Intelligent campus service pushing method and system based on artificial intelligence

Abstract

The invention provides an intelligent campus service pushing method and system based on artificial intelligence, and the method comprises the steps: obtaining a service pushing training sample marked with a service acceptance description portrait, carrying out the feature extraction of service use track data and campus service resource data in the service pushing training sample, and generating a plurality of service knowledge vectors; and obtaining an attention coefficient of each service knowledge vector by using an attention network, and performing service acceptance portrait estimation in combination with the estimation network to generate a service acceptance estimation portrait corresponding to the plurality of service push nodes. And by calculating a network training error, optimizing neuron weight information in the service acceptance portrait estimation network until the network converges. According to the method, the intelligent campus service pushing decision can be accurately made for the target pushing object, the service pushing accuracy and the user satisfaction degree are improved, and efficient and personalized pushing of the intelligent campus service is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence. Specifically, it relates to a method and system for pushing intelligent campus services based on artificial intelligence. Background Art

[0002] With the rapid development of information technology and the wide application of artificial intelligence technology, the construction of intelligent campuses has become an important trend in the development of modern education. Intelligent campuses integrate advanced information technology and intelligent management systems to provide more convenient and efficient service experiences for teachers and students. However, in terms of pushing intelligent campus services, how to accurately and personalized push the required services to users has always been a difficult problem in the construction of intelligent campuses.

[0003] Traditional service push methods often rely on simple rules or users' historical behavior data. Although these methods can realize the automatic push of services to a certain extent, they lack a deep understanding of users' current needs and service acceptance. Therefore, the pushed services are often not accurate enough to meet users' personalized needs. Summary of the Invention

[0004] In view of the above-mentioned problems, in combination with the first aspect of the present invention, embodiments of the present invention provide a method for pushing intelligent campus services based on artificial intelligence. The method includes:

[0005] Obtain service push training examples of intelligent campus services, where the service push training examples are labeled with a service acceptance description portrait corresponding to the target service push node of the service push training examples;

[0006] Extract features from the service usage trajectory data and campus service resource data in the service push training examples to generate multiple service knowledge vectors included in the service push training examples;

[0007] Obtain attention coefficients of each service knowledge vector through an attention network, and estimate the service acceptance portrait through an estimation network based on the multiple service knowledge vectors included in the service push training examples and the attention coefficients of each service knowledge vector, and generate a service acceptance estimation portrait corresponding to each service push node among the multiple service push nodes of the service push training examples;

[0008] Calculate the network training error based on the service acceptance description portrait corresponding to the target service push node of the service push training example and the service acceptance estimation portrait corresponding to each service push node of the service push training example;

[0009] Optimize the neuron weight information in the service acceptance portrait estimation network for estimating the service push training example corresponding to the target service push node based on the network training error until the network convergence requirement is met. Then, make a decision on the intelligent campus service push for the target push object according to the service acceptance portrait estimation network, where the service acceptance portrait estimation network includes the attention network and the estimation network.

[0010] In another aspect, an embodiment of the present invention further provides an intelligent campus service push system based on artificial intelligence, including a processor and a machine-readable storage medium. The machine-readable storage medium is connected to the processor. The machine-readable storage medium is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the machine-readable storage medium to implement the above method.

[0011] Based on the above aspects, the embodiments of the present application obtain service push training examples marked with service acceptance description portraits, extract features from the service usage trajectory data and campus service resource data in the service push training examples, and generate multiple service knowledge vectors. Use the attention network to obtain the attention coefficients of each service knowledge vector, and combine the estimation network to estimate the service acceptance portrait, and generate a service acceptance estimation portrait corresponding to multiple service push nodes. By calculating the network training error, optimize the neuron weight information in the service acceptance portrait estimation network until the network converges. This method can accurately make a decision on the intelligent campus service push for the target push object, improve the accuracy of service push and user satisfaction, and realize the efficient and personalized push of intelligent campus services. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] Figure 1 is a schematic execution flowchart of an intelligent campus service push method based on artificial intelligence provided by an embodiment of the present invention.

[0013] Figure 2 is a schematic hardware architecture diagram of an intelligent campus service push system based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0014] The present invention will be specifically described below with reference to the accompanying drawings of the specification. Figure 1 is a schematic flowchart of an intelligent campus service push method based on artificial intelligence provided by an embodiment of the present invention. The intelligent campus service push method based on artificial intelligence will be introduced in detail below.

[0015] Step S110, obtain a service push training example of an intelligent campus service, where the service push training example is marked with a service acceptance description portrait corresponding to the service push training example for a target service push node.

[0016] In this embodiment, in the environment of a smart campus, there are various services, such as online course learning services, campus activity notification services, library borrowing services, etc. Suppose a smart campus service push system needs to be constructed, and a large number of service push training examples are required to train the model. Taking the online course learning service as an example, the service push training examples can be the historical records of pushing online course learning services to different student groups in the past period. These historical records contain a lot of information, such as the basic information of the students being pushed (grade, major, etc.), the push time, and the campus network environment at that time.

[0017] The service acceptance description portrait is a description of the acceptance degree of each service push training example at the target service push node (such as a specific course type push node, such as the advanced mathematics course push node). For the training examples of the online course learning service, the service acceptance description portrait may include multi-dimensional descriptions such as whether the student clicks to view the course details, whether the student actually registers to study the course, and the learning duration. If a student, after receiving the advanced mathematics course push, not only clicks to view the course details but also registers and continues to study for a long time, then the service acceptance description portrait at this advanced mathematics course push node will show a high acceptance degree. By collecting numerous such service push training examples and their corresponding service acceptance description portraits of the target service push nodes, it provides basic data for subsequent model training.

[0018] Step S120, perform feature extraction on the service usage trajectory data and campus service resource data in the service push training examples to generate multiple service knowledge vectors included in the service push training examples.

[0019] Continuing with the example of the online course learning service, the service usage trajectory data may include the operation trajectories related to courses of students on the smart campus platform. For example, the time when students log in to the platform, the stay duration on different course pages, and whether they participate in the interaction in the course-related discussion area. First, clean these service usage trajectory data to remove possible error records (such as abnormal login time records caused by network fluctuations), and perform denoising processing to remove some meaningless operation records (such as extremely short page stay times caused by frequent misclicks), and then perform normalization processing to convert data of different magnitudes into the same magnitude for subsequent analysis. Then, extract key behavior features from these processed trajectory data. For example, if the average stay duration of a student on a specific course page exceeds a certain threshold, this may indicate that the student has a high degree of attention to this course, and this is a key behavior feature.

[0020] Meanwhile, in terms of campus service resource data, taking the resources related to online course learning services as an example, it includes course videos, course handouts, course assignments, etc. Semantic analysis is performed on these resource data. For example, the explanatory content in the course video is analyzed to extract the key knowledge points, and keyword extraction is carried out. For example, keywords such as "limit" and "derivative" are extracted from the higher mathematics course video, and then these keywords are converted into vectorized representations to generate corresponding resource feature vectors. Finally, the key behavior features extracted from the service usage trajectory data and the resource feature vectors obtained from the campus service resource data are fused to generate each service knowledge vector containing service information and resource semantics. For example, a service knowledge vector may include the degree of attention of students to the higher mathematics course (obtained from the service usage trajectory data) and the vector representation of the key knowledge points in the course (obtained from the campus service resource data). Such a service knowledge vector can comprehensively reflect the relevant information in the service push training example.

[0021] In step S130, attention coefficients of each service knowledge vector are obtained through an attention network, and an acceptance portrait estimation of the service is performed by an estimation network based on the multiple service knowledge vectors included in the service push training example and the attention coefficients of each service knowledge vector, so as to generate an acceptance estimation portrait of the service push training example corresponding to each service push node among the multiple service push nodes.

[0022] In the scenario of online course learning services in a smart campus, assuming there are multiple service knowledge vectors, these service knowledge vectors contain various information related to students and courses. The attention network starts to work to capture the relationships between these service knowledge vectors. Using the global attention mechanism of the attention network, for each service knowledge vector, similarity calculations are performed with all other service knowledge vectors. For example, a service knowledge vector representing the degree of attention of students to the higher mathematics course and the course resource characteristics is calculated for similarity with another similar vector representing the students' attention to the English course. Through such calculations, a global similarity matrix is obtained, and this matrix can reflect the importance of each service knowledge vector in the global context. If a service knowledge vector has a high similarity with multiple other vectors related to the improvement of academic performance, then it has a high importance in the global context.

