition has occurred compared to the usual acquisition standard.
[0046] By providing these functions, the present invention enables operators or teachers to compare how quickly learners have actually achieved results under the same number of sessions or exposure levels, and allows the information to be utilized for performance management in camp / school class operations.
[0047] (6) Areas that are lacking can be automatically identified and reinforcements can be automatically suggested.
[0048] Specifically, the present invention can automatically identify keywords or UnitID word groups whose calculated AI indicators are below a threshold value through the provision of a feedback loop management unit, and subsequently enable reinforcement designs such as providing additional activity packages, re-recommending keywords, and adjusting difficulty levels to be performed for the identified areas.
[0049] By providing these functions, the present invention enables operators or teachers to immediately know which word groups need to be reinforced based on data, rather than experience or intuition.
[0050] (7) By using Pre / Post evaluations together, it is possible to present "actual score improvement" in an integrated manner.
[0051] Specifically, the present invention can store and analyze not only log-based AI indicators but also pre / post evaluation results (△Score) together.
[0052] By providing these functions, the present invention enables learning process indicators (AI / EffExp) and final achievement indicators (Pre / Post improvement) to be presented together to teachers or operators, thereby allowing teachers or operators to verify learning results in a form that explains "why they improved."
[0053] (8) It can be applied equally to various operating environments such as classroom lessons, camps, and online platforms.
[0054] In other words, the configuration (technology) of the present invention is not limited to specific textbooks or age groups, and since it utilizes book text-based input, UnitID-based word group structures, and log-based analysis, it can be implemented and expanded in various educational operational environments, such as camp programs, school classes, and online learning platforms.
[0055] However, the effects obtainable from the present invention are not limited to those described above, and other effects may exist. Brief explanation of the drawing
[0056] FIG. 1 is a diagram showing the schematic configuration of a vocabulary learning system that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs according to one embodiment of the present invention. FIG. 2 is a diagram showing the main symbols and modifiers used in describing a vocabulary learning system and device that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs according to an embodiment of the present invention. FIG. 3 is a flowchart of a vocabulary learning method that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs according to an embodiment of the present invention. Specific details for implementing the invention
[0057] Embodiments of the present invention are described below with reference to the attached drawings so that those skilled in the art can easily implement the invention. However, the present invention may be embodied in various different forms and is not limited to the embodiments described herein. Furthermore, in order to clearly explain the present invention in the drawings, parts unrelated to the explanation have been omitted, and similar parts throughout the specification are denoted by similar reference numerals.
[0058] Throughout the specification of the present invention, when a part is described as being "connected" to another part, this includes not only cases where they are "directly connected," but also cases where they are "electrically connected" or "indirectly connected" with other elements interposed between them.
[0059] Throughout the specification of the present invention, when a member is described as being located "on," "on the upper," "on the top," "under," "on the lower," or "on the bottom" of another member, this includes not only cases where the member is in contact with the other member, but also cases where another member exists between the two members.
[0060] Throughout the specification of the present invention, when a part is described as "comprising" a certain component, this means that, unless specifically stated otherwise, it does not exclude other components but may include additional components.
[0061] Throughout the specification of the present invention, some of the operations or functions described as being performed by a terminal, device, or device may instead be performed by a server connected to said terminal, device, or device. Likewise, some of the operations or functions described as being performed by a server may also be performed by a terminal, device, or device connected to said server.
[0062] Throughout the specification of the present invention, the term "at least one" may be defined as a term including both singular and plural forms, and it will be obvious that even if the term "at least one" is not present, each component may exist in a singular or plural form and may mean singular or plural. Furthermore, whether each component is provided in a singular or plural form may be changed according to the embodiment.
[0063] <Definitions of Key Terms and Symbols in This Specification>
[0064] In addition, in describing the present invention, UnitID may refer to a word group identifier, EffExp to an effective impression value, AI to an acceleration metric, Transfer to a transfer event, Morphological Transfer to a shape-based transfer, and Contextual Transfer to a context-based transfer. Additionally, in the present invention, β may refer to a VPS calculation weight, γ to an EffExp calculation weight, and E* to a reference impression count. The explanation of each term can be more easily understood through the description provided below.
[0065] FIG. 1 is a diagram showing the schematic configuration of a vocabulary learning system (100) that verifies and optimizes vocabulary acceleration effects based on an etymology cluster and a learning log according to one embodiment of the present invention.
[0066] For convenience of explanation, the vocabulary learning system (100) that verifies and optimizes vocabulary acceleration effects based on an etymology cluster and a learning log according to one embodiment of the present invention will be referred to as the present system (100). In addition, the details shown (described) in the drawings of the present invention, including FIG. 1, may be applied equally to the description of the present system (100) or the server (10, the device) included in the present system (100), even if such details are omitted below.
[0067] Referring to FIG. 1, the system (100) may include a server (10), an administrator terminal (30), a teacher terminal (40), and a learner terminal (50).
[0068] At this time, the server (10) may refer to a vocabulary learning server (device) that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs, and for convenience of explanation, it will be referred to as the device (10) below.
[0069] The present system (100) and the present device (10) may be a vocabulary learning-related system and device (server) for optimizing learning efficiency by analyzing the acceleration of a learner's vocabulary acquisition based on logs and evaluation results generated during the vocabulary (word) learning process and reflecting the results in learning content and parameter settings. The present device (10) may perform functions such as extracting and confirming keywords from book text, collecting and analyzing learning logs to calculate the Effective Exposure Value (EffExp) and Acceleration Indicator (AI), and updating weights and reference values based on evaluation results and operation logs.
[0070] The device (10) may be a device or server that provides at least one of a webpage (homepage), program, application (app), service, and platform related to 'vocabulary learning that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs'. At this time, the webpage (homepage), program, application, service, and platform related to 'vocabulary learning that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs' provided by the device (10) may be referred to as the webpage (this homepage), the program, the app, the service, and the platform, respectively, for convenience of explanation below. In the example illustrated in FIG. 1, the device (10) is described as being provided in the form of a server capable of transmitting and receiving data through terminals (30, 40, 50) and a network (5).
[0071] Users of the device (10) may include administrators, teachers, and learners. Users may use the service by accessing the program, the app, the website, etc., using their own terminals (especially accessing after installation or without installation), and may also use the service without registering (logging in).
[0072] The administrator terminal (30) may refer to a terminal possessed by the administrator. The administrator may be a person who develops and distributes the system (100) and the device (10), and who operates and manages the system (100) and the device (10). The administrator may be referred to by other terms such as operator.
[0073] The teacher terminal (40) may refer to a terminal possessed by a teacher. The teacher terminal (40) may be a terminal for checking the keyword learning status, acceleration indicator (AI), and evaluation results for each learner through the device (10), and for performing the configuration or reinforcement of learning activity packages. The teacher may be, for example, a person who guides and manages learners of the learner terminal (50).
[0074] The learner terminal (50) may refer to a terminal possessed by a learner (student). The learner terminal (50) may be a terminal for performing learning (learning about the target book) using keyword learning content provided by the device (10), and for transmitting exposure, usage, and transfer events occurring during the learning process to the device (10).
[0075] In FIG. 1, the present invention (100) is described as an example in which one teacher terminal (40) and one learner terminal (50) are each included, but this is merely an example to aid in understanding the present invention and is not limited thereto, and preferably, multiple such configurations may be provided. That is, the present invention (100) may include multiple teacher terminals possessed by multiple teachers and multiple learner terminals possessed by multiple learners, and in describing the present invention below, the description of the teacher terminal (40) and the learner terminal (50) may be applied equally to the description of each of the multiple teacher terminals and each of the multiple learner terminals, even if the description is omitted below.
[0076] In the present invention, the terminal (i.e., each of the administrator terminal (30), teacher terminal (40) and learner terminal (50)) may include, for example, all types of wired and wireless communication devices such as PCS (Personal Communication System), GSM (Global System for Mobile communication), PDC (Personal Digital Cellular), PHS (Personal Handyphone System), PDA (Personal Digital Assistant), IMT (International Mobile Telecommunication)-2000, CDMA (Code Division Multiple Access)-2000, W-CDMA (WCode Division Multiple Access), Wibro (Wireless Broadband Internet) terminal, smartphone, smartpad, tablet PC, laptop, wearable device, desktop PC, etc., but is not limited thereto.
[0077] In addition, the device (10) can transmit and receive data by being linked with each terminal (i.e., each of the administrator terminal (30), teacher terminal (40), and learner terminal (50)) through a network (5), and can control the operation of each terminal (e.g., screen display operation).
[0078] Here, the network (5) may include, for example, a 3GPP (3rd Generation Partnership Project) network, an LTE (Long Term Evolution) network, a WIMAX (World Interoperability for Microwave Access) network, the Internet, a LAN (Local Area Network), a Wireless LAN (Wireless Local Area Network), a WAN (Wide Area Network), a PAN (Personal Area Network), a Bluetooth network, a NFC (Near Field Communication) network, a satellite broadcasting network, an analog broadcasting network, a DMB (Digital Multimedia Broadcasting) network, but is not limited thereto and may include various wired / wireless communication networks. A more detailed description of the device (10) is as follows.
[0079] Referring to FIG. 1, the device (10) may include an input unit (11), a preprocessing unit (12), a candidate keyword generation unit (13), an etymology cluster matching unit (14), a keyword confirmation unit (15), a word family pack generation unit (16), an activity package generation unit (17), a log management unit (18), an acceleration effect analysis unit (19), an evaluation unit (20), a feedback loop management unit (21), a control unit (22), and a database unit (23).
[0080] The components included in the device (10) may be implemented as a single or combined software / hardware module and may be executed on one or more servers, cloud environments, or learner terminals (tablets, smartphones, etc.).
[0081] The device (10) may be configured to have a computer-based data processing structure that implements vocabulary learning, and may be configured to have a continuous processing structure of, for example, 'data input → preprocessing → matching → log collection → numerical calculation → normalization → feedback loop'.
[0082] The input unit (11) can receive the text of the target book. Here, the target book may refer to, for example, a book that a teacher of the teacher terminal (40) intends to teach (or have learners learn). The input unit (11) can receive the text of the target book from the teacher terminal (40), but is not limited thereto and may also receive it from the administrator terminal (30).
[0083] The above book text may refer to the original text of the target book. When the input unit (11) receives the above book text, it may receive it in any one of the forms, for example, e-book text, OCR recognition-based text, or document file-based text.
[0084] The preprocessing unit (12) can perform preprocessing and normalization processing on the book text of the target book input from the input unit (11). When performing preprocessing and normalization, the preprocessing unit (12) can perform at least one of sentence segmentation, tokenization, POS tagging, lemmatization, and stopword removal on the input book text, for example, and thereby convert the input book text into an analyzable form.
[0085] The candidate keyword generation unit (13) can generate (extract) candidate keywords from the book text preprocessed by the preprocessing unit (12). In particular, the candidate keyword generation unit (13) can automatically generate (extract) at least one candidate keyword including a unigram and / or a phrase from the preprocessed book text. That is, the candidate keyword includes a unigram and a phrase, and the phrase can be generated by syntactic analysis or n-gram extraction.
[0086] At this time, the candidate keyword generation unit (13) may assign metadata to each of the generated candidate keywords, including at least one of the frequency (appearance frequency, occurrence frequency), position, context, sentence ID, surrounding vocabulary (surrounding word (window)), and part of speech information of the candidate keyword within the book text.
[0087] According to this, 'Keyword' in the present invention may refer to a candidate vocabulary (word) or phrase extracted (generated) through a preprocessing process for book text and a candidate keyword generation process by the candidate keyword generation unit (13). Additionally, 'Candidate Keyword' in the present invention may be defined as a keyword automatically generated by the candidate keyword generation unit (13), to which metadata such as frequency, location of occurrence, context, and sentence ID is assigned.
[0088] The candidate keyword generation unit (13) can be configured so that the number of generated candidate keywords varies according to the length of the preprocessed book text and the candidate keyword generation conditions set by the manager. Here, the candidate keyword generation conditions may include, for example, an n-gram range, a part-of-speech-based extraction condition, a phrase length limit, etc.
[0089] The candidate keyword generation unit (13) can perform refinement of valid candidate keywords by, for example, removing duplicate candidates (including integration based on headwords), removing function words and unnecessary expressions, and filtering based on parts of speech after generating candidate keywords, and thereby derive and provide (output) refined candidate keywords (refined valid candidate keywords).
[0090] According to this, the candidate keywords generated by the candidate keyword generation unit (13) may refer to refined candidate keywords that are filtered by performing secondary preprocessing on candidate keywords generated primarily according to the length of the book text and pre-set candidate keyword generation conditions.
