Campus personalized service recommendation system based on AI agent

Abstract

The application discloses a kind of campus individualized service recommendation system based on AI intelligent agent, it is related to wisdom campus and artificial intelligence technical field, including: S1: detecting key events from campus multi-source data, generating active interaction instruction, S2: receiving instruction or user input, S3: intent recognition and slot extraction, judge whether it meets the recommendation condition, S4: when not satisfied, access campus service knowledge graph, based on the difference between the distribution entropy and conditional entropy of intent node associated service entity calculates the information gain of each to be clarified slot, S5: select the slot of maximum information gain to generate guiding question, update portrait slot after receiving reply and return S3, S6: when satisfied, generate individualized service recommendation result, S7: track execution state, when failed, retrieve alternative resources and re-enter S2.The application realizes the efficient individualized recommendation of campus service by active event triggering, information gain driven clarification and closed-loop alternative mechanism.

Description

Technical Field

[0001] This invention relates to the fields of smart campus and artificial intelligence technology, specifically to a personalized campus service recommendation system based on AI intelligent agents. Background Technology

[0002] With the advancement of smart campus construction, a massive amount of data on learning resources, daily services, and club activities has accumulated on campus. To help students efficiently obtain the services they need from this vast amount of information, personalized recommendation technology is being gradually introduced into campus scenarios.

[0003] Existing campus recommendation systems mostly employ methods such as collaborative filtering, content-based recommendation, or knowledge graph reasoning to achieve information matching and delivery within a single domain. For example, a learning resource recommendation system might recommend course videos and exercises to students, while a life service recommendation system might recommend cafeteria dishes and library resources. These systems typically operate in a passive response mode, meaning they receive explicit user queries and then output a recommendation list.

[0004] However, the above solution has the following shortcomings:

[0005] First, the system only works when the user actively initiates a request. It cannot proactively identify abnormalities or changes in the user's needs in academics, life, etc. When the user does not initiate a request due to a lack of clear goals, the system is in a silent state and cannot play a proactive service role.

[0006] Second, some systems attempt to refine user needs by asking follow-up questions when user intent is ambiguous, but the generation of clarifying questions often relies on fixed templates or predetermined order, failing to dynamically adjust the questioning strategy according to the current information status, resulting in more interaction rounds and a limited user experience.

[0007] Third, when recommended service resources fail or user operations fail, the system cannot detect this and provide alternative solutions, and the service completion rate depends entirely on the user to resolve the issue themselves.

[0008] The aforementioned problems result in significant deficiencies in the existing campus recommendation system in terms of initiative, interaction efficiency, and result assurance, making it difficult for it to truly assume the role of personalized services in the form of an intelligent entity. Summary of the Invention

[0009] The present invention aims to at least partially solve one of the technical problems in the above-mentioned technologies.

[0010] To achieve the above objectives, the first aspect of this invention proposes a campus personalized service recommendation method based on AI intelligent agents, comprising the following steps: S1: Obtain at least one of academic performance data, campus card consumption data, library borrowing data, and academic calendar information from a multi-source heterogeneous data stream on campus; detect key events based on preset event rules or anomaly detection models; and generate an active interaction instruction for a specific user when a key event that meets preset triggering conditions is detected; S2: Receive the active interaction instruction or the initial natural language expression input by the user, and output corresponding prompts or recommendation results; S3: Perform intent recognition and slot extraction on the initial natural language expression or subsequent dialogue, update the user profile slots, and determine whether the current slot status meets preset recommendation conditions; S4: When the recommendation conditions are not met, access the campus service knowledge graph, locate the intent node in the knowledge graph, and obtain a set of slots to be clarified associated with the current intent. The knowledge graph includes entities such as courses, activities, life services, skills, time, and location, as well as their relationships. For each slot to be clarified, the information gain is calculated based on the difference between the distribution entropy of the service entity associated with the intent node and the conditional entropy given the slot value. S5: Select the slot with the largest information gain, generate and output the corresponding guiding clarification question, receive the user's supplementary response to the guiding clarification question, extract the slot value and update the user profile slot, and return to S3. S6: When the slot status meets the recommendation conditions, query and match based on the current user profile slot and the campus service knowledge graph, generate personalized service recommendation results and output them. S7: Track the execution status of the personalized service recommendation results. If execution failure or unavailable recommendation resources are detected, retrieve alternative resources in the campus service knowledge graph, generate new active interaction instructions based on the alternative resources, and re-enter S2 to initiate an alternative recommendation dialogue.