[0023] Then, use this global similarity matrix to perform a weighted sum on the service knowledge vectors to obtain enhanced service knowledge vectors that incorporate global context information. Next, segment each enhanced service knowledge vector to generate multiple sub-vectors. For example, segment an enhanced service knowledge vector regarding a higher mathematics course along dimensions such as course content, study time, and learning effect. Apply the local attention function of the local attention mechanism based on the attention network to each sub-vector, and identify local attention foci by calculating the correlations between the elements within the sub-vector. For instance, in the sub-vector regarding study time, identify that the period from 7 to 9 pm every day is the concentrated study period for students taking the higher mathematics course as the local attention focus. Combine the local attention foci of each sub-vector to generate a local attention focus vector for the entire service knowledge vector.

[0024] Then, adopt the multi-level attention fusion strategy of the attention network to define multiple levels of attention heads, such as the semantic level (focusing on the semantic information of the course content), the behavior level (focusing on the learning behavior of students), and the resource level (focusing on the utilization of course resources). Apply these attention heads to each local attention focus vector simultaneously to obtain multiple levels of attention representations, and fuse these multiple levels of attention representations through weighted summation to generate the final service knowledge vector that incorporates multi-level attention information. Then, apply the predefined dynamic adjustment rules to the attention coefficients of each final service knowledge vector. For example, based on the time decay adjustment rule, if the course learning represented by a service knowledge vector occurred a long time ago, its attention coefficient may decrease according to time decay; based on the user preference adjustment rule, if a student indicates in their profile that they are not interested in mathematics, then the attention coefficient of the service knowledge vector related to the mathematics course will be adjusted accordingly; based on the service popularity adjustment rule, if a certain course is very popular on campus recently, the attention coefficient of the service knowledge vector related to that course may increase. Calculate the sum or maximum value of all dynamically adjusted attention coefficients, and divide each attention coefficient by this sum or maximum value to obtain the normalized attention coefficients.

[0025] For service acceptance portrait estimation through an estimation network, for each service knowledge vector, the service knowledge vector is weighted with the corresponding attention coefficient to generate a target weighted knowledge vector corresponding to each service knowledge vector. Assume that the first network functional unit in the estimation network is a multi-dimensional fusion output unit, and the second network functional unit is a sequence response unit. Through the multi-dimensional fusion output unit, service acceptance portrait estimation is performed based on multiple target weighted knowledge vectors corresponding to service push training examples, generating a first service acceptance estimation portrait corresponding to each service push node (such as different course push nodes) of the service push training example. At the same time, through the sequence response unit, service acceptance portrait estimation is performed based on multiple target weighted knowledge vectors corresponding to service push training examples, generating a second service acceptance estimation portrait corresponding to each service push node of the service push training example. In this way, the service acceptance estimation portraits corresponding to each service push node among multiple service push nodes of the service push training example are obtained.

[0026] Step S140, calculate the network training error based on the service acceptance description portrait corresponding to the target service push node of the service push training example and the service acceptance estimation portraits corresponding to each service push node of the service push training example.

[0027] Still taking the online course learning service in a smart campus as an example, assume that the target service push node is the advanced mathematics course push node. The service acceptance description portrait corresponding to this target service push node of the service push training example contains actual student acceptance situations, such as the proportion of students actually registering for the advanced mathematics course, the learning progress of students in the course, and other multi-dimensional accurate information. The service acceptance estimation portraits corresponding to each service push node (including the advanced mathematics course push node and other course push nodes) of the service push training example are estimated values obtained through the previous estimation network.

[0028] For example, at the advanced mathematics course push node, the service acceptance description portrait shows that actually 30% of the students have registered for the course and the average learning progress has reached 50%, while the first service acceptance estimation portrait obtained through the estimation network shows that it is expected that 25% of the students will register for the course and the average learning progress is 45%, and the second service acceptance estimation portrait shows that it is expected that 28% of the students will register for the course and the average learning progress is 48%. Based on these actual values and estimated values, a specific error calculation method (such as mean square error, etc.) is used to calculate the network training error. The error is calculated for each dimension (registration proportion, learning progress, etc.), and these errors are combined to obtain the network training error at the advanced mathematics course push node. Similarly, such calculations are also performed for other course push nodes, so as to obtain the network training errors of the entire service push training example at each service push node.

[0029] Step S150, optimize the neuron weight information in the service acceptance portrait estimation network for estimating the service push training example corresponding to the target service push node based on the network training error. After meeting the network convergence requirement, make a decision on the intelligent campus service push for the target push object according to the service acceptance portrait estimation network, where the service acceptance portrait estimation network includes the attention network and the estimation network.

[0030] In the scenario of intelligent campus online course learning service, optimize the neuron weight information in the service acceptance portrait estimation network based on the network training error calculated on the advanced mathematics course push node. Assume that the attention network in the service acceptance portrait estimation network contains multiple neurons, and these neurons play different roles in the process of processing the service knowledge vector, such as in the process of calculating the global similarity matrix, local attention focus, etc. The first network functional unit (multi-dimensional fusion output unit) and the second network functional unit (sequence response unit) in the estimation network also contain many neurons.

[0031] First, mask and label the target neuron weight information in the service acceptance portrait estimation network. Here, the target neuron weight information is other neuron weight information except the neuron weight information in the fully connected mapping unit of the estimation network for estimating the service push training example corresponding to the advanced mathematics course push node. For example, in the attention network, mask and label the neuron weight information related to other courses (non-advanced mathematics courses). Then, optimize the neuron weight information in the service acceptance portrait estimation network that is not masked and labeled according to the network training error. For example, if it is found that the error in calculating the global similarity matrix for the service knowledge vector related to the advanced mathematics course is large, adjust the weights of the relevant neurons in the attention network to improve the calculation accuracy in this regard.

[0032] Continuously perform such an optimization process until the network convergence requirement is met, that is, the network training error is reduced to an acceptable range. Once the network convergence requirement is met, the network can make a decision on the intelligent campus service push for the target push object (such as a specific student group) based on the service acceptance portrait. For example, for a new batch of students, obtain the service usage trajectory vectors of them (such as their learning trajectories in other courses) and the semantic vectors of campus service resource data (such as semantic vector representations of different online course resources) of multiple campus service resources to be pushed (such as different online course resources). The service acceptance portrait estimation network generates an estimated portrait of service acceptance corresponding to each service push node (such as different course push nodes) for the campus service resource data to be pushed based on these vectors. Based on these estimated portraits of service acceptance, determine the target campus service resource data (such as the most suitable online course for a certain student group) and the service push node corresponding to the target campus service resource data (such as the push timing and method of this course) from multiple campus service resource data to be pushed. Finally, based on the target campus service resource data and the service push node corresponding to the target campus service resource data, perform an intelligent campus service push for the target push object, for example, push the most suitable online course to a specific student group at the right time and in the right way.

[0033] Based on the above steps, in the embodiment of the present application, by obtaining service push training examples marked with service acceptance description portraits, and extracting features from the service usage trajectory data and campus service resource data in the service push training examples, multiple service knowledge vectors are generated. The attention network is used to obtain the attention coefficients of each service knowledge vector, and combined with the estimation network, the service acceptance portrait is estimated to generate an estimated portrait of service acceptance corresponding to multiple service push nodes. By calculating the network training error, the neuron weight information in the service acceptance portrait estimation network is optimized until the network converges. This method can accurately make a decision on the intelligent campus service push for the target push object, improve the accuracy of service push and user satisfaction, and achieve efficient and personalized push of intelligent campus services.

[0034] In a possible implementation manner, step S150 includes:

[0035] Step S151, perform masked annotation on the target neuron weight information in the service acceptance portrait estimation network, where the target neuron weight information is other neuron weight information except the neuron weight information used to estimate the service push training example corresponding to the target service push node in the fully connected mapping unit of the estimation network.

[0036] Step S152, optimize the neuron weight information in the service acceptance portrait estimation network that is not masked and annotated according to the network training error.

