An english vocabulary learning system based on corpus collocation frequency and context construction

Abstract

The application discloses an English vocabulary learning system based on corpus collocation frequency and context construction, and belongs to the technical field of information processing, which comprises a corpus labeling module, a group feedback module, a context updating module and a path generation module.The corpus labeling module is used for analyzing and extracting based on a source corpus, obtaining corpus segments, and performing feature labeling and assignment of teaching value coefficients to generate a labeled corpus set.The group feedback module is used for collecting interactive behaviors of learners and the corpus segments to generate group feedback data.The context updating module is used for dynamically updating the teaching value coefficients to form a dynamic context field.The path generation module is used for performing content retrieval from the dynamic context field based on historical learning data of a target learner to generate an individualized learning path.The application adopts a cooperative working mode of corpus labeling, group feedback, context updating and path generation, can dynamically update the teaching value of learning materials, identify weak points of learners, and generate an individualized learning path, thereby improving the efficiency and effect of English vocabulary learning.

Description

Technical Field

[0001] This invention relates to the field of information processing technology, and in particular to an English vocabulary learning system based on corpus collocation frequency and context construction. Background Technology

[0002] With the rapid development of information technology, especially the improvement of computer hardware performance and the popularization of the Internet, digital information processing systems are increasingly widely used in various industries. In the field of education, particularly in language learning, various computer-aided instruction systems, such as online dictionaries, e-textbooks, and language practice platforms, have become important tools for assisting language learning. These systems typically involve computer information processing technologies such as language data processing, user interface design, and the organization and presentation of learning content.

[0003] In the construction of existing English vocabulary learning systems, the systems tend to use pre-set learning paths or modular learning units, with learners following a predetermined order or category. While some systems collect learners' answers, this is largely limited to statistically analyzing accuracy rates or marking incorrect words, providing only simple feedback on learning progress or effectiveness. Regarding the dynamism and personalization of content, content recommendation typically relies on manual updates or rather coarse algorithmic rules.

[0004] However, existing technical solutions have significant technical shortcomings. First, vocabulary learning systems based on fixed corpora often lack the ability to assess and adaptively adjust the real-time teaching effectiveness of the corpora. Once the pre-set teaching value is determined, it is difficult to dynamically revise it based on learners' actual feedback. This results in some corpora being included in the recommendation sequence even though their actual effectiveness is poor, or high-value corpora not being fully utilized due to insufficient initial weight. Second, existing learning path generation methods are often insufficient in terms of personalization, recommending content based solely on learners' limited tags or simple accuracy statistics. They fail to deeply explore learners' specific knowledge weaknesses and do not incorporate the real-time teaching effectiveness of learning materials into the recommendation consideration, leading to a disconnect between learning content and learners' needs. Furthermore, many systems do not form a complete learning feedback loop, resulting in a lack of self-optimization capabilities in the content database, making it difficult to continuously improve its teaching quality and adapt to constantly changing learning environments. Summary of the Invention

[0005] To address the aforementioned issues, this invention provides an English vocabulary learning system based on corpus collocation frequency and contextual construction. It employs a multi-module collaborative approach involving corpus annotation, group feedback, context updates, and path generation. This allows for dynamic updates to the pedagogical value of learning materials, accurate identification of learners' weaknesses, and the generation of highly personalized learning paths, thereby improving the efficiency and effectiveness of English vocabulary learning.

[0006] The above objectives can be achieved through the following approach: An English vocabulary learning system based on corpus collocation frequency and context construction includes: a corpus annotation module for analyzing and extracting corpus fragments from a pre-set source corpus, and assigning feature annotations and teaching value coefficients to the corpus fragments to generate an annotated corpus set; a group feedback module for collecting interactive behaviors generated when learners interact with the corpus fragments in the annotated corpus set, and generating group feedback data; a context update module for dynamically updating the teaching value coefficient of each corpus fragment based on the group feedback data, forming a dynamic context field, wherein the dynamic context field represents the mapping relationship between the corpus fragments and the real-time teaching value coefficients; and a path generation module for retrieving content from the dynamic context field based on the target learner's historical learning data, and generating a personalized learning path for the target learner.

[0007] Optionally, the corpus annotation module includes: a feature extraction unit, used to extract features based on each corpus segment to obtain static features; a value calculation unit, used to calculate based on the static features to obtain a teaching value coefficient; and a corpus annotation unit, used to associate the static features and the teaching value coefficient with the corresponding corpus segments to form an annotated corpus set.

[0008] Optionally, the group feedback module includes: a task presentation unit, used to present a learning task to the target learner based on the corpus segment; an interaction recording unit, used to record interaction indicators generated by the target learner when completing the learning task, including the correctness of the answer, the length of the answer time, and the type of error; and a group analysis unit, used to aggregate the interaction indicators of multiple learners for the same corpus segment and calculate a group performance index, wherein the group performance index constitutes group feedback data.

[0009] Optionally, the context update module includes: an indicator extraction unit, used to extract from the group feedback data to obtain the group performance indicator for the first corpus segment; a coefficient evolution unit, used to combine the current teaching value coefficient of the first corpus segment with the group performance indicator to perform value evolution and calculate the updated teaching value coefficient; and a context update unit, used to replace the current teaching value coefficient of the first corpus segment with the updated teaching value coefficient to update the dynamic context field.

[0010] Optionally, the system further includes: visually mapping the dynamic context field to generate a context space topology map; identifying hotspot regions based on the context space topology map; and analyzing the static features of corpus fragments within the hotspot regions to identify common learning difficulty topics.

[0011] Optionally, the step of combining the current teaching value coefficient of the first corpus segment with the group performance index to perform value evolution includes: identifying typical learning difficulty types revealed by the first corpus segment based on the error type distribution in the group performance index; evaluating the teaching diagnostic efficacy of the first corpus segment based on the answer time length and answer correctness in the group performance index; and calculating the updated teaching value coefficient based on the typical learning difficulty types and the teaching diagnostic efficacy.

[0012] Optionally, the path generation module includes: a weakness identification unit, used to identify knowledge weakness features from the target learner's historical learning data; a corpus retrieval unit, used to retrieve candidate corpus fragments that match the knowledge weakness features and have a teaching value coefficient higher than a preset coefficient threshold in the dynamic context, forming a candidate learning material set; and a path generation unit, used to sort and organize the candidate learning material set according to the teaching value coefficient of the candidate corpus fragments and their matching degree with the knowledge weakness features, to generate a personalized learning path.