[0011] In addition, the campus personalized service recommendation method based on AI intelligent agents proposed above according to the present invention may also have the following additional technical features:

[0012] Furthermore, in S1, the anomaly detection model adopts an anomaly scoring algorithm based on isolated forest or Gaussian mixture model to calculate anomaly scores on the user's multidimensional behavior vector. When the anomaly score exceeds a dynamic threshold, it is determined that the preset triggering conditions are met. The key events include at least one of the following: make-up exam arrangements released by the academic affairs system, warnings of failing grades, the start of course selection, the amount or frequency of campus card consumption being lower than the historical average, and long-term unborrowed items from the library.

[0013] Furthermore, in S4, the calculation of information gain specifically involves: for the slot to be clarified... Its information gain ,in This represents the distribution entropy of the service entities associated with the intent node in the knowledge graph. Let v be the probability that slot S takes the value v. For the conditional entropy of the intention given a slot value v, select The issue of guiding and clarifying the generation of slots corresponding to the maximum value.

[0014] Furthermore, after S1 generates the active interaction instruction, in S4, the slots that are directly related to the key events that trigger the instruction in the knowledge graph are identified as the initial slots to be clarified, and the information gain calculation for the slots is skipped to directly generate guiding clarification questions.

[0015] Furthermore, in S7, retrieving alternative resources from the campus service knowledge graph includes: using the substitution relationship edges or attribute similarity of entities in the knowledge graph to find available resources that have the same speaker, the same topic, or similar attributes as the currently unavailable resources, and generating alternative recommendations by sorting them by relevance.

[0016] Furthermore, in S3, the confidence level of the current intent is calculated. When the confidence level of the intent is lower than the preset threshold, even if the slot value has been partially filled, it is still determined that the recommendation conditions are not met, and proceed to S4.

[0017] Furthermore, the personalized service recommendation result generated by S6 includes at least two cross-domain service combinations from learning resources, life services, and community activities. When generating the cross-domain service combination, matching learning resource nodes and life service nodes are retrieved from the knowledge graph, and time conflicts or location conflicts are detected between the two types of nodes. If there is no conflict, the combination is output. If there is a conflict, one of them is retained according to a preset priority and a conflict prompt is output to the user. At the same time, for the services not retained, non-conflicting alternative resources are retrieved from the knowledge graph and output.

[0018] To achieve the above objectives, a second aspect of the present invention proposes a campus personalized service recommendation system based on an AI intelligent agent to implement the first method described above, comprising: a campus event monitoring module for executing S1; a multimodal interaction module for executing S2; an intent recognition and slot status judgment module for executing S3; a campus service knowledge graph module for constructing and storing the knowledge graph; a knowledge graph-driven clarification module for executing S4 and S5; a recommendation generation module for executing S6; and a result tracking and closed-loop processing module for executing S7.

[0019] Furthermore, the knowledge graph-driven clarification module is specifically used to calculate the information gain according to the following formula: , and select The issue of guiding and clarifying the generation of slots corresponding to the maximum value.

[0020] To achieve the above objectives, a third aspect of the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements a campus personalized service recommendation method based on an AI intelligent agent, as described in the first aspect.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1. By proactively detecting key events and generating proactive interaction commands through the campus event monitoring module, the system is able to proactively perceive users' potential needs, and the interaction mode is transformed from "people looking for services" to "services looking for people";

[0023] 2. By using a knowledge graph-driven clarification module to dynamically select the optimal clarification question based on information gain, the system can pinpoint the user's true needs with the fewest interaction rounds, significantly improving clarification efficiency.

[0024] 3. The execution status of the recommendation results is tracked through the result tracking and closed-loop processing module, and alternative resources are retrieved and alternative recommendations are initiated proactively when failure occurs, forming a complete service closed loop and improving the service completion rate;

[0025] 4. By combining services across different domains and detecting conflicts, we can achieve unified and coordinated recommendations for services in multiple domains such as learning resources, life services, and club activities, thus avoiding time or location conflicts. Attached Figure Description

[0026] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0027] Figure 1 This is a block diagram of the overall architecture of a campus personalized service recommendation system based on an AI intelligent agent according to an embodiment of the present invention;

[0028] Figure 2 This is a core interactive control flowchart of a campus personalized service recommendation system based on an AI intelligent agent according to an embodiment of the present invention;

[0029] Figure 3 This is a partial structural diagram of a campus service knowledge graph of a campus personalized service recommendation system based on an AI agent according to an embodiment of the present invention.