[0037] In this embodiment, in the context of the online course learning service in a smart campus, after the service acceptance portrait estimation network obtains the network training error through the previous steps, it begins to optimize the neuron weight information of the service push node used to estimate the service push training example, where the target service push node is a target service push node (such as the advanced mathematics course push node).

[0038] First, mask and label the target neuron weight information in the service acceptance portrait estimation network. The service acceptance portrait estimation network is a complex structure, including parts such as an attention network and an estimation network. There is a fully connected mapping unit in the estimation network. When estimating the service push training example corresponding to the advanced mathematics course push node, specific neuron weight information plays a key role. And the target neuron weight information to be masked and labeled here is the other neuron weight information except for these specific neuron weight information. For example, in the attention network part, there are some neurons used to process the service knowledge vector calculation related to other courses (such as English courses, physics courses, etc.). These neuron weight information is not directly related to the estimation of the advanced mathematics course push node, so it needs to be masked and labeled.

[0039] Next, based on the network training error, optimize the weight information of the neurons in the service acceptance portrait estimation network that are not masked and labeled. Since the network training error has been calculated previously, this error reflects the difference between the estimated value and the actual service acceptance description portrait at the advanced mathematics course push node. If, during the previous service acceptance portrait estimation process, it is found that for the service knowledge vector related to the advanced mathematics course, the results obtained in some links (such as the link of calculating the global similarity matrix under the global attention mechanism or the link of determining the local attention focus under the local attention mechanism) deviate significantly from the actual situation, this indicates that the relevant neuron weights may need to be adjusted. For example, if it is found that in the calculation of the global similarity matrix for the service knowledge vector related to the advanced mathematics course, due to the setting of some neuron weights, the importance attached to the factor of course difficulty is too high or too low, thus affecting the overall service acceptance estimation, then the weight of these unmasked and labeled neurons can be optimized and adjusted according to the feedback of the network training error. In the first network functional unit (such as the multi-dimensional fusion output unit) and the second network functional unit (such as the sequence response unit) in the estimation network, there is also neuron weight information related to the estimation of the advanced mathematics course push node. If it is found that a certain neuron weight in these units causes a large error when fusing information from different dimensions or processing information in sequence, it can also be optimized based on the network training error. By optimizing the weight information of the unmasked and labeled neurons in this targeted manner, the accuracy of the service acceptance portrait estimation network in estimating the advanced mathematics course push node can be gradually improved, thereby improving the performance of the entire intelligent campus service push system.

[0040] In a possible implementation manner, the estimation network includes a first network functional unit and a second network functional unit, and step S120 includes:

[0041] Step S121, clean, denoise, and normalize the service usage trajectory data, extract key behavior features, and at the same time perform semantic analysis, keyword extraction, and vectorization representation on the campus service resource data to generate corresponding resource feature vectors. Then fuse the key behavior features and the resource feature vectors to generate each service knowledge vector containing service information and resource semantics.

[0042] Step S130 includes:

[0043] Step S131, capture the global dependency relationships between each service knowledge vector by using the global attention mechanism of the attention network. Specifically, calculate the similarity between each service knowledge vector and all other service knowledge vectors to obtain a global similarity matrix, which reflects the importance of each service knowledge vector in the global context. Then, use the global similarity matrix to perform weighted summation on the service knowledge vectors to obtain an enhanced service knowledge vector that integrates global context information.

[0044] Step S132, segment each enhanced service knowledge vector to generate multiple sub-vectors, and apply a local attention function to each sub-vector based on the local attention mechanism of the attention network. This local attention function identifies local attention foci by calculating the correlations between the internal elements of the sub-vectors, and combines the local attention foci of each sub-vector to generate a local attention focus vector for the entire service knowledge vector.

[0045] Step S133, adopt the multi-level attention fusion strategy of the attention network to define multiple levels of attention heads, and each attention head is used to capture information at the corresponding level.

[0046] Step S134, apply the attention heads to each local attention focus vector simultaneously to obtain multiple levels of attention representations, and fuse the multiple levels of attention representations by weighted summation to generate a final service knowledge vector that integrates multi-level attention information. The levels include the semantic level, the behavior level, and the resource level.

[0047] Step S135, apply a predefined dynamic adjustment rule to the attention coefficients of each final service knowledge vector to obtain dynamically adjusted attention coefficients. The dynamic adjustment rules include an adjustment rule based on time decay, an adjustment rule based on user preferences, and an adjustment rule based on service popularity.

[0048] Step S136, calculate the sum or maximum value of all dynamically adjusted attention coefficients, and divide each attention coefficient by the sum or maximum value to obtain normalized attention coefficients.

[0049] Step S137, for each service knowledge vector, perform weighted calculation on the service knowledge vector and the attention coefficient corresponding to the service knowledge vector to generate a target weighted knowledge vector corresponding to each service knowledge vector.

[0050] Step S138, estimate the service acceptance portrait through the first network functional unit based on the multiple target weighted knowledge vectors corresponding to the service push training examples, and generate the first service acceptance estimation portrait corresponding to each service push node for the service push training examples.

[0051] Step S139: According to the multiple target weighted knowledge vectors corresponding to the service push training examples by the second network function unit, estimate the service acceptance portrait, and generate the second service acceptance estimation portrait corresponding to each service push node for the service push training example.

[0052] Step S140 may include: calculating the network training error according to the service acceptance description portrait of the service push node corresponding to the service push training example, the first service acceptance estimation portrait corresponding to each service push node for the service push training example, and the second service acceptance estimation portrait corresponding to each service push node for the service push training example.

[0053] Step S150 may include: optimizing the neuron weight information of the attention network based on the network training error, optimizing the neuron weight information in the first network function unit for estimating the service push training example corresponding to the target service push node, and optimizing the neuron weight information of the second network function unit.

[0054] In this embodiment, on the smart campus platform, the service usage trajectory data for the online course learning service contains a lot of information. For example, the time record of students logging in to the smart campus online course platform, the browsing duration on different course pages, the timestamps of operations related to the course (such as clicking on course materials, submitting assignments, participating in course discussions, etc.). First, clean this service usage trajectory data to remove those obviously incorrect records, such as abnormal login times caused by system failures or unreasonable operation records caused by misoperations. Then perform denoising processing, which is to exclude those small fluctuation data that may interfere with the analysis. For example, occasional rapid page switching may be a misoperation rather than a real learning behavior and needs to be removed. Then perform normalization processing to convert data of different magnitudes into the same magnitude. For example, the login time may be in seconds, while the assignment submission time interval may be in days. Through normalization, it can be analyzed on the same scale. Extract key behavior features from the processed service usage trajectory data. For example, a student spends more than a certain duration (such as 5 hours) per week on course material browsing and assignment completion in a specific online course (such as advanced mathematics course). This is a key behavior feature, indicating that the student has a high degree of participation in this course.

[0055] Meanwhile, for campus service resource data, taking the resources related to online course learning as an example, it includes course videos, course handouts, course assignments, etc. Conduct semantic analysis on the course videos, which involves recognizing the speech content in the videos and converting it into text, and then analyzing the semantic information in these texts, such as identifying key concepts, theories, etc. in the video explanations. For the course handouts, extract keywords, such as extracting keywords closely related to the core content of the course like "function", "limit", "derivative", etc. from the advanced mathematics course handout. Then, vectorize the extracted keywords and relevant information in the course assignments to generate corresponding resource feature vectors. For example, convert the keyword "function" into a vector, and the dimension of this vector may be related to the knowledge system of the entire smart campus, and each element in the vector represents the eigenvalue of this keyword in different knowledge dimensions. Finally, fuse the key behavior features extracted from the service usage trajectory data and the resource feature vectors obtained from the campus service resource data to generate each service knowledge vector containing service information and resource semantics. For example, a service knowledge vector may include the student's participation in the advanced mathematics course (key behavior feature) and the vector representation of the core knowledge points in the course (such as functions, limits, etc.) (resource feature vector). Such a service knowledge vector can comprehensively reflect the relevant information in the service push training examples and provide a rich data basis for subsequent analysis.