[0091] Subsequently, the etymology cluster matching unit (14) can calculate the matching reliability based on assigning a word group identifier (UnitID) to the candidate keywords through matching with the etymology cluster database unit (231) by matching the candidate keywords generated by the candidate keyword generation unit (13) with the etymology cluster database unit (231). That is, the etymology cluster matching unit (14) can assign a word group identifier (UnitID) to the candidate keywords through matching with the etymology cluster database unit (231) and calculate the matching reliability. The etymology cluster matching unit (14) may be referred to by other terms such as the etymology matching unit.
[0092] That is, the etymology cluster matching unit (14) can assign a UnitID to each candidate keyword by performing matching with the etymology cluster database unit (231) by referring to the etymology cluster database unit (231) for each candidate keyword (especially refined candidate keywords) generated by the candidate keyword generation unit (13), and then calculate the matching confidence of the matching result for each candidate keyword.
[0093] Here, the etymology cluster database section (231) refers to a database section in which etymology clusters defined by an administrator are stored, and this may be a configuration included within the database section (23). In the present invention, 'Etymology Cluster' may be defined as a structure in which word families based on the same etymology are grouped and stored under a specific word family identifier (UnitID). In the etymology cluster database section (231), UnitID information and information on a list of variant words corresponding to the UnitID information may be stored in a linked manner.
[0094] At this time, UnitID refers to an identifier (word group identifier) that identifies a group of words based on the same etymology within the etymology cluster database section (231), and words belonging to UnitID can be connected based on the same root or etymology.
[0095] The Variant Forms list may refer to a list of variations (a list of variants) that includes inflectional forms, derivative forms, spelling variations, homonym processing results, etc., of words belonging to UnitID.
[0096] The etymology cluster matching unit (14) can assign a UnitID to each candidate keyword by matching with the etymology cluster database unit (231), for example, by comparing the candidate keyword with the list of variant words within the etymology cluster database unit (231). For example, the etymology cluster matching unit (14) can identify a UnitID within the etymology cluster database unit (231) that includes a keyword identical to the first candidate keyword in the list of variant words in relation to the first candidate keyword, which is one of the candidate keywords, and then assign the identified UnitID to the first candidate keyword.
[0097] Subsequently, the etymology cluster matching unit (14) can calculate the matching reliability for each candidate keyword to which a Unit ID (in particular, information on the Unit ID and matching type) is assigned (i.e., calculate the matching reliability for each candidate keyword). In particular, after assigning a Unit ID to each candidate keyword, the etymology cluster matching unit (14) can additionally assign information on the matching type (i.e., Unit ID matching type) for the Unit ID assigned to each candidate keyword to each candidate keyword, and then calculate the matching reliability for each candidate keyword.
[0098] Here, the above-mentioned Match Type refers to information regarding the matching type (matching method) of the UnitID, and can be classified into multiple (four) types including Exact matching type, Lemma matching type, Variant matching type, and Partial matching type. The Etymology Cluster Matching Unit (14) may assign one of the above multiple types of matching types as information for the UnitID matching type to each of the candidate keywords. That is, the above-mentioned Match Type refers to the type of UnitID matching and may include at least one of Exact, Lemma, Variant, and Partial matching.
[0099] At this time, the Exact matching type may refer to a matching method in which the candidate keyword and the keyword stored in the etymology cluster database (231) have completely identical spelling, form, and string.
[0100] The above Lemma matching type may refer to a matching method in which a candidate keyword is normalized into a lemma form, and the normalized candidate keyword matches a lemma stored in the etymology cluster database section (231).
[0101] The above Variant matching type may refer to a matching method that corresponds to the case where a candidate keyword matches any one of a derivative, variant, or morphological variant belonging to the same UnitID.
[0102] The above Partial Matching Type is a matching method corresponding to cases where a part of the string of a candidate keyword is similar to or partially matches a part of the string of a keyword stored in the etymology cluster database (231), and may include string similarity-based matching.
[0103] The above-mentioned matching confidence refers to a numerical value indicating the degree to which a candidate keyword matches a specific UnitID, and may be information calculated using matching types (Exact / Lemma / Variant / Partial), edit distance, part-of-speech matching, ambiguity check results, etc. Accordingly, matching confidence may refer to the confidence score of UnitID matching for a candidate keyword.
[0104] The etymology cluster matching unit (14) can calculate the matching reliability for each candidate keyword, for example, by using at least one of i) a basic weight for each matching type that is predefined (set) by the manager, ii) string similarity (e.g., edit distance or normalized similarity), iii) part of speech consistency, and iv) ambiguity check result.
[0105] Here, the basic weights for each matching type are intended to reflect the possibility of false positives in the matching method. For example, the Exact matching type is assigned a first weight, which is the highest weight (i.e., the highest weight), because the spelling and form match perfectly; the Lemma matching type is assigned a second weight, which is a high weight (i.e., the second highest weight), because the match occurs after lemma normalization; the Variant matching type is assigned a third weight, which is an intermediate weight (i.e., the third highest weight), because it includes matching of derivatives / variants within the same UnitID; and the Partial matching type is assigned a fourth weight, which is a relatively low weight (i.e., the lowest weight), because there is a possibility of false positives based on string similarity.
[0106] The basic weights for each matching type may be set as shown in Table 1 below, for example. In this example of the present invention, regarding the values of the basic weights for each matching type, the first weight is set to 1.0, the second weight to 0.9, the third weight to 0.85, and the fourth weight to 0.6, for example; however, this is merely an example to aid in understanding the present invention and is not limited thereto. Preferably, the values of the basic weights for each matching type may be adjusted (changed) according to the quality of the etymology cluster database section (231), the learner level, the noise level of the input data (STT / OCR), or the calibration data, and are not limited to fixed values.
[0107] [Table 1]
[0108]
[0109] The above string similarity may be information calculated using at least one of, for example, edit distance, n-gram-based similarity, and embedding-based similarity. When calculating the matching reliability, the etymology cluster matching unit (14) may reflect (set) that the matching reliability increases as the value of string similarity increases. That is, the value of matching reliability may increase as the value of string similarity increases.
[0110] The above ambiguity check is a check to distinguish between the candidates (UnitID candidates) when there is a possibility that a candidate keyword may be matched to two or more UnitID candidates simultaneously, and can be set (arranged) so that if the difference in confidence (confidence score) between multiple candidates (UnitID candidates) is less than a threshold, they are classified as uncertain or a penalty is assigned.
[0111] Thus, the etymology cluster matching unit (14) can calculate the matching reliability for each candidate keyword using at least one of the basic weights for each matching type, string similarity, part-of-speech matching, and the result of an ambiguity check. At this time, the etymology cluster matching unit (14) can determine the candidate keywords by excluding keywords (some candidate keywords) among the candidate keywords generated by the candidate keyword generation unit (13) from the candidate keywords or by applying a penalty term (Penalty) to those keywords whose matching reliability is less than a threshold value (θ) pre-set by the administrator. Here, the pre-set threshold value may be set to 0.70 as an example, but is not limited thereto and may be set and changed in various ways by the administrator.
[0112] That is, the etymology cluster matching unit (14) can calculate the matching confidence when matching Unit IDs for candidate keywords, and then, based on the calculated matching confidence for each candidate keyword, exclude keywords (candidate keywords) whose calculated matching confidence is less than a threshold value from the candidate keywords or apply a penalty, thereby confirming (finally selecting) the candidate keywords. In the present invention, the penalty may refer to a penalty value that reflects factors that reduce the learning value of candidate keywords, such as a decrease in matching confidence, excessive difficulty, excessive duplicate Unit IDs, or the inclusion of function words.
[0113] Subsequently, the keyword confirmation unit (15) confirms a set of keywords (especially a final set of keywords) from the book text of the target book, and in particular, can confirm the set of keywords based on the keyword priority score (VPS) calculated for each candidate keyword. Before confirming the set of keywords, a processing process by the input unit (11), preprocessing unit (12), candidate keyword generation unit (13), and etymology cluster matching unit (14) may be performed.
[0114] Specifically, once the candidate keywords are confirmed based on the matching reliability of each candidate keyword, the keyword confirmation unit (15) can then confirm a final keyword set that includes some of the confirmed (finally selected) candidate keywords as the keyword set.
[0115] At this time, the keyword determination unit (15) may include a keyword value calculation unit (151) that calculates a keyword value for each candidate keyword in order to determine the final keyword set. Here, the keyword value may include at least one element among i) Contribution to Book Understanding (CIS), ii) Etymology Transfer Efficiency (EFS), iii) Level Fit, iv) Novelty, and v) Penalty.
[0116] Here, i) Comprehension Impact Score (CIS) refers to a score indicating the degree to which a candidate keyword contributes to book comprehension. ii) Etymology Family Efficiency Score (EFS) refers to a score indicating transfer learning efficiency by reflecting the Family Size and variant structure of the UnitID word group to which the candidate keyword belongs. iii) Level Fit refers to a score evaluating whether the candidate keyword is at an appropriate level for a specific learner using learner level (L1–L4) and word difficulty information. iv) Novelty refers to a score evaluating the degree of novelty of the candidate keyword by comparing it with the past exposure / learning history of the learner or the system. v) Penalty refers to a deduction value reflecting factors that reduce the learning value of the candidate keyword, such as reduced matching reliability, excessive difficulty, excessive duplicate UnitIDs, and the inclusion of function words.
[0117] Specifically, i) Contribution to Book Comprehension (CIS) refers to the degree to which a candidate keyword contributes to book comprehension, and this may be a value calculated by reflecting the keyword's contextual importance, position of appearance (e.g., title, key paragraph), role within the sentence (topic word / action word), and contextual diversity. ii) Etymological Transferability (EFS) is a score reflecting the potential for derivative expansion of the UnitID to which the candidate keyword belongs, and this may be a value calculated by reflecting FamilySize (number of derivatives), number of Variant Forms, and transferability. iii) Level Fit may be a value calculated using learner levels (L1–L4) and word difficulty (e.g., CEFR / Lexile-based difficulty, abstractness, part of speech, etc.). iv) Novelty may be a value calculated regarding the novelty (degree of first encounter) to the learner based on past learning or exposure history. v) The penalty term may be a value calculated by reflecting penalty factors such as reduced reliability, excessive difficulty, excessive duplicate Unit IDs, or function words / unnecessary expressions.
[0118] To summarize again, i) the above-mentioned contribution to book comprehension (CIS) may refer to information indicating the degree to which a candidate keyword contributes to the understanding of the content of a book (target book). The keyword value calculation unit (151) can calculate the contribution to book comprehension (CIS) as a keyword value for the candidate keyword. At this time, the keyword value calculation unit (151) can calculate the contribution to book comprehension (CIS) by considering at least one of the importance of the context in which the candidate keyword is used within the book, the location where the candidate keyword appears (e.g., title, subtitle, key paragraph, etc.), the role within the sentence (topic word, action word, etc.), and contextual diversity based on whether it is repeatedly used in different contexts.
[0119] ii) The above Etymological Transfer Efficiency (EFS) may refer to information indicating the efficiency of vocabulary expansion that can be derived based on the Unit ID to which the candidate keyword belongs. The keyword value calculation unit (151) can calculate the Etymological Transfer Efficiency (EFS) as a keyword value for the candidate keyword. At this time, the keyword value calculation unit (151) can calculate the Etymological Transfer Efficiency (EFS) by reflecting at least one of the number of derivative words (FamilySize) included in the corresponding Unit ID, the number of variant forms, and the possibility of transfer to other words or concepts. At this time, for example, the composition of the calculation of the Etymological Transfer Efficiency (EFS) may be a composition calculated through the keyword value calculation unit (151), or a composition calculated through the Etymological Cluster Matching Unit (14).
[0120] iii) The above Level Fit may refer to information indicating the suitability between the learner's learning level and the difficulty level of the candidate keyword. The keyword value calculation unit (151) can calculate the Level Fit for the candidate keyword as a keyword value using learner level information and word difficulty information. At this time, the keyword value calculation unit (151) can calculate the Level Fit by considering at least one of the learner level (L1 to L4), the difficulty index of the word (e.g., CEFR or Lexile-based difficulty), the abstractness of the word, and part of speech information.
[0121] iv) The above novelty may refer to information indicating the degree to which a candidate keyword is newly exposed to a learner. The keyword value calculation unit (151) can calculate the novelty of a candidate keyword as a keyword value based on the learner's past learning history or keyword exposure history. At this time, the novelty may be calculated as a higher value for keywords that the learner has not previously learned or that have a low exposure frequency.