[0030] Figure 4 This is a flowchart of a campus personalized service recommendation system based on AI intelligent agents according to an embodiment of the present invention, which is a dynamic generation flowchart of clarification questions based on information gain;

[0031] Figure 5This is a schematic diagram illustrating the dynamic update of user profile slot status in a campus personalized service recommendation system based on AI intelligent agents according to an embodiment of the present invention.

[0032] As shown in the figure:

[0033] 1. Campus event monitoring module; 2. Multimodal interaction module; 3. Intent recognition and slot status judgment module; 4. Campus service knowledge graph module; 5. Knowledge graph-driven clarification module; 6. Recommendation generation module; 7. Result tracking and closed-loop processing module. Detailed Implementation

[0034] Embodiments of the present invention are described in detail below, with examples of the embodiments illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0035] like Figure 1 As shown in the figure, an embodiment of the present invention provides a campus personalized service recommendation system based on AI intelligent agents, including a campus event monitoring module 1, a multimodal interaction module 2, an intent recognition and slot status judgment module 3, a campus service knowledge graph module 4, a knowledge graph-driven clarification module 5, a recommendation generation module 6, and a result tracking and closed-loop processing module 7.

[0036] The campus incident monitoring module 1 is connected to the campus academic affairs system, campus card system, library management system and academic calendar database to obtain academic performance data, campus card consumption data, library borrowing data and academic calendar information in real time or at regular intervals.

[0037] It should be noted that the campus event monitoring module 1 has built-in preset event rules and anomaly detection model. The preset event rules are used to match events such as exam weeks and course selection starts in the school calendar. The anomaly detection model uses the isolated forest algorithm to score anomalies in students' multi-dimensional behavior vectors.

[0038] Specifically, when a critical event or abnormal score is detected to exceed a dynamic threshold, an active interaction instruction for the student is generated and sent to the multimodal interaction module 2.

[0039] Multimodal interaction module 2 serves as the interaction interface between the system and students. It can be integrated into the campus APP or mini-program and supports text and voice input and output.

[0040] It should be noted that the multimodal interaction module 2 receives the active interaction instructions sent by the campus event monitoring module 1, or the initial natural language expression actively entered by the student, and presents the prompts, clarification questions or recommendation results generated by the system in the form of text or rich text cards.

[0041] Campus service knowledge graph module 4 pre-constructs a campus service knowledge graph, such as Figure 3 As shown, the graph contains entity nodes such as courses, activities, life services, skills, time, and location.

[0042] Specifically, entities are connected through various relational edges, such as the "cultivation" relationship between "course" and "skill", the "event time" relationship between "activity" and "time", the "substitution" relationship between "activity" and "activity", and the "prerequisite" relationship between "course" and "course".

[0043] It should be noted that the Campus Service Knowledge Graph Module 4 supports real-time querying and updating. For example, changes in the remaining number of spots for an activity can be synchronized to the corresponding entity attributes in the graph in real time.

[0044] The intent recognition and slot status judgment module 3 performs natural language understanding on user input, recognizes user intent and extracts slot values, and updates the structured user profile slots.

[0045] It should be noted that user profile slots include, but are not limited to: skill direction, current level, available time, location preference, areas of interest, and learning goals.

[0046] Furthermore, the intent recognition and slot status judgment module 3 also calculates the confidence level of the current intent and combines the slot filling degree and confidence level to determine whether the recommendation conditions are met.

[0047] Knowledge graph-driven clarification module 5 is activated when the recommendation conditions are not met. Its workflow is as follows: Figure 4 As shown, firstly, the intent node is located in the knowledge graph, and the set of slots to be clarified associated with the intent is obtained. For each slot to be clarified, its information gain is calculated. The slot with the largest information gain is selected to generate a guiding clarification question and output it. After receiving the user's supplementary reply, the user profile slot is updated, and the intent recognition and slot status judgment module 3 is triggered to judge again.

