Student learning condition analysis method and system based on hybrid engine

Abstract

The invention provides a student learning condition analysis method and system based on a hybrid engine, and relates to the field of education information processing, and the system comprises a data collection module which is used for obtaining learning data of students from a plurality of data sources, carrying out the preprocessing of the learning data of the students, and generating the preprocessed learning data; the question-and-answer mode module comprises a question-and-answer interface based on natural language interaction, and the question-and-answer interface is used for receiving questions of the user; the hybrid engine module is used for performing structured query and intelligent response on the user questions received by the question and answer interface through a hybrid engine based on the preprocessed learning data, and generating query results corresponding to the user questions; and the question and answer mode module is also used for displaying a query result corresponding to the question asked by the user. The method has the advantage of improving the intelligence level of learning condition analysis.

Description

Technical Field

[0001] This invention relates to the field of educational information processing, and in particular to a student learning analysis method and system based on a hybrid engine. Background Technology

[0002] With the continuous advancement of educational informatization, various teaching management platforms, online homework systems, and examination systems have been widely deployed, and educational data resources are constantly being enriched. Current student learning analysis technologies primarily rely on the data provided by these platforms to summarize and statistically analyze student attendance, academic performance, activity participation, and other behaviors.

[0003] However, in practical teaching applications, this type of technology still has significant shortcomings: 1. Data is scattered and fragmented: Teaching-related data is scattered across multiple platforms and lacks unified integration. Teachers need to manually switch between multiple systems and copy and paste data, which is inefficient and prone to errors.

[0004] 2. Limited analytical dimensions and delayed response: Existing systems mostly remain at the level of static charts or grade lists, making it difficult to achieve diagnosis and tracking at the knowledge point level. Moreover, they are mostly post-class summary analyses, lacking the ability to provide dynamic feedback during the teaching process.

[0005] 3. Lack of intelligent interaction methods: Most systems require teachers to set query conditions and manually select fields, which has a high operating threshold and makes it difficult for teachers without technical backgrounds to use them efficiently, thus affecting their widespread adoption.

[0006] 4. Unable to support personalized teaching decisions: The current system is unable to generate individualized teaching suggestions based on student profiles, and cannot proactively identify problems or assist in teaching intervention.

[0007] Therefore, there is a need to provide a student learning analysis method and system based on a hybrid engine to improve the intelligence level of learning analysis. Summary of the Invention

[0008] This invention provides a student learning analysis system based on a hybrid engine, comprising: a data acquisition module for acquiring student learning data from multiple data sources and preprocessing the student learning data to generate preprocessed learning data; a question-and-answer module, including a question-and-answer interface based on natural language interaction, the question-and-answer interface being used to receive user questions; a hybrid engine module for performing structured queries and intelligent responses to user questions received by the question-and-answer interface based on the preprocessed learning data through a hybrid engine, and generating query results corresponding to the user questions; the question-and-answer module is also used to display the query results corresponding to the user questions.

[0009] Furthermore, the student's learning data includes the student's behavioral trajectory during the learning process, learning outcomes, knowledge mastery assessment, and teacher-student interaction feedback data.

[0010] Furthermore, the data acquisition module preprocesses the students' learning data to generate preprocessed learning data, including: standardizing fields, normalizing timestamps, associating user identities with courses, and mapping the students' learning data to generate preprocessed learning data.

[0011] Furthermore, the hybrid engine module, through the hybrid engine, performs structured queries and intelligent responses on user questions received from the question-and-answer interface based on preprocessed learning data, and generates query results corresponding to the user questions. This includes: performing intent recognition on user questions received from the question-and-answer interface and generating intent recognition results; performing entity recognition and standardization on user questions received from the question-and-answer interface and generating entity recognition results; performing template matching based on intent recognition results and obtaining matching templates; filling in the matching templates based on entity recognition results and generating structured query statements; and generating query results corresponding to the user questions based on the structured query statements.

[0012] Furthermore, the hybrid engine module performs intent recognition on the user questions received by the question-and-answer interface and generates intent recognition results, including: performing text preprocessing on the user questions received by the question-and-answer interface to generate text preprocessing results; understanding the user intent based on the text preprocessing results using a large language model; if the large language model returns a valid intent and the intent is in a predefined list, then generating intent recognition results based on the output of the large language model; if the large language model does not return a valid intent or the intent is not in the predefined list, then matching is performed using a predefined mapping relationship between keywords and intents to generate intent recognition results.

