Method and device for fusing knowledge questions and answers and course recommendation based on knowledge tree and large model
By constructing knowledge trees and large models, and combining user data with various recommendation algorithms, the problem of low knowledge organization and user matching in enterprise learning platforms has been solved, enabling personalized course recommendations and intelligent Q&A, thereby improving learning efficiency and user experience.
Patent Information
- Application Number
- CN202511297504.9
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-09-11
- Publication Date
- 2025-12-16
AI Technical Summary
Existing enterprise learning platforms suffer from problems such as flattened knowledge organization, low user matching, insufficient personalized services, and lagging knowledge updates, resulting in inadequate intelligence and user experience in course recommendation and Q&A systems.
By constructing knowledge trees and large models, combined with user data and pre-trained models, personalized course recommendations and question-and-answer services are achieved. Knowledge graphs are used to display the hierarchical structure of information, and hybrid recommendation algorithms and vector databases are used for searching and deduplication. Multiple recommendation algorithms are integrated to improve personalization and accuracy.
It has enabled the systematic organization of course resources, enhanced the hierarchy and logic of knowledge expression, improved user interaction experience and learning efficiency, supported dynamic knowledge updates, reduced manual intervention, and improved the intelligence level of the enterprise learning platform.
Smart Images
Figure CN121144508A_ABST
Abstract
Description
TECHNICAL FIELD
[0001] The present application relates to the technical field of computer, more particularly, it relates to a method and device for fusing knowledge question and answer and course recommendation based on knowledge tree and large model. BACKGROUND
[0002] With the rapid development of enterprise digital learning platform, knowledge resources present the characteristics of mass, multi-source and heterogeneous, and enterprise learning systems such as State Grid School face new challenges in knowledge services. The existing course recommendation and question and answer system generally has the following technical bottlenecks: (1) The knowledge organization method is flat, lacks systematic and hierarchical knowledge structure support, which leads to weak correlation between course content, and it is difficult to reflect the progressive relationship and system logic of professional knowledge; (2) The matching degree of user question and course is low, and the traditional keyword matching or collaborative filtering recommendation method cannot understand the deep semantic intention, which leads to low accuracy of question and answer and one-sided or repeated recommendation results; (3) The individual service capability is insufficient, and it is unable to dynamically model combined with multi-dimensional information such as user post, learning history and knowledge mastery, which affects the scientificity and effectiveness of learning path; (4) The knowledge update and recommendation system are disconnected, and new courses are difficult to quickly integrate into the existing knowledge system after being put online, which leads to lagging recommendation and knowledge island problems.
[0003] Therefore, how to break through the technical limitations of weak semantic understanding and single recommendation logic of traditional systems, improve the intelligent level of knowledge service and user experience, and realize the transformation from "resource push" to "cognitive guidance" is a technical problem that needs to be solved by the technical personnel in the field. SUMMARY
[0004] Therefore, the present application provides a method and device for fusing knowledge question and answer and course recommendation based on knowledge tree and large model, which solves the problems in the background art.
[0005] In order to achieve the above purpose, the present application provides the following technical scheme: A method for fusing knowledge question and answer and course recommendation based on knowledge tree and large model, comprising the following steps: S1, combining user viewing time and homework score, using a hybrid recommendation algorithm to recommend courses; S2, based on knowledge enhancement and pre-training large model, completing Query intention recognition; S3, using a knowledge graph to construct a field knowledge tree, connecting knowledge points and concepts in a graphical way to show the hierarchical structure and mutual relationship between information; S4, similarity search, clustering and courseware deduplication of high-dimensional vectors are performed through the vector database, and finally the knowledge Q&A result and the personalized course recommendation result are obtained.
[0006] Optionally, S1 specifically comprises the following steps: S11, interest and engagement are evaluated by collecting user viewing time, and the learning ability and mastery of specific courses of the user are measured by using homework scores; S12, a user portrait including learning preferences, learning efficiency and knowledge mastery level, and a course portrait including course difficulty, theme and prerequisite requirements are constructed; S13, combined with collaborative filtering algorithm, content-based recommendation algorithm, time series analysis, context-aware recommendation algorithm and reinforcement learning or online learning algorithm, the personalized course recommendation result is generated.