[0122] v) The above penalty term may refer to information for reflecting penalty factors when calculating the keyword value of a candidate keyword. The keyword value calculation unit (151) can calculate a penalty term for a candidate keyword. At this time, the keyword value calculation unit (151) can calculate the penalty term by reflecting at least one of a decrease in matching reliability, an excessive difficulty level relative to the learner level, excessive duplication of the same Unit ID, or the presence of function words or unnecessary expressions.
[0123] After the keyword value (in particular, the keyword value for multiple elements including book comprehension contribution, etymology transfer efficiency, level suitability, novelty, and deduction items) is calculated for each candidate keyword by the keyword value calculation unit (151), the keyword confirmation unit (15) can then calculate the keyword priority score (VPS) for each candidate keyword by adding the keyword values for multiple elements calculated for each candidate keyword based on the following mathematical formula 1, and then can confirm the final keyword set based on the calculated VPS for each candidate keyword.
[0124] At this time, the keyword determination unit (15) may include, for example, a VPS calculation unit (VPS calculation engine, not shown) for calculating VPS. The keyword determination unit (15) can calculate VPS for each of the candidate keywords based on the following mathematical formula 1.
[0125] In the present invention, the Vocabulary Priority Score (VPS) is a final score calculated by including CIS, EFS, LevelFit, Novelty, and Penalty to quantify the learning value of candidate keywords, and keywords with a high VPS may be preferentially selected as final keywords.
[0126] [Mathematical Formula 1]
[0127]
[0128] Here, k represents a candidate keyword, and each of β1 to β5 may represent a weight value for each of the above multiple elements (i.e., contribution to book comprehension, etymological transfer efficiency, level suitability, novelty, and penalty term). That is, in the above mathematical formula 1, k represents a candidate keyword, CIS represents contribution to book comprehension, EFS represents etymological transfer efficiency, LevelFit represents level suitability, Novelty represents novelty, Penalty represents a penalty term, and each of β1 to β5 may represent a weight for each element.
[0129] The keyword confirmation unit (15) can identify the top N candidate keywords with the highest VPS values among the candidate keywords as final keywords based on the calculated VPS for each candidate keyword, and then can generate and confirm (select) a final keyword set including the identified final keywords. Here, the top N may be, for example, 20, and the number of top N may be set and changed in various ways by the administrator.
[0130] That is, the keyword determination unit (15) can be configured to calculate keyword values for multiple elements including Contribution to Book Understanding (CIS), Etymology Transfer Efficiency (EFS), Level Fit, Novelty, and Penalty for each candidate keyword, calculate keyword priority scores for each candidate keyword by weighting the calculated keyword values, and then determine a pre-set number (top N) of candidate keywords with high keyword priority scores among the candidate keywords as the final keyword set.
[0131] The final keywords within the final keyword set determined by the keyword determination unit (15) may mean 'words of high teaching value' in the book text, and the keyword determination unit (15) can automatically select these final keywords.
[0132] For example, the keyword confirmation unit (15) can provide information about the final keyword set that has been confirmed (finally selected) to the teacher terminal (40), for example, and can also provide the data that served as the basis for selecting (confirming) the final keyword set. In this way, the keyword confirmation unit (15) can provide the teacher with the basis for the selection (confirmation) of the final keywords (i.e., final keywords that are ‘words of high teaching value’ in the book text) that have been confirmed (selected) by the keyword confirmation unit (15) in the form of score decomposition and basis logs, thereby reducing the teacher's preparation burden and enabling consistency between lessons.
[0133] Once the final keyword set is determined by the keyword determination unit (15), the word family pack generation unit (16) can then generate a word family pack including a list of derivatives, a semantic connection structure, a list of variant forms, etc., based on the Unit IDs of the final keywords within the final keyword set. For example, the word family pack generation unit (16) can generate a word family pack based on the Unit IDs for the final keyword set. The generated word family pack may include derivatives, inflectional forms, spelling variations, and semantic extension information matched to the Unit IDs.
[0134] For reference, a Word Family Pack refers to a data package that structures and groups derivatives, variant forms, semantic relations, and / or related word information corresponding to a single final keyword, based on Unit IDs assigned to identify words sharing the same etymology or semantic unit. The aforementioned Word Family Pack can be utilized as a word group-based learning resource to support the extended understanding and use of the target word.
[0135] The activity package generation unit (17) can generate an activity package (learning activity package) related to learning activities, including search activities, semantic inference, usage (output) tasks, and transfer tasks, according to the characteristics of the generated word family pack and final keywords. In particular, the activity package generation unit (17) can automatically generate (configure) a learning activity package using a rule-based template engine and / or a generative AI model (LLM) based on the word family pack and learner level information.
[0136] The above activity package (learning activity package) may refer to a bundle of learning tasks comprising a plurality of learning activity elements designed to improve a learner's word comprehension, usage ability, and transfer ability based on the characteristics of the above word family pack and final keyword. The above activity package may include at least one of a search activity, a semantic inference activity, a usage (output) task, and a transfer task.
[0137] That is, the word family pack generation unit (16) generates a word family pack based on the final keywords within the final keyword set, and the activity package generation unit (17) can generate a learning activity package based on the final keywords.
[0138] At this time, for example, the control unit (22) within the device (10) may provide a word family pack generated by the word family pack generation unit (16) and / or a learning activity package generated by the activity package generation unit (17) to the learner terminal (50). Subsequently, the learner terminal (50) may perform learning on the target book based on the word family pack and / or learning activity package provided by the device (10), and transmit event logs related to exposure, usage, and transfer occurring during the learning process to the device (10). In this way, the log management unit (18) within the device (10) may collect (acquire) the event logs provided by the learner terminal (50).
[0139] For example, the word family pack and / or learning activity package provided to the learner terminal (50) by the control unit (22) may be referred to by other terms such as content, learning content, keyword learning content, etc. That is, the control unit (22) may provide at least one of the word family pack and the learning activity package to the learner terminal (50) as keyword learning content.
[0140] The log management unit (18) can collect (acquire) event logs (especially logs of multiple events) that occur during the learning process (learning activity process) targeting the book and record (store) and manage them in the database unit (23). The log management unit (18) can collect and manage event logs that occur during the learning process of the book using the above-mentioned keyword set (confirmed keyword set) confirmed by the keyword confirmation unit (15).
[0141] According to this, the above learning process may refer to a learning process for a target book that is performed using keyword learning content (i.e., keyword learning content such as a word family pack or a learning activity package, which is generated based on keywords (final keywords) included in the keyword set (final keyword set) determined by the keyword determination unit (15). In the present invention, the 'learning process' may be referred to by other terms such as a learning activity process, a vocabulary (word) learning process, etc.
[0142] Here, the above events can be classified into three types, including a first event, a second event, and a third event. The first event refers to an exposure event, the second event refers to a use evidence event, and the third event refers to a transfer event.
[0143] Here, the aforementioned exposure may refer to an event in which a learner encounters the relevant keyword or a word related to the UnitID through text or learning activities. That is, the log of the first event is an Exposure Log, which may refer to a log of the event in which a learner encounters the keyword or a group of UnitID words.
[0144] The above-mentioned Use Evidence may refer to an event in which a learner uses the corresponding keyword semantically appropriately during output processes such as speaking, writing, or presentation. That is, the log of the second event is a Use Evidence Log (or Use Log), which may refer to a log of an event in which a learner uses the target keyword semantically appropriately in speaking, writing, or presentation. In other words, the Use Evidence event includes an event in which the final keyword is determined to have been used semantically appropriately in the learner's speech or writing result, and the log management unit (18) generates an automatic or semi-automatic determination result (candidate) for the use event through UnitID matching and / or context analysis, and can confirm and record it through verification / correction by a teacher or operator if necessary.
[0145] The above-mentioned transfer may refer to an event in which a learner uses (i) a morphological transfer in which a derivative, inflectional form, or variant belonging to the same UnitID as the target keyword in the output, and / or (ii) a contextual transfer in which the target keyword is extended and used within a new semantic category, discourse function, or syntactic structure. Here, morphological transfer refers to an event in which a derivative, inflectional form, or variant belonging to the same UnitID as the target keyword is used in the output. The above-mentioned contextual transfer may refer to an event in which the target keyword or a word group belonging to the same UnitID is extended and used within a new semantic category, discourse function, or syntactic structure. That is, the log of the third event, as a Transfer Log, may refer to a log of events including (i) form-based transfers in which a learner uses a derivative / variant within the same UnitID in the output and / or (ii) context-based transfers in which a target keyword or a word family within the same UnitID is extended in a new semantic category, discourse function, or syntactic structure.
[0146] In other words, a transfer event may include an event in which a learner uses (i) a derivative, inflectional form, or variant belonging to the same UnitID as the final keyword in a speech or writing result (form-based transfer) and / or (ii) an event in which the final keyword or a group of words belonging to the same UnitID is expanded and used in a new semantic category, discourse function, or syntactic structure (context-based transfer). The log management unit (18) of the device (10) can generate an automatic or semi-automatic determination result (candidate) for the transfer event through UnitID matching and / or context analysis, and can confirm and record it through verification / correction by a teacher or operator if necessary.
[0147] In the present invention, the term 'output (output data)' refers to a result generated by a learner during the course of learning activities, and may include voice data, video data, image data, or text data generated through speaking, writing, presentation, or equivalent output acts. For example, the output may include the learner's spoken voice, presentation video, handwritten notes, worksheets, evaluation answer sheets, or images of such materials, and the log management unit (18) may utilize the output as an input target for determining Use events and Transfer events.
[0148] That is, in the present invention, the 'output (output data)' may include text generated by performing STT from a learner's speech or presentation voice, text generated by performing OCR from a learner's handwriting or output image, and sentences used by students entered by a teacher.
[0149] In this way, the log management unit (18) can collect logs of multiple events (first to third events) occurring during the learning process (learning activity process) using the target book, record them in the database unit (23), and manage them. At this time, the collection of the logs of the multiple events can be performed in an automatic or semi-automatic manner (e.g., including teacher input). That is, the log management unit (18) can control the collection and recording of the logs to be performed in a semi-automatic manner based on STT / OCR automatic extraction, input by a teacher or administrator (operator), or a combination thereof.
[0150] The logs (log data) collected from the log management unit (18) can be used to calculate the effective exposure value (EffExp) and acceleration indicator (AI) described later. The logs collected from the log management unit (18) (i.e., logs of multiple events) may be referred to by other terms such as learning logs.
[0151] In particular, the log management unit (18) can collect (receive input) logs related to the second event (i.e., usage event) and the third event (i.e., transition event) based on at least one of (i) STT converted text of voice or video data, (ii) OCR recognized text of captured worksheet / evaluation data, and (iii) teacher input memo (memo information entered by a teacher regarding student utterance / sentence recording). In other words, the input channel of the logs (log data) related to the second event and the third event may include at least one of the STT converted text of voice / video, the OCR recognized text of the worksheet / evaluation, and the teacher input memo.
[0152] The log management unit (18) receives text data (text output) generated from STT (Speech-To-Text), OCR (Optical Character Recognition), or teacher input memos, or input / uploaded by an administrator (operator) as input values, and then tokenizes and lemmatizes the received text data, and then performs matching with the UnitID in the candidate keyword and etymology cluster database unit (231), thereby generating an automatic or semi-automatic judgment result (candidate) for use and transfer events (i.e., first event and second event).
[0153] In the present invention, 'text data' refers to character-based data obtained by directly generating from output or converting based on output, and may mean data received by the log management unit (18) as an input value for determining events. That is, the text data may include at least one of text generated by performing STT (Speech-To-Text) from a learner's speech or presentation voice, text generated by performing OCR (Optical Character Recognition) from a learner's handwriting or output image, and text memos entered by a teacher or administrator.
[0154] Here, the above judgment result may refer to information regarding the judgment result that the log management unit (18) automatically determines on its own whether the received text data is a usage event or a transition event. Since this judgment result is not confirmed information regarding whether it is a usage / transition event, but rather information temporarily determined by automatic determination (i.e., information temporarily determined as a candidate for a usage / transition event), it may be referred to by other terms such as candidate result. The log management unit (18) may, if necessary, provide the above judgment result to a teacher terminal (40) or an administrator terminal (30) so that the teacher or administrator can verify and correct the above judgment result, and based on this, the determination of the judgment result may be made. Based on the judgment result determined by the input of the teacher or administrator, the log management unit (18) may record the received text data in the database unit (23) as a log of the confirmed event (second event or third event).
[0155] Additionally, when the log management unit (18) receives the output text generated through the STT or OCR as an input value, it can then perform a normalization process on the received output text, including noise removal, spelling correction, lemmatization, and part-of-speech tagging, and then perform matching with the UnitID in the etymology cluster database unit (231) to generate the judgment result (candidate) for the Use and Transfer events (i.e., it can be used to generate judgment candidates for Use and Transfer events, thereby generating a judgment result). Afterward, the log management unit (18) can confirm the generated judgment result through verification and correction by a teacher or administrator if necessary, as mentioned above, and then record the received output text as an event log in the database unit (23) based on the confirmed judgment result.