[0056] In the context of online course learning services in a smart campus, there are multiple service knowledge vectors, each containing information in different aspects. The global attention mechanism of the attention network is adopted to capture the global dependencies among various service knowledge vectors. For example, there is a service knowledge vector representing the learning behavior of students in advanced mathematics courses and the characteristics of course resources, and another service knowledge vector representing similar information about students' English courses. By calculating the similarity between each service knowledge vector and all other service knowledge vectors, a global similarity matrix is constructed. Specifically, for the service knowledge vector representing the advanced mathematics course, calculate its similarity with the service knowledge vectors related to all other courses such as English courses, physics courses, computer courses, etc. This similarity calculation may be based on distance measurement methods between vectors, such as Euclidean distance or cosine similarity, etc. The obtained global similarity matrix reflects the importance of each service knowledge vector in the global context. If a service knowledge vector has a high similarity with multiple other vectors related to the improvement of learning performance, for example, a service knowledge vector representing that students actively participate in course discussions and have a high utilization rate of course resources has a high similarity with other similar actively learning-related service knowledge vectors, then it has a high importance in the global context. Then, use this global similarity matrix to perform a weighted sum on the service knowledge vectors to obtain an enhanced service knowledge vector that integrates global context information. For example, for the service knowledge vector of the advanced mathematics course, according to its importance weight in the global similarity matrix, its original vector representation is weighted and adjusted to obtain an enhanced service knowledge vector that contains global context information (such as the relationship with the learning behavior and resource utilization of other courses).

[0057] Segment each enhanced service knowledge vector. For example, segment an enhanced service knowledge vector about a higher mathematics course according to dimensions such as course content, learning time, and learning effect to generate multiple sub-vectors. Apply the local attention function to each sub-vector based on the local attention mechanism of the attention network. Taking the sub-vector of learning time as an example, this sub-vector contains internal elements such as the time distribution of students learning the higher mathematics course at different time periods. The local attention function identifies the local attention focus by calculating the correlation between the internal elements of the sub-vector. For example, by analysis, it is found that the proportion of time for students to learn the higher mathematics course from 7 to 9 pm every day is the highest, and the learning efficiency during this time period is relatively high (which can be reflected by indicators such as the quality of homework completion), then this time period is the local attention focus of the learning time sub-vector. Perform such operations on each sub-vector, and then merge the local attention focuses of each sub-vector to generate a local attention focus vector for the entire service knowledge vector. For example, merge the local attention focuses of sub-vectors such as course content, learning time, and learning effect to form a local attention focus vector that comprehensively reflects the key information of the higher mathematics course service knowledge vector at the local level.

[0058] Adopt the multi-level attention fusion strategy of the attention network to define multiple levels of attention heads, where the levels include the semantic level, the behavior level, and the resource level. Taking the online course learning service as an example, the attention heads at the semantic level mainly focus on the semantic information of the course content. For example, in a higher mathematics course, they focus on the semantic relationships between mathematical concepts and theorems in the course; the attention heads at the behavior level focus on the learning behaviors of students, such as the learning time arrangement of students and the frequency of participating in course discussions; the attention heads at the resource level focus on the utilization of course resources, such as the number of times the course video is viewed and the number of times the course handout is downloaded. Apply these attention heads to each local attention focus vector simultaneously to obtain attention representations at multiple levels. For example, for the local attention focus vector of the higher mathematics course, obtain the attention representation regarding the semantics of the course content through the attention head at the semantic level, obtain the attention representation regarding learning behaviors through the attention head at the behavior level, and obtain the attention representation regarding the utilization of course resources through the attention head at the resource level. Then fuse these attention representations at multiple levels through weighted summation to generate the final service knowledge vector that integrates multi-level attention information. This final service knowledge vector synthesizes information in multiple aspects such as the semantics of the course content, students' learning behaviors, and the utilization of course resources, and can more comprehensively and accurately reflect the relevant situation of the online course learning service.

[0059] Apply a predefined dynamic adjustment rule to the attention coefficients of each final service knowledge vector. In the online course learning service of a smart campus, the dynamic adjustment rules include an adjustment rule based on time decay, an adjustment rule based on user preferences, and an adjustment rule based on service popularity. Regarding the adjustment rule based on time decay, if the course learning represented by a service knowledge vector occurred a long time ago, for example, a student had a learning behavior record for a certain online course one year ago, as time goes by, the importance of this record may decrease, and its attention coefficient is adjusted accordingly according to time decay. For the adjustment rule based on user preferences, if a student indicates in their personal profile that they are not interested in mathematics, then the attention coefficient of the service knowledge vector related to the mathematics course will be adjusted accordingly. For example, if a service knowledge vector is related to the advanced mathematics course and the student's preference shows that they are not interested in mathematics, then the attention coefficient of this service knowledge vector will be reduced to reflect the impact of such user preferences on service acceptance. For the adjustment rule based on service popularity, if a certain course is very popular on campus recently, for example, a newly emerging computer technology-related course has received a lot of attention and learning from students on campus, the attention coefficient of the service knowledge vector related to this course may increase. Calculate the sum or maximum value of all the dynamically adjusted attention coefficients, and divide each attention coefficient by this sum or maximum value to obtain the normalized attention coefficient. The normalized attention coefficient obtained in this way can reflect the relative importance of each service knowledge vector on a unified scale, providing accurate weight information for the subsequent estimation of the service acceptance portrait.

[0060] For each service knowledge vector in the online course learning service of the smart campus, perform a weighted calculation on the service knowledge vector and the attention coefficient corresponding to the service knowledge vector to generate a target weighted knowledge vector corresponding to each service knowledge vector. For example, for a service knowledge vector related to the advanced mathematics course, which contains information such as course participation and course resource characteristics, multiply this service knowledge vector by the attention coefficient obtained in the previous steps to obtain the target weighted knowledge vector. This target weighted knowledge vector highlights to a certain extent the importance weight of the information related to the advanced mathematics course in the entire service knowledge system, and is the result of comprehensively considering the content of the service knowledge vector itself and its importance in the global and local environments.

[0061] Suppose the first network function unit is a multi-dimensional fusion output unit. Through this multi-dimensional fusion output unit, the service acceptance portrait estimation is carried out based on multiple target weighted knowledge vectors corresponding to the service push training examples, and the first service acceptance estimation portrait corresponding to each service push node (such as different course push nodes) of the service push training example is generated. In the scenario of the intelligent campus online course learning service, the multi-dimensional fusion output unit will comprehensively consider the multi-dimensional information in each target weighted knowledge vector, such as the learning behaviors of students in different courses, the characteristics of course resources, and the importance weights of this information, etc. For the advanced mathematics course push node, the multi-dimensional fusion output unit will fuse the information in these vectors according to multiple target weighted knowledge vectors related to the advanced mathematics course, and output the service acceptance estimation portrait on this node. This portrait may contain multi-dimensional estimation information such as the acceptance probability of students for the advanced mathematics course push and the possible learning participation degree, etc.

[0062] Through the second network function unit (suppose it is a sequence response unit), the service acceptance portrait estimation is carried out based on multiple target weighted knowledge vectors corresponding to the service push training examples, and the second service acceptance estimation portrait corresponding to each service push node of the service push training example is generated. In the online course learning service, the sequence response unit will process the target weighted knowledge vectors in a certain order. For example, in the order of the learning time of the course or the increasing order of the course difficulty, etc. For the advanced mathematics course push node, the sequence response unit will process the target weighted knowledge vectors related to the advanced mathematics course according to this order, considering the response of students to the course push at different learning stages or different course difficulties, so as to generate the second service acceptance estimation portrait on this node. This portrait may be different from the first service acceptance estimation portrait in some aspects. For example, it may focus more on the influence of sequential factors in the learning process on the service acceptance.

[0063] Calculate the network training error based on the service acceptance description portrait of the service push node corresponding to the service push training example (such as the advanced mathematics course push node), the first service acceptance estimation portrait corresponding to each service push node of the service push training example, and the second service acceptance estimation portrait corresponding to each service push node of the service push training example. In the scenario of the intelligent campus online course learning service, the service acceptance description portrait contains accurate information in multiple dimensions such as the actual student acceptance situation, such as the proportion of students actually registering for the advanced mathematics course and the actual learning progress of students in the course. The first service acceptance estimation portrait and the second service acceptance estimation portrait are estimated values obtained through the previous estimation network. For example, on the advanced mathematics course push node, the service acceptance description portrait shows that actually 30% of the students have registered for the course and the average learning progress has reached 50%, while the first service acceptance estimation portrait shows that it is expected that 25% of the students will register for the course and the average learning progress is 45%, and the second service acceptance estimation portrait shows that it is expected that 28% of the students will register for the course and the average learning progress is 48%. According to these actual values and estimated values, a specific error calculation method (such as mean square error, etc.) is used to calculate the network training error. The error is calculated for each dimension (proportion of registered students, learning progress, etc.), and these errors are integrated to obtain the network training error on the advanced mathematics course push node. Similarly, such calculations are also performed on other course push nodes, so as to obtain the network training error of the entire service push training example on each service push node.