[0007] Optionally, the specific content of S13 is: Through the collaborative filtering algorithm, other user groups similar to the target user's interest are identified and the courses liked by the user groups are recommended; Using the content-based recommendation algorithm, courses are recommended according to the matching degree of the user portrait and the course portrait; Introducing time series analysis to capture the changes of user interest and demand over time and the evolution of course relevance over time; Using the context-aware recommendation algorithm, combined with the user's current learning stage, learning environment and device information, the recommendation result is refined; Considering the data changes of test scores, using reinforcement learning or online learning algorithm, dynamically adjusting the recommendation strategy to respond to the improvement or decline of user performance.
[0008] Optionally, the specific steps of S2 are: using model parameter to initialize the large model part, and recalling the most similar knowledge cards and similar Queries by calculating the cosine similarity between vectors.
[0009] Optionally, in S4, courseware deduplication is achieved by calculating the similarity of courseware content vectors, and when the similarity exceeds the preset threshold, it is determined as a duplicate courseware.
[0010] Optionally, the method further comprises: processing the pre-trained large model through SFT and prompt technology to complete sensitive content filtering and intelligent correction.
[0011] Optionally, the method further comprises: using the Voting method of ensemble learning to determine the final recommendation result combined with the prediction results of multiple classifiers.
[0012] The application discloses a device for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model, and is used for realizing the method for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model. The personalized recommendation module is used for realizing personalized course recommendation by using a hybrid recommendation algorithm according to user watching time length and homework scores. The intention recognition module is used for completing Query intention recognition according to knowledge enhancement and pre-training of a large model. The knowledge organization module is used for connecting knowledge points and concepts in a graphical manner to display the hierarchical structure and mutual relationship between information by constructing a field knowledge tree based on a knowledge graph. The data retrieval and deduplication module is used for finally obtaining knowledge question answering results and personalized course recommendation results by performing similarity search, clustering and courseware deduplication of high-dimensional vectors based on a vector database.
[0013] Compared with the prior art, the application discloses a method and a device for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model, and has the following beneficial effects: (1) The application realizes systematic organization and semantic association of course resources by constructing a structured field knowledge tree, effectively improves the hierarchical nature and logicality of knowledge expression, and solves the problems of knowledge fragmentation and isolation in a traditional learning platform. (2) The application fuses the powerful natural language understanding and generation capability of a large language model, can accurately analyze the deep semantic intention of user questions, provides more accurate, coherent and interpretable intelligent question answering services, and significantly improves user interactive experience. (3) The application realizes personalized course recommendation and adaptive learning path planning by combining multi-dimensional portrait information such as user post roles, learning behaviors and knowledge mastery progress, and improves learning efficiency and resource matching degree. (4) The application can support dynamic updating of a knowledge tree and automatic collection of course resources, guarantee the consistency of a recommendation system and time sequence knowledge evolution, and avoid knowledge islands. (5) The overall technical scheme reduces the dependence on artificial intervention, improves the knowledge service capability and intelligent level of an enterprise learning platform, and has good scalability and application prospect. BRIEF DESCRIPTION OF DRAWINGS
[0014] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or the prior art description. Obviously, the drawings in the following description only constitute the embodiments of the application, and for those skilled in the art, other drawings can be obtained without creative labor on the basis of the provided drawings.
[0015] Figure 1 A method flowchart for knowledge question answering and course recommendation based on knowledge tree and large model fusion is provided. DETAILED DESCRIPTION
[0016] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.
[0017] The embodiments of the present application disclose a method for knowledge question answering and course recommendation based on knowledge tree and large model fusion, as shown in the following figure, comprising the following steps: Figure 1 S1, combining user viewing time and homework score, using a hybrid recommendation algorithm to recommend courses; S2, based on knowledge enhancement and pre-training large model, completing Query intent recognition; S3, using a knowledge graph to construct a domain knowledge tree, connecting knowledge points and concepts in a graphical manner to show the hierarchical structure and mutual relationship between information; S4, performing similarity search, clustering and courseware deduplication of high-dimensional vectors through a vector database, and finally obtaining knowledge question answering results and personalized course recommendation results.