[0156] In the present invention, 'output text' may refer to text among the text data that is processed by the log management unit (18) as data subject to event determination. That is, output text may refer to text data generated through STT or OCR or input / uploaded by an administrator (operator), which undergoes a normalization process including tokenization, lemmatization, part-of-speech tagging, and noise removal, and is then used to determine Use events and Transfer events through matching with UnitIDs within the candidate keyword and etymology cluster database unit (231).
[0157] Additionally, the logic (process) for automated determination of transfer events may be as follows. The log management unit (18) may define the transfer event as an event that includes i) a morphological transfer corresponding to when the learner uses a variant form or derivational / inflectional variant belonging to the same UnitID as the target keyword in the output, and / or ii) a contextual transfer corresponding to when the learner uses the target keyword or the same UnitID word group in the output by extending it in a new semantic category, discourse function, or syntactic structure.
[0158] The log management unit (18) can generate a judgment result (candidate) for the transition event by performing an automatic or semi-automatic judgment based on the output text generated from the output for the transition event, and can confirm the transition event through verification and correction by a teacher or operator if necessary.
[0159] The log management unit (18) can derive a UnitID (w) or perform UnitID-based matching by referring to the etymology cluster database unit (231) after performing lemmatization and part-of-speech tagging for each word (w) found in the output text. That is, the log management unit (18) can derive information of the UnitID corresponding to each word (w) in the output text through matching with the etymology cluster database unit (231).
[0160] At this time, the log management unit (18) can determine that the specific word (w) is a candidate for morphological transfer (corresponding to) the transfer information (i.e., UnitID(w)) derived for a specific word (w, any one word) among the words in the output text, if the UnitID assigned to the target keyword is the same as the UnitID (i.e., “UnitID(w) = UnitID(Target Keyword”)) and the word (w) is different from the target keyword itself (i.e., “w ≠ Target Keyword”).
[0161] Meanwhile, the log management unit (18) can determine that the use of the target keyword is a candidate for contextual transfer (corresponding to) even if the target keyword itself is used in the output text, if the way the target keyword is used satisfies at least one of the predefined context extension rules, collocation patterns, syntax patterns, or semantic extension rules.
[0162] For reference, in the present invention, 'Target Keyword' may refer to a keyword set as a standard keyword for determining whether to use or transfer in a learner's learning activity, output task, or evaluation process, among the keywords (final keywords) included in the final keyword set determined by the keyword determination unit (15) (i.e., a keyword set as a standard target for a specific learning activity, evaluation, or log determination).
[0163] The log management unit (18) can compare the determined transfer candidate (i.e., form-based transfer candidate and / or context-based transfer candidate) with the log of the usage event (second event) and the log of the transfer event (third event) that are stored in the database unit (23), i) if it does not exist in the previous log, classify the determined transfer candidate as a New Transfer and record (store) and manage it in the database unit (23), and ii) if it exists in the previous log, classify the determined transfer candidate as a Repeated Transfer and record and manage it. That is, the log management unit (18) can compare (determine) whether the determined transfer candidate exists in the previous log already recorded in the database unit (23), classify it as a New Transfer if it does not exist in the previous log, and classify it as a Repeated Transfer if it exists in the previous log and record and manage it.
[0164] The log management unit (18) is a word corresponding to the determined transition candidate. Transition candidate wordsIf the matching confidence (Confidence(w)) for the aforementioned specific word w (that is, the word determined to be a transition candidate) is below a preset threshold (θ), or if the part-of-speech tagging result is inconsistent with the part-of-speech conditions of the variant forms associated with the corresponding UnitID, the transition candidate word can be classified and recorded as an uncertain state (uncertain transition). Subsequently, a penalty can be applied to the transition candidate word classified as uncertain, or the transition candidate word can be classified and recorded as an event requiring verification by a teacher or administrator.
[0165] Additionally, the log management unit (18) can calculate the context-based transition candidate (i.e., the use of a transition candidate determined to correspond to a context-based transition in the usage of a target keyword or the same UnitID word group) and / or the semantic-based transition candidate, and if the calculated context-based transition candidate or semantic-based transition candidate is below a preset threshold, the context-based transition candidate can be classified as an uncertain state (uncertain transition) and then controlled to be confirmed through verification and correction by a teacher or manager.
[0166] Subsequently, the log management unit (18) can generate an explanation log containing the judgment result for the above-mentioned transition event, the transition candidate information (e.g., transition candidate words for form-based transitions and / or transition candidate usage (or transition candidate context) for context-based transitions), UnitID information, matching confidence and / or context fit scores, judgment type (i.e., new transition, repeated transition, uncertain transition), evidence sentence and / or evidence phrase, and store it in the database unit (23).
[0167] Subsequently, the acceleration effect analysis unit (19) can calculate the effective exposure value (EffExp) and acceleration indicator (AI) regarding the learner using the logs (log data, event logs) collected by the log management unit (18) and recorded in the database unit (23). The acceleration effect analysis unit (19) can be configured to calculate the effective exposure value and acceleration indicator regarding the learner using the recorded logs, by calculating the effective exposure value by weighting the number of events related to exposure, usage, and transfer, and by normalizing the result to the standard number of exposures to calculate the acceleration indicator.
[0168] The acceleration effect analysis unit (19) can calculate an effective exposure value (EffExp) for a learner (u) based on the following mathematical formula 2 and an acceleration indicator (AI) based on the following mathematical formula 3, using the above-mentioned recorded log (i.e., event log related to exposure, use and transfer).
[0169] [Mathematical Formula 2]
[0170]
[0171] [Mathematical Formula 3]
[0172]
[0173] The acceleration effect analysis unit (19) can calculate the effective exposure value by weighting exposure, use, and transfer according to the learning contribution. According to this, in the present invention, the effective exposure value (EffExp), which represents the actual acquisition strength, does not treat all events of exposure, use, and transfer equally, but can be considered by assigning different weights according to the type of each event.
[0174] That is, in mathematical formula 2, the Effective Exposure (EffExp) is a value calculated by weighting the Exposure, Use, and Transfer related events according to their respective relative learning contributions, and can represent an indicator of actual learning strength rather than simple exposure amount.
[0175] The above Effective Exposure Value (EffExp) may be a value indicating how much a learner has performed learning behaviors important for acquisition (i.e., actual acquisition intensity) by reflecting learning events related to exposure, use, and transfer according to learning contribution. In other words, the above Effective Exposure Value may be a value representing actual acquisition intensity (actual vocabulary acquisition intensity) by reflecting the level of learning engagement, including the learner's output and extension behaviors, rather than simply the number of exposures.
[0176] In Equation 2, γ is a weight and can be set according to the relative contribution of the learning behavior. In particular, γ1 is a weight for the Exposure event, and for example, the initial value (default) can be set to 1.0. γ2 is a weight for the Use Evidence event, and for example, the initial value (default) can be set to 2.0. γ3 is a weight for the Transfer event, and for example, the initial value (default) can be set to 3.0.
[0177] In mathematical formula 2, the initial value of the weight can be configured to be updated based on operational experience and subsequent pilot data-based calibration. The acceleration effect analysis unit (19) can set the weight to satisfy, for example, the relationship 'γ1 < γ2 < γ3'.
[0178] In addition, in Equation 2, Exp(u), Use(u), and Trans(u) may represent the number of Exposure events, the number of Use Evidence events, and the number of Transfer events, respectively. u may represent information identifying the learner of the learner terminal (50), i.e., a learner identifier.
[0179] In particular, Trans(u) represents the number of transfer events occurring for learner u, and transfer events may include (i) morphological transfers and / or (ii) contextual transfers. That is, Trans(u) can be defined as the number of occurrences of transfer events including morphological transfers and / or contextual transfers. Alternatively, Trans(u) can be defined as the number of transfer events occurring for learner u, as the sum of the number of morphological transfers (Trans_morph(u)) and the number of contextual transfers (Trans_ctx(u)).
[0180] Here, since the learning contributions of Trans_morph(u) and Trans_ctx(u) may differ, separate weights (different weights) may be applied when calculating EffExp, or the same weights may be applied. For example, γ3a may be applied as a weight to Trans_morph(u), and γ3b may be applied as a weight to Trans_ctx(u).
[0181] As previously mentioned, this Trans(u) may be information aggregated (derived, produced) by generating automatic or semi-automatic judgment results (candidates) for Transfer events through UnitID matching and context-based judgment logic based on STT / OCR and teacher input (teacher memo) related data (output text), and confirming the events through verification / correction by a teacher or operator if necessary. Since Trans(u) can be used with the same meaning as TransferCount in this invention, it may be referred to (notated) as TransferCount.
[0182] In summary, in the above mathematical formula 3, the Acceleration Index (AI) is a value obtained by normalizing the Effective Exposure Value (EffExp) by the Reference Exposure Count (E*), and may refer to a quantitative indicator indicating how quickly (acceleratedly) a learner has acquired vocabulary relative to the reference. In the present invention, "AI" refers to the abbreviation for "Acceleration Index," and does not mean "Artificial Intelligence."
[0183] Here, the threshold exposure count (E*) refers to the threshold exposure count required for typical vocabulary acquisition, and this can be set and adjusted according to learner levels (L1–L4) or word characteristics (abstraction, word length, Family Size, etc.). In Equation 3, the threshold exposure count E* is the threshold value (E*) per learner level level It can be saved as ) and can be adjusted to reflect word characteristics (abstraction, length, FamilySize, etc.).
[0184] In other words, the baseline exposure count can be stored as different values depending on the learner level (L1–L4). It can be set so that the baseline exposure count (E*) decreases as the learner level increases. The baseline exposure count (E*) can be adjusted to reflect word characteristics (abstraction, length, UnitID family size).
[0185] For example, adjusted reference exposure count (E* adj ) is the equation [E* adj =E* level It can be expressed as Х(1+ λ1·Abstractness+ λ2·Length- λ3·FamilySize)], which can mean an expression in which the reference number of exposures is adjusted based on word characteristics. In other words, the reference number of exposures can be adjusted by the aforementioned expression.
[0186] In other words, the reference exposure count (E*) can refer to a comparative reference value representing the 'average number of reference exposures required for a typical learner to acquire the word,' and if EffExp (actual effective exposure value) is less than E*, it means the word was acquired quickly, and if EffExp is greater than E*, it means the word was acquired slowly.
[0187] For example, reference values by learner level (E* level In relation to ), the E* of a learner at level L1 (beginner) may be set to 12 times, the E* of a learner at level L2 (intermediate) may be set to 9 times, the E* of a learner at level L3 (advanced) may be set to 6 times, and the E* of a learner at level L4 (proficient) may be set (saved) to 4 times.
[0188] As such, the standard exposure count (E*) is the standard amount of exposure required for a typical learner to acquire words, and can be set to different values depending on the learner's level. For example, the standard exposure count (E*) may be stored in the database unit (23) as a standard value (E*level) for each learner level, and when calculating the acceleration indicator (AI), the standard value corresponding to the learner's level may be selected and applied.
[0189] To summarize, in the above mathematical formula 2, u is a learner identifier, γ1 is the weight of an exposure event in the recorded log, γ2 is the weight of a usage event in the recorded log, and γ3 is the weight of a transition event in the recorded log, wherein the weights are set to satisfy the relationship γ1 < γ2 < γ3, and Trans(u) is the number of times a learner's transition event occurs, which can be defined as the sum of the number of form-based transitions and the number of context-based transitions.
[0190] In addition, in the above mathematical formula 3, E* represents the reference number of exposures, and if the calculated acceleration index exceeds 1 (when AI > 1), it represents delayed acquisition, and if the calculated acceleration index is less than 1 (when AI < 1), it represents accelerated acquisition. At this time, the reference number of exposures can be set according to the learner level, for example, L1: 12 times, L2: 9 times, L3: 6 times, L4: 4 times, etc.
[0191] Here, the aforementioned "Accelerated Acquisition" may refer to a state in which learning occurs faster than the average acquisition speed by reaching acquisition with only a small number of valid exposures compared to the reference number of exposures when acquiring specific learning vocabulary. On the other hand, the aforementioned "Delayed Acquisition" may refer to a state in which learning occurs slower than the average acquisition speed because a number of valid exposures exceeding the reference number of exposures is required when acquiring specific learning vocabulary.