[0064] Optimize the neuron weight information in the service acceptance portrait estimation network for estimating the service push training example corresponding to the target service push node (such as the advanced mathematics course push node) based on the network training error. Specifically, optimize the neuron weight information of the attention network because the attention network plays a key role in processing service knowledge vectors, calculating attention coefficients, etc. If it is found in the previous analysis that there are errors in the attention network in calculating the global similarity matrix or local attention focus vectors, etc., resulting in inaccurate final service acceptance estimation, the neuron weights in the attention network are adjusted according to the network training error. For example, if it is found that the weight setting of a certain neuron is unreasonable when calculating the global similarity between the advanced mathematics course and other courses, resulting in a deviation in the overall similarity calculation and thus affecting the service acceptance estimation, the weight of this neuron is optimized.

[0065] Meanwhile, optimize the neuron weight information in the first network functional unit (multi-dimensional fusion output unit) for estimating the service push training example corresponding to the advanced mathematics course push node. If there is an error when the multi-dimensional fusion output unit generates the first service acceptance estimation portrait by fusing multi-dimensional information, for example, the neuron weights for some key dimensions (such as the dimension of the impact of course difficulty on service acceptance) are set improperly, resulting in a large deviation between the estimated value and the actual service acceptance description portrait, then adjust these neuron weights according to the network training error.

[0066] In addition, optimize the neuron weight information of the second network functional unit (sequence response unit). If there is an error when the sequence response unit processes the target weighted knowledge vector in order to generate the second service acceptance estimation portrait, for example, in considering the impact of the course learning order on service acceptance, some neuron weights are set inaccurately, resulting in the estimated result not matching the actual situation, then adjust these neuron weights according to the network training error. By comprehensively optimizing the neuron weight information in the service acceptance portrait estimation network in this way, the accuracy of the network in estimating the service acceptance for each service push node can be gradually improved, thereby enhancing the overall performance of the intelligent campus service push system.

[0067] In a possible implementation manner, the first network functional unit is a multi-dimensional fusion output unit, and the second network functional unit is a sequence response unit or a feature interaction relationship unit.

[0068] In a possible implementation manner, the estimation network further includes a third network functional unit, and step S130 may further include:

[0069] Estimate the service acceptance portrait based on the multiple target weighted knowledge vectors corresponding to the service push training example by the third network functional unit, and generate the third service acceptance estimation portrait corresponding to the service push training example for each service push node, where the second network functional unit is a sequence response unit and the third network functional unit is a feature interaction relationship unit.

[0070] Calculating the network training error based on the service acceptance description portrait of the service push node corresponding to the service push training example, the first service acceptance estimation portrait of the service push training example corresponding to each service push node, and the second service acceptance estimation portrait of the service push training example corresponding to each service push node includes: calculating the network training error based on the service acceptance description portrait of the service push training example corresponding to the service push node, the first service acceptance estimation portrait of the service push training example corresponding to each service push node, the second service acceptance estimation portrait of the service push training example corresponding to each service push node, and the third service acceptance estimation portrait of the service push training example corresponding to each service push node.

[0071] In this embodiment, in the online course learning service scenario of a smart campus, for the case where the first network functional unit is a multi-dimensional fusion output unit, the second network functional unit is a sequence response unit or a feature interaction relationship unit, and the estimation network further includes a third network functional unit (feature interaction relationship unit), the following is a detailed example of each step.

[0072] First, in the stage of generating the service acceptance estimation portrait, take the advanced mathematics course push node as an example. The multi-dimensional fusion output unit estimates the service acceptance portrait based on multiple target weighted knowledge vectors corresponding to the service push training example, and generates the first service acceptance estimation portrait. Suppose the service push training example contains online course learning related data of multiple students. For example, the target weighted knowledge vector of student A includes his participation in the advanced mathematics course (such as weekly study duration, homework completion situation, etc.) and the course resource utilization situation (such as the number of times of watching course videos, the number of times of downloading lecture notes, etc.). The multi-dimensional fusion output unit comprehensively considers this multi-dimensional information. For the advanced mathematics course push node, it may analyze the first service acceptance estimation portrait at this node based on the similar target weighted knowledge vectors of multiple students. For example, the portrait shows that it is expected that 25% of the students will register for the course and the average learning progress is 45%.

[0073] The sequence response unit estimates the service acceptance portrait based on multiple target weighted knowledge vectors corresponding to the same service push training example, and generates the second service acceptance estimation portrait. Since the sequence response unit processes the target weighted knowledge vectors in a certain order, such as in the time order of course learning. For the advanced mathematics course, it will consider the responses of students to the course push at different learning stages. For example, the responses of students in different time orders such as the early course promotion stage, the mid-term in-depth course learning stage, and the late review stage. Based on these order-related analyses, the second service acceptance estimation portrait generated for the advanced mathematics course push node may show that it is expected that 28% of the students will register for the course and the average learning progress is 48%.

[0074] The third network function unit (feature interaction relationship unit) also estimates the service acceptance portrait based on multiple target weighted knowledge vectors corresponding to the service push training examples, and generates a third service acceptance estimation portrait. In the intelligent campus online course learning service, the feature interaction relationship unit mainly focuses on the impact of the interaction relationship between different features on the service acceptance. Taking the advanced mathematics course as an example, it analyzes how the interaction relationship between the students' learning foundation (such as the grades of previous related courses) and the course resources (such as the course difficulty, the richness of course materials, etc.) affects the students' acceptance of the course push. By analyzing the relevant data of many students, the third service acceptance estimation portrait generated for the advanced mathematics course push node may show that it is expected that 26% of the students will register for the course and the average learning progress is 46%.

[0075] Then, in the stage of calculating the network training error, the network training error is calculated based on the service acceptance description portrait corresponding to the service push training example for the service push node (here is the advanced mathematics course push node) and the first service acceptance estimation portrait, the second service acceptance estimation portrait, and the third service acceptance estimation portrait corresponding to each service push node of the service push training example. The service acceptance description portrait contains the actual student acceptance situation, such as the fact that actually 30% of the students register for the course and the average learning progress reaches 50%. For the dimension of the registration number ratio, the first service acceptance estimation portrait is expected to be 25%, which deviates from the actual 30%; the second service acceptance estimation portrait is expected to be 28%, which also has a difference from the actual situation; the third service acceptance estimation portrait is expected to be 26% and there is also an error. The situation is similar for the learning progress dimension. Using specific error calculation methods such as the mean square error, the error of each dimension is calculated separately, and then these errors are combined to obtain the network training error on the advanced mathematics course push node. For other course push nodes, such as the English course, the physics course, etc., the calculation is also carried out in the same way, so as to obtain the network training error of the entire service push training example on each service push node. By comprehensively considering the comparison between multiple service acceptance estimation portraits and the actual service acceptance description portrait to calculate the network training error, it can more accurately reflect the accuracy of the estimation network in different aspects and provide an accurate basis for optimizing the neuron weight information in the subsequent stage.

[0076] In a possible implementation manner, step S150 includes:

[0077] If the campus service resource label in the service push training example is the target campus service resource label, optimize the neuron weight information of the neurons in the service acceptance portrait estimation network that are used to estimate the service push node of the service push training example as the target service push node according to the network training error.

[0078] If the campus service resource label in the service push training example is not the target campus service resource label, optimize the weight information of at least one neuron in the service acceptance portrait estimation network according to the network training error.

[0079] In this embodiment, taking the advanced mathematics course as an example of the target service push node, assume that there are many service push training examples, and each example contains relevant information such as campus service resource labels. When the campus service resource label in the service push training example is the target campus service resource label (such as labels of specific textbook versions, specific teaching video series directly related to the advanced mathematics course), optimize the weight information of the neuron in the service acceptance portrait estimation network for estimating the service push node of the advanced mathematics course for this service push training example according to the network training error.

