Learning performance analysis and personalized recommendation system based on AI intelligent agent

Abstract

The invention provides a learning performance analysis and personalized recommendation system based on an AI agent, which comprises a fusion engine, the fusion engine is internally provided with a space-time alignment algorithm for fusing, organizing and mapping various learning data, and the space-time alignment algorithm can analyze the data and generate behavior representation vectors beneficial to learning. Relates to the technical field of learning performance analysis and recommendation. According to the method, front-end objective data and rear-end subjective data of a user are collected in real time, timestamp unification and spatial feature mapping are performed on heterogeneous data through a space-time alignment algorithm, a cross-modal behavior representation vector is generated, long-term learning dependence is captured by using an AI portrait model and adopting an LSTM network, key features are extracted in combination with a self-attention mechanism, and a cross-modal behavior representation vector is obtained. And dynamically generating a user capability portrait, and realizing self-adaptive iteration, thereby solving data space-time inconsistency and ensuring key feature dominant decisions.

Description

Technical Field

[0001] The present invention relates to the field of learning effectiveness analysis and recommendation technology, and specifically to a learning effectiveness analysis and personalized recommendation system based on AI intelligent agents. Background Art

[0002] Currently, personalized learning recommendation systems primarily rely on single-dimensional data analysis, such as test scores or user click behavior, and are implemented using traditional algorithms such as collaborative filtering, content-based recommendations, or shallow neural networks such as fully connected networks. These systems are typically built on static models, processing multi-source heterogeneous data (including academic performance, subjective evaluations, and behavioral logs) through simple time series analysis methods such as sliding averages or feature concatenation, ultimately generating recommendations based on fixed rules. Regarding temporal processing, existing technologies often downsample data using a unified time window or simply ignore timestamp differences, resulting in significant loss of temporal information. Regarding multimodal data fusion, manually setting feature weights, such as linear weighting or concatenating low-dimensional vectors, is difficult to effectively bridge the semantic gap between heterogeneous data types, such as numerical and textual data. Furthermore, existing methods often rely on historical behavior similarity matching, such as the K-nearest neighbor algorithm or predefined knowledge paths, lacking the ability to deeply model and respond to users' dynamic learning states in real time.

[0003] Existing technologies have significant defects that directly affect system performance and practicality.

[0004] First, there is a problem of spatiotemporal separation in the fusion of multi-source data, that is, the collection frequencies of different data sources, such as test scores and subjective evaluations, are inconsistent. Simple alignment methods such as interpolation and filling are prone to introduce noise and cannot accurately characterize the user's long-term learning patterns, such as stage-by-stage progress or regression. At the same time, multimodal data such as numerical and text types lack a unified representation space, and direct splicing or weighted fusion can easily lead to feature redundancy and semantic confusion. For example, it is impossible to effectively resolve the contradictory state of high scores and negative evaluations.

[0005] Secondly, there are problems with the system's static nature and low interpretability. That is, traditional recommendation strategies rely on fixed rules or static model parameters, making it difficult to dynamically adjust recommendation priorities based on real-time user feedback, such as resource click-through rates or knowledge point mastery, resulting in solutions lagging behind actual needs. Although deep learning-based systems can improve accuracy, the decision-making process is opaque. For example, the implicit state of LSTM is difficult to parse, and a credible explanation of the basis for recommendations cannot be provided.

[0006] Finally, there is the problem of fragmented data processing, that is, existing methods focus on a single data source or local features, such as relying solely on test scores or short-term behavior, and ignore the synergistic effect of multi-source data. For example, high-frequency practice and inefficient learning need to be jointly diagnosed with spatiotemporal features; in addition, due to the lack of a dynamic weight distribution mechanism, abnormal data such as occasional low scores or erroneous operation logs can easily interfere with the model, causing the recommendation results to deviate from the user's real needs. Summary of the Invention

[0007] The purpose of the present invention is to provide a learning effectiveness analysis and personalized recommendation system based on AI intelligent agents to solve the problems raised in the background technology in the above content and overcome the technical defects therein.