[0192] Accordingly, the above acceleration indicator (AI) is an indicator representing the relative learning efficiency of a learner. If the AI value is less than 1 (AI < 1), the learner is determined to be in a 'accelerated acquisition' state, having acquired the vocabulary within fewer times than the standard exposure amount. If the AI value exceeds 1 (AI > 1), the learner is determined to be in a 'delayed acquisition' state, having acquired the vocabulary through repeated learning exceeding the standard exposure amount. Meanwhile, if the AI value is 1, the learner's acquisition speed is determined to be in a state that conforms to a standard acquisition model.
[0193] The values of the effective exposure value (EffExp) and acceleration indicator (AI) calculated in the acceleration effect analysis unit (19) can be utilized in conjunction with the evaluation unit (20) and the feedback loop management unit (21).
[0194] Additionally, the acceleration effect analysis unit (19) can provide information on the calculated effective exposure value and acceleration indicator to, for example, a teacher terminal (40) and / or a learner terminal (50), thereby enabling quantitative verification of how quickly the learner has acquired compared to normal learning.
[0195] The evaluation unit (20) can perform a pre-test and a post-test on learners (especially for each learner), and perform validation on the reliability of the acceleration indicator (AI) calculated by the acceleration effect analysis unit (19) based on the results of the pre-test and post-test.
[0196] In particular, the evaluation unit (20) can perform evaluations including a pre-test and a post-test for each learner. Here, the pre-test refers to an evaluation performed before the start of learning to determine the learner's initial vocabulary comprehension, usage ability, and transfer level, and the post-test refers to an evaluation performed after the completion of learning to verify the degree of vocabulary acquisition, changes in usage ability, and transfer performance according to the learning process.
[0197] The above assessment (pre-assessment and post-assessment) may include multiple assessment items to evaluate the learner's semantic recognition ability, contextual usage ability, output-based usage ability, and UnitID-based transfer ability. That is, in order to comprehensively assess the learner's vocabulary acquisition level, the assessment may include semantic recognition assessment items, contextual usage assessment items, output-based usage assessment items, and UnitID-based transfer assessment items.
[0198] The evaluation unit (20) can store the pre-evaluation score (PreScore) derived by performing a pre-evaluation on learners and the post-evaluation score (PostScore) derived by performing a post-evaluation in the database unit (23), and can calculate the score change (△Score) through the difference between the pre-evaluation score and the post-evaluation score. That is, the score change can be calculated (derived) based on, for example, the formula [△Score = PreScore - PostScore ].
[0199] The evaluation unit (20) can verify the reliability of the acceleration indicator (AI) by calculating (deriving) the correlation or predictive power between the acceleration indicator (AI) calculated by the acceleration effect analysis unit (19) and the score change (△Score) calculated from the prior evaluation and the post-evaluation.
[0200] The evaluation unit (20) may calculate (derive) a correlation coefficient or a predictive power indicator to analyze the relationship between the acceleration indicator (AI(u)) and the score change (△Score(u)) calculated for each learner. Here, the correlation coefficient may be calculated using a statistical correlation analysis technique to quantitatively represent the relationship between the acceleration indicator (AI) and the score change (△Score) calculated for each learner. Additionally, the predictive power indicator may be calculated as a value that quantitatively represents the predictive performance of the acceleration indicator (AI) by applying a prediction model that takes the acceleration indicator (AI) as an input variable and the score change (△Score) as an output value.
[0201] The evaluation unit (20) can evaluate the increasing trend of the score change (△Score(u)) as an increase in the acceleration index (AI(u)) by calculating at least one of the Pearson correlation coefficient (r) or the Spearman rank correlation coefficient (ρ) as the correlation coefficient. That is, the evaluation unit (20) can evaluate whether there is a tendency for the score change to increase as the acceleration index increases.
[0202] The evaluation unit (20) can determine whether the acceleration indicator (AI(u)) is valid as a predictor variable for score change (△Score(u)) by using a regression analysis or a classification model-based prediction procedure, and the coefficient of determination (R) as a predictor power indicator 2 At least one of the mean absolute error (MAE) or classification accuracy can be calculated.
[0203] In particular, the evaluation unit (20) can calculate a correlation coefficient or predictive power indicator between an acceleration indicator (AI) and a score change (△Score), and if the calculated correlation coefficient or predictive power indicator satisfies a preset threshold value, it can verify the reliability of the acceleration indicator (AI) by determining that the acceleration indicator (AI) calculated based on logs is consistent with the learner's actual achievement change. Here, the preset threshold value may refer to a standard condition that is set in advance by an administrator and stored in the database unit (23) to determine the validity of the correlation coefficient or predictive power indicator.
[0204] The evaluation unit (20) may determine that the acceleration indicator (AI) has a significant correlation with the score change (△Score) when the correlation coefficient is greater than or equal to a preset threshold value, and the preset threshold value may represent a minimum standard value for determining whether the correlation coefficient or the predictive power indicator satisfies the reliability of the acceleration indicator (AI).
[0205] Additionally, the evaluation unit (20) may determine that the acceleration indicator (AI) is valid as a predictor variable for score change (△Score) when the above-mentioned predictive power indicator satisfies the standard condition, and the above-mentioned standard condition may mean a judgment condition pre-set by the manager based on at least one of prediction accuracy, error range, or explanatory power.
[0206] In this way, the evaluation unit (20) can determine whether the acceleration indicator (AI) calculated based on logs is consistent with the learner's actual achievement change through the results of the correlation analysis or predictive power analysis. If the calculated correlation coefficient or predictive power indicator satisfies a preset threshold value, the evaluation unit (20) can verify the reliability of the acceleration indicator (AI) by determining that the acceleration indicator (AI) calculated based on logs is consistent with the learner's actual achievement change.
[0207] The feedback loop management unit (21) can manage a feedback-based optimization loop that automatically identifies keywords or UnitID areas with poor learning performance based on acceleration indicators (AI) calculated by the acceleration effect analysis unit (19), and performs re-recommendation and parameter updates for them.
[0208] The feedback loop management unit (21) can manage a feedback-based optimization loop that automatically identifies keywords or UnitID ranges to which the acceleration indicator (AI) belongs that are less than a preset threshold value (α) by referring to the acceleration indicator (AI) calculated based on the learning log collected by the log management unit (18) and stored in the database unit (23), and performs keyword re-recommendation and parameter update therefor.
[0209] At this time, the feedback loop management unit (21) can identify cases where the AI calculated by keyword or by the UnitID to which the keyword belongs is less than the threshold value among the keywords used in actual learning (learning process) that are included in the final keyword set confirmed by the keyword confirmation unit (15).
[0210] In other words, the feedback loop management unit (21) can automatically identify a keyword (final keywords) among the keywords (final keywords) included in the final keyword set confirmed by the keyword confirmation unit (15) whose acceleration indicator (AI) is less than a preset threshold value (α) or the UnitID area to which the keyword belongs. Subsequently, the feedback loop management unit (21) can perform feedback for the identified keyword or UnitID area, including at least one of keyword re-recommendation and reinforcement of an activity package (learning activity package). The feedback loop management unit (21) can provide feedback on keyword re-recommendation and activity package reinforcement to, for example, a teacher terminal (40) or an administrator terminal (30).
[0211] That is, the feedback loop management unit (21) can perform keyword re-recommendation or activity package reinforcement for areas below a preset threshold value. Here, when keyword re-recommendation, VPS recalculation or UnitID coverage constraints may be applied.
[0212] Subsequently, the feedback loop management unit (21) can reflect (consider) the result of the feedback (i.e., the result of keyword re-recommendation and reinforcement of the activity package) and set or update at least one of the weight (β) considered when calculating VPS in Equation 1, the weight (γ) applied to the effective exposure value (EffExp), and the reference exposure count (E*) through a calibration procedure (optimization procedure based on operation log) based on pilot operation data or operation logs. In the present invention, β or γ or E* can be calibrated by regression analysis or exploratory optimization (Grid Search) based on pilot data or operation logs.
[0213] Here, the pilot operation data may refer to data including learning logs, evaluation results, and acceleration indicator (AI) calculation results collected during a learning operation process in which the device (10) is performed experimentally on a limited number of learners or limited content prior to official operation.
[0214] Additionally, the calibration procedure based on the above-mentioned operation log may mean a procedure for analyzing a learning log collected by the log management unit (18) in the actual operating environment of the system (100) and accumulated and stored in the database unit (23), and adjusting at least one of the weights (β, γ) or reference exposure count (E*) to improve the calculation accuracy or prediction performance of the acceleration indicator (AI).
[0215] That is, in the present invention, calibration may refer to a process in which the VPS calculation weight (β), EffExp calculation weight (γ), reference exposure count (E*), etc., are not arbitrary values but are calculated or updated by an optimization procedure (regression, exploratory optimization, etc.) based on pilot data or operational logs and achievement data.
[0216] Accordingly, the feedback loop management unit (21) can control the acceleration indicator (AI) calculation and keyword recommendation structure to be continuously optimized according to data accumulated during the operation process (operation data) without relying only on the initial set parameters (initial set values), and thereby provide the present system (100), which is a vocabulary acceleration system that is continuously optimized according to the operation data.
[0217] The control unit (22) can control the operation of the device (10) (in particular, the operation of each part included in the device). In addition, the control unit (22) can control the operation (e.g., screen display operation, etc.) of each terminal (i.e., administrator terminal (30), teacher terminal (40), learner terminal (50)) connected to the device (10) through the network (5).
[0218] For example, the control unit (22) may store the score breakdown for the previously calculated VPS, AI, or EffExp in the database unit (23). Here, score breakdown may mean calculating the contribution of a plurality of sub-indicators or factors constituting a composite score such as the VPS, AI, or EffExp, and showing the influence of each factor on the composite score in a distinguishable manner.
[0219] Additionally, the control unit (22) can store an explanation log explaining the basis for selecting the keyword (final keyword) in the keyword confirmation unit (15) in the database unit (23).
[0220] Additionally, the control unit (22) can control the generation of a report or dashboard, for example, including a final keyword set, Word Family Pack, EffExp, AI, and Pre / Post evaluation results, to be output (displayed) on the screen of the administrator terminal (30) and / or the teacher terminal (40).
[0221] The database section (23) can store various data transmitted and received from the device (10). The database section (23) may include an etymology cluster database section (231) as previously mentioned.
[0222] Meanwhile, FIG. 2 is a diagram showing the main symbols and modifiers used in describing a vocabulary learning system and device (i.e., the system (100) and the device (10)) that verify and optimize vocabulary acceleration effects based on an etymology cluster and a learning log according to an embodiment of the present invention. Referring to FIG. 2, the main symbols and modifiers used in the specification to describe the present invention may be the same as those shown in FIG. 2.
[0223] In summary, the present invention (i.e., the present system (100) and the present device (10)) can provide a technology for calculating etymology cluster-based keyword optimization and vocabulary acquisition acceleration indicators. In particular, the present invention relates to a book-based vocabulary learning system that automatically extracts candidate keywords from book texts, matches the candidate keywords to word group identifiers (UnitID: identifiers identifying word groups based on the same etymology) in an etymology cluster database to optimally select core vocabulary with high learning value, collects exposure, use evidence, and transfer logs generated during the learning process, calculates an effective exposure value (EffExp: effective exposure value obtained by weighting exposure / use / transfer) and a vocabulary acquisition acceleration index (AI: vocabulary acquisition acceleration index obtained by normalizing EffExp by the reference exposure count E*) based on these, and verifies and reports the acceleration effect through pre / post assessment (i.e., automatically analyzes and reports the achievement and efficiency of vocabulary learning by quantifying the vocabulary acquisition acceleration index and pre / post achievement changes).
[0224] The present invention can perform preprocessing and normalization including sentence segmentation, tokenization, part-of-speech tagging, lemmatization, and stop word removal after inputting book text, generate a list of candidate keywords including single words and phrases, calculate etymological cluster (UnitID) matching and matching confidence for the candidate keywords, calculate Book Comprehension Contribution (CIS), Etymological Transfer Efficiency (EFS), Level Fit, Novelty, and Penalty, and then determine the final keyword set by calculating a Keyword Priority Score (VPS) as a weighted sum thereof.
[0225] In addition, the present invention can generate a UnitID-based Word Family Pack and an activity package for a final keyword, collect exposure / usage / transfer events as logs during the performance of learning activities, calculate EffExp, and quantify the acceleration effect of vocabulary acquisition by the learner by calculating AI (=EffExp / E*) relative to the threshold exposure count (E*: Threshold Exposure Count). Here, transfer events may include morphological transfer, which uses the same UnitID derivative / inflectional / variant in the output, and / or contextual transfer, which uses the target keyword or the same UnitID word group by extending it in a new semantic category, discourse function, or syntactic structure.