[0013] Furthermore, the hybrid engine module performs entity recognition and standardization on user questions received from the question-and-answer interface to generate entity recognition results, including: extracting entities from user questions received from the question-and-answer interface; performing text standardization on the extracted entities to generate text standardization results; and mapping the text standardization results to database primary keys or codes.

[0014] Furthermore, the hybrid engine module performs template matching based on the intent recognition result to obtain the matched template, including: determining a set of candidate templates based on the intent recognition result; ranking the templates included in the set of candidate templates by score; and obtaining the matched template based on the score ranking result.

[0015] Furthermore, the hybrid engine module fills the matching template based on the entity recognition results to generate a structured query statement, including: filling the slots in the matching template based on the entity recognition results to generate a structured query statement.

[0016] Furthermore, the question-and-answer module displays the query results corresponding to the user's question, including: structuring the query results into a table, trend chart, or heatmap, and displaying the structured results.

[0017] This invention provides a student learning analysis method based on a hybrid engine, applied to the aforementioned student learning analysis system based on a hybrid engine. The method includes: acquiring student learning data from multiple data sources and preprocessing the student learning data to generate preprocessed learning data; a natural language interactive question-and-answer interface for receiving user questions; using the hybrid engine, based on the preprocessed learning data, performing structured queries and intelligent responses on the user questions received by the question-and-answer interface, and generating query results corresponding to the user questions; and displaying the query results corresponding to the user questions.

[0018] Compared with existing technologies, the student learning analysis method and system based on a hybrid engine provided by this invention has at least the following beneficial effects: This system introduces a natural language question-and-answer mechanism into the learning analysis system for the first time. The question-and-answer module is equipped with a question-and-answer interface based on natural language interaction. Teachers do not need to master database or report operation skills. They can make requests through the question-and-answer interface simply by "dialogue". This greatly improves the interactivity and usability of the system and lowers the threshold for use.

[0019] The hybrid engine module introduces a hybrid architecture of NL2SQL / NL2Code for the education field, combining SQL templates with Pandas code generation. This ensures high question coverage and execution flexibility, effectively handling complex educational query needs and providing accurate structured queries and intelligent responses to user questions received by the Q&A interface.

[0020] The data acquisition module obtains and preprocesses student learning data from multiple data sources, and the hybrid engine module operates based on this data. Simultaneously, the system constructs a unified abstraction and semantic fusion model for multi-source heterogeneous data, enabling unified modeling and intelligent aggregation of various learning data. Furthermore, the question-and-answer module supports displaying query results in various formats such as heatmaps, trend charts, and tables, and also supports follow-up questions and interactive recommendations, helping teachers comprehensively understand student situations and improve the efficiency of teaching decision-making.

[0021] A data privacy protection scheme based on a sandbox mechanism was designed. Query execution is carried out in a controlled environment, and sensitive fields are anonymized, which complies with the requirements of education data security specifications and ensures the security of students' learning data. Attached Figure Description

[0022] This specification will be further described by way of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting; in these embodiments, the same reference numerals denote the same structures, wherein: Figure 1 This is a schematic diagram of a hybrid engine-based student learning analysis system according to some embodiments of this specification; Figure 2 This is a flowchart illustrating the process of generating query results corresponding to user queries, based on some embodiments of this specification. Figure 3 This is a schematic diagram illustrating the entity extraction results according to some embodiments of this specification; Figure 4 This is a flowchart illustrating a student learning analysis method based on a hybrid engine, according to some embodiments of this specification. Detailed Implementation

[0023] To more clearly illustrate the technical solutions of the embodiments in this specification, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are merely some examples or embodiments of this specification. For those skilled in the art, these drawings can be applied to other similar scenarios without creative effort. Unless obvious from the context or otherwise specified, the same reference numerals in the drawings represent the same structures or operations.

[0024] Figure 1 This is a schematic diagram of a hybrid engine-based student learning analysis system according to some embodiments of this specification, such as... Figure 1 As shown, a student learning analysis system based on a hybrid engine may include a data acquisition module, a question-and-answer module, and a hybrid engine module.

[0025] The data acquisition module is used to obtain students' learning data from multiple data sources, preprocess the students' learning data, and generate preprocessed learning data.

[0026] The student learning data includes students' behavioral patterns during the learning process, learning outcomes, knowledge mastery assessments, and teacher-student interaction feedback data.