[0080] For example, in the attention network part, some neurons are responsible for processing the service knowledge vector calculation related to the advanced mathematics course. For example, in the link of calculating the global similarity matrix, if it is found that for the service knowledge vector related to the advanced mathematics course, due to the setting of some neuron weights, a large error occurs when comparing its similarity with other service knowledge vectors (such as vectors related to other courses), thus affecting the overall service acceptance estimation. For example, when comparing the similarity between the service knowledge vector of the advanced mathematics course and the service knowledge vector of the physics course, due to an unreasonable setting of a certain neuron weight, two vectors that should have a low similarity are determined to have a high similarity, thereby affecting the accuracy of the enhanced service knowledge vector that fuses the global context information obtained by subsequent weighted summation. According to the network training error feedback, optimize and adjust the neuron weights related to the advanced mathematics course to improve the calculation accuracy in this regard.

[0081] In the estimation network, if the first network functional unit multi-dimensional fusion output unit sets the weights of neurons for some key dimensions (such as the dimension of the impact of course difficulty on service acceptance) improperly when fusing multi-dimensional information to generate the first service acceptance estimation portrait, resulting in a large deviation from the actual service acceptance description portrait (such as multi-dimensional accurate information such as the actual proportion of students registering for the advanced mathematics course and the actual learning progress of students in the course). For example, in the multi-dimensional fusion output unit, the dimension of course difficulty originally has a greater impact on service acceptance, but due to too low neuron weight setting, the influence of this factor is not accurately reflected in the generated first service acceptance estimation portrait. For example, in the actual situation, the high course difficulty leads to a low registration proportion, but this trend is not reflected in the estimation portrait. Optimize the weight information of the neurons in the multi-dimensional fusion output unit for estimating the service push node of the advanced mathematics course in the service push training example according to the network training error.

[0082] Similarly, for the second network functional unit (such as the sequence response unit or the feature interaction relationship unit), if there is an error in the sequential processing of the target weighted knowledge vector to generate the second or third service acceptance estimation portrait, for example, when the target weighted knowledge vector related to the advanced mathematics course is processed in the sequence response unit according to the chronological order of course learning, some neuron weights are not set accurately, resulting in the consideration of the impact of different learning stages on service acceptance not being consistent with the actual situation. For example, in the early promotion stage of the course, there was not enough course guidance information, so the number of registered people was small, but this was not reflected in the estimated portrait. These neuron weights are adjusted according to the network training error.

[0083] When the campus service resource label in the service push training example is not the target campus service resource label (for example, it is a resource label related to English courses, and the target is the advanced mathematics course push node), at least one neuron weight information of the service acceptance portrait estimation network is optimized according to the network training error. Since the entire service acceptance portrait estimation network is a complex whole, there are interrelationships between each part. Although the campus service resource label is not related to the target service push node, some parts of the network may indirectly affect the estimation of the target service push node.

[0084] For example, in the attention network, some general neurons are responsible for processing the basic relationship calculations between different courses. If the weights of these neurons are not set properly, it may affect the performance of the overall network, and thus affect the estimation of the push node for advanced mathematics courses. Even if the campus service resource labels of the current service push training examples are not related to advanced mathematics courses, the unreasonable weights of these neurons may cause deviations in the calculation of the global similarity matrix and other links, thereby affecting the entire network's estimation of the service acceptance of the push node for advanced mathematics courses. According to the network training error, the neuron weight information that may affect the overall network performance is optimized.

[0085] In the estimation network, if the multidimensional fusion output unit of the first network functional unit has a problem with setting the neuron weight when processing other service push training samples, it may affect the estimation of the advanced mathematics course push node through the internal association mechanism of the network. For example, when the multidimensional fusion output unit processes service push training samples related to English courses, due to the unreasonable setting of a certain neuron weight, an error occurs when fusing multidimensional information. This error may affect the estimation of the advanced mathematics course push node through the information transmission and interaction mechanism within the network. Based on the network training error, the neuron weight information in the multidimensional fusion output unit that may affect the estimation of the advanced mathematics course push node is optimized.

[0086] Similarly, for the second network function unit (such as the sequence response unit or the feature interaction relationship unit), the error caused by improper neuron weight setting when processing other service push training examples may affect the estimation of the higher mathematics course push node through the association between networks. For example, when the sequence response unit processes the target weighted knowledge vector related to the English course in sequence, there is an incorrect neuron weight setting, resulting in a deviation in analyzing the impact of the learning sequence on service acceptance. This deviation may indirectly affect the estimation of the higher mathematics course push node through the internal association of the network. According to the network training error, optimize the neuron weight information that may have an impact, so as to improve the accuracy of the entire service acceptance portrait estimation network when estimating the higher mathematics course push node.

[0087] In a possible implementation manner, step S150 includes:

[0088] Step S151, obtain the service usage trajectory vector of the target push object and the campus service resource semantic vectors of multiple campus service resource data to be pushed.

[0089] Step S152, through the service acceptance portrait estimation network, generate service acceptance estimation portraits corresponding to each service push node for the campus service resource data to be pushed according to the service usage trajectory vector and the campus service resource semantic vectors of each campus service resource data to be pushed.

[0090] Step S153, according to the service acceptance estimation portraits corresponding to each campus service resource data to be pushed for each service push node, determine the target campus service resource data and the service push node corresponding to the target campus service resource data from the multiple campus service resource data to be pushed.

[0091] Step S154, based on the target campus service resource data and the service push node corresponding to the target campus service resource data, perform intelligent campus service push on the target push object.

[0092] In a possible implementation manner, step S151 includes:

[0093] Step S1511, obtain a service push task, where the object index information and service demand knowledge points of the target push object are marked in the service push task.

[0094] Step S1512, based on the service push task, obtain the service usage trajectory vector of the target push object.

[0095] Step S1513, based on the service demand knowledge points, obtain multiple campus service resource data to be pushed, and extract the campus service resource semantic vectors of each campus service resource data to be pushed.

[0096] In a possible implementation manner, step S153 includes: according to the service acceptance estimation portraits corresponding to each campus service resource data to be pushed for each service push node, extracting the campus service resource data corresponding to the service acceptance estimation portraits whose probability values meet the set conditions as the target campus service resource data, and using the service push node corresponding to the service acceptance estimation portrait whose probability value meets the set conditions in the target campus service resource data as the service push node corresponding to the target campus service resource data.

[0097] In this embodiment, when obtaining a service push task, the service push task is marked with the object index information of the target push object and the service demand knowledge points. For example, in the service push task, the target push object is a specific grade student group of a certain major, and its object index information clearly identifies the characteristics of this student group, such as the major code and grade number where they are located. The service demand knowledge points are the knowledge demands related to the students' academic development or campus life, such as the in-depth learning demand for advanced mathematics courses or the knowledge demand for campus cultural activity organization, etc.

[0098] Then, taking the online course learning service as an example, the service usage trajectory vector of the target push object includes various operation record information related to the courses of this student group on the smart campus platform. For example, the login time of the student group on different online course platforms, the stay duration on the advanced mathematics course page, the interaction frequency in the course-related discussion area, the assignment submission time and quality, etc. These information comprehensively reflect the behavior patterns and participation degrees of the target push object in the past online course learning.

[0099] Meanwhile, if the service demand knowledge point is the in-depth learning of advanced mathematics courses, then the campus service resource data to be pushed may include electronic materials supporting different versions of advanced mathematics textbooks, advanced mathematics course videos recorded by different teachers, online tutoring courses, etc. For each campus service resource data to be pushed, its campus service resource semantic vector is extracted. Taking the advanced mathematics course video as an example, by identifying the voice content in the video and converting it into text, then analyzing the semantic information in these texts, extracting concepts closely related to the core content of the course, such as limits, derivatives, integrals, etc. as keywords, and then converting these keywords into vector form. The dimension of this vector is related to the knowledge system of the entire smart campus, and each element in the vector represents the eigenvalue of this keyword in different knowledge dimensions, so as to obtain the campus service resource semantic vector of this course video. For other campus service resource data to be pushed, such as electronic textbooks and online tutoring courses, etc., their campus service resource semantic vectors are also extracted in a similar way.