[0226] At this time, the input of the use and transfer log may include at least one of (i) the result of STT conversion of voice or video data, (ii) OCR recognized text captured from a worksheet or evaluation sheet, and (iii) a teacher input memo. The present invention generates an automatic or semi-automatic determination result (candidate) for use and transfer events by performing UnitID-based word group matching and / or context analysis on the text output. The candidate may include confidence, context fit, and evidence. If necessary, the event can be confirmed and recorded in the log through verification / correction by a teacher or operator.
[0227] That is, the present invention comprises: 1) a configuration for automatically generating candidate keywords from book text, performing etymological cluster (UnitID) matching on the candidate keywords, and calculating matching confidence; 2) a configuration for calculating Book Comprehension Contribution (CIS), Etymological Transfer Efficiency (EFS), Level Fit, Novelty, and Penalty on the candidate keywords, and calculating a weighted sum-based Keyword Priority Score (VPS) to select a final keyword set; 3) a configuration for generating a UnitID-based Word Family Pack on the final keyword set, and automatically generating / configuring a learning activity package using a rule-based template engine and / or a generative AI model (LLM) based on the said Word Family Pack and learner level information; 4) a configuration for collecting Exposure, Use Evidence, and Transfer event logs during the learning activity process, wherein collection can be performed in an automatic or semi-automatic manner (e.g., including teacher input), and from this log data, the Effective Exposure Value (EffExp) and Acceleration A configuration for calculating an indicator (AI); 5) including at least one of (i) voice / video STT converted text, (ii) worksheet / assessment OCR recognized text, and (iii) teacher input memo as an input channel for Use and Transfer logs, and performing UnitID matching on the text output to generate automatic determination results (candidates) for Use and Transfer events, wherein the candidates may be provided including confidence and supporting evidence, and are recorded as logs through verification / correction by a teacher or operator if necessary; 6) a configuration for performing Pre / Post Assessment to calculate score changes and analyzing the correlation with the AI indicator (calculating correlation or predictive power) to verify the reliability of the log-based acceleration indicator.7) It may include a configuration including a feedback-based optimization loop in which the AI automatically identifies keywords or UnitID ranges below a threshold value to perform keyword re-recommendation and weight updates.
[0228] To summarize, the device (10) can select keywords for vocabulary learning based on book text and construct a UnitID-based word group (i.e., perform keyword optimization and word family pack generation). To this end, it can perform the following steps: i) generating candidate keywords from book text; ii) matching the candidate keywords to an etymology cluster database to assign a UnitID and calculating the matching confidence; iii) calculating the Contribution to Book Understanding (CIS), Etymology Transfer Efficiency (EFS), Level Fit, Novelty, and Penalty for the candidate keywords, and calculating the Keyword Priority Score (VPS) as a weighted sum thereof; iv) selecting a final set of keywords based on the VPS; and v) generating a UnitID-based Word Family Pack for the final set of keywords.
[0229] In addition, the device (10) can log Use and Transfer events and calculate acceleration indicators (i.e., perform automatic judgment of Use / Transfer logs and calculation of EffExp / AI) using the learner's output data in a book-based vocabulary learning process. To this end, the device can perform the following steps: i) receiving STT converted text, OCR recognized text, and / or teacher input memo as the learner's output data; ii) tokenizing and headwording the output data and then rematching it to UnitID; iii) generating automatic or semi-automatic judgment results (candidates) for final keyword use (Use Evidence) events and transfer (Transfer) events based on the rematching results, and, if necessary, confirming the events through verification / correction by a teacher or operator and recording them as logs; iv) calculating an effective exposure value (EffExp) based on the event logs and calculating an acceleration indicator (AI) relative to the reference exposure count (E*); and v) verifying the correlation with the change in pre / post evaluation scores (△Score) and / or automatically identifying areas where AI is insufficient.
[0230] < Theoretical Basis and Basis for Numerical Setting >
[0231] Meanwhile, in describing the present invention, the theoretical basis and the basis for numerical setting are explained as follows.
[0232] The present invention is not merely an "educational method," but a data processing-based computer implementation system that follows the sequence of 'book text input → keyword value calculation → collection of learning logs (including automatic or operator input-based) and automated judgment / quantitative analysis → calculation of acceleration indicators → feedback optimization.' Each mathematical formula and setting value applied in the present invention is not an arbitrary value but can be designed (prepared, set) to be set and updated by calibration based on literature grounds and pilot data.
[0233] [[1. Regarding Exposure-based Vocabulary Acquisition Basis and E* Dynamic Adjustment Logic]]
[0234] (1) Repeated exposure has a significant effect on improving vocabulary knowledge.
[0235] Studies have reported that vocabulary acquisition is not sufficient with a single exposure, and that vocabulary knowledge in various aspects, such as spelling, meaning, grammatical function, and association, is significantly improved as the number of repeated exposures increases. For example, Webb (2007) showed that learning achievement increases as the number of repetitions increases through a controlled experiment with conditions of 1, 3, 7, and 10 repeated exposures.
[0236] In addition, although there is no single agreed-upon standard for the number of repeated exposures, a review of studies has reported various ranges such as "6, 10, 12, and 20 or more times," suggesting that repeated exposure is an important foundational condition for vocabulary acquisition.
[0237] (2) E* is not a "fixed value", but can be defined as a standard value by level / word characteristic.
[0238] In this invention, E* is defined as the "standard exposure amount required for general vocabulary acquisition" and can be set differentially according to the learner level and word characteristics. This is because beginner learners require more exposure due to a heavy burden of form-meaning connection and low contextual reasoning ability, whereas as learners progress to higher levels, the standard exposure amount may decrease as contextual reasoning and transfer learning abilities increase.
[0239] For example, in the present invention, an example of an E* setting (initial value) for each level of a learner may be as shown in Table 2 below. In this case, the E* value shown in Table 2 below is an example of an initial reference value, and the E* value may be adjusted in actual operation by pilot and operation log-based calibration.
[0240] [Table 2]
[0241]
[0242] (3) Word characteristic-based E* dynamic adjustment logic
[0243] The present invention does not set E* as a simple fixed value per level, but can dynamically adjust it by reflecting word characteristic information.
[0244] In other words, E*, the adjusted reference exposure count explained earlier adj (adjusted E*) is the standard number of exposures per learner level (E* level It refers to a value adjusted according to word characteristics (abstraction, length, UnitID-based word group size, etc.), and can be used as a reference value to calculate AI by normalizing EffExp. The adjusted reference exposure count can be expressed as shown in Equation 4 below.
[0245] [Mathematical Formula 4]
[0246]
[0247] In Equation 4, Abstractness refers to the degree of abstraction, and as the degree of abstraction increases, meaning formation becomes more difficult, so E* can be set to increase. Length refers to length, and as the length increases, the burden of form processing increases, so E* can be set to increase. FamilySize refers to the scale of UnitID derivatives, and as FamilySize increases, E* can be set to decrease to reflect the transfer learning efficiency.
[0248] The present invention can reflect standard exposure more realistically by reflecting "learning difficulty and transferability" through such E* dynamic adjustment, and in the case of word feature-based data, it can be obtained from external databases such as WordNet and MRC Psycholinguistic Database or from system-built databases.
[0249] [[2. Regarding grounds for weighting "Use" and "Transfer"]]
[0250] (1) Output-based use is more beneficial for long-term memory than simple exposure.
[0251] Numerous studies have reported that output activities involving memory retrieval, such as testing, speaking, and writing, strengthen long-term memory more effectively than simple repetition of reading or watching (study). For example, Karpicke & Roediger (2008) experimentally demonstrated that repeated retrieval (testing) is more effective for long-term memory retention than repeated learning.
[0252] Therefore, the Use event considered in the present invention can be modeled as having a greater learning contribution than simple Exposure.
[0253] (2) Transfer is a learning behavior that represents the "expansion effect" of vocabulary acquisition.
[0254] Transfer is the act of a learner not merely memorizing specific words, but inferring and generating derived words based on the same UnitID or applying them to new contexts. This can be regarded as a higher-order learning outcome that goes beyond simple receptive knowledge and includes production and expansion.
[0255] Furthermore, Laufer & Hulstijn's Involvement Load Hypothesis (ILH) explains that vocabulary learning is more effective the more a learning activity requires "need / search / evaluation," and writing / construction tasks are reported to have a high involvement load.
[0256] Since the transition activity strongly involves the "exploration + evaluation" of this involvement load, the transition event in the present invention can be modeled as the event with the highest learning contribution.
[0257] In addition, in the present invention, transfer is not limited to the use of simple derived words but may include context-based transfers such as the expansion of semantic categories of the same keyword or UnitID word group, the expansion of discourse functions, and the expansion of syntactic structures, and such transfer events may be modeled as events having a high learning contribution as higher-order outputs and extended learning behaviors.
[0258] (3) Therefore, the EffExp considered in the present invention does not treat exposure / use / transfer equally.
[0259] That is, the present invention can calculate EffExp by weighting exposure / use / transfer according to the learning contribution. In this case, the weight for each event type can be set according to the relative contribution of the learning behavior, and the initial value of the weight can be based on operational experience, and can be configured (set) so that it can be updated through subsequent calibration based on pilot data.
[0260] [[3. Regarding the Basis for Setting VPS Weights (β)]]
[0261] (1) VPS simultaneously reflects the goals required in book-based vocabulary learning.
[0262] When selecting words in book-based learning, relying solely on simple frequency leads to problems such as: i) excessive inclusion of function words / redundant words, ii) words that do not match the level appearing at the top, iii) failure to consider word groups capable of transferability, iv) failure to reflect novelty based on learner history, and v) inability to explain why the words were selected.
[0263] To solve these problems, the present invention may include multi-purpose elements in the VPS. Here, the multi-purpose elements may include: i) CIS: contribution to book comprehension ("important in reading context"), ii) EFS: etymological transfer expansion efficiency ("word group expansion"), iii) LevelFit: suitability to learner level ("learnability"), iv) Novelty: novelty ("novelty compared to past learning history"), and v) Penalty: penalty term ("low reliability / duplication / inappropriateness"). Additionally, the calculation formula for the VPS may be the same as the aforementioned Equation 1.
[0264] (2) The β value may not be an arbitrary value but may represent the "relative importance of the objective function."
[0265] The weight (β) considered when calculating VPS in the present invention is a parameter that expresses the "importance of the learning objective," and while the initial value may be set based on operational experience, the present invention can be designed (or provided) to calculate an optimal value based on data through a calibration procedure.
[0266] For example, the initial values of the weights (β) in Equation 1 may be set to 0.3 for the weight β1 representing the CIS weight, 0.3 for the weight β2 representing the EFS weight, 0.2 for the weight β3 representing the LevelFit weight, 0.1 for the weight β4 representing the Novelty weight, and 0.1 for the weight β5 representing the Penalty weight.
[0267] [[4. Regarding the Calibration Procedure (Technical Parameter Correction)]]
[0268] In the present invention, "calibration" refers to a technical correction process in which parameters such as β, γ, and E* are not arbitrarily fixed but are calculated / updated through an optimization procedure based on pilot or operational data. That is, the expression "can be calibrated" may mean that parameter correction is possible by automatically calculating and storing β, γ, and E* through regression / grid search, etc., to maximize predictive power or performance improvement in pilot / operational data, and then updating them according to operational logs.
[0269] (1) Calibration procedure (example) in the present invention
[0270] In the present invention, the calibration procedure may be carried out as follows: for example, i) operating 2 to 4 sessions with a pilot training group (e.g., 20 people), ii) collecting (CIS, EFS, LevelFit, Novelty, Penalty) and result variables (Y) for each keyword / UnitID, iii) using post-test scores, Use accuracy, Transfer rate, or AI as result variables (Y), iv) calculating the combination of β and γ with maximum predictive power through regression analysis or exploratory optimization (Grid Search), v) saving the calculated β and γ as system default values, and subsequently performing an Adaptive Update according to the operation log.
[0271] (2) Calibration result table (example)
[0272] The performance of each β / γ combination in the calibration process performed in the present invention can be evaluated by indicators such as predictive power (R²), error (MAE), and average AI growth rate.
[0273] Predictive power (R²) indicates the extent to which a log-based metric (AI or EffExp) explains changes in pre- and post-evaluation scores or achievement data (Y); a higher value may indicate a better predictive fit. Error (MAE) refers to the mean absolute error between predicted and actual values; a smaller value may indicate higher accuracy. Average AI growth rate represents the average increase or growth rate of AI in the learner group when the corresponding parameter combination is applied; a higher value may indicate a greater learning acceleration effect.
[0274] An example of a β / γ calibration result (deriving the optimal combination) may be as shown in Table 3 below, for example.