[0027] Specifically, the data acquisition module undertakes the crucial task of centrally collecting student learning-related data, aiming to build a comprehensive and unified learning data system. It interfaces with multiple data sources corresponding to different functional modules within the same teaching platform, specifically covering online assignments, exam management, class attendance, course progress, and interactive Q&A. Through deep integration with these functional modules, this module can comprehensively and in real-time extract various types of student data during the learning process.

[0028] Student learning data encompasses a rich variety of content. Behavioral trajectory data details students' specific operations and activities during the learning process, such as homework completion status (submission time, quality, etc.); answer paths (reflecting students' thought processes and logical order when solving problems); attendance information (reflecting students' punctuality and frequency of participation in class); and interaction frequency (covering the number of interactions between students and teachers, and among classmates on the learning platform). Learning outcome data focuses on various test and exam scores, intuitively presenting students' mastery of knowledge and learning effectiveness at different learning stages. Knowledge mastery assessment data is a comprehensive evaluation and analysis of students' knowledge mastery based on their performance in homework, exams, and interactions, using scientific algorithms and models. Teacher-student interaction feedback data includes students' evaluations and suggestions on teachers' teaching methods, as well as teachers' feedback on students' learning status and problem-solving during interactions.

[0029] Preferably, the data acquisition module preprocesses the students' learning data to generate preprocessed learning data, including: The student learning data is standardized by fields, normalized by timestamps, associated with user identities, and mapped to course structures to generate preprocessed learning data.

[0030] Specifically, field standardization is performed to ensure that data fields collected from different functional modules have unified naming rules and formats, facilitating subsequent data processing and analysis. Timestamp normalization converts time information from various data sources into a standardized time format, eliminating discrepancies in time records and ensuring data consistency across the time dimension. User identity association and course structure mapping are performed to clearly identify the student's identity and corresponding course information, enabling accurate association between data and specific students and courses. This ensures semantic consistency across multiple modules and allows for correlated analysis. These preprocessing operations provide high-quality data support for subsequent functions such as intelligent question answering, user profile modeling, and visual feedback, helping the teaching platform achieve more precise and personalized teaching services and decision support.

[0031] The question-and-answer module includes a natural language-based interactive interface for receiving user questions.

[0032] For example, teachers can directly input questions such as: "What is the attendance rate of students in a certain class?" or "How well have students mastered the knowledge points in Unit 3?" The hybrid engine module is used to perform structured queries and intelligent responses to user questions received from the question-and-answer interface based on preprocessed learning data, and to generate corresponding query results. The hybrid engine can be an NL2SQL / NL2Code hybrid engine.

[0033] Figure 2 This is a flowchart illustrating the process of generating query results corresponding to user queries, as shown in some embodiments of this specification. Figure 2 As shown, it specifically includes: Perform intent recognition on user questions received from the Q&A interface and generate intent recognition results; Perform entity recognition and standardization on user questions received from the question-and-answer interface, and generate entity recognition results; Based on the intent recognition results, template matching is performed to obtain the matching template; Based on the entity recognition results, the matched template is populated to generate a structured query statement; Based on structured query statements, generate query results corresponding to user queries.

[0034] Preferably, the hybrid engine module performs intent recognition on user questions received by the question-and-answer interface and generates intent recognition results, including: Perform text preprocessing on user questions received from the Q&A interface to generate text preprocessing results; Understanding user intent through large language models based on text preprocessing results; If the large language model returns a valid intent and the intent is in the predefined list, then an intent recognition result is generated based on the output of the large language model. If the large language model does not return a valid intent or the intent is not in the predefined list, then the predefined keyword-intent mapping relationship is used for matching to generate intent recognition results.

[0035] Specifically, text preprocessing of user questions received by the Q&A interface can include: Sentence segmentation: A long text input by the user is logically divided into multiple independent sentences according to their sentence structure. This helps to more accurately analyze the meaning of each sentence and avoids misunderstandings caused by excessively long or complex sentences. For example, if a user inputs "I want to know the course schedule and the deadline for submitting assignments," after sentence segmentation, it will result in two independent sentences: "I want to know the course schedule" and "The deadline for submitting assignments," facilitating subsequent intent analysis.