[0013] Optionally, the system further includes: delivering the personalized learning path to the target learner; collecting new interactive behaviors generated by the target learner when interacting with the personalized learning path; incorporating the new interactive behaviors into the generation process of the group feedback data; and initiating the next round of updates to the dynamic context field.

[0014] Optionally, the system further includes: periodically identifying corpus fragments in the dynamic context whose teaching value coefficient is continuously lower than a preset elimination threshold, and obtaining eliminated corpus fragments; archiving the eliminated corpus fragments from the labeled corpus set, and introducing new corpus fragments from the source corpus to the labeled corpus set.

[0015] Based on the same inventive concept, this invention also provides an English vocabulary learning method based on corpus collocation frequency and context construction. The method includes: analyzing and extracting corpus fragments from a pre-set source corpus, and assigning feature annotations and teaching value coefficients to the corpus fragments to generate an annotated corpus set; collecting interactive behaviors generated when learners interact with the corpus fragments in the annotated corpus set to generate group feedback data; dynamically updating the teaching value coefficient of each corpus fragment based on the group feedback data to form a dynamic context field, wherein the dynamic context field represents the mapping relationship between the corpus fragments and the real-time teaching value coefficients; and retrieving content from the dynamic context field based on the target learner's historical learning data to generate a personalized learning path for the target learner.

[0016] Compared with the prior art, the present invention has the following advantages: 1. This invention, through dynamic corpus annotation and a teaching value evaluation mechanism, endows each vocabulary corpus fragment with multi-dimensional and computable teaching attributes. This refined data processing enables the system to establish a high-quality, standardized teaching content library, overcoming the limitations of static content and subjective evaluation in traditional learning materials. This lays a solid foundation for subsequent intelligent recommendation and learning process optimization, improving the efficiency and accuracy of content utilization.

[0017] 2. This invention introduces a group feedback mechanism and a dynamic context update algorithm, enabling the system to capture and quantify learners' performance in actual interactions in real time, and feed this group data back into the dynamic adjustment of the teaching value coefficient. This cyclical update process ensures the activity and adaptability of the system content, enabling timely responses to changes in the teaching environment and common learning difficulties among learners, thus ensuring that the learning content provided by the system remains highly relevant and effective to current teaching needs.

[0018] 3. This invention constructs a highly personalized learning path generation capability. By accurately identifying learners' knowledge weaknesses and combining them with dynamically updated corpus teaching value, it tailors the most suitable learning content sequence for each learner. This not only greatly improves learners' learning efficiency and focus, helping them effectively overcome individual learning obstacles, but also effectively stimulates learning interest and autonomy by providing appropriately challenging content, thereby enhancing the overall learning experience and vocabulary mastery.

[0019] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures pointed out in the description, claims and drawings. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a framework diagram of an English vocabulary learning system based on corpus collocation frequency and context construction, according to an embodiment of the present invention.

[0022] Figure 2This is a schematic diagram of the structure of an English vocabulary learning system based on corpus collocation frequency and context construction according to an embodiment of the present invention.

[0023] Figure 3 This is a distribution diagram of group error types in the corpus fragments of an embodiment of the present invention.

[0024] Figure 4 This is a contextual space topology map and a hotspot region identification map according to an embodiment of the present invention.

[0025] Figure 5 This is a flowchart illustrating an English vocabulary learning system based on corpus collocation frequency and context construction, according to an embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0027] Reference Figure 1 One embodiment of the present invention proposes an English vocabulary learning system based on corpus collocation frequency and context construction. It adopts a multi-module collaborative working method of corpus annotation, group feedback, context update and path generation, which can dynamically update the teaching value of learning materials, accurately identify learners' weaknesses, and generate highly personalized learning paths for them, thereby improving the efficiency and effectiveness of English vocabulary learning.

[0028] The system described in this embodiment specifically includes: The corpus annotation module is used to analyze and extract corpus fragments based on a preset source corpus, and to assign feature annotations and teaching value coefficients to the corpus fragments to generate an annotated corpus set. The group feedback module is used to collect the interactive behaviors generated when learners interact with the corpus fragments in the annotated corpus and generate group feedback data. The context update module is used to dynamically update the teaching value coefficient of each of the corpus segments based on the group feedback data to form a dynamic context field, wherein the dynamic context field represents the mapping relationship between the corpus segments and the real-time teaching value coefficient. The path generation module is used to retrieve content from the dynamic context field based on the target learner's historical learning data and generate a personalized learning path for the target learner.

[0029] Optionally, the corpus annotation module includes: A feature extraction unit is used to extract features based on each of the corpus segments to obtain static features; A value calculation unit is used to calculate the teaching value coefficient based on the static characteristics. The corpus annotation unit is used to associate the static features with the teaching value coefficients to the corresponding corpus segments, thereby forming an annotated corpus set.

[0030] Specifically, such as Figure 2 As shown, the static feature extraction unit first extracts static features. Its input is a single corpus fragment, i.e., a text containing the target vocabulary and its collocations, typically between 15 and 50 words in length. The feature extraction unit integrates a natural language processing pipeline, which sequentially analyzes the corpus fragment. This pipeline first performs lexical analysis to obtain the word frequency of the target vocabulary in a benchmark corpus, such as the Corpus of Contemporary American English, and uses mutual information algorithms or T-score tests to determine the collocation strength of core collocations, selecting strong collocations with mutual information values ​​higher than 3.0 as key features. Next, a dependency parser performs dependency parsing to extract syntactic complexity indicators, such as average dependency distance and maximum syntactic tree depth. Finally, topic modeling analysis is performed to categorize the corpus fragment into at least one of 10 pre-defined teaching topics. All extracted indicators together constitute a numerical vector, serving as the static feature of the corpus fragment, which remains unchanged throughout the learning lifecycle. Then, the value calculation unit calculates the teaching value coefficient based on the extracted static features. The value calculation unit receives the static feature vector output by the feature extraction unit as input. To eliminate the dimensional differences between different features, all feature values ​​are first normalized to the interval between 0 and 1. This is used for calculating the teaching value coefficient. ,have: ; in, This represents the mutual information value of the core collocations in the corpus segment; , These are the minimum and maximum mutual information values ​​among all candidate corpus segments; This is the maximum depth of the syntax tree, usually greater than or equal to 1; The word frequency of the target vocabulary; The language complexity after max-min normalization; As a content relevance feature, the corpus fragments are classified into preset teaching themes, and the preset weight of the theme in the teaching syllabus is directly used as the numerical value. , , These are preset weighting coefficients for collocation, complexity, and content relevance, set by language teaching experts, for example... The weights are set between 0.4 and 0.6 to highlight the importance of collocation learning, and the sum of the three weight coefficients is 1. Finally, the corpus annotation unit generates the annotated corpus. The corpus annotation unit integrates the original text of each corpus segment, its corresponding static feature vector, and the calculated teaching value coefficient into a data record, such as a JSON object. This record is assigned a unique identifier and stored in the database, forming the annotated corpus. Each record in this set not only contains the learning content itself but also encapsulates its initial teaching metadata, laying the foundation for subsequent group feedback collection, dynamic updates of teaching value, and the generation of personalized learning paths.