[0275] [Table 3]
[0276]
[0277] In Table 3, "Predictive Power (R²)" is an indicator representing how well a calibrated β, γ combination explains the outcome variable (Y) (e.g., PostScore, △Score, or AI), and a higher value may indicate higher explanatory power. "Error (MAE)" refers to the mean absolute error between the predicted value and the actual value, and a lower value may indicate higher prediction accuracy. "Average AI Growth Rate" refers to the ratio or amount of increase in the average AI of the operational group compared to the baseline or existing combination when the corresponding β, γ combination is applied, and can be used as an indicator to quantitatively express the learning acceleration effect when combined with achievement improvement.
[0278] As such, the β and γ values considered in the present invention are not "arbitrary values," but may be values calculated (set) as a combination that shows the highest predictive power and achievement improvement in pilot and operational data.
[0279] [[5. Regarding the Definition of AI (Acceleration Metrics) and the Logic of "Quantifying Acceleration Effects"]]
[0280] In the present invention, AI is a value obtained by normalizing EffExp to E*, and may represent an indicator indicating how quickly (acceleratedly) a learner has met the vocabulary acquisition conditions compared to a conventional standard. The calculation formula for AI may be the same as the aforementioned mathematical formula 3.
[0281] At this time, in mathematical formula 3, i) if AI(u) > 1.0, it means acceleration relative to the standard, ii) if AI(u) = 1.0, it means that the standard level has been achieved, and iii) if AI(u) < 1.0, it means that the standard has not been met (subject to reinforcement).
[0282] The AI produced in this invention is not a simple score but an indicator based on the learning process log, so it can be considered information that enables the explanation of "why achievement improved" when combined with Pre / Post evaluations.
[0283] Hereinafter, an embodiment of the present invention will be described in more detail. Specifically, an embodiment (experimental example) for verifying the performance of the present invention will be described below, and for convenience of explanation, this embodiment will be referred to as the "main experiment" below.
[0284] < Example of 'If I Build a School' Book-Based VAS Operation and Quantitative Comparison of Acceleration Effects >
[0285] [[1. Implementation Conditions and Input Data]]
[0286] In this experiment, the implementation conditions and input data can be set as follows, for example.
[0287] [ (1) Book (Target Book): If I Build a School
[0288] (2) Target learner level: L2(A2)
[0289] (3) Number of participants: A total of 10 people (set as follows: Operation group (VAS applied): 5 people (u1~u5), Control group (Non-VAS): 5 people (c1~c5))
[0290] (4) Operation Period and Session Structure: Total of 4 sessions (once a week, 60 minutes each)
[0291] (5) Standard exposure count (E*): L2 standard 12 (Applying the initial value example from Table 1)
[0292] (6) EffExp Weights (γ): γ1=1.0, γ2=2.0, γ3=3.0 (Example of initial values)
[0293] That is, for example, the input section (11) can receive the text of the book 'If I Build a School' as the book text of the target book, and can also receive setting information for initial values such as the standard exposure count and EffExp weight from the administrator in advance.
[0294] [[2. Keyword Candidate Generation and Preprocessing]]
[0295] In the process, the preprocessing unit (12) can receive book text as input and perform preprocessing and normalization including sentence segmentation, tokenization, POS tagging, lemmatization, and stopword removal.
[0296] Afterward, the candidate keyword generation unit (13) can automatically generate candidate keywords including unigrams and phrases from the preprocessed text, and at this time, metadata such as frequency, position, context, sentence ID, surrounding words (window), and part of speech information may be assigned to the candidate keywords.
[0297] Here, the number of candidate keywords generated may vary depending on the length of the input text and candidate generation conditions (e.g., n-gram range, part-of-speech-based extraction conditions, phrase length limit, etc.), and in this experiment, about 280 candidate keywords were generated.
[0298] Subsequently, the candidate keyword generation unit (13) refined valid candidate keywords by removing duplicate candidates (including integration based on headwords), removing function words and unnecessary expressions, and filtering based on parts of speech, and in this experiment, the number of valid candidate keywords refined after preprocessing was about 165. That is, the number of candidate keywords generated (finally selected, confirmed) by the candidate keyword generation unit (13) in this experiment may be, for example, 165.
[0299] [[3. UnitID Matching and Matching Confidence]]
[0300] Afterwards, the etymology cluster matching unit (14) can perform UnitID matching for 165 candidate keywords by referring to the etymology cluster database unit (231). At this time, in this experiment, as a result of performing the matching, i) the number of successful UnitID matchings was 120 (i.e., the number of keywords that successfully matched UnitID out of 165 candidate keywords was 120), and ii) candidate keywords with a matching confidence of less than 0.7 were excluded from the candidate keywords or penalized (i.e., matching confidence < 0.70: 15 candidates were excluded or penalized).
[0301] For example, in this experiment, the result of UnitID matching by the etymology cluster matching unit (14) may be as shown in Table 4 below.
[0302] [Table 4]
[0303]
[0304] [[4. VPS Calculation and Final Keyword Selection]]
[0305] Afterward, the keyword determination unit (15) can calculate CIS, EFS, LevelFit, Novelty, and Penalty for each candidate keyword k, and calculate VPS for each candidate keyword according to the mathematical formula 1 described above. For example, an example of the calculation result of VPS through the keyword determination unit (15) may be as shown in Table 5 below.
[0306] [Table 5]
[0307]
[0308] At this time, the keyword confirmation unit (15) can confirm the top-N (e.g., 20) keywords among the candidate keywords as final keywords based on the VPS values calculated for each candidate keyword. At this time, the keyword confirmation unit (15) can store the Score Breakdown and Explanation Log of each keyword in the database unit (23). The keyword confirmation unit (15) can generate (confirm) a final keyword set that includes the top 20 candidate keywords as final keywords.
[0309] [[5. Activity Package Organization and Operation]]
[0310] Afterwards, the word family pack generation unit (16) can generate a UnitID-based Word Family Pack for the final set of keywords confirmed above, and the activity package generation unit (17) can generate an activity package (learning activity package) based on the generated Word Family Pack and the final set of keywords confirmed.
[0311] At this time, the activity package generation unit (17) can generate and provide a 4-session package that includes the following activities as an activity package, for example. That is, the activity package generation unit (17) can generate and provide 4 activity packages that include, for example, i) 1st activity package (S1): Keyword Hunt + Meaning Inference, ii) 2nd activity package (S2): Word Family Expansion + Sorting, iii) 3rd activity package (S3): Output Use Task (Speaking / Writing) + Retrieval Quiz, iv) 4th activity package (S4): Transfer Challenge (New Derivative Word Inference / Generation) + Mini Presentation.
[0312] [[6. Log Collection (Exposure / Use / Transfer)]]
[0313] Subsequently, in this experiment, the generated activity packages (i.e., four activity packages) are provided to each of the operation group and the control group, and learning of the target book based on the activity packages can be performed. At this time, the log management unit (18) can collect, record, and manage event logs related to Exposure, Use, and Transfer that occur during the performance of learning (i.e., the performance of the learning activity process) for both the operation group and the control group.
[0314] In this experiment, the operation group organized activities using VAS keyword selection and UnitID-based Word Family Packs, while the control group performed the same number of sessions of activities based on a simple frequency-based list.
[0315] Considering that the above event log is difficult to automatically collect completely in the learner's actual classroom environment, the log management unit (18) in the present invention can collect the event log using one or more of the following: (i) STT converted text of the learner's speech or presentation voice, (ii) OCR recognized text of an image of a learning result (worksheet / evaluation sheet, etc.), and (iii) input memo of a teacher or operator. Subsequently, the recording of the event log can be controlled in a semi-automatic manner, in which an automatic judgment result (candidate) is generated and then confirmed through a verification / correction procedure by a teacher or operator.
[0316] [[7. EffExp and AI Output (Operations) ]]
[0317] Afterwards, the acceleration effect analysis unit (19) can calculate EffExp and AI for the learners of the operation group based on logs collected from the operation group, and in this case, EffExp can be calculated based on the mathematical formula 2 described above, and AI can be calculated based on the mathematical formula 3 described above.
[0318] For example, in this experiment, it is assumed that the log collected / recorded for one learner (learner u1) among the learners in the operation group is [ i) ExposureCount = 10, ii) UseCount = 4, iii) TransferCount = 3 (e.g., 2 form-based transfers + 1 context-based transfer), iv) γ1=1.0, γ2=2.0, γ3=3.0, v) EffExp(u1)= 10 + 2 × 4 + 3×3 = 27, vi) E* = 12, vii) AI(u1)=27 / 12 = 2.25 ].
[0319] For example, examples of logs collected / recorded for an operational group (VAS applied) and values of EffExp and AI calculated for the operational group may be as shown in Table 6 below.
[0320] [Table 6]
[0321]
[0322] [[8. EffExp and AI Output for Control Group (Non-VAS)]]
[0323] In this experiment, the same number of sessions were operated for the control group, but the experiment was conducted without a UnitID-based transfer learning design and a re-recommendation loop.
[0324] The acceleration effect analysis unit (19) can calculate EffExp and AI for the learners in the control group based on logs collected from the control group.
[0325] For example, the logs collected / recorded for the control group (Non-VAS) and the values of EffExp and AI calculated for the control group may be as shown in Table 7 below.
[0326] [Table 7]
[0327]
[0328] [[9. Comparison of Average AI and Acceleration Scales between Operational and Control Groups]]
[0329] According to this experiment, the average EffExp and average AI calculated for the operational and control groups were as shown in Table 8 below.
[0330] [Table 8]
[0331]
[0332] According to Table 8, although the average exposure in this experiment was the same (9.4), it was confirmed that the VAS operating group showed increased Use and Transfer events, resulting in significantly higher EffExp and AI values compared to the control group. This can be said to quantitatively indicate that the UnitID-based Word Family Pack and activity design of the present invention improve acquisition efficiency compared to simple exposure-based learning.
[0333] [[10. Pre / Post Test-Based Effectiveness Verification (Pre-starting Point vs. Post-achievement)]]
[0334] The evaluation department (20) conducted evaluations on the same items (e.g., form-meaning matching, context blanks, production tasks, etc.) before and after operation (i.e., operation of a learning (learning activity) process in which a target book is taught using keyword learning content) for the operation group and the control group, and the results of the evaluation may be as shown in Table 9 below, for example. That is, Table 9 shows examples of average scores for pre-evaluation and post-evaluation.
[0335] [Table 9]
[0336]
[0337] [[ 11. Integrated Results Summary (Log + AI + Pre / Post) ]]
[0338] For example, Table 10 below shows a table that integrates and compares the results of the operational group and the control group in this experiment. Specifically, Table 10 below shows the results of an integrated comparison of log-based metrics (Exposure / Use / Transfer), effective exposure values (EffExp), acceleration metrics (AI), and pre-→post evaluation improvement (Pre→Post) between the operational group (VAS applied) and the control group (Non-VAS).
[0339] [Table 10]
[0340]
[0341] Referring to Table 10 above, the average Exposure (9.4) of the operational and control groups was the same in this experiment; however, it was observed that the operational group's average Use increased by +1.6 and average Transfer increased by +2.0 due to the operation of the UnitID-based Word Family Pack and activity package. Consequently, it was confirmed that the operational group's average EffExp increased by +9.2 (21.0 vs 11.8) and its average AI also increased by +0.77 (1.75 vs 0.98). Furthermore, in the pre- and post-evaluations, the operational group improved by an average of +3.9 points, while the control group improved by an average of +1.7 points, indicating that the operational group showed a greater improvement of +2.2 points compared to the control group.
[0342] These results suggest that even if the simple exposure itself remains the same, an increase in Use and Transfer event logs leads to an increase in EffExp and AI, which can cause significant differences in Pre / Post achievement changes.
[0343] In addition, in the present invention, the evaluation unit (20) analyzes the correlation between the AI(u) for each learner and the change in pre / post evaluation score △Score(u) using a correlation coefficient or a predictive power indicator (R 2 It is possible to calculate (MAE), and through this, it is possible to verify that the log-based acceleration metric (AI) can be consistent with actual achievement changes.
[0344] [[ 12. Calculate Effect Size ]]
[0345] In this experiment, Cohen's d effect size indicator was calculated using the Pre / Post evaluation improvement values (△Score) of the operation group and the control group to evaluate the statistically significant effect size of the difference in achievement improvement between the two groups.
[0346] Cohen's d is the combined standard deviation (SD) of the mean difference between two groups. pooled It is an indicator used to express the effect of achievement improvement as a standardized scale, as a value normalized by ).
[0347] In this experiment, individual improvement values were constructed as example data to match the average improvement values in [Table 10] described above, and the improvement values for the operational and control groups are, for example, ["Operational group improvement value △VAS: [5, 4, 3, 4, 3] → average 3.8, SD 0.84", "Control group improvement △Control: [2, 2, 1, 1, 1] → Mean 1.4, SD It can be equal to 0.55 "].