[0100] Next, taking the data of campus service resources to be pushed related to the advanced mathematics course as an example, the attention network in the service acceptance portrait estimation network first captures the relationship between the service usage trajectory vector and the campus service resource semantic vector. For example, when calculating the global similarity matrix, the historical learning behavior of the target push object in the advanced mathematics course (part of the information in the service usage trajectory vector) is compared with the campus service resource semantic vector of the advanced mathematics course video to be pushed to determine the similarity between them, reflecting the importance of the course video for the target push object in the global context. Each piece of campus service resource data to be pushed is processed in this way to obtain an enhanced service knowledge vector that integrates global context information.

[0101] Then, for example, for the enhanced service knowledge vector of the advanced mathematics course video, it is segmented according to dimensions such as course content difficulty, learning time requirements, and expected learning effects. Based on the local attention mechanism, the local attention focus of each sub-vector is determined, and then the final service knowledge vector is obtained through multi-level attention fusion (such as semantic-level attention to the mathematical concept relationship of the course content, behavior-level attention to the learning behavior habits of the target push object, and resource-level attention to the utilization method of the course resources, etc.). According to the predefined dynamic adjustment rules (such as the adjustment rule based on time decay, if the target push object last studied the advanced mathematics course a long time ago, the attention coefficient of the relevant service knowledge vector will decrease; the adjustment rule based on user preference, if the target push object shows no interest in mathematics in previous learning behaviors, the attention coefficient of the relevant service knowledge vector will also decrease; the adjustment rule based on service popularity, if the current advanced mathematics course is very popular on campus, the attention coefficient of the relevant service knowledge vector will increase), the attention coefficient is adjusted to obtain a normalized attention coefficient.

[0102] After that, the first network functional unit (multi-dimensional fusion output unit), the second network functional unit (such as the sequence response unit), and the third network functional unit (such as the feature interaction relationship unit) in the estimation network estimate the service acceptance portrait based on multiple target weighted knowledge vectors corresponding to the service push training examples, and generate service acceptance estimation portraits of the campus service resource data to be pushed corresponding to each service push node (such as different teaching stage push nodes of advanced mathematics courses). The multi-dimensional fusion output unit comprehensively considers multi-dimensional information in the target weighted knowledge vectors, such as the learning behaviors of the target push object in different courses, the characteristics of course resources, and the importance weights of this information, etc. For different push nodes of the advanced mathematics course (such as the course preview stage, the middle stage of course learning, the course review stage, etc.), it fuses the information in these vectors and outputs the service acceptance estimation portrait for these nodes. The sequence response unit processes the target weighted knowledge vectors in a certain order (such as in the time order of course learning or in the increasing order of course difficulty, etc.). For different push nodes of the advanced mathematics course, considering the response of the target push object to the course push at different learning stages or different course difficulties, it generates the service acceptance estimation portrait for these nodes. The feature interaction relationship unit focuses on the impact of the interaction relationship between different features on the service acceptance, analyzes how the interaction relationship between the learning foundation of the target push object (such as the grades of previous related courses) and the course resources (such as course difficulty, richness of course materials, etc.) affects the acceptance of different push nodes of the advanced mathematics course by the target push object, and thus generates the corresponding service acceptance estimation portrait.

[0103] Finally, for the campus service resource data to be pushed for the advanced mathematics course, including electronic materials of different versions of textbooks, course videos recorded by multiple teachers, and online tutoring courses, etc., there are corresponding service acceptance estimation portraits at different service push nodes (such as the course preview stage, the middle stage of course learning, the course review stage). Assuming that the set condition is that the registration probability value in the service acceptance estimation portrait at the course preview stage is the highest and the learning participation prediction value is good, then the campus service resource data to be pushed that meets this set condition (such as the course preview video recorded by a certain teacher) is determined as the target campus service resource data, and the service push node corresponding to this set condition in this target campus service resource data (such as the course preview stage) is determined as the service push node corresponding to this target campus service resource data.

[0104] In the above example, based on the determined target campus service resource data (a course preview video recorded by a certain teacher) and the corresponding service push node (course preview stage), the intelligent campus service push system will push the course preview video recorded by this teacher to this student group when the target push object (a specific grade student group of a certain major) is in the course preview stage. Such a push decision is made based on the analysis of the service usage trajectory vector of the target push object, the campus service resource semantic vector of the campus service resource data to be pushed, and the service acceptance estimation portrait generated by the service acceptance portrait estimation network, thus realizing the precise push of intelligent campus services, improving the effectiveness and pertinence of service push, meeting the service demand knowledge points of the target push object, and helping to improve the learning experience and learning effect of the target push object in the intelligent campus.

[0105] Figure 2 FIG. shows the hardware structure diagram of the intelligent campus service push system 100 based on artificial intelligence for implementing the above-mentioned intelligent campus service push method based on artificial intelligence, as Figure 2 shown, the intelligent campus service push system 100 based on artificial intelligence may include a processor 110, a machine-readable storage medium 120, a bus 130, and a communication unit 140.

[0106] The machine-readable storage medium 120 may store data and / or instructions. In some embodiments, the machine-readable storage medium 120 may store data obtained from an external terminal. In some embodiments, the machine-readable storage medium 120 may store the data and / or instructions used by the intelligent campus service push system 100 based on artificial intelligence to execute or use to complete the exemplary methods described in the present invention.

[0107] In a specific implementation process, one or more processors 110 execute the computer-executable instructions stored in the machine-readable storage medium 120, so that the processor 110 can execute the intelligent campus service push method based on artificial intelligence in the above method embodiments. The processor 110, the machine-readable storage medium 120, and the communication unit 140 are connected through the bus 130, and the processor 110 can be used to control the transceiver actions of the communication unit 140.

[0108] The specific implementation process of the processor 110 can refer to the various method embodiments executed by the intelligent campus service push system 100 based on artificial intelligence above. The implementation principle and technical effects are similar, and will not be elaborated here in this embodiment.

[0109] In addition, an embodiment of the present invention further provides a readable storage medium, in which computer-executable instructions are preset. When the processor executes the computer-executable instructions, the above-mentioned intelligent campus service push method based on artificial intelligence is implemented.

[0110] It should be noted that, in order to simplify the description of the present invention disclosure and thus help the understanding of one or more embodiments of the invention, in the foregoing description of the embodiments of the present invention, sometimes multiple features are merged into one embodiment, drawing or description thereof.

Claims

1. A method for pushing smart campus services based on artificial intelligence, characterized in that: The method comprises: Obtaining a service push training sample of a smart campus service, wherein the service push training sample is annotated with a service acceptance description portrait corresponding to a target service push node of the service push training sample; Performing feature extraction on the service usage trajectory data and campus service resource data in the service push training sample to generate a plurality of service knowledge vectors included in the service push training sample; Obtaining the attention coefficient of each service knowledge vector through the attention network, and estimating the service acceptance portrait according to the multiple service knowledge vectors included in the service push training sample and the attention coefficient of each of the service knowledge vectors through the estimation network, to generate a service acceptance estimation portrait of each service push node in the multiple service push nodes corresponding to the service push training sample; Calculate the network training error based on the service acceptance description portrait of the target service push node corresponding to the service push training sample and the service acceptance estimation portrait of each service push node corresponding to the service push training sample; The neuron weight information of the service push training sample corresponding to the target service push node in the service acceptance portrait estimation network is optimized based on the network training error until the network convergence requirement is met, and then a smart campus service push decision is made for the target push object based on the service acceptance portrait estimation network, wherein the service acceptance portrait estimation network includes the attention network and the estimation network.

2. The method for pushing smart campus services based on artificial intelligence according to claim 1, characterized in that: The step of optimizing the neuron weight information of the service acceptance portrait estimation network for estimating the service push training sample corresponding to the target service push node based on the network training error includes: Shielding and marking target neuron weight information in the service acceptance portrait estimation network, wherein the target neuron weight information is other neuron weight information except for the neuron weight information used in the fully connected mapping unit of the estimation network to estimate the neuron weight information corresponding to the target service push node corresponding to the service push training sample; Based on the network training error, the unshielded and labeled neuron weight information in the service acceptance portrait estimation network is optimized.