[0018] Further, in the present embodiment, S1 specifically comprises the following steps: S11, by collecting user viewing time to evaluate interest and participation, using homework score to measure the user's mastery of specific courses and learning ability; S12, constructing a user portrait containing learning preference, learning efficiency and knowledge mastery level, and a course portrait containing course difficulty, theme and prerequisite requirements; S13, combining collaborative filtering algorithm, content-based recommendation algorithm, time series analysis, context-aware recommendation algorithm and reinforcement learning or online learning algorithm to generate personalized course recommendation results.
[0019] Further, in the present embodiment, the specific content of S13 is: Through the collaborative filtering algorithm, identify other user groups similar to the target user's interest and recommend the courses that the user groups like; Using the content-based recommendation algorithm, recommend courses according to the matching degree of user portrait and course portrait; Introducing time series analysis to capture the changes of user interest and demand over time and the evolution of course relevance over time, ensuring the timeliness of the recommended content; With the context-aware recommendation algorithm, the current learning stage, learning environment and device information of the user are combined to refine the recommendation results, ensuring that the recommendation results match the current situation of the user. Considering the data changes of the examination scores, a reinforcement learning or online learning algorithm is used to dynamically adjust the recommendation strategy to respond to the improvement or decline of the user's performance.
[0020] Further, Query intent recognition is an important task in Query understanding, which has a great influence on the relevance and accuracy of downstream search. In enterprise search, Query intent recognition can be achieved through methods based on knowledge enhancement and pre-trained large models. In this embodiment, the specific steps of S2 are: initializing the large model part using model parameters, and recalling the most similar knowledge cards and similar Queries by calculating the cosine similarity between vectors.
[0021] Further, the knowledge graph plays a key role in knowledge organization as a key tool for understanding and managing complex information. It clearly shows the hierarchical structure and mutual relationship between information by connecting knowledge points and concepts in a graphical way. This structured representation method not only helps users quickly grasp the overall framework of the knowledge system, but also reveals the internal relationship between concepts, promotes cross-domain cognition and association, and improves learning efficiency and deep understanding.
[0022] Further, in S4 of the present embodiment, courseware deduplication is achieved by calculating the similarity of courseware content vectors, and when the similarity exceeds a preset threshold, it is determined as a duplicate courseware. In addition, the present embodiment supports searching large-scale data sets (up to billion-level vectors) and can handle vector sets that do not fit in a single machine memory. The core advantage is to provide multiple algorithms to search vector sets of any size, including exact search and approximate search; some core algorithms implemented in the vector database also provide GPU-accelerated versions to further improve search efficiency.
[0023] Still further, the technical solution of the present embodiment further includes: processing the pre-trained large model through SFT and prompt technology to complete sensitive content filtering and intelligent correction.
[0024] Still further, the technical solution of the present embodiment further includes: adopting the Voting method of ensemble learning to determine the final recommendation result in combination with the prediction results of multiple classifiers.
[0025] With Figure 1 Corresponding to the method, the embodiment of the present application further provides a device for fusing knowledge question and answer and course recommendation based on knowledge tree and large model, which is used for Figure 1In specific implementation of the method, the device for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model provided by the embodiment of the application can be applied to a computer terminal or various mobile devices, and specifically includes: The personalized recommendation module is configured to use a hybrid recommendation algorithm to realize personalized course recommendation according to the user viewing time length and the homework score. The intention recognition module is configured to complete Query intention recognition according to the knowledge enhancement and the pre-training large model. The knowledge organization module is configured to connect knowledge points and concepts in a graphical manner to display the hierarchical structure and mutual relationship between information by constructing a domain knowledge tree through a knowledge graph. The data retrieval and deduplication module is configured to finally obtain knowledge question answering results and personalized course recommendation results by performing similarity search, clustering and courseware deduplication of high-dimensional vectors through a vector database.