[0031] For example, suppose the source corpus contains a passage: “Despitenumeroussetbacks,herresilientspiritallowedhertoovercomeadversity.” The corpus annotation module begins processing this passage. First, the feature extraction unit performs feature extraction. This unit analyzes and finds that the target word “resilient” has a frequency of 15 times per million words in the Contemporary American English Corpus, belonging to the low-to-medium frequency category. Through the mutual information algorithm, the mutual information value (MI) of the core collocation “resilientspirit” is found to be 4.5, which is higher than the preset value of 3.0, and therefore it is identified as a strong collocation. Next, through dependency parsing, it is found that the maximum depth of the syntactic tree of this sentence is 5, indicating that it is a relatively complex sentence structure. Finally, topic modeling analysis categorizes this passage into the teaching topic of “personal qualities.” These indicators together constitute the static features of this passage. Subsequently, the value calculation unit calculates the teaching value coefficient based on these static features. Assuming that the minimum mutual information value (MImin) is 1.0 and the maximum mutual information value (MImax) is 7.0 among all candidate corpus fragments, and that the content relevance feature is preset by the teaching syllabus, with the weight of the "personal qualities" theme being 0.8. The preset weight coefficients are: collocation weight = 0.5, complexity weight = 0.2, and content relevance weight = 0.3. First, the normalized language complexity is calculated. The value, after max-min normalization, is 0.6. Initial teaching value coefficient. Finally, the corpus annotation unit integrates the original text "Despitenumeroussetbacks,herresilientspiritallowedhertoovercomeadversity.", its static feature vector, and the calculated pedagogical value coefficient of 0.752 into a single data record, assigns it a unique identifier, and stores it in the database, forming a record in the annotated corpus. Through this process, the unstructured original text is transformed into a structured learning resource containing rich metadata. The calculation of the pedagogical value coefficient comprehensively considers multiple dimensions such as lexical collocation, syntactic complexity, and content theme, providing a scientific and quantitative initial basis for subsequent teaching decisions.

[0032] Optionally, the group feedback module includes: A task presentation unit is used to present a learning task to the target learner based on the corpus fragments. An interactive recording unit is used to record interactive indicators generated by the target learner when completing the learning task, including the correctness of the answer, the length of the answer time, and the type of error. The group analysis unit is used to aggregate the interaction metrics of multiple learners for the same corpus segment and calculate the group performance metrics, wherein the group performance metrics constitute group feedback data.

[0033] Specifically, the task presentation unit first presents learning tasks to the target learners. When a learner logs into the system and selects a learning session, suitable corpus fragments are retrieved from the current dynamic context, and various learning tasks are generated based on these fragments, such as multiple-choice questions, fill-in-the-blank questions, translation exercises, or collocation exercises. Each learning task typically revolves around a core corpus fragment, such as a sentence or paragraph, requiring learners to understand the usage of specific words or their collocation relationships. The task presentation unit ensures a concise and clear task interface, a timing mechanism accurate to the millisecond level, and that the task type can be dynamically adjusted based on preset strategies or the learner's historical performance. For example, simple discrimination tasks are presented to beginner learners, while open-ended output tasks are presented to advanced learners. Then, the interaction recording unit records the interaction metrics generated by the target learners when completing the learning tasks. When learners interact with the tasks provided by the task presentation unit, the interaction recording unit collects the following key metrics in real time: First, the correctness of the answer, recorded as a Boolean value: 1 indicates correct, 0 indicates incorrect. Second, the length of time taken to answer, accurate to the second, representing the time spent by the learner from the display of the task to the submission of the answer. Third, error types: if an answer is incorrect, it will be categorized and marked according to preset error classification rules, such as spelling errors, grammatical errors, misuse of words, and incorrect collocations. These raw interaction metrics, such as {User ID, Corpus Segment ID, Task ID, Answer Correctness, Answer Time, Error Type, Timestamp}, are persistently stored in a real-time behavior log database as the source of raw group feedback data. Figure 3 As shown in the bar chart, this chart displays the percentage of various errors made by learners when responding to the corpus segment "resilientspirit". The chart clearly shows that the proportion of "collocation errors" is as high as 67%, far exceeding other error types such as word semantic misuse; this indicates that the corpus segment can very effectively expose learners' common weaknesses in the collocation use of the word "resilient". Finally, the group analysis unit aggregates the interaction metrics of multiple learners for the same corpus segment to calculate the group performance metrics. The group analysis unit periodically or in real-time retrieves the interaction records of all learners for a specific corpus segment from the behavior log database within a specified time window, such as the past 24 hours or the past week. For each corpus segment, the following group performance metrics are calculated: average accuracy, average response time, and error type distribution. For calculating the average accuracy... ,have: ; in, The total number of learners who completed the task for this corpus segment; Let represent the correctness of the j-th learner's answer, with 0 indicating incorrect and 1 indicating correct. The average response time is calculated using... ,have: ; in, The total number of learners who have completed the task and have sufficient time to answer questions; Let be the response time of the k-th learner. For calculating the error type distribution of error x... ,have: ; in, For certain error types, such as spelling errors, misuse of words, and incorrect collocations; The number of times a specific error type occurs; This represents the total number of errors. These aggregated group performance metrics collectively constitute the group feedback data for this corpus segment, which is stored in a structured manner as input for subsequent value evolution in the context update module.