[0048] The recommendation generation module 6 is activated when the slot status meets the recommendation conditions. Based on the current user profile slot, it queries and matches in the knowledge graph. When the recommendation result is a cross-domain service combination, it also performs time conflict and location conflict detection on each service node in the combination.

[0049] The result tracking and closed-loop processing module 7 obtains the execution status of the recommendation results by calling relevant campus service APIs. If it detects that the user has not completed the operation or the service resource has expired, it retrieves alternative resources in the knowledge graph and generates new proactive interaction instructions.

[0050] like Figure 2 As shown in the figure, an embodiment of the present invention provides a campus personalized service recommendation method based on an AI intelligent agent, which includes the following steps:

[0051] S1: Obtain at least one of the following from the multi-source heterogeneous data stream of the campus: academic performance data, campus card consumption data, library borrowing data, and academic calendar information; detect key events based on preset event rules or anomaly detection models; and generate proactive interaction instructions for specific users when a key event that meets preset trigger conditions is detected.

[0052] It should be noted that the anomaly detection model uses an anomaly scoring algorithm based on isolated forest or Gaussian mixture model to calculate anomaly scores on the user's multidimensional behavior vector. When the anomaly score exceeds the dynamic threshold, it is determined that the preset triggering conditions are met. Key events include at least one of the following: make-up exam arrangements released by the academic affairs system, warnings of failing grades, the start of course selection, the amount or frequency of campus card consumption being lower than the historical average, and library items not being borrowed for a long time.

[0053] S2: Receives active interaction instructions or initial natural language expressions input by the user, and outputs corresponding prompts or recommendations.

[0054] S3: Perform intent recognition and slot extraction on the initial natural language expression or subsequent dialogue, update the user profile slots, and determine whether the current slot status meets the preset recommendation conditions.

[0055] Furthermore, when determining whether the current slot status meets the preset recommendation conditions, the confidence level of the current intention is also calculated. When the confidence level of the intention is lower than the preset threshold, even if the slot value has been partially filled, it is still determined that the recommendation conditions are not met, and proceed to S4.

[0056] Specifically, the confidence level of intent is calculated as follows: based on the probability distribution output by the intent recognition and slot status judgment module 3, the maximum probability value is taken as the confidence level of the current intent; or, the confidence level is calculated based on the ratio of the number of filled slots to the minimum number of slots required for recommendation; or, the confidence level is calculated by weighting the output probability of the intent recognition model with the slot filling ratio.

[0057] S4: When the recommendation conditions are not met, access the campus service knowledge graph, locate the intent node in the knowledge graph, and obtain the set of slots to be clarified associated with the current intent. The knowledge graph contains entities such as courses, activities, life services, skills, time, and location, as well as their relationships. For each slot to be clarified, calculate the information gain based on the difference between the distribution entropy of the service entity associated with the intent node and the conditional entropy after giving the slot value.

[0058] It should be noted that the calculation of information gain specifically involves: for the slot to be clarified... Its information gain ,in This represents the distribution entropy of the service entities associated with the intent node in the knowledge graph. Let v be the probability that slot S takes the value v. For the conditional entropy of the intention given a slot value v, select The issue of guiding and clarifying the generation of slots corresponding to the maximum value.

[0059] It should be noted that after S1 generates the active interaction instruction, in S4, the slots in the knowledge graph that have a direct correlation with the key event that triggered the instruction are identified as the initial slots to be clarified, and the information gain calculation for the slots is skipped to directly generate the guiding clarification question.

[0060] S5: Select the slot with the highest information gain, generate and output the corresponding guidance clarification question, receive the user's supplementary response to the guidance clarification question, extract the slot value and update the user profile slot, and return to S3.

[0061] S6: When the slot status meets the recommendation conditions, query and match based on the current user profile slot and campus service knowledge graph to generate personalized service recommendation results and output them.

[0062] It should be noted that the generated personalized service recommendation results include at least two cross-domain service combinations from learning resources, life services, and community activities. When generating cross-domain service combinations, matching learning resource nodes and life service nodes are retrieved from the knowledge graph respectively. The system checks whether there are time conflicts or location conflicts between the two types of nodes. If there are no conflicts, the combination is output. If there are conflicts, one of them is retained according to the preset priority and a conflict prompt is output to the user. At the same time, for the services that are not retained, conflict-free alternative resources are retrieved from the knowledge graph and output.