[0036] Stop-word removal: Stop words refer to words that frequently appear in text but do not contribute substantially to expressing the core intention, such as "的" (de), "是" (shi), "在" (zai), etc. Removing these stop words can reduce noise interference in the text, highlight key information, and improve the efficiency and accuracy of intention recognition. For example, for the sentence "我想知道这门课程的考试时间" (I want to know the exam time of this course), removing the stop word "的" (de) does not affect the expression of the core intention of the sentence and instead makes the text more concise and clear.

[0037] Spelling correction: Considering that users may make spelling mistakes during input, the hybrid engine module checks and corrects the spelling in the text. By comparing with the built-in dictionary or spelling rules, it corrects incorrect spellings into correct words to ensure the accuracy and readability of the text. For example, if the user inputs "我想查寻课程资料" (I want to search for course materials), the module will correct "查寻" (chaxun) to "查询" (chaxun) to avoid incorrect intention understanding due to spelling mistakes. After these preprocessing operations, a text preprocessing result is generated, providing clean and standardized data for subsequent intention understanding.

[0038] After completing the text preprocessing, the hybrid engine module inputs the text preprocessing result into the large language model.凭借其强大的语言理解和生成能力,对用户提问进行深度分析,尝试理解用户所表达的意图。大语言模型通过学习海量的文本数据,掌握了丰富的语言知识和语义关系,能够根据输入的文本上下文,推断出用户可能想要实现的目标或获取的信息。例如,对于预处理后的文本“查询明天的课程安排”,大语言模型可以理解为用户希望获取关于明天课程的具体信息,如课程名称、上课时间、上课地点等。

[0039] If the large language model returns a valid intention and this intention is in the predefined intention list, it means that the large language model has accurately understood the intention of the user's question and that this intention is a common intention type that the system can handle and respond to. In this case, the hybrid engine module will directly generate the final intention recognition result based on the output of the large language model. For example, if the predefined list contains the intention of "查询课程安排" (querying the course schedule), when the large language model recognizes that the intention of the user's question is "查询课程安排" (querying the course schedule), the module will generate the corresponding intention recognition result, providing a clear direction for subsequent answer generation.

[0040] It should be noted that there is an incomplete sentence in the translation of . The Chinese part "凭借其强大的语言理解和生成能力,对用户提问进行深度分析,尝试理解用户所表达的意图。" is missing in the English translation. The corrected English translation for should be: After completing the text preprocessing, the hybrid engine module inputs the text preprocessing result into the large language model. With its powerful language understanding and generation capabilities, it conducts in-depth analysis on the user's question and attempts to understand the intention expressed by the user. The large language model has mastered rich language knowledge and semantic relationships through learning a vast amount of text data and can infer the goals that the user may want to achieve or the information they may want to obtain based on the input text context. For example, for the preprocessed text "查询明天的课程安排" (Query the course schedule for tomorrow), the large language model can understand that the user hopes to obtain specific information about tomorrow's courses, such as course names, class times, and class locations.If the large language model fails to return a valid intent, it may be because the user's question is expressed in a unique, ambiguous, or manner that exceeds the model's comprehension; or the intent returned by the large language model may not be in the predefined list, meaning the intent is a type that the system has not yet clearly defined or processed. In this case, the hybrid engine module will use the predefined keyword-intent mapping relationship for matching. By extracting keywords from the user's question and comparing them with the pre-set keyword-intent correspondence, the most matching intent is found, and an intent recognition result is generated. For example, if a predefined mapping relationship exists between the keyword "grades" and the intent "check exam grades," when a user asks "I want to see my grades," the large language model may fail to accurately identify the intent, but the module can extract the keyword "grades" and generate the intent recognition result "check exam grades" based on the mapping relationship.

[0041] Understandably, by combining a large language model with predefined keyword mapping relationships, the hybrid engine module can more comprehensively and accurately understand the user's intent in asking questions, improve the intelligence and practicality of the question-and-answer system, and provide users with more accurate and effective answers.

[0042] For example, a teacher query (such as "What is Zhang San's error rate in English?") can be directly mapped to "Query Error Rate" in the following predefined intent name: 1. Mastery (check the error rate and average score); 2. Progress (Check unit completion rate, learning progress); 3. Attendance (check attendance rate, number of late arrivals); 4. Exam (check score distribution and class ranking); 5. Assignments (query submission rate, average assignment score).

[0043] Preferably, the hybrid engine module performs entity recognition and standardization on user questions received by the question-and-answer interface, generating entity recognition results, including: Extract entities from user questions received from the Q&A interface; The extracted entities are text-normalized to generate text-normalized results; Map the text standardization results to database primary keys or codes.