3. The method for pushing smart campus services based on artificial intelligence according to claim 1, characterized in that: The estimation network includes a first network function unit and a second network function unit. The step of extracting features from the service usage trajectory data and campus service resource data in the service push training sample to generate a plurality of service knowledge vectors included in the service push training sample includes: Cleaning, denoising and normalizing the service usage trajectory data to extract key behavior features, and performing semantic analysis, keyword extraction and vectorization on the campus service resource data to generate corresponding resource feature vectors, and fusing the key behavior features with the resource feature vectors to generate various service knowledge vectors containing service information and resource semantics; The step of obtaining the attention coefficient of each service knowledge vector through the attention network includes: The global attention mechanism of the attention network is used to capture the global dependency between each service knowledge vector. Specifically, a global similarity matrix is ​​obtained by calculating the similarity between each service knowledge vector and all other service knowledge vectors. The global similarity matrix reflects the importance of each service knowledge vector in the global context. Then, the service knowledge vectors are weighted and summed using the global similarity matrix to obtain an enhanced service knowledge vector that integrates global context information. Segment each enhanced service knowledge vector to generate multiple sub-vectors, and apply a local attention function to each sub-vector based on the local attention mechanism of the attention network, wherein the local attention function identifies a local focus by calculating the correlation between internal elements of the sub-vectors, and merges the local focus of each sub-vector to generate a local focus vector for the entire service knowledge vector; A multi-level attention fusion strategy of the attention network is adopted to define multiple levels of attention heads, each of which is used to capture information at a corresponding level; Applying the attention head to each local focus vector simultaneously to obtain multiple levels of attention representation, and fusing the multiple levels of attention representation by weighted summation to generate a final service knowledge vector that fuses multiple levels of attention information, wherein the levels include semantic level, behavior level and resource level; Applying a predefined dynamic adjustment rule to the attention coefficient of each final service knowledge vector to obtain a dynamically adjusted attention coefficient, wherein the dynamic adjustment rule includes an adjustment rule based on time decay, an adjustment rule based on user preference, and an adjustment rule based on service popularity; Calculate the sum or maximum value of all dynamically adjusted attention coefficients, and divide each attention coefficient by the sum or maximum value to obtain a normalized attention coefficient; The step of performing service acceptance portrait estimation based on a plurality of service knowledge vectors included in the service push training sample and an attention coefficient of each of the service knowledge vectors through an estimation network to generate a service acceptance estimation portrait of each of the plurality of service push nodes corresponding to the service push training sample comprises: For each service knowledge vector, weighted calculation is performed on the service knowledge vector and the attention coefficient corresponding to the service knowledge vector to generate a target weighted knowledge vector corresponding to each service knowledge vector; The first network function unit estimates a service acceptance profile according to a plurality of target weighted knowledge vectors corresponding to the service push training sample, and generates a first service acceptance estimation profile corresponding to each service push node of the service push training sample; The second network function unit estimates the service acceptance profile according to the multiple target weighted knowledge vectors corresponding to the service push training sample, and generates a second service acceptance estimation profile corresponding to each service push node of the service push training sample; The step of calculating the network training error based on the service acceptance description portrait corresponding to the target service push node of the service push training sample and the service acceptance estimation portrait corresponding to each service push node of the service push training sample comprises: Calculate the network training error according to the service acceptance description portrait of the service push node corresponding to the service push training sample, the first service acceptance estimation portrait of each service push node corresponding to the service push training sample, and the second service acceptance estimation portrait of each service push node corresponding to the service push training sample; The step of optimizing the neuron weight information of the service acceptance portrait estimation network for estimating the service push training sample corresponding to the target service push node based on the network training error includes: Based on the network training error, the neuron weight information of the attention network is optimized, the neuron weight information used to estimate the service push training sample corresponding to the target service push node in the first network function unit is optimized, and the neuron weight information of the second network function unit is optimized.

4. The method for pushing smart campus services based on artificial intelligence according to claim 3 is characterized in that: The first network function unit is a multi-dimensional fusion output unit, and the second network function unit is a sequence response unit or a feature interaction relationship unit.

5. The method for pushing smart campus services based on artificial intelligence according to claim 3 is characterized in that: The estimation network further includes a third network function unit, and the step of performing service acceptance portrait estimation based on the multiple service knowledge vectors included in the service push training sample and the attention coefficient of each of the service knowledge vectors through the estimation network to generate a service acceptance estimation portrait of each of the multiple service push nodes corresponding to the service push training sample also includes: According to the third network function unit, a service acceptance profile is estimated according to a plurality of target weighted knowledge vectors corresponding to the service push training sample, and a third service acceptance estimation profile of the service push training sample corresponding to each service push node is generated, wherein the second network function unit is a sequence response unit, and the third network function unit is a feature interaction relationship unit; The calculating of the network training error according to the service acceptance description portrait of the service push node corresponding to the service push training sample, the first service acceptance estimation portrait of each service push node corresponding to the service push training sample, and the second service acceptance estimation portrait of each service push node corresponding to the service push training sample includes: The network training error is calculated based on the service acceptance description portrait of the service push node corresponding to the service push training sample, the first service acceptance estimation portrait of each service push node corresponding to the service push training sample, the second service acceptance estimation portrait of each service push node corresponding to the service push training sample, and the third service acceptance estimation portrait of each service push node corresponding to the service push training sample.

6. The method for pushing smart campus services based on artificial intelligence according to any one of claims 1 to 5, characterized in that: The step of optimizing the neuron weight information of the service acceptance portrait estimation network for estimating the service push training sample corresponding to the target service push node based on the network training error includes: If the campus service resource label in the service push training sample is the target campus service resource label, optimizing the neuron weight information in the service acceptance portrait estimation network for estimating the service push node of the service push training sample as the target service push node based on the network training error; If the campus service resource label in the service push training example is not the target campus service resource label, at least one neuron weight information of the service acceptance portrait estimation network is optimized based on the network training error.

7. The method for pushing smart campus services based on artificial intelligence according to any one of claims 1 to 5, characterized in that: The step of making a smart campus service push decision for the target push object based on the service acceptance profile estimation network includes: Acquire a service usage trajectory vector of a target push object and a campus service resource semantic vector of multiple campus service resource data to be pushed; Generate a service acceptance estimation portrait of the campus service resource data to be pushed corresponding to each service push node through the service acceptance portrait estimation network according to the service usage trajectory vector and the campus service resource semantic vector of each campus service resource data to be pushed; Determine the target campus service resource data and the service push node corresponding to the target campus service resource data from the plurality of campus service resource data to be pushed according to the service acceptance estimation portrait corresponding to each service push node of each campus service resource data to be pushed; Based on the target campus service resource data and the service push node corresponding to the target campus service resource data, smart campus service is pushed to the target push object.

8. The method for pushing smart campus services based on artificial intelligence according to claim 7 is characterized in that: The step of obtaining the service usage trajectory vector of the target push object and the campus service resource semantic vectors of the plurality of campus service resource data to be pushed comprises: Acquire a service push task, in which the object index information of the target push object and the service requirement knowledge points are marked; Acquire a service usage trajectory vector of the target push object based on the service push task; Based on the service demand knowledge point, a plurality of campus service resource data to be pushed are acquired, and a campus service resource semantic vector of each campus service resource data to be pushed is extracted.

9. The method for pushing smart campus services based on artificial intelligence according to claim 7 is characterized in that: The step of determining the target campus service resource data and the service push node corresponding to the target campus service resource data from the plurality of campus service resource data to be pushed according to the service acceptance estimation portrait corresponding to each service push node of each campus service resource data to be pushed includes: Based on the service acceptance estimation portraits corresponding to each service pushing node of each campus service resource data to be pushed, the campus service resource data to be pushed corresponding to the service acceptance estimation portraits whose probability values ​​meet the set conditions are extracted as the target campus service resource data, and the service pushing nodes corresponding to the service acceptance estimation portraits whose probability values ​​meet the set conditions in the target campus service resource data are used as the service pushing nodes corresponding to the target campus service resource data.

10. A smart campus service push system based on artificial intelligence, characterized in that: The artificial intelligence-based smart campus service push system includes a processor and a memory, the memory is connected to the processor, the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to implement the artificial intelligence-based smart campus service push method described in any one of claims 1 to 9 above.

Images