[0026] To sum up, the embodiment proposes a method and device for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model, realizes the structured organization and semantic association of course content by constructing a domain knowledge tree model, accurately analyzes the user natural language question intention by fusing the semantic understanding ability of a large language model, and realizes personalized and interpretable knowledge question answering and course recommendation by combining the user portrait and the knowledge mastery state. Therefore, the technical solution breaks through the technical limitations of weak semantic understanding and single recommendation logic of the traditional system, improves the intelligent level of knowledge service and the user experience, and realizes the change from "resource pushing" to "cognitive guidance".
[0027] The embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed by the embodiments, since it corresponds to the method disclosed by the embodiments, the description is relatively simple, and the related parts can be referred to the method part.
[0028] The above description of the disclosed embodiments enables a person skilled in the art to implement or use the present application. Various modifications to the embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.
Claims
1. A method for fusing knowledge question answering and course recommendation based on knowledge tree and large model, characterized in that, Comprise the following steps: S1, combine user viewing time and homework score, use mixed recommendation algorithm for course recommendation; S2, based on knowledge enhancement and pre-training large model, complete Query intent recognition; S3, use knowledge graph to build domain knowledge tree, connect knowledge points and concepts in graphical way, show the hierarchical structure and mutual relationship between information; S4, through the vector database for high-dimensional vector similarity search, clustering and courseware deduplication, finally get knowledge Q & A results and personalized course recommendation results.
2. The method of claim 1, wherein the method is characterized by, S1 specifically includes the following steps: S11, through collecting user viewing time to evaluate interest and participation, using homework score to measure user's mastery of specific courses and learning ability; S12, build user portrait including learning preference, learning efficiency, knowledge mastery level, and course portrait including course difficulty, theme and prerequisite; S13, combined with collaborative filtering algorithm, content-based recommendation algorithm, time series analysis, context-aware recommendation algorithm and reinforcement learning or online learning algorithm, generate personalized course recommendation results.
3. The method of claim 2, wherein the method further comprises: The specific content of S13 is: Through collaborative filtering algorithm, identify other user groups similar to the target user's interest and recommend the courses that the user groups like; Use content-based recommendation algorithm to recommend courses according to the matching degree of user portrait and course portrait; Introduce time series analysis to capture the changes of user interest and demand over time and the evolution of course relevance over time; Use context-aware recommendation algorithm to refine the recommendation results combined with user's current learning stage, learning environment and device information; Consider the data changes of test scores, use reinforcement learning or online learning algorithm to dynamically adjust the recommendation strategy to respond to the improvement or decline of user performance.
4. The method of claim 1, wherein the method is characterized by, The specific steps of S2 are: use model parameter initialization large model part, recall the most similar knowledge cards and similar Query by calculating the cosine similarity between vectors.
5. The method of claim 1, wherein the method is characterized by, In S4, courseware deduplication is achieved by calculating the similarity of courseware content vectors. When the similarity exceeds the preset threshold, it is determined as duplicate courseware.
6. The method of claim 1, wherein, Also includes: Through SFT and prompt technology to process the pre-training large model, complete sensitive content filtering and intelligent correction.
7. The method of claim 1, wherein, Also includes: Adopt Voting method of ensemble learning, combine the prediction results of multiple classifiers to determine the final recommendation results.
8. An apparatus for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model, for implementing the method for fusing knowledge question answering and course recommendation based on a knowledge tree and a large model according to any one of claims 1-7, characterized in that, Includes: Personalized recommendation module, used for using mixed recommendation algorithm to realize personalized course recommendation according to user viewing time and homework score; Intent recognition module, used for completing Query intent recognition according to knowledge enhancement and pre-training large model; Knowledge organization module, used for connecting knowledge points and concepts in graphical way through knowledge graph to build domain knowledge tree, show the hierarchical structure and mutual relationship between information; Data retrieval and deduplication module, used for finally getting knowledge Q & A results and personalized course recommendation results through high-dimensional vector similarity search, clustering and courseware deduplication in vector database.