[0044] Specifically, the large language model leverages natural language processing capabilities to perform deep analysis of user queries. Based on learning from large amounts of text data, the large language model identifies entities with specific meanings in the query and returns a list of entities in a structured JSON format.

[0045] After extracting entities, to ensure that entities have consistent semantics in different contexts and expressions and facilitate accurate matching with data in the system database, the hybrid engine module performs text normalization on the extracted entities. This process mainly involves two aspects: Normalization of time range: For relative time expressions like "the last two weeks", the module calls the date engine and combines the current system time to convert it into specific start and end dates in the format of "YYYY - MM - DD". For example, if the current date is October 15, 2024, "the last two weeks" will be converted to "2024 - 10 - 01" to "2024 - 10 - 15".

[0046] Normalization of term mapping: To unify the terms used in the system and avoid data matching errors caused by inconsistent expressions, the hybrid engine module performs term mapping processing.

[0047] For some terms with clear corresponding relationships, the hybrid engine module will perform conversion according to the predefined exact matching vocabulary. For example, it will exactly map "Class 3, Grade 8" to "Grade 8 Class 3" and "unit2" to "Unit 2".

[0048] When encountering terms that cannot be exactly matched, the hybrid engine module will adopt fuzzy matching techniques, such as calculating the edit distance and using word vector similarity, etc., to find the most similar term to the original entity from the candidate vocabulary as the normalization result. For example, if the user enters "English class", the hybrid engine module may normalize it to "English" through fuzzy matching.

[0049] After text normalization, the hybrid engine module needs to associate the normalized entities with the data in the system database to obtain accurate information to answer user questions. This process is achieved by mapping the text normalization result to the database primary key or code.

[0050] If a corresponding record can be found in the database for the normalized entity, the hybrid engine module will map the entity to the corresponding database primary key or code. For example, it will map the student name "Zhang San" to the student ID "student_id = 10023" in the database and the unit name "Unit 2" to the unit code "unit_code = \"U002\"". Through this mapping, the hybrid engine module can quickly locate the data related to the user's question in the database, improving the query efficiency and accuracy.

[0051] If the standardized entity fails to find a match in the database, it means the system may not have a record for that entity, or the user's question may be unclear. In this case, the hybrid engine module will retain the original entity information and initiate a multi-round clarification mechanism. By further interacting with the user and requesting more relevant information, it clarifies the user's intent and ensures an accurate answer to the user's question.

[0052] For example, extract all the required entities from the query: 1. Student (student_name); 2. Class; 3. Subject; 4. Unit / knowledge point; 5. Exam / Assignment Name; 6. Attendance events (attendance_event); 7. Time range (date_range).

[0053] As an example only, entity extraction results can be as follows: Figure 3 As shown.

[0054] Preferably, the hybrid engine module performs template matching based on the intent recognition result to obtain the matched template, including: Based on the intent recognition results, determine the set of candidate templates; The templates in the candidate template set are scored and sorted. Based on the ranking results, obtain the matching template.

[0055] Specifically, after identifying the user's intent, the hybrid engine module retrieves and loads the corresponding template set from a pre-built template library based on the identified intent name (intent_name), denoted as T = {T1, T2, …}. The template library is carefully categorized and organized according to different intents, with multiple placeholder templates stored under each intent. These templates are standardized query or answer frameworks designed for specific intents, and the placeholders are used to subsequently populate entity information extracted from the user's question to generate a personalized answer. For example, when the intent identification result is "query student error rate," the module loads the template set related to "query student error rate" under the "Mastery" intent group from the template library. In this way, the hybrid engine module can quickly locate the template set related to the user's intent, providing a foundation for subsequent filtering and matching.

[0056] After obtaining the candidate template set, the hybrid engine module will adopt different processing strategies based on the number of templates in the set: Single template case: If there is only one template in the candidate template set, the hybrid engine module will directly use that template without scoring or sorting. This is because there are no other templates available for comparison, making it the only usable option.