[0008] To solve the above technical problems, the technical solution adopted by the present invention is: a learning effectiveness analysis and personalized recommendation system based on AI intelligent body, including a fusion engine, which has a built-in spatiotemporal alignment algorithm that fuses, organizes and maps various types of learning data, and the spatiotemporal alignment algorithm can analyze the data and generate behavioral representation vectors that are conducive to learning; an AI portrait model, which has a built-in deep neural network that intelligently analyzes learning parameters, and the deep neural network can automatically select data as the processing layer and iteratively upgrade the model; and a multidimensional generator, which uses a built-in framework to perform multi-dimensional matching of the AI portrait model and the fusion engine and generate a customized composite recommendation solution.

[0009] A learning effectiveness analysis and personalized recommendation system based on an AI agent and a method for using the same include:

[0010] Step S1: Real-time acquisition of multi-source learning data of users, including objective scores at the front end and subjective evaluations at the back end;

[0011] Step S2: Using the fusion engine's spatiotemporal alignment algorithm, the heterogeneous data is uniformly managed with timestamps and spatial feature mapping to generate a cross-modal behavior representation vector.

[0012] Step S3: The AI profiling model uses the behavior representation vector as input, captures temporal dependencies through LSTM, extracts key features using the self-attention mechanism, and iteratively generates a user capability profile.

[0013] Step S4: The multidimensional generator matches the user profile with the knowledge graph and resource library, and generates a composite recommendation solution including learning paths, resources and strategies based on reinforcement learning;

[0014] Step S5: Based on the user's feedback on the execution effect of the recommended solution, dynamically adjust the weight of the spatiotemporal alignment algorithm and the network parameters of the AI portrait model to achieve closed-loop optimization.

[0015] As a further solution of the present invention: a learning effectiveness analysis and personalized recommendation system based on AI intelligent body, the spatiotemporal alignment algorithm includes a time dimension alignment module, a spatial feature mapping module and a calibration module, the time dimension alignment module performs unified timestamp management on the data through a sliding window mechanism, the spatial feature mapping module establishes a unified representation space for behavior representation vectors, and the calibration module adjusts the calculation weights within the spatiotemporal alignment algorithm. The time dimension alignment module, the spatial feature mapping module and the calibration module can be combined to calculate the spatiotemporal joint function of multi-source data that concretizes the spatiotemporal alignment algorithm.

[0016] As a further solution of the present invention: a learning effectiveness analysis and personalized recommendation system based on AI intelligent body, the learning data includes front-end data and back-end data, the front-end data includes objective learning performance, and the back-end data includes subjective evaluation performance. The fusion engine uses a spatiotemporal alignment algorithm to perform heterogeneous data fusion on the front-end data and the back-end data and generate a behavior representation vector.

[0017] As a further solution of the present invention: a learning efficiency analysis and personalized recommendation system based on AI intelligent agent, the AI portrait model adopts a hybrid architecture combining self-attention mechanism and LSTM network.

[0018] As a further solution of the present invention: a learning efficiency analysis and personalized recommendation system based on AI intelligent body, the deep neural network processes correlation data in parallel with the LSTM network through the self-attention mechanism, the deep neural network constructs learning dependencies through the LSTM network and autonomously selects and updates the AI model based on the self-attention mechanism.

[0019] As a further solution of the present invention: a learning efficiency analysis and personalized recommendation system based on AI intelligent body, the processing layer includes a pooling layer and a convolution layer, the pooling layer has a built-in extraction unit, the convolution layer has a built-in reasoning unit, the extraction unit parses the front-end data and the back-end data through a spatiotemporal alignment algorithm, and the reasoning unit calculates the behavior representation vector through GRU modeling.

[0020] As a further solution of the present invention: a learning efficiency analysis and personalized recommendation system based on AI agent, the multi-source data spatiotemporal joint function F is

[0021]

[0022] Among them, T k (X k ) is the time series feature vector generated by the k-th data source after its module processing in the time dimension, S k (Y k) is the spatial feature vector generated by the k-th data source after being processed by the spatial feature mapping module, λ k (time weight) and μ k (spatial weight) is the time and space weight coefficient calculated by the calibration module, and λ k and μ k The sigma functions evaluate to 1.