[0034] For example, a corpus fragment about "resilientspirit" is selected, and the task presentation unit packages it into a learning task, such as "Please select the most suitable word to fill in 'her___spirit': A. strong, B. resilient, C. happy," and presents it to ten target learners. The interaction recording unit then records the interaction metrics generated by these ten learners. For example, User 1 selects B in 8 seconds; User 2 selects A in 12 seconds, and the error type is marked as "incorrect collocation"; User 3 selects B in 6 seconds, and so on. After all ten learners complete the task, the group analysis unit aggregates these interaction metrics. If seven out of ten learners answer correctly and three answer incorrectly, then the average accuracy rate is... Assuming the effective response times for these ten learners are 8, 12, 6, 9, 15, 7, 10, 8, 11, and 13 seconds respectively, then the average response time... (seconds). Of the three incorrect answers, two were classified as "collocation errors" and one as "word meaning misuse". If the number of occurrences of "collocation errors" is 2, and the total number of errors is 3, then the error type distribution of "collocation errors" calculated by Edist[collocation errors] yields a distribution ratio of approximately 0.67. These group performance indicators, namely the average accuracy rate of 0.7, the average response time of 9.9 seconds, and the error type distribution, together constitute the group feedback data for this corpus segment. This process transforms fragmented individual learning behavior data into statistically significant group performance indicators. The group feedback data objectively reflects the difficulty, discrimination, and error-prone points of this corpus segment in a real teaching environment. For example, an accuracy rate of 0.7 indicates that it has a moderate difficulty level, while the high proportion of "collocation errors" accurately reveals the learners' common weaknesses in this language point.

[0035] Optionally, the context update module includes: The indicator extraction unit is used to extract the group performance indicators from the group feedback data to obtain the group performance indicators for the first corpus segment. The coefficient evolution unit is used to combine the current teaching value coefficient of the first corpus segment with the group performance index to perform value evolution and calculate the updated teaching value coefficient. The context update unit is used to replace the current teaching value coefficient of the first corpus segment with the updated teaching value coefficient to update the dynamic context field.

[0036] Specifically, the indicator extraction unit first extracts group performance indicators for the first corpus segment from the group feedback data. When a context update cycle is triggered, or when new group feedback data becomes available, the indicator extraction unit retrieves the corresponding group performance indicators from the group feedback data storage generated by the group feedback module, using a specific corpus segment—for example, a corpus segment recently studied by a large number of learners—as the query condition. These indicators include average accuracy, average response time, and error type distribution. The indicator extraction unit passes this aggregated data as parameters to the subsequent coefficient evolution unit. Then, the coefficient evolution unit combines the current teaching value coefficient of the first corpus segment with the group performance indicators to perform value evolution and calculate the updated teaching value coefficient. Finally, the context update unit replaces the current teaching value coefficient of the first corpus segment with the updated teaching value coefficient to update the dynamic context field. The context update unit submits the identifier of the specific corpus segment and its newly calculated updated teaching value coefficient as an update instruction to the dynamic context field database. The dynamic context field is a key-value pair storage structure, where the key is a unique identifier for a corpus fragment, and the value is its corresponding real-time teaching value coefficient. Through this update operation, the "priority" of the corpus fragment in the system is adjusted in real time, directly affecting the subsequent path generation module's retrieval and recommendation of learning content. This process is typically executed asynchronously in the background to minimize the impact on the user's learning experience, and the update frequency can be configured according to system load and data change rate, such as updating hourly or whenever 100 new interaction data points are accumulated. Furthermore, each corpus fragment in the annotated corpus is periodically traversed, and the teaching value coefficient is dynamically evolved and replaced one by one based on group feedback data, ensuring that the real-time teaching effectiveness of all corpus fragments can be continuously evaluated and accurately adjusted.

[0037] For example, when updating the context, the initial instructional value coefficient for the corpus segment about "resilientspirit," i.e., the first corpus segment, is 0.752. First, the indicator extraction unit extracts the group performance indicators from the group feedback data: an average accuracy rate of 0.7, an average response time of 9.9 seconds, and a "collocation error" proportion of 0.67% in the error type distribution. Next, the coefficient evolution unit performs value evolution based on these indicators. This unit first identifies typical learning difficulty types based on the error type distribution. Since the "collocation error" proportion of 0.67% far exceeds the preset significance threshold of 0.3, "collocation learning" is identified as the typical learning difficulty type revealed by this corpus segment. Then, the instructional diagnostic efficacy is evaluated. The average accuracy rate of 0.7 is within the ideal range of 40% to 70%, indicating that this corpus segment can effectively distinguish learners' levels and has high instructional diagnostic efficacy. Finally, the updated instructional value coefficient is calculated based on these analyses. Ultimately, the context update unit adopts this updated teaching value coefficient, replacing the current teaching value coefficient of the corpus segment in the dynamic context field. By introducing group feedback for value evolution, the teaching value of a corpus segment is no longer a static initial value, but a dynamic value that reflects its true teaching effectiveness. Even if the initial evaluation value is high, the value will be adjusted accordingly if the group performance is poor. In this example, although the accuracy rate is moderate, the value is maintained at a high level because an important "collocation" difficulty was effectively diagnosed, making the value assessment more scientific and closer to reality. The dynamic context field can therefore more accurately reflect the current teaching effectiveness of all corpora.

[0038] Optionally, the system further includes: The dynamic context field is visualized and mapped to generate a context space topology map; Hotspot regions are identified based on the aforementioned contextual spatial topology map. The static features of the corpus fragments in the hotspot area are analyzed to identify common learning difficulties.

[0039] Specifically, the dynamic context field is first visualized and mapped to generate a context space topology map. All active corpus fragments and their corresponding current pedagogical value coefficients are retrieved periodically or as needed from the dynamic context field. Next, using methods such as Principal Component Analysis (PCA) or t-distributed random neighborhood embedding, the high-dimensional static feature vector of each corpus fragment is mapped to a coordinate point in two-dimensional or three-dimensional space. Simultaneously, the pedagogical value coefficient of the corpus fragment is overlaid onto this coordinate point as information about the point's color depth, size, or height. For example, a lower pedagogical value coefficient may result in a redder or larger point. Finally, using a graphics rendering engine such as Plotly or D3.js, all mapped corpus fragment points, along with their pedagogical value information, are drawn into an interactive topology map. Each point in the map represents a corpus fragment, and its visual attributes, such as color and size, reflect its current pedagogical value. Then, hotspot regions are identified based on the context space topology map. Clustering algorithms, such as density-based spatial clustering using DBSCAN, are applied to the generated context space topology map to identify high-density or low-density regions. Specifically, the focus is on identifying corpus fragments with a teaching value coefficient below a preset threshold. For example, this could be one standard deviation below the average teaching value coefficient of all corpus fragments in a dynamic context, or a fixed value such as 30 points, ranging from 0 to 100. Next, the spatial distribution of these low-value corpus fragments on a topological map is analyzed. If a large number of such corpus fragments are geographically adjacent (i.e., in the feature space), this adjacent area is identified as a "hotspot region." A hotspot region is defined by the set of corpus fragments it contains. For example... Figure 4 As shown, this scatter plot simulates a contextual space topology, where each point represents a corpus fragment. Its position in the plot is determined by dimensionality reduction of its high-dimensional static features. The color intensity of the point represents its current teaching value coefficient; the lighter the color, the higher the value. A densely packed area of ​​dark points appears in the lower left corner of the plot, identified as a "hotspot area." Finally, the static features of corpus fragments within hotspot areas are analyzed to identify common learning difficulties. Once one or more hotspot areas are identified, the original static features of all corpus fragments within these areas are traced back. Cross-analysis and statistical aggregation are performed on the features of these corpus fragments, such as the frequency and collocation strength of core vocabulary, syntactic complexity, text topic classification, and the distribution of specific error types. By performing statistical frequency analysis or topic model analysis on these features—for example, identifying a topic tag or syntactic structure that appears abnormally frequently in low-value areas—common language difficulties or teaching themes that cause widespread learning difficulties can be revealed. For example, if the corpus fragments in a certain hot area generally involve verb phrase collocations or complex sentence structures of the subjunctive mood, then the system will identify "poor mastery of verb phrase collocations" or "difficulty in understanding the subjunctive mood" as common learning difficulties.