[0063] S7: Track the execution status of personalized service recommendation results. If execution failure or unavailable recommendation resources are detected, retrieve alternative resources from the campus service knowledge graph, generate new proactive interaction instructions based on the alternative resources, and re-enter S2 to initiate an alternative recommendation dialogue.

[0064] Specifically, retrieving alternative resources in the campus service knowledge graph includes: using the substitution relationship edges or attribute similarity of entities in the knowledge graph to find available resources that have the same speaker, the same topic, or similar attributes as the currently unavailable resources, and generating alternative recommendations by sorting them by relevance.

[0065] Example 1: Active Triggering and Closed-Loop Replacement Scenarios

[0066] This embodiment illustrates the complete interaction process of the system in a failing grade warning scenario.

[0067] A student failed the final exam for "Advanced Mathematics". After the academic affairs system released the make-up exam arrangements, the campus event monitoring module 1 detected this critical event and generated an active interaction command.

[0068] The multimodal interaction module 2 proactively prompts the student: "Student, we have detected that you need to retake the 'Advanced Mathematics' exam on March 15th. Would you like me to recommend some review materials and study partners for you?"

[0069] The student replied "Okay". The intent recognition and slot status judgment module 3 recognized the intent as "make-up exam review". The slot included the course name "Advanced Mathematics", but the learning method slot was missing. The confidence level was medium, and it was determined that the recommendation conditions were not met.

[0070] The knowledge graph-driven clarification module 5 locates the "make-up exam review" intent node in the knowledge graph and obtains the associated clarification slots, including "learning method" (values: video courses, exercise sets, study groups) and "available time" (values: weekday evenings, weekends). Calculations show that the information gain IG for "learning method" is 0.68, and the information gain IG for "available time" is 0.31. Selecting "learning method" generates a guiding question: "Would you prefer to review through video courses, do exercise sets, or join a study group?"

[0071] The student replies "Study Group," and the slot "Study Method" is filled with "Study Group." However, "Available Time" is still missing. Returning to S3 for further evaluation, the condition is still not met. The information gain of the remaining slots is recalculated, and the question "Available Time" is generated: "When is it convenient for you to participate? Weekday evenings or weekends?"

[0072] The student replied "weekend," and the slot status met the recommendation criteria. The recommendation generation module 6 searched for a match in the knowledge graph and recommended "Advanced Mathematics Make-up Exam Intensive Study Group (Saturday morning 9:00~11:00, Library Seminar Room 3)" and "Advanced Mathematics Exercises Explained Video (Online Resource)."

[0073] The student attempted to register for the study group, but all spots were filled. As a result, the tracking and closed-loop processing module 7 detected the registration failure. In the knowledge graph, it searched for "Advanced Mathematics Make-up Exam Q&A Workshop (Sunday afternoon 14:00~16:00, Teaching Building 301)" through "substitution" relationship edges and "same topic" attribute similarity. A new active interaction instruction was generated, and the multimodal interaction module 2 sent an alternative recommendation to the student: "The study group is full. I found a make-up exam Q&A workshop with the same topic for you, Sunday afternoon 14:00~16:00, in Teaching Building 301. Do you want to register?" After the student confirmed, the registration was completed.

[0074] Example 2: Information Gain Calculation Case

[0075] This embodiment uses specific data to illustrate the calculation process of information gain in S4.

[0076] When a user inputs "I want to improve myself", the intent recognition and slot status judgment module 3 locates the "ability improvement" intent node in the knowledge graph. This intent node is associated with three service entities: Python courses, data analysis workshops, and lecture clubs. Based on historical recommendation data and student group preferences, the probability distribution of the three entities is (0.4, 0.35, 0.25). Therefore, the intent entropy H(Intent) = -0.4×log20.4-0.35×log20.35-0.25×log20.25≈1.56.

