Nursing teaching scheme recommendation algorithm based on AI

Abstract

The invention relates to the technical field of nursing teaching scheme recommendation, and discloses an AI-based nursing teaching scheme recommendation algorithm, which comprises the steps of 1, performing information extraction on a current hospital to obtain hospital characteristics; dynamically evaluating the importance weight of each hospital feature according to the scene of the current hospital; secondly, Prompt is dynamically generated according to the scene of a current hospital, a large language model is called to analyze the Prompt, and a matched scheme is obtained; and step 3, performing multi-dimensional analysis on the matched scheme according to the importance weight to obtain a matching degree score, adjusting the matching degree score based on a business rule to obtain a final score, sorting according to a descending order of the final score, and performing diversity adjustment on a sorting result to obtain a recommendation result. Through the hospital feature extraction process, the dynamic Prompt generation process, the multi-dimensional analysis process and the secondary adjustment process, the most accurate context information can be obtained for reasoning.

Description

Technical Field

[0001] This application relates to the technical field of nursing teaching program recommendation, and in particular to AI-based nursing teaching program recommendation algorithms. Background Technology

[0002] Against the backdrop of rapid development in healthcare information technology, nursing teaching program databases have become an important tool for hospital nursing training, quality management, and the implementation of special programs. With the rapid increase in the number of programs, a large number of teaching programs applicable to different scenarios have been accumulated. How to enable users to quickly find a program suitable for their hospital has become a prominent issue.

[0003] The selection and recommendation of existing nursing teaching programs mainly involve the following methods: First, manual retrieval and screening. Users search the program library by keyword or browse by category, manually reviewing the details of each program and judging its suitability for their hospital based on experience. Second, simple tag filtering. Programs are pre-tagged (e.g., province, hospital level, hospital type), and users filter by selecting tags. The system then returns a list of programs that meet the criteria. However, manual retrieval and screening is inefficient. When the program library contains hundreds or even thousands of programs, users need to spend a significant amount of time browsing and filtering, with an average search time exceeding 20 minutes. The first problem is the lack of objective data. Users struggle to determine which solution is most suitable, easily overlooking high-quality options. Faced with a massive number of solutions, users often give up on the solution library due to the difficulty of selection. Furthermore, relying on personal experience makes it difficult for new users or managers transferred from other regions to quickly find suitable solutions. The second problem is the simple tag filtering method, which is based on precise tag matching and cannot handle "partial matching" scenarios. The filtered results are usually sorted by upload time or in a fixed order, failing to reflect the relevance to the current hospital. Moreover, all hospitals see the same sorted results, without considering the differences in hospital characteristics. This requires manual tagging of each solution, and batch updates are needed when the tag system changes.

[0004] With the application of AI technology, new options have been provided for the selection and recommendation of nursing teaching programs. However, the recommendation process of nursing teaching programs involves multi-dimensional information and is not a simple keyword matching. Therefore, the fundamental problem to be solved by this invention is how to use the semantic understanding ability of Large Language Model (LLM) to deeply understand the semantic relationship between hospital needs and program content and accurately calculate the matching degree between hospitals and programs. Summary of the Invention

[0005] To leverage the semantic understanding capabilities of Large Language Models (LLMs) to deeply understand the semantic relationships between hospital needs and program content and accurately calculate the matching degree between hospitals and programs, this application provides an AI-based nursing teaching program recommendation algorithm, employing the following technical solution: AI-based nursing teaching program recommendation algorithms include: Step 1: Extract information from the current hospital to obtain hospital characteristics; dynamically evaluate the importance weight of each hospital characteristic based on the current hospital scenario; Step 2: Dynamically generate a Prompt based on the current hospital scenario, call the large language model to analyze the Prompt, and obtain matching solutions; Step 3: Perform multi-dimensional analysis on the matched solutions based on importance weights to obtain a matching score. Adjust the matching score based on business rules to obtain the final score. Sort the final scores in descending order and then perform diversity adjustments on the sorting results to obtain the recommendation results.

[0006] Optionally, the process of dynamically generating the Prompt includes: Establish a basic Prompt template, which includes hospital information, a list of candidate solutions, tasks, and output format; Determine the corresponding recommendation strategy based on the current hospital scenario, add the recommendation strategy to the basic Prompt template, and obtain the Prompt.

[0007] Optionally, the hospital scenario determination process includes: Determine whether the user search process provides constraints: If yes, then it is determined to be a specific demand scenario; If not, compare the hospital's historical usage count of the plan with the preset range [n1, n2]: If the number of times the hospital's solution has been used in the past is less than n1, then it is judged as a new hospital scenario; If the number of times the hospital's solution has been used in the past is greater than n2, then it is judged as an active hospital scenario; If the hospital's historical usage count is ∈ [n1, n2], then it is judged as a typical hospital scenario.