[0040] For example, a dynamic context field containing hundreds of corpus fragments is visualized and mapped. The high-dimensional static feature vector of the corpus fragment concerning "resilientspirit" is reduced to two-dimensional coordinates (25, 68) using a t-distributed random neighborhood embedding algorithm. On the generated context space topology map, this point is displayed with a specific color and size; for example, the higher the value, the more green the color. Subsequently, hotspot region identification is performed based on this map. The DBSCAN clustering algorithm is run, with the filtering criterion being corpus fragments with a teaching value coefficient below 0.4. In a certain region of the topology map, for example, within the coordinate range of x from 10 to 30 and y from -5 to -20, 15 corpus fragment points with teaching value coefficients all below 0.4 are densely distributed; these points are visually presented as a red area. This set of 15 corpus fragments is then identified as a hotspot region. Finally, the static features of the corpus fragments within this hotspot region are analyzed. Statistical analysis of the static features of these 15 corpus fragments revealed that the core collocation type in 12 of them was phrasal verbs, such as "give up," "look forward to," and "figure out." Furthermore, group feedback data for these fragments generally showed "collocation errors" or "grammatical errors" as the primary error types. Based on this analysis, it was concluded that these learners generally have difficulty using phrasal verbs correctly, thus identifying "poor mastery of phrasal verb collocations" as a common learning challenge faced by the current group. By transforming abstract teaching data into an intuitive visual topology map, educational administrators can readily identify prevalent teaching and learning problems. By identifying hotspots and tracing their static features, common learning difficulties hidden behind numerous learning behaviors can be automatically diagnosed, providing guidance for optimizing course content, adjusting teaching focus, and developing new specialized exercises.

[0041] Optionally, the step of combining the current teaching value coefficient of the first corpus segment with the group performance index to perform value evolution includes: Based on the error type distribution in the group performance indicators, identify the typical learning difficulties revealed by the first corpus segment; The teaching diagnostic efficacy of the first corpus segment is obtained by evaluating the response time and the correctness of the responses based on the group performance indicators. The updated teaching value coefficient is obtained by calculating based on the typical learning difficulty types and the teaching diagnostic effectiveness.

[0042] Specifically, firstly, based on the error type distribution in the group performance index, the typical learning difficulties revealed by the first corpus segment are identified. When the coefficient evolution unit processes the group performance index of a first corpus segment, it focuses on analyzing its error type distribution. It checks which error type has the highest proportion in the error type distribution vector. If this proportion exceeds a preset significance threshold, such as 20% to 40%, then this error type is considered a typical learning difficulty type that this corpus segment can effectively expose. For example, if for a corpus segment, "collocation error" accounts for 35% of all errors, then "collocation error" is marked as a typical learning difficulty type of the corpus segment. Then, the teaching diagnostic efficacy of the first corpus segment is obtained based on the response time and answer accuracy in the group performance index. The teaching diagnostic efficacy is defined as a comprehensive index. It is calculated by evaluating the average accuracy and average response time of the group. An ideal diagnostic material should not have an average accuracy rate that is too high, such as above 90%, or too low, such as below 30%, but rather fall within a moderate range, such as 40%-70%. This is because such material can effectively identify knowledge gaps in learners of average ability. At the same time, a longer average response time may indicate that learners need to engage in deeper thinking, suggesting that the material is challenging. This relates to the diagnostic effectiveness of computational instruction. ,have: ; in, , The accuracy rate weighting coefficient and the response time weighting coefficient are preset by language teaching experts based on the teaching objectives, and their sum is 1. Finally, the updated teaching value coefficient is calculated based on typical learning difficulty types and teaching diagnostic effectiveness. The calculation of the updated teaching value coefficient... ,have: ; in, This is the historical value weighting coefficient, typically set to 0.7 to 0.9. The new feedback weighting coefficient is typically set to 0.1 to 0.3, and the sum of the two is 1. Entropy value for error types, measuring the proportion of low-level errors; , The weights are preset, and the sum of the three is 1. The updated teaching value coefficient is calculated within an effective range, such as 0 to 100. Through this formula, a corpus segment that can effectively diagnose important learning difficulties may have its teaching value coefficient improved even if its average accuracy is not high, because it plays an important "probe" role in the generation of personalized learning paths.

[0043] For example, the coefficient evolution unit is processing a first corpus segment about "resilient spirit," whose current instructional value coefficient is 0.752. First, based on the error type distribution in the group performance indicators, the typical learning difficulty type revealed by this corpus segment is identified. From the group feedback data, it was found that the learning task for this segment produced 40 errors, of which 30 were "collocation errors" and 10 were "word meaning misuse." Based on this, the proportion of "collocation errors" is calculated to be 75%, which significantly exceeds the preset significance threshold of 30%. Therefore, "collocation learning" is marked as a typical learning difficulty type for this corpus segment. Then, the group average accuracy rate is obtained as 0.6, and the average response time is 15 seconds. If the preset accuracy rate weight coefficient is 0.6 and the response time weight coefficient is 0.4, the average response time of 15 seconds, after horizontal comparison and maximum-minimum normalization across all corpus segments, yields a normalized response time of 0.7. Instructional diagnostic effectiveness. Finally, the updated teaching value coefficients are calculated. If the historical value weight coefficient is 0.8, the new feedback weight coefficient is 0.2; the teaching diagnostic effectiveness weight coefficient is 0.7, and the error concentration weight coefficient is 0.3, then the error type entropy value is obtained by analyzing the error type distribution. For a distribution where "collocation errors" account for 75% and "word meaning misuse" accounts for 25%, the entropy value is approximately 0.81. Updated teaching value coefficients. By employing a more refined value evolution model, the assessment of the teaching value of text fragments is deepened from a single consideration of difficulty to a comprehensive evaluation of their diagnostic efficacy. This reduces the problem of erroneously discarding valuable diagnostic materials due to suboptimal accuracy.