[0057] In the case of multiple templates: when the candidate template set contains multiple templates, the hybrid engine module will enter the scoring and ranking stage to determine which template matches the user's question the most. The specific scoring and ranking process is as follows: Question Vectorization: The module first vectorizes the teacher's input question (which, in practical applications, can also be extended to the user's input question) using the bge-embedding model. The bge-embedding model is an advanced text vectorization model that can convert text into high-dimensional vectors that capture the semantic information of the text. By converting the question into vector form, the hybrid engine module can represent and analyze the question in vector space. Next, the hybrid engine module also uses the same bge-embedding model to vectorize each template in the candidate template set, obtaining the vector corresponding to each template. The module calculates the cosine similarity between the vector of the input question and the vector of each candidate template, and the hybrid engine module checks the highest similarity value obtained. If the highest similarity is lower than a pre-set threshold (e.g., 0.8), it is considered that no template matches the user's question, and the hybrid engine module will enter the exception handling process, such as prompting the user to re-enter the question or providing more detailed information. If the highest similarity is greater than or equal to the threshold, the template with the highest similarity is selected as the best matching template.

[0058] After scoring and sorting, the hybrid engine module obtains matching templates based on the sorting results. If a template exists that meets the matching requirements of the user's question (i.e., the highest similarity is greater than or equal to the threshold), the hybrid engine module will select that template as the final matching template.

[0059] When the system encounters situations where the semantics cannot match any SQL template, such as complex problems (involving multiple conditions, multiple data table relationships, and complex logic) or data linkage problems (the answer to one problem depends on the calculation result of another problem, forming a linkage relationship between data), the system will fall back to the path from natural language to Pandas code. Before actually executing the Pandas code generated by the large language model, the system will place it in a sandbox environment. In the sandbox environment, the system provides a specific, access-controlled DataFrame dataset. This dataset is filtered and processed, containing only data relevant to the current problem, and access and manipulation of the data are strictly restricted. For example, if the problem only involves student grades, then the DataFrame dataset in the sandbox environment will only contain columns related to student grades and will not contain other irrelevant sensitive information, such as students' home addresses and contact information. The access control mechanism ensures that the Pandas code can only perform legal operations on the DataFrame dataset in the sandbox environment. For example, the code can only read data from the dataset and cannot modify or delete the original data; it can only perform specific data analysis operations, such as calculating averages, sums, and filtering, and cannot perform operations that may compromise system stability or leak sensitive information, such as file read / write or network access. In the sandbox environment, the system executes the Pandas code generated by the large language model and monitors the execution process. If an error occurs or access control rules are violated during code execution, the system will terminate execution promptly and return the corresponding error message. If the code executes successfully, the system will extract the calculation results from the sandbox environment and return them to the user as an answer to their questions.

[0060] Preferably, the hybrid engine module populates the matched template based on the entity recognition results to generate a structured query statement, including: The entity recognition results are used to fill the slots in the matching template to generate a structured query statement.

[0061] Specifically, the hybrid engine module extracts entity information from the user's question and fills it into placeholders in the template to generate a structured query statement in response to the user's question. For example, in the "query student error rate" example above, if template 1 has the highest cosine similarity and is greater than or equal to 0.8, the hybrid engine module will fill the placeholders in template 1 with entity information such as "student_id", "subject", "start_date", and "end_date" extracted from the user's question. In this way, the hybrid engine module can accurately match the appropriate template based on the intent recognition result and generate an answer that meets the user's needs, improving the intelligence level and answer quality of the question-answering system.

[0062] For example, the original question: "What is the recent average math score of Class 3, Grade 8?" Matching template: SELECT AVG(score) FROM exam_results WHERE class='[Class]' AND subject='[Subject]' AND date>= '[Time]'; After autofill: SELECT AVG(score) FROM exam_results WHERE class='8th Grade Class 3' AND subject='Mathematics' AND date>= '2025-04-01'; The database is then queried in a controlled environment, and structured results are returned.

[0063] The question-and-answer module is also used to display the query results corresponding to the user's question.

[0064] Specifically, it includes: Structure the query results into tables, trend charts, or heatmaps to display the structured results.

[0065] Figure 4 This is a flowchart illustrating a student learning analysis method based on a hybrid engine, as shown in some embodiments of this specification. Figure 4 As shown, a student learning analysis method based on a hybrid engine includes the following steps: Student learning data is obtained from multiple data sources and preprocessed to generate preprocessed learning data. A question-and-answer interface based on natural language interaction is used to receive user questions. Through the hybrid engine, based on the pre-processed learning data, the system performs structured queries and intelligent responses to user questions received by the question-and-answer interface, and generates query results corresponding to the user questions. Displays the query results corresponding to the user's question.