[0348] In this experiment, the combined standard deviation (SD) was calculated using the standard deviations of the two groups mentioned above. pooled After calculating ) as in Equation 5 below, Cohen's d (effect size) was calculated as in Equation 6 below.
[0349] [Mathematical Formula 5]
[0350]
[0351] [Mathematical Formula 6]
[0352]
[0353] For reference, in mathematical formula 5, SD pooled ε is the combined standard deviation of the two groups' variability, n1 is the sample size of the first group (operation group), n2 is the sample size of the second group (control group), SD1 is the standard deviation of the first group, SD2 is the standard deviation of the second group, SD1 2 is the variance of the first group, SD2 2 represents the variance of the second group, (n1-1) and (n2-1) are the degrees of freedom correction values, and (n1+n2-2) represents the total degrees of freedom.
[0354] In Equation 6, d is Cohen's d (standardized effect size), M1 is the mean of the first group, M2 is the mean of the second group, M1-M2 is the mean difference between the two groups, and SD pooled represents the combined standard deviation of the two groups.
[0355] According to this, SD represents variability within each group, and SD pooled is a value obtained by integrating the variability of the two groups with degrees of freedom correction, and Cohen's d is an effect size indicator calculated by standardizing the difference in means between the two groups based on the above integrated standard deviation (i.e., an indicator that standardizes the difference in achievement improvement between the two groups).
[0356] In this experiment, d It was calculated as 3.38, confirming that the operation group showed a very large effect compared to the control group. An example of the effect size based on pre / post improvement may be as shown in Table 11 below.
[0357] [Table 11]
[0358]
[0359] [[ 13. β / γ Calibration Results ]]
[0360] The β and γ weights used in the present invention are not "arbitrary values" as previously mentioned, but may be values calculated as an optimal combination by pilot data-based calibration. An example of the β / γ calibration result (an example of deriving an optimal combination) according to the present experiment may be as shown in Table 12 below.
[0361] [Table 12]
[0362]
[0363] According to the experiment based on one embodiment of the present invention described above, EffExp and AI presented in the present invention are values calculated based on data such as learner exposure / use / transfer event logs, and can be considered quantitative indicators produced through computer implementation rather than simple subjective evaluations. In particular, it was confirmed that EffExp and AI increased significantly as the Use and Transfer events of the operational group increased compared to the control group, and that the improvement of the operational group was greater in the Pre / Post evaluations.
[0364] Therefore, the present invention can be said to have feasibility and industrial applicability as a technical system that includes not only a simple vocabulary list, but also text-based keyword optimization, UnitID-based etymology cluster matching, learning log-based acceleration indicator calculation, and a feedback optimization loop.
[0365] The present invention can accelerate vocabulary (word) acquisition by extracting keywords of camp books based on etymology clusters, identifying students' prior and subsequent awareness of these keywords, and allowing students (learners) to experience the content of the target book in various contexts based on these keywords (finally determined keywords) at camps, etc.
[0366] Below, we will briefly examine the operation flow of the present invention based on the details described above.
[0367] FIG. 3 is a flowchart of a vocabulary learning method that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs according to an embodiment of the present invention.
[0368] A vocabulary learning method that verifies and optimizes vocabulary acceleration effects based on the etymology clusters and learning logs illustrated in FIG. 3 can be performed by the device (10) described above. Therefore, even if the content described below is omitted, the description of the device (10) can be equally applied to the description of the vocabulary learning method that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs.
[0369] Referring to FIG. 3, in step S11, the keyword determination unit can determine a set of keywords from the book text of the target book.
[0370] Next, in step S12, the log management unit can record and manage event logs that occur during the learning process of the target book using the above keyword set.
[0371] Next, in step S13, the acceleration effect analysis unit can calculate the effective exposure value and acceleration indicator for the learner using the recorded log.
[0372] In the description above, steps S11 to S13 may be further divided into additional steps or combined into fewer steps, depending on an embodiment of the present invention. Additionally, some steps may be omitted as necessary, and the order between steps may be changed.
[0373] A vocabulary learning method for verifying and optimizing vocabulary acceleration effects based on etymology clusters and learning logs according to an embodiment of the present invention may be implemented in the form of program instructions that can be executed through various computer means and recorded on a computer-readable medium. The computer-readable medium may include program instructions, data files, data structures, etc., either individually or in combination. The program instructions recorded on the medium may be those specifically designed and configured for the present invention, or they may be those known and available to those skilled in the art of computer software. Examples of computer-readable recording media include magnetic media such as hard disks, floppy disks, and magnetic tapes; optical recording media such as CD-ROMs and DVDs; magneto-optical media such as floptical disks; and hardware devices specifically configured to store and execute program instructions, such as ROM, RAM, and flash memory. Examples of program instructions include machine code, such as that generated by a compiler, as well as high-level language code that can be executed by a computer using an interpreter, etc. The above hardware device may be configured to operate as one or more software modules to perform the operation of the present invention, and vice versa.
[0374] In addition, the vocabulary learning method for verifying and optimizing vocabulary acceleration effects based on the aforementioned etymology clusters and learning logs can also be implemented in the form of a computer program or application executed by a computer and stored on a recording medium.
[0375] The foregoing description of the present invention is for illustrative purposes only, and those skilled in the art will understand that other specific forms can be easily modified without altering the technical spirit or essential features of the present invention. Therefore, the embodiments described above should be understood as illustrative in all respects and not restrictive. For example, each component described as a single unit may be implemented in a distributed manner, and components described as distributed may likewise be implemented in a combined form.
[0376] The scope of the present invention is defined by the claims set forth below rather than by the detailed description above, and all modifications or variations derived from the meaning and scope of the claims and equivalent concepts thereof should be interpreted as being included within the scope of the present invention.
Claims
Claim 1 An input unit that receives the book text of a target book; a preprocessing unit that performs preprocessing on the book text; a candidate keyword generation unit that generates candidate keywords from the book text; an etymology cluster matching unit that calculates matching reliability based on assigning word group identifiers to the candidate keywords through matching with an etymology cluster database; a keyword confirmation unit that confirms a keyword set from the book text of the target book, wherein the keyword set is confirmed based on keyword priority scores calculated for each candidate keyword; a log management unit that records and manages event logs generated during the learning process of the target book using the keyword set; an acceleration effect analysis unit configured to calculate an effective exposure value and an acceleration indicator regarding a learner using the recorded logs, wherein the effective exposure value is calculated by weighting the number of events related to exposure, usage, and transfer, and the acceleration indicator is calculated by normalizing by the reference number of exposures; and a word family pack generation unit that generates a word family pack for the keyword set confirmed by the keyword confirmation unit, wherein the keyword priority score is calculated based on the following mathematical formula 1, [Mathematical Formula 1] In the above mathematical formula 1, k represents a candidate keyword, CIS represents book comprehension contribution, EFS represents etymological transfer efficiency, LevelFit represents level suitability, Novelty represents novelty, Penalty represents a penalty term, and β1 to β5 represent the weights of each element, and the above acceleration effect analysis unit calculates the above effective exposure value based on the following mathematical formula 2, and calculates the above acceleration index based on the following mathematical formula 3, [Mathematical Formula 2] [Mathematical Formula 3] A vocabulary learning device that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs, wherein in the above mathematical formula 2, u is a learner identifier, γ1 is the weight of an exposure event in a recorded log, γ2 is the weight of a use event in a recorded log, and γ3 is the weight of a transition event in a recorded log, wherein the weights are set to satisfy the relationship γ1 < γ2 < γ3, Exp(u) is the number of exposure events of learner u, Use(u) is the number of use events of learner u, and Trans(u) is the number of transition events of the learner, defined as the sum of the number of form-based transitions and the number of context-based transitions, and in the above mathematical formula 3, E* represents the reference number of exposures, and if the calculated acceleration index exceeds 1, it is determined to be in a delayed acquisition state, and if the calculated acceleration index is less than 1, it is determined to be in an accelerated acquisition state. Claim 2 A vocabulary learning device for verifying and optimizing vocabulary acceleration effects based on etymology clusters and learning logs, wherein, in claim 1, the keyword determination unit is configured to calculate keyword values for multiple elements including Contribution to Book Understanding (CIS), Etymology Transfer Efficiency (EFS), Level Fit, Novelty, and Penalty for each candidate keyword, calculate keyword priority scores for each candidate keyword by weighting the calculated keyword values, and determine a pre-set number of top candidate keywords ranked based on the value of the keyword priority score among the candidate keywords as the final keyword set. Claim 3 delete Claim 4 A vocabulary learning device that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs, further comprising an evaluation unit that performs verification of the reliability of the acceleration indicator based on the results of pre-evaluation and post-evaluation in claim 1. Claim 5 A vocabulary learning device that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs, further comprising a feedback loop management unit that manages a feedback-based optimization loop that performs keyword re-recommendation and parameter update by referring to the acceleration indicator in claim 1. Claim 6 A vocabulary learning method for verifying and optimizing vocabulary acceleration effects based on etymology clusters and learning logs, comprising: an input unit receiving the text of a target book; a preprocessing unit performing preprocessing on the text of the book; a candidate keyword generation unit generating candidate keywords from the text of the book; an etymology cluster matching unit calculating a matching reliability based on assigning word group identifiers to the candidate keywords through matching with an etymology cluster database; a keyword confirmation unit confirming a keyword set from the text of the target book, wherein the keyword set is confirmed based on keyword priority scores calculated for each candidate keyword; a log management unit recording and managing event logs generated during the learning process of the target book using the keyword set; and an acceleration effect analysis unit calculating an effective exposure value and an acceleration indicator regarding a learner using the recorded log, wherein the effective exposure value is calculated by weighting the number of events related to exposure, use, and transfer, and the acceleration indicator is calculated by normalizing the number of standard exposures. and the word family pack generation unit includes the step of generating a word family pack for the keyword set determined by the keyword determination unit, and the keyword priority score is calculated based on the following mathematical formula 1, [Mathematical Formula 1] In the above mathematical formula 1, k represents a candidate keyword, CIS represents book comprehension contribution, EFS represents etymological transfer efficiency, LevelFit represents level suitability, Novelty represents novelty, Penalty represents a penalty term, and β1 to β5 represent the weights of each element, and the above acceleration effect analysis unit calculates the above effective exposure value based on the following mathematical formula 2, and calculates the above acceleration index based on the following mathematical formula 3, [Mathematical Formula 2] [Mathematical Formula 3] A vocabulary learning method that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs, wherein in the above mathematical formula 2, u is a learner identifier, γ1 is the weight of an exposure event in a recorded log, γ2 is the weight of a use event in a recorded log, and γ3 is the weight of a transition event in a recorded log, wherein the weights are set to satisfy the relationship γ1 < γ2 < γ3, Exp(u) is the number of exposure events of learner u, Use(u) is the number of use events of learner u, and Trans(u) is the number of transition events of the learner, defined as the sum of the number of form-based transitions and the number of context-based transitions, and in the above mathematical formula 3, E* represents the reference number of exposures, and if the calculated acceleration index exceeds 1, it is determined to be in a delayed acquisition state, and if the calculated acceleration index is less than 1, it is determined to be in an accelerated acquisition state. Claim 7 In claim 6, the keyword determination unit calculates keyword values for multiple elements including Contribution to Book Comprehension (CIS), Etymological Transfer Efficiency (EFS), Level Fit, Novelty, and Penalty for each candidate keyword, calculates keyword priority scores for each candidate keyword by weighting the calculated keyword values, and determines a pre-set number of top candidate keywords ranked based on the value of the keyword priority score among the candidate keywords as the final keyword set, in a vocabulary learning method that verifies and optimizes vocabulary acceleration effects based on etymology clusters and learning logs. Claim 8 delete Claim 9 A vocabulary learning method for verifying and optimizing vocabulary acceleration effects based on etymology clusters and learning logs, further comprising, in claim 6, a step in which an evaluation unit performs verification of the reliability of the acceleration indicator based on the results of pre-evaluation and post-evaluation. Claim 10 A vocabulary learning method for verifying and optimizing vocabulary acceleration effects based on etymology clusters and learning logs, further comprising, in claim 6, a step in which a feedback loop management unit manages a feedback-based optimization loop that performs keyword re-recommendation and parameter update by referring to the acceleration indicator. Claim 11 A computer-readable recording medium having a program for executing the method of paragraph 6 on a computer.
Citation Information
Patent Citations
The System For Providing Korean Language Education System with Animation
KR102188553B1
Method and system for providing a natural language question and answer service based on information retrieval augmentation for a building energy analytics platform
KR102811533B1
AI-based System for Providing Advertising Media Marketplace Platform
KR102913182B1