[0044] Optionally, the path generation module includes: A weakness identification unit is used to identify knowledge weakness features from the target learner's historical learning data; The corpus retrieval unit is used to retrieve candidate corpus fragments that match the characteristics of the knowledge gaps and have a teaching value coefficient higher than a preset coefficient threshold in the dynamic context field, thereby forming a candidate learning material set. The path generation unit is used to sort and organize the candidate learning material set based on the teaching value coefficient of the candidate corpus fragments and the matching degree with the knowledge weakness features, and generate personalized learning paths.

[0045] Specifically, the weakness identification unit first identifies knowledge weakness features from the target learner's historical learning data. This unit receives all the learner's historical learning records within the system, including the accuracy rate of completed tasks, response time, error type, learning progress, and mastery of specific knowledge points. Statistical analysis, such as response theory, is used to assess the learner's mastery level of the language skills represented by each static feature, such as collocation of specific vocabulary, syntactic understanding, and thematic vocabulary mastery. For example, if a learner consistently exhibits "collocation errors" in multiple corpus segments involving verb collocations, and the average accuracy rate of these errors is below a preset proficiency threshold, such as 60%, then "verb collocation comprehension" will be identified as a knowledge weakness feature for that learner. Knowledge weakness features are represented in a structured manner, such as a list containing weakness tags and corresponding proficiency scores. Then, the corpus retrieval unit retrieves candidate corpus segments that match the knowledge weakness features and have a teaching value coefficient higher than a preset threshold within a dynamic context, forming a candidate learning material set. The corpus retrieval unit receives knowledge weakness features output by the weakness identification unit as query conditions, and simultaneously retrieves the current teaching value coefficients of all corpus segments from a dynamically updated context. For each corpus segment in the dynamic context, it first checks whether its static features contain or match a learner's weakness feature. For example, if the weakness feature is "verb collocation comprehension," it filters out all corpus segments whose static features contain "verb collocation" or whose themes are strongly related to verb collocation. Next, it further filters those corpus segments whose current teaching value coefficients are higher than a preset threshold. For example, this threshold can be set as the average of the teaching value coefficients of all corpus segments in the dynamic context minus one standard deviation, or a fixed value such as 60 points, ranging from 0 to 100. These corpus segments, after two rounds of filtering, constitute a candidate learning material set. They are highly relevant to learners' weaknesses and have been confirmed by group feedback to have good teaching effectiveness or diagnostic potential. Finally, the path generation unit sorts and organizes the candidate learning materials set based on the teaching value coefficient of the candidate corpus fragments and their matching degree with the characteristics of knowledge gaps, generating personalized learning paths. The path generation unit receives the candidate learning materials set as input and calculates a comprehensive recommendation score for each candidate corpus fragment. ,have: ; in, , These are the weighting coefficients for teaching effectiveness and personalized matching, typically set to 0.7 and 0.3 respectively, with a sum of 1. This score represents the matching degree between the corpus segment and the learner's knowledge gap features. This score is obtained by calculating the cosine similarity between the static features of the corpus segment and the gap feature features. The path generation unit sorts the candidate learning material set in descending order based on the comprehensive recommendation score. Then, based on the learner's expected current learning session length and cognitive load management strategies (e.g., avoiding the continuous presentation of overly complex corpora), it selects the top N corpus segments (N is typically between 5 and 15) and organizes them into a sequence, forming a personalized learning path. This path is delivered to the learner in a structured form.

[0046] For example, a learner named Xiaoming wants to improve his vocabulary. First, the weakness identification unit analyzes Xiaoming's learning data from the past week and finds that his average accuracy rate in exercises involving word collocation is only 45%, and the frequency of "collocation error" is much higher than other error types. Therefore, "mastery of word collocation" is identified as Xiaoming's weak knowledge feature. Next, the corpus retrieval unit starts working, searching in a dynamic context using "word collocation" as the keyword. It retrieves the aforementioned corpus fragment about "resilientspirit," which matches the weak knowledge feature because its static features contain strong collocations. At the same time, the updated teaching value coefficient of this corpus fragment is 0.716, higher than the preset coefficient threshold of 0.6. Therefore, this corpus fragment is added to the candidate learning material set. In addition, four other corpus fragments that are also related to collocation and whose teaching value coefficients meet the standard are also selected. Finally, the path generation unit sorts these five candidate corpus fragments. For the "resilientspirit" segment, its teaching value coefficient is 0.716, and its static feature matching degree (Mscore) with Xiaoming's knowledge weakness features, calculated using cosine similarity, is 0.9. Let the teaching effectiveness weight coefficient be 0.7 and the personalized matching weight coefficient be 0.3. Its comprehensive recommendation score... After calculating the comprehensive recommendation score for all five candidate text segments, the path generation unit sorted the scores from highest to lowest and selected the top three to form a sequence, generating a personalized learning path targeting Xiaoming's weakness in "word collocation." By calculating the comprehensive recommendation score, the generated learning path balanced relevance and effectiveness, improving learning efficiency and reducing learners' blind practice amidst massive amounts of content.

[0047] Optionally, the system further includes: The personalized learning path is delivered to the target learner, and new interactive behaviors generated by the target learner when interacting with the personalized learning path are collected. The new interactive behavior is incorporated into the process of generating the group feedback data to initiate the next round of updates to the dynamic context field.