[0023] Compared with the prior art, the present invention has the following beneficial effects:

[0024] This application first collects the user's front-end objective data (such as grades) and back-end subjective data (such as evaluations) in real time, and uses the spatiotemporal alignment algorithm to unify the timestamps of heterogeneous data (sliding window management time series) and spatial feature mapping (high-dimensional space fusion of multimodal data) to generate a cross-modal behavior representation vector. Then, the AI portrait model uses the LSTM network to capture long-term learning dependencies, combines the self-attention mechanism to extract key features, and dynamically generates a user ability portrait. Then, the multidimensional generator matches the user portrait with the knowledge graph, and uses reinforcement learning to optimize the composite recommendation plan of learning path, resources and strategy. Finally, the lambda is dynamically adjusted according to user feedback. k (time weight) and μ k (spatial weights) and model parameters to achieve adaptive iteration, thereby resolving the spatiotemporal inconsistency of data and ensuring that key features dominate decision-making. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] The disclosure of the present invention is illustrated with reference to the accompanying drawings. It should be understood that the drawings are for illustrative purposes only and are not intended to limit the scope of protection of the present invention. In the accompanying drawings, the same reference numerals are used to refer to the same components. Among them:

[0026] Figure 1 The figure schematically shows a flow chart of a method proposed according to one embodiment of the present invention. DETAILED DESCRIPTION

[0027] It is easy to understand that according to the technical solution of the present invention, without changing the essential spirit of the present invention, a person skilled in the art can propose a variety of interchangeable structural modes and implementation modes. Therefore, the following specific embodiments and drawings are only exemplary descriptions of the technical solution of the present invention and should not be regarded as the entire invention or as a limitation or restriction of the technical solution of the present invention.

[0028] According to one embodiment of the present invention, in combination with the accompanying drawings, a learning efficiency analysis and personalized recommendation system based on AI intelligent agents is shown, including a fusion engine, the fusion engine has a built-in spatiotemporal alignment algorithm that fuses, organizes and maps various types of learning data, the spatiotemporal alignment algorithm can analyze data and generate a behavior representation vector that is conducive to learning; an AI portrait model, the AI portrait model has a built-in deep neural network that intelligently analyzes learning parameters, the deep neural network can automatically select data as a processing layer and iteratively upgrade the model; and a multidimensional generator, the multidimensional generator uses a built-in framework to perform multi-dimensional matching between the AI portrait model and the fusion engine and generates a customized composite recommendation solution, the spatiotemporal alignment algorithm includes a time dimension alignment module, a spatial feature mapping module and a calibration module, the time dimension alignment module performs unified timestamp management on the data through a sliding window mechanism, the spatial feature mapping module establishes a unified representation space for the behavior representation vector, the calibration module adjusts the calculation weights in the spatiotemporal alignment algorithm, and the time dimension alignment module performs unified timestamp management on the data through a sliding window mechanism, the spatial feature mapping module establishes a unified representation space for the behavior representation vector, and the calibration module adjusts the calculation weights in the spatiotemporal alignment algorithm. The inter-dimensional alignment module, the spatial feature mapping module and the calibration module can be combined to calculate the spatiotemporal joint function of the multi-source data of the concrete spatiotemporal alignment algorithm. The learning data includes front-end data and back-end data. The front-end data includes objective learning results, and the back-end data includes subjective evaluation results. The fusion engine uses the spatiotemporal alignment algorithm to perform heterogeneous data fusion on the front-end data and the back-end data and generate a behavior representation vector. The AI portrait model adopts a hybrid architecture combining the self-attention mechanism and the LSTM network. The deep neural network processes the correlation data in parallel with the LSTM network through the self-attention mechanism. The deep neural network constructs learning dependencies through the LSTM network and autonomously selects and updates the AI model based on the self-attention mechanism. The processing layer includes a pooling layer and a convolution layer. The pooling layer has an extraction unit built in, and the convolution layer has an inference unit built in. The extraction unit parses the front-end data and the back-end data through the spatiotemporal alignment algorithm, and the inference unit calculates the behavior representation vector through GRU modeling.