[0077] The slot S1, whose "skill direction" needs clarification, has three possible values: "programming," "data analysis," and "public speaking." When the value is "programming," it is associated with a Python course, with a determined conditional probability distribution: H(Intent|S1=programming)=0. When the value is "data analysis," it is associated with a data analysis workshop, H(Intent|S1=data analysis)=0. When the value is "public speaking," it is associated with a public speaking community, H(Intent|S1=public speaking)=0. The probabilities for each value are estimated to be 0.4, 0.35, and 0.25 respectively, based on the weights of the associated edges in the knowledge graph and group statistics. Therefore, the conditional entropy H(Intent|S1)=0.4×0+0.35×0+0.25×0=0, and the information gain IG(S1)=1.56-0=1.56.

[0078] The slot S2 "Time Investment" to be clarified has two values: "Short-term" and "Long-term". When the value is "Short-term", the conditional probability distribution of the associated Python course and data analysis workshop is (0.53, 0.47), and H(Intent|S2=Short-term)≈1.0. When the value is "Long-term", the conditional probability distribution of the associated three entities is (0.2, 0.2, 0.6), and H(Intent|S2=Long-term)≈1.37. P(Short-term)=0.65, P(Long-term)=0.35. Therefore, the conditional entropy H(Intent|S2)=0.65×1.0+0.35×1.37≈1.13, and the information gain IG(S2)=1.56-1.13=0.43.

[0079] Since IG(S1) > IG(S2), the system prioritizes the "Skill Direction" slot to generate the question: "Do you want to improve your programming skills, data analysis skills, or presentation skills?" When the student replies "programming", the "Skill Direction" slot is filled with "programming", which greatly reduces the uncertainty of intent. At this point, the information gain of the remaining slots is calculated, and the optimal slot is selected to ask questions until the recommendation conditions are met.

[0080] like Figure 5As shown, the dynamic filling process of user profile slot values ​​during multiple rounds of clarification is as follows: as the conversation progresses, the "skill direction" and "time investment" slots are filled in sequence, and the confidence of intent gradually increases. When all key slots are filled and the confidence exceeds the threshold, the system determines that the recommendation conditions are met and enters the recommendation generation step.

[0081] It should be noted that S1 and S2 above refer to the slots to be clarified and do not involve any methodological steps.

[0082] Example 3: Cross-domain combination and conflict detection

[0083] This embodiment illustrates the implementation method of cross-domain service composition and conflict detection in S6.

[0084] After the student profile slots are filled, the recommendation generation module 6 searches for matching learning resource nodes and life service nodes in the knowledge graph. For example, it retrieves "Python Data Analysis Workshop (Saturday 9:00~11:00, Information Building 201)" and "Library Seminar Reservation (Saturday 9:00~11:00, Library 3rd Floor)". Both nodes have a time slot of Saturday 9:00~11:00, which is a time conflict.

[0085] After the recommendation generation module 6 detects the conflict, it retains the "Python Data Analysis Workshop" according to the preset priority rules (such as learning services taking precedence over life services), outputs a conflict prompt to the user, and searches the knowledge graph for a conflict-free alternative resource for "Library Seminar". It finds "Library Seminar Reservation (Saturday 15:00~17:00, 3rd Floor of the Library)", which does not conflict with the already retained workshop. The two resources are combined and output. The final recommendation result is: "We recommend the Python Data Analysis Workshop (Saturday 9:00~11:00) for you, and have reserved the Library Seminar for you from 15:00 to 17:00 on Saturday for after-class practice."

[0086] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art can make various improvements and modifications to the present invention without departing from its principles, and these improvements and modifications should also be considered within the scope of protection of the present invention.

[0087] In the description of this specification, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this application, "multiple" means at least two, such as two, three, etc., unless otherwise explicitly specified.

[0088] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0089] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A method for recommending personalized campus services based on AI intelligent agents, characterized in that, Includes the following steps: S1: Obtain at least one of the following from the campus multi-source heterogeneous data stream: academic performance data, campus card consumption data, library borrowing data, and academic calendar information; detect key events based on preset event rules or anomaly detection models; and generate proactive interaction instructions for specific users when a key event that meets preset trigger conditions is detected. S2: Receive the active interaction command or the initial natural language expression input by the user, and output the corresponding prompts or recommendation results; S3: Perform intent recognition and slot extraction on the initial natural language expression or subsequent dialogue, update the user profile slots, and determine whether the current slot status meets the preset recommendation conditions; S4: When the recommendation conditions are not met, access the campus service knowledge graph, locate the intent node in the knowledge graph, and obtain the set of slots to be clarified associated with the current intent. The knowledge graph contains entities such as courses, activities, life services, skills, time, and location, as well as their relationships. For each slot to be clarified, calculate the information gain based on the difference between the distribution entropy of the service entity associated with the intent node and the conditional entropy after giving the slot value. S5: Select the slot with the highest information gain, generate and output the corresponding guidance clarification question, receive the user's supplementary response to the guidance clarification question, extract the slot value and update the user profile slot, and return to S3; S6: When the slot status meets the recommendation conditions, query and match based on the current user profile slot and the campus service knowledge graph to generate and output personalized service recommendation results; S7: Track the execution status of the personalized service recommendation results. If execution failure or unavailable recommendation resources are detected, retrieve alternative resources from the campus service knowledge graph and generate new active interaction instructions based on the alternative resources. Re-enter S2 to initiate an alternative recommendation dialogue.