[0048] Specifically, the system first delivers a personalized learning path to the target learner and collects new interactive behaviors generated by the learner while interacting with the personalized learning path. After the path generation module generates a personalized learning path for the target learner, the task presentation unit is responsible for displaying the corpus fragments contained in the path to the learner in the form of interactive learning tasks. At the same time, the interaction recording unit is activated to capture every click, every input, and every answer made by the learner when completing these tasks. Specifically, for each learning task corresponding to each corpus fragment in the path, the following indicators are recorded in real time: answer correctness, i.e., whether the learner has completed the task correctly, usually recorded as a Boolean value, 1 for correct and 0 for incorrect; response time length, i.e. the time spent by the learner from the start of the task to the submission of the answer, accurate to milliseconds; and error type, if the task is not completed correctly, the error will be identified and marked according to the preset error classification criteria. All these detailed interaction data, including the learner's unique ID, the ID of the corpus fragment, the task type, the result, and the timestamp, are collected in real time and stored in the system's raw behavior log database. Then, the new interactive behaviors are incorporated into the group feedback data generation process, initiating the next round of updates to the dynamic context field. Once the interaction recording unit completes the collection and storage of new interactive behavior data, these new records are identified by an internal event-triggered mechanism. Instead of immediately initiating an update to the context field for each individual interactive behavior, these new raw interactive data are aggregated into the data stream of the group feedback module. This means that the group analysis unit periodically—for example, every hour, every half-day, or whenever the amount of new data reaches a preset threshold, such as 1000 records—scans the raw behavior log database and aggregates all relevant data, including these newly collected interactive behaviors, for each corpus segment, thereby updating the group performance indicators for each corpus segment, such as average accuracy, average response time, and error type distribution. Once these group performance indicators are updated, they become the input to the context update module. In this way, the performance of an individual learner on their personalized path, through the group aggregation mechanism, influences the teaching value coefficient of the entire dynamic context field, thereby indirectly triggering a new round of updates to the dynamic context field.

[0049] For example, the aforementioned personalized learning path was delivered to the target learner, Xiaoming. The first learning task in the path was a fill-in-the-blank question designed around the corpus fragment "resilientspirit". After thinking for 8 seconds, Xiaoming correctly filled in "resilient". The task presentation unit presented this learning process, and at the same time, the interaction recording unit was activated and collected a new set of interaction behaviors, which were recorded as {User ID: "Xiaoming", Corpus Fragment ID: "resilient_spirit_001", Correctness: 1, Response Time: 8.0, Error Type: null, Timestamp: "2023-10-27T10:00:08Z"}. This new interaction behavior data was immediately stored in the original behavior log database. Subsequently, this data was not immediately used to update the dynamic context field. Instead, when the preset group analysis period, such as one hour later, arrived, the group analysis unit was activated, and it scanned all new interaction behaviors generated during this period, including Xiaoming's new record. It aggregates Xiaoming's correct record with interaction records from other learners regarding the same corpus segment. This results in slight changes to group performance metrics such as the average accuracy and average response time for that corpus segment. These updated group performance metrics then serve as input to the context update module, initiating the next round of updates to the dynamic context field. By incorporating new interactive behaviors into the batch processing flow of group feedback, both the timeliness of the data and the smoothing of random fluctuations in individual behaviors through aggregation ensure the stability and statistical significance of the dynamic context field updates. This mechanism enables continuous and adaptive learning and evolution, with the effectiveness of the teaching content base constantly increasing with user engagement.

[0050] Optionally, the system further includes: Periodically identify corpus fragments in the dynamic context whose teaching value coefficient is consistently lower than a preset elimination threshold, and obtain eliminated corpus fragments; The eliminated corpus fragments are archived from the annotated corpus, and new corpus fragments are introduced from the source corpus into the annotated corpus.

[0051] Specifically, the system periodically identifies corpus segments in a dynamic context whose teaching value coefficient consistently falls below a preset elimination threshold, thus eliminating these segments. A background daemon process scans the dynamic context at a preset interval, such as weekly or monthly. During each scan, it iterates through all corpus segments in the dynamic context. For each segment, it queries its historical teaching value coefficient record. If the real-time teaching value coefficient of a corpus segment remains below a preset elimination threshold for multiple consecutive update cycles, such as four consecutive weeks, this threshold can be set as a fixed value between 10% and 20% of the system's teaching value coefficient range (e.g., 20 points, ranging from 0 to 100), then the corpus segment is marked as an "eliminated corpus segment." This continuous judgment is to avoid erroneously eliminating potentially valuable corpus segments due to short-term fluctuations. The eliminated corpus segments are then archived from the labeled corpus set. Once a corpus segment is identified as an eliminated corpus segment, the following operations are performed. First, it updates the status field of the corpus fragment in the annotated corpus, marking it as "archived" or "inactive." This means that the path generation module will no longer select these archived corpus fragments when performing content retrieval. Simultaneously, these archived corpus fragments, along with their static features and historical teaching value coefficient data, are migrated to a dedicated "archived database" or logically isolated to free up resources in the active annotated corpus. This archiving rather than complete deletion strategy allows for a re-evaluation of the potential value of these corpus fragments in the future, based on new teaching strategies or large-scale data retrospective analysis. Finally, new corpus fragments are introduced from the source corpus into the annotated corpus. After corpus elimination and archiving are completed, a corpus replenishment mechanism is triggered. This mechanism extracts a batch of new original corpus fragments from the initial, unannotated source corpus—which must continuously receive new, high-quality text materials, such as the latest English news corpus, academic papers, or popular culture texts—randomly or according to a preset strategy, such as prioritizing corpus related to current hot topics. These new corpus fragments are then fed into the corpus annotation module for feature extraction and calculation of pedagogical value coefficients, and are ultimately added as entirely new entries to the annotated corpus. The number of entries introduced is usually slightly more than the number of entries removed, for example, 1.1 to 1.5 times the number of entries removed, in order to gradually expand and optimize the overall size and coverage of the annotated corpus.

[0052] For example, a weekly corpus maintenance task is performed. First, the background daemon periodically identifies obsolete corpus fragments. The process scans the dynamic context field and finds a fragment about the outdated slang "thecat'spajamas," whose pedagogical value coefficient over the past four weeks is 0.18, 0.15, 0.16, and 0.14, all consistently below the preset obsolescence threshold of 0.2. Therefore, this fragment is identified as obsolete. Next, this obsolete fragment is archived from the labeled corpus. Specifically, in the labeled corpus database, the status field of this fragment is updated from "active" to "archived," and its data record is migrated to the archive database. This way, the path generation module will automatically ignore this fragment in subsequent searches. Finally, to fill content gaps, new corpus fragments are introduced from the source corpus. The source corpus recently added a batch of technology news articles about artificial intelligence. A new corpus fragment, "The new algorithm demonstrates remarkable efficiency in processing large datasets," was extracted. This new fragment was then automatically fed into the corpus annotation module, beginning its lifecycle of feature extraction, initial value calculation, and annotation. Ultimately, it was added to the annotated corpus as a brand new, active entry, thus maintaining the corpus's size and relevance. By automatically discarding low-value, outdated corpora, the bloated and rigid nature of teaching resources was reduced.