[0029] Other embodiments of the present invention: A learning efficiency analysis and personalized recommendation system based on AI agent, wherein the multi-source data spatiotemporal joint function F is

[0030]

[0031] Among them, T k (X k ) is the time series feature vector generated by the k-th data source after its module processing in the time dimension, S k (Y k ) is the spatial feature vector generated by the k-th data source after being processed by the spatial feature mapping module, λ k (time weight) and μ k(spatial weight) is the time and space weight coefficient calculated by the calibration module, and λ k and μ k The sigma functions evaluate to 1.

[0032] In this embodiment, the core goal of the multi-source data spatiotemporal joint function F is to solve the inconsistency problem of multi-source heterogeneous learning data (such as grades, behaviors, subjective evaluations, etc.) in the time dimension and spatial characteristics, and to achieve it through time alignment, space mapping and dynamic weight allocation. Specifically, time alignment is achieved through T k (X k ) extracts temporal features (such as the temporal regularity of learning behavior), unifies the timestamps of different data sources, eliminates the temporal deviation caused by sampling frequency differences, and spatial mapping is achieved through S k (Y k ) maps different modal data (such as numerical scores and text evaluations) into a unified high-dimensional feature space to resolve the semantic barriers of heterogeneous data. Dynamic weight allocation is achieved through λ k (time weight) and μ k The spatial weight calibration module adaptively adjusts the contribution of different data sources to the final behavior representation vector, ensuring that key data drives model decisions, improving data fusion quality and recommendation accuracy. This addresses the pain points of data fragmentation and spatiotemporal inconsistencies in traditional recommendation systems, providing the underlying technical support for generating high-precision, explainable, personalized recommendations. Its dynamic weighting mechanism further adapts to complex and changing learning scenarios and is the core algorithmic support for the system's "intelligent agent" attributes.

[0033] Other embodiments of the present invention include a learning effectiveness analysis and personalized recommendation system based on an AI agent and a method for using the system, including:

[0034] Step S1: Real-time acquisition of multi-source learning data of users, including objective scores at the front end and subjective evaluations at the back end;

[0035] Step S2: Using the fusion engine's spatiotemporal alignment algorithm, the heterogeneous data is uniformly managed with timestamps and spatial feature mapping to generate a cross-modal behavior representation vector.

[0036] Step S3: The AI profiling model uses the behavior representation vector as input, captures temporal dependencies through LSTM, extracts key features using the self-attention mechanism, and iteratively generates a user capability profile.

[0037] Step S4: The multidimensional generator matches the user profile with the knowledge graph and resource library, and generates a composite recommendation solution including learning paths, resources and strategies based on reinforcement learning;

[0038] Step S5: Based on the user's feedback on the execution effect of the recommended solution, dynamically adjust the weight of the spatiotemporal alignment algorithm and the network parameters of the AI portrait model to achieve closed-loop optimization.

[0039] Working principle: This application first collects the user's front-end objective data (such as grades) and back-end subjective data (such as evaluations) in real time, and uses the spatiotemporal alignment algorithm to unify the timestamps of heterogeneous data (sliding window management time series) and spatial feature mapping (high-dimensional space fusion of multimodal data) to generate a cross-modal behavior representation vector. Then, the AI portrait model uses the LSTM network to capture long-term learning dependencies, combines the self-attention mechanism to extract key features, and dynamically generates a user ability portrait. The multidimensional generator then matches the user portrait with the knowledge graph, and uses reinforcement learning to optimize the composite recommendation plan of learning paths, resources and strategies. Finally, the lambda is dynamically adjusted according to user feedback. k (time weight) and μ k (spatial weights) and model parameters to achieve adaptive iteration, thereby resolving the spatiotemporal inconsistency of data and ensuring that key features dominate decision-making.

[0040] The technical scope of the present invention is not limited to the contents of the above description. Those skilled in the art can make various deformations and modifications to the above embodiments without departing from the technical idea of the present invention, and these deformations and modifications should all fall within the protection scope of the present invention.