2. The campus personalized service recommendation method based on AI intelligent agents according to claim 1, characterized in that, In S1, the anomaly detection model uses an anomaly scoring algorithm based on isolated forest or Gaussian mixture model to calculate anomaly scores on the user's multidimensional behavior vectors. When the anomaly score exceeds the dynamic threshold, it is determined that the preset triggering condition is met. The key events include at least one of the following: make-up exam arrangements released by the academic affairs system, warnings of failing grades, the start of course selection, the amount or frequency of campus card transactions being lower than the historical average, and library items not being borrowed for a long period of time.

3. The campus personalized service recommendation method based on AI intelligent agents according to claim 1, characterized in that, In S4, the calculation of information gain specifically refers to: for the slot to be clarified... Its information gain ,in This represents the distribution entropy of the service entities associated with the intent node in the knowledge graph. Let v be the probability that slot S takes the value v. For the conditional entropy of the intention given a slot value v, select The issue of clarifying the generation guidelines for the slot corresponding to the maximum value.

4. The campus personalized service recommendation method based on AI intelligent agents according to claim 1, characterized in that, After S1 generates the active interaction instruction, in S4, the slots that are directly related to the key events that trigger the instruction in the knowledge graph are identified as the initial slots to be clarified, and the information gain calculation for the slots is skipped to directly generate guiding clarification questions.

5. The campus personalized service recommendation method based on AI intelligent agents according to claim 1, characterized in that, In S7, retrieving alternative resources from the campus service knowledge graph includes: using the substitution relationship edges or attribute similarity of entities in the knowledge graph to find available resources that have the same speaker, the same topic, or similar attributes as the currently unavailable resources, and generating alternative recommendations by sorting them by relevance.

6. The campus personalized service recommendation method based on AI intelligent agents according to claim 1, characterized in that, In S3, the confidence level of the current intent is also calculated. When the confidence level of the intent is lower than the preset threshold, even if the slot value has been partially filled, it is still determined that the recommendation conditions are not met, and proceed to S4.

7. The campus personalized service recommendation method based on AI intelligent agents according to claim 1, characterized in that, The personalized service recommendation result generated by S6 includes at least two cross-domain service combinations from learning resources, life services, and community activities. When generating the cross-domain service combination, matching learning resource nodes and life service nodes are retrieved in the knowledge graph, and time conflicts or location conflicts are detected between the two types of nodes. If there is no conflict, the combination is output. If there is a conflict, one of them is retained according to a preset priority and a conflict prompt is output to the user. At the same time, for the services not retained, non-conflicting alternative resources are retrieved in the knowledge graph and output.

8. A system for implementing the AI-based personalized campus service recommendation method according to claims 1-7, characterized in that, include: The campus incident monitoring module (1) is used to execute S1; The multimodal interaction module (2) is used to execute S2; The intent recognition and slot status judgment module (3) is used to execute S3; The campus service knowledge graph module (4) is used to construct and store the knowledge graph; Knowledge graph-driven clarification module (5) is used to execute S4 and S5; It is recommended to generate module (6) for executing S6; The result tracking and closed-loop processing module (7) is used to execute S7.

9. A campus personalized service recommendation system based on AI intelligent agents according to claim 8, characterized in that, The knowledge graph-driven clarification module (5) is specifically used to calculate the information gain according to the following formula: , and select The issue of clarifying the generation guidelines for the slot corresponding to the maximum value.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the campus personalized service recommendation method based on any one of claims 1 to 7.

Images