[0053] Based on the same inventive concept, such as Figure 5 As shown, the present invention also provides an English vocabulary learning method based on corpus collocation frequency and context construction, the method comprising: Based on a pre-set source corpus, analysis and extraction are performed to obtain corpus fragments, and feature annotation and teaching value coefficients are assigned to the corpus fragments to generate an annotated corpus set; Collect the interactive behaviors generated when learners interact with the corpus fragments in the annotated corpus, and generate group feedback data; Based on the group feedback data, the teaching value coefficient of each of the corpus segments is dynamically updated to form a dynamic context field, wherein the dynamic context field represents the mapping relationship between the corpus segments and the real-time teaching value coefficient. Based on the target learner's historical learning data, content is retrieved from the dynamic context field to generate a personalized learning path for the target learner.

[0054] It should be noted that the electrical connections between the various units described above do not necessarily represent direct or indirect connections. Any indirect connection method can be applied to the embodiments of the present invention as long as it achieves the purpose of the present invention. The above descriptions are merely exemplary embodiments of the present invention and should not be construed as limiting the scope of the present invention.

[0055] All equivalent changes and modifications made in accordance with the teachings of this invention are still within the scope of this invention. Those skilled in the art will readily conceive of other embodiments of this invention upon considering the disclosure in this specification. This application is intended to cover any variations, uses, or adaptations of this invention that follow the general principles of this invention and include common knowledge or conventional techniques in the art not described herein.

Claims

1. An English vocabulary learning system based on corpus collocation frequency and context construction, characterized in that, The system includes: The corpus annotation module is used to analyze and extract corpus fragments based on a preset source corpus, and to assign feature annotations and teaching value coefficients to the corpus fragments to generate an annotated corpus set. The group feedback module is used to collect the interactive behaviors generated when learners interact with the corpus fragments in the annotated corpus and generate group feedback data. The context update module is used to dynamically update the teaching value coefficient of each of the corpus segments based on the group feedback data to form a dynamic context field, wherein the dynamic context field represents the mapping relationship between the corpus segments and the real-time teaching value coefficient. The path generation module is used to retrieve content from the dynamic context field based on the target learner's historical learning data and generate a personalized learning path for the target learner.

2. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 1, characterized in that, The corpus annotation module includes: A feature extraction unit is used to extract features based on each of the corpus segments to obtain static features; A value calculation unit is used to calculate the teaching value coefficient based on the static characteristics. The corpus annotation unit is used to associate the static features and the teaching value coefficient with the corresponding corpus segments to form an annotated corpus set.

3. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 2, characterized in that, The group feedback module includes: A task presentation unit is used to present a learning task to the target learner based on the corpus fragments. An interactive recording unit is used to record interactive indicators generated by the target learner when completing the learning task, including the correctness of the answer, the length of the answer time, and the type of error. The group analysis unit is used to aggregate the interaction metrics of multiple learners for the same corpus segment and calculate the group performance metrics, wherein the group performance metrics constitute group feedback data.

4. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 3, characterized in that, The context update module includes: The indicator extraction unit is used to extract the group performance indicators from the group feedback data to obtain the group performance indicators for the first corpus segment. The coefficient evolution unit is used to combine the current teaching value coefficient of the first corpus segment with the group performance index to perform value evolution and calculate the updated teaching value coefficient. The context update unit is used to replace the current teaching value coefficient of the first corpus segment with the updated teaching value coefficient to update the dynamic context field.

5. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 4, characterized in that, The system also includes: The dynamic context field is visualized and mapped to generate a context space topology map; Hotspot regions are identified based on the aforementioned contextual spatial topology map. The static features of the corpus fragments in the hotspot area are analyzed to identify common learning difficulties.

6. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 4, characterized in that, The value evolution process, which combines the current teaching value coefficient of the first corpus segment with the group performance index, includes: Based on the error type distribution in the group performance indicators, identify the typical learning difficulties revealed by the first corpus segment; The teaching diagnostic efficacy of the first corpus segment is obtained by evaluating the response time and the correctness of the responses in the group performance indicators. The updated teaching value coefficient is obtained by calculating based on the typical learning difficulty types and the teaching diagnostic effectiveness.

7. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 1, characterized in that, The path generation module includes: The weakness identification unit is used to identify knowledge weakness features from the target learner's historical learning data; The corpus retrieval unit is used to retrieve candidate corpus fragments that match the characteristics of the knowledge gaps and have a teaching value coefficient higher than a preset coefficient threshold in the dynamic context field, thereby forming a candidate learning material set. The path generation unit is used to sort and organize the candidate learning material set based on the teaching value coefficient of the candidate corpus fragments and the matching degree with the knowledge weakness features, and generate personalized learning paths.

8. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 7, characterized in that, The system also includes: The personalized learning path is delivered to the target learner, and new interactive behaviors generated by the target learner when interacting with the personalized learning path are collected. The new interactive behavior is incorporated into the process of generating the group feedback data to initiate the next round of updates to the dynamic context field.

9. The English vocabulary learning system based on corpus collocation frequency and context construction according to claim 1, characterized in that, The system also includes: Periodically identify corpus fragments in the dynamic context whose teaching value coefficient is consistently lower than a preset elimination threshold, and obtain eliminated corpus fragments; The eliminated corpus fragments are archived from the annotated corpus, and new corpus fragments are introduced from the source corpus into the annotated corpus.

10. An English vocabulary learning method based on corpus collocation frequency and context construction, applied to an English vocabulary learning system based on corpus collocation frequency and context construction as described in any one of claims 1-9, characterized in that, The method includes: Based on a pre-set source corpus, analysis and extraction are performed to obtain corpus fragments, and feature annotation and teaching value coefficients are assigned to the corpus fragments to generate an annotated corpus set; Collect the interactive behaviors generated when learners interact with the corpus fragments in the annotated corpus, and generate group feedback data; Based on the group feedback data, the teaching value coefficient of each of the corpus segments is dynamically updated to form a dynamic context field, wherein the dynamic context field represents the mapping relationship between the corpus segments and the real-time teaching value coefficient. Based on the target learner's historical learning data, content is retrieved from the dynamic context field to generate a personalized learning path for the target learner.

Images