[0008] Optionally, the hospital characteristics include hospital level, hospital type, region, historical preferences, and content relevance; The process of the multidimensional analysis includes: The matching scores for hospital level, hospital type, region, historical preferences, and content relevance are obtained separately. The matching scores of hospital features are weighted and summed according to their importance to obtain the matching score.

[0009] Optionally, the process of adjusting the matching score based on business rules includes: The score adjustment factor is obtained according to the business rules, and the sum of the matching score and the adjustment factor is used as the final score. The scores are then sorted according to the final scores. The business rules include: Bonus points will be awarded if the new plan has been online for less than 30 days; Bonus points will be awarded if the proposal receives an expert recommendation; Deduct points if the proposal is a popular one; Bonus points for proposals from the same region.

[0010] Optionally, the process of adjusting the diversity includes: Starting from the second-ranked result, the similarity between the current-ranked solution and the solutions before it is determined sequentially based on the large language model. If there is a similarity score greater than a preset threshold, the solution corresponding to the current-ranked solution is excluded, and a preset number of candidate solutions are obtained as the recommendation result.

[0011] Optionally, the algorithm further includes: Step 4: Obtain user feedback information in real time, dynamically obtain the attention weight of each hospital feature based on the feedback information, recalculate the matching score based on the attention weight and importance weight, and obtain new recommendation results based on the matching score.

[0012] Optionally, the process of obtaining the attention weight includes: Based on the feedback information, user click scenarios are obtained. The similarity of hospital features between any two scenarios is judged sequentially based on the large language model. All similarities of the same hospital feature are summed to obtain the consistency score of the hospital feature. The consistency score ratio of the hospital feature is used as the attention weight.

[0013] In summary, this application includes at least one of the following beneficial technical effects: This invention, through a process of hospital feature extraction, dynamic prompt generation, multi-dimensional analysis, and secondary adjustment, can dynamically construct the information (Prompt) sent to AI based on the characteristics of different hospitals and different recommendation scenarios. This ensures that AI can obtain the most relevant and accurate contextual information for reasoning, deeply understand the semantic relationship between hospital needs and solution content, rather than simple keyword matching, accurately calculate the matching degree between hospitals and solutions, and accurately parse key information such as matching degree scores and recommendation reasons from the unstructured text returned by AI. Based on multi-dimensional scores, it performs reasonable sorting and generates a final recommendation list, prioritizing high automation, accurate matching, and high recommendation efficiency. Attached Figure Description

[0014] Figure 1 This is a flowchart illustrating the steps of the AI-based nursing teaching program recommendation algorithm of this invention. Detailed Implementation

[0015] The embodiments of this application are described in detail below, and examples of the embodiments are shown in the accompanying drawings.

[0016] In the description of this specification, the references to "certain embodiments," "one embodiment," "some embodiments," "illustrative embodiment," "example," "specific example," or "some examples" refer to specific features, structures, materials, or characteristics described in connection with the described embodiment or example, which are 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.

[0017] This application discloses an AI-based nursing teaching program recommendation algorithm, referring to... Figure 1 The process includes: Step 1: Extracting information about the current hospital and acquiring hospital features. This involves extracting hospital features from basic hospital attributes (region, level, type, size), historical behavioral characteristics (used solutions, business preferences), and contextual information (current business scope, user role). In this embodiment, five hospital features are analyzed: hospital level, hospital type, region, historical preferences, and content relevance. Hospital level includes Grade III Class A and Grade II Class A; hospital type includes general hospitals and specialized hospitals; and historical preferences are the percentage of each type of solution used, for example: quality control solutions: 60%, training solutions: 30%, special programs: 10%. Then, the importance weight of each hospital feature is dynamically evaluated based on the current hospital scenario. Hospital scenarios include new hospital scenarios, active hospital scenarios, general hospital scenarios, and specific demand scenarios. The importance weight of each hospital feature is determined according to different scenarios. For example, in a new hospital scenario, hospital level, hospital type, and region have higher weights; in an active hospital scenario, historical preferences have higher weights. This allows for more intelligent extraction of key information and its structured organization into a format suitable for AI understanding to support subsequent intelligent recommendations.

[0018] Next, step two is performed: dynamically generating a Prompt based on the current hospital scenario, and calling a large language model to analyze the Prompt and obtain matching solutions. The dynamic generation process of the Prompt includes: first, establishing a basic Prompt template, which includes hospital information, a list of candidate solutions, tasks, and output format; determining the corresponding recommendation strategy based on the current hospital scenario, adding the recommendation strategy to the basic Prompt template, and obtaining the Prompt. The following is the Prompt in this embodiment. "You are a senior nursing education expert who is skilled at recommending the most suitable nursing education programs for different hospitals."

[0019] [Hospital Information] - Hospital Name: {hospital_name} - Hospital grade: {hospital_grade} - Hospital type: {hospital_type} - Location: {province} {city} - Hospital size: Number of beds, number of nurses {dynamic_sections} [List of Candidate Solutions