Claims

1. A learning effectiveness analysis and personalized recommendation system based on AI agents, characterized by: A fusion engine includes a built-in spatiotemporal alignment algorithm that fuses, organizes, and maps various types of learning data. The spatiotemporal alignment algorithm analyzes the data and generates a vector containing behavioral representations that are conducive to learning. AI portrait model, which has a built-in deep neural network for intelligent analysis and learning parameters. The deep neural network can automatically select data as the processing layer and iteratively upgrade the model; as well as A multi-dimensional generator uses a built-in framework to perform multi-dimensional matching between the AI portrait model and the fusion engine and generate a customized composite recommendation solution.

2. The AI agent-based learning effectiveness analysis and personalized recommendation system according to claim 1, characterized in that: The spatiotemporal alignment algorithm includes a time dimension alignment module, a spatial feature mapping module and a calibration module. The time dimension alignment module performs unified timestamp management on the data through a sliding window mechanism. The spatial feature mapping module establishes a unified representation space for behavior representation vectors. The calibration module adjusts the calculation weights within the spatiotemporal alignment algorithm. The time dimension alignment module, the spatial feature mapping module and the calibration module can be combined to calculate the spatiotemporal joint function of multi-source data that visualizes the spatiotemporal alignment algorithm.

3. The AI agent-based learning effectiveness analysis and personalized recommendation system according to claim 2, characterized in that: The learning data includes front-end data and back-end data, the front-end data includes objective learning results, and the back-end data includes subjective evaluation results. The fusion engine uses a spatiotemporal alignment algorithm to perform heterogeneous data fusion on the front-end data and the back-end data and generate a behavior representation vector.

4. The AI agent-based learning effectiveness analysis and personalized recommendation system according to claim 3, characterized in that: The AI portrait model adopts a hybrid architecture that combines the self-attention mechanism with the LSTM network.

5. The AI agent-based learning effectiveness analysis and personalized recommendation system according to claim 4, characterized in that: The deep neural network processes association data in parallel with the LSTM network through the self-attention mechanism. The deep neural network builds learning dependencies through the LSTM network and autonomously selects and updates the AI model based on the self-attention mechanism.

6. The AI agent-based learning effectiveness analysis and personalized recommendation system according to claim 5, characterized in that: The processing layer includes a pooling layer and a convolution layer. The pooling layer has a built-in extraction unit, and the convolution layer has a built-in reasoning unit. The extraction unit parses the front-end data and the back-end data through a spatiotemporal alignment algorithm, and the reasoning unit calculates the behavior representation vector through GRU modeling.

7. The AI agent-based learning effectiveness analysis and personalized recommendation system according to claim 6, characterized in that: The multi-source data spatiotemporal joint function F is: Among them, T k (X k ) is the time series feature vector generated by the k-th data source after its module processing in the time dimension, S k (Y k ) is the spatial feature vector generated by the k-th data source after being processed by the spatial feature mapping module, λ k (time weight) and μ k (spatial weight) is the time and space weight coefficient calculated by the calibration module, and λ k and μ k The sigma functions evaluate to 1.

8. The AI agent-based learning effectiveness analysis and personalized recommendation system and its use method according to claim 7, characterized in that: include: Step S1: Real-time acquisition of multi-source learning data of users, including objective scores at the front end and subjective evaluations at the back end; Step S2: Using the fusion engine's spatiotemporal alignment algorithm, the heterogeneous data is uniformly managed with timestamps and spatial feature mapping to generate a cross-modal behavior representation vector. Step S3: The AI profiling model uses the behavior representation vector as input, captures temporal dependencies through LSTM, extracts key features using the self-attention mechanism, and iteratively generates a user capability profile. Step S4: The multidimensional generator matches the user profile with the knowledge graph and resource library, and generates a composite recommendation solution including learning paths, resources and strategies based on reinforcement learning; Step S5: Based on the user's feedback on the execution effect of the recommended solution, dynamically adjust the weight of the spatiotemporal alignment algorithm and the network parameters of the AI portrait model to achieve closed-loop optimization.

Images