[0066] A student learning analysis method based on a hybrid engine can be applied to a student learning analysis system based on a hybrid engine, which will not be elaborated here.

[0067] Finally, it should be understood that the embodiments described in this specification are merely illustrative of the principles of the embodiments described herein. Other variations may also fall within the scope of this specification. Therefore, alternative configurations of the embodiments described herein are intended to be illustrative rather than limiting, and should be considered consistent with the teachings of this specification. Accordingly, the embodiments described herein are not limited to those explicitly introduced and described herein.

Claims

1. A student learning analysis system based on a hybrid engine, characterized in that, include: The data acquisition module is used to obtain students' learning data from multiple data sources, preprocess the students' learning data, and generate preprocessed learning data. The question-and-answer module includes a question-and-answer interface based on natural language interaction, which is used to receive user questions; The hybrid engine module is used to perform structured queries and intelligent responses to user questions received by the question-and-answer interface based on preprocessed learning data, and generate query results corresponding to the user questions. The question-and-answer module is also used to display the query results corresponding to the user's question.

2. The student learning analysis system based on a hybrid engine according to claim 1, characterized in that, The student's learning data includes the student's behavioral trajectory during the learning process, learning outcomes, knowledge mastery assessment, and teacher-student interaction feedback data.

3. The student learning analysis system based on a hybrid engine according to claim 1, characterized in that, The data acquisition module preprocesses the students' learning data to generate preprocessed learning data, including: The student learning data is standardized by fields, normalized by timestamps, associated with user identities, and mapped to course structures to generate preprocessed learning data.

4. The student learning analysis system based on a hybrid engine according to claim 1, characterized in that, The hybrid engine module, based on preprocessed learning data, performs structured queries and intelligent responses to user questions received from the question-and-answer interface, and generates query results corresponding to the user questions, including: Perform intent recognition on user questions received from the Q&A interface and generate intent recognition results; Perform entity recognition and standardization on user questions received from the question-and-answer interface, and generate entity recognition results; Based on the intent recognition results, template matching is performed to obtain the matching template; Based on the entity recognition results, the matched template is populated to generate a structured query statement; Based on structured query statements, generate query results corresponding to user queries.

5. A student learning analysis system based on a hybrid engine according to claim 4, characterized in that, The hybrid engine module performs intent recognition on user questions received by the question-and-answer interface and generates intent recognition results, including: Perform text preprocessing on user questions received from the Q&A interface to generate text preprocessing results; Understanding user intent through large language models based on text preprocessing results; If the large language model returns a valid intent and the intent is in the predefined list, then an intent recognition result is generated based on the output of the large language model. If the large language model does not return a valid intent or the intent is not in the predefined list, then the predefined keyword-intent mapping relationship is used for matching to generate intent recognition results.

6. The student learning analysis system based on a hybrid engine according to claim 4, characterized in that, The hybrid engine module performs entity recognition and standardization on user questions received from the question-and-answer interface, generating entity recognition results, including: Extract entities from user questions received from the Q&A interface; The extracted entities are text-normalized to generate text-normalized results; Map the text standardization results to database primary keys or codes.

7. A student learning analysis system based on a hybrid engine according to claim 4, characterized in that, The hybrid engine module performs template matching based on the intent recognition result to obtain the matched template, including: Based on the intent recognition results, determine the set of candidate templates; The templates in the candidate template set are scored and sorted. Based on the ranking results, obtain the matching template.

8. A student learning analysis system based on a hybrid engine according to claim 4, characterized in that, The hybrid engine module, based on entity recognition results, populates the matched templates and generates structured query statements, including: The entity recognition results are used to fill the slots in the matching template to generate a structured query statement.

9. A student learning analysis system based on a hybrid engine according to any one of claims 1-8, characterized in that, The question-and-answer module displays the query results corresponding to the user's question, including: Structure the query results into tables, trend charts, or heatmaps to display the structured results.

10. A student learning analysis method based on a hybrid engine, characterized in that, The student learning analysis system based on a hybrid engine, as described in any one of claims 1-9, comprises: Student learning data is obtained from multiple data sources and preprocessed to generate preprocessed learning data. A question-and-answer interface based on natural language interaction, wherein the question-and-answer interface is used to receive user questions; Through a hybrid engine, based on preprocessed learning data, structured queries and intelligent responses are performed on user questions received by the question-and-answer interface, and corresponding query results are generated. Displays the query results corresponding to the user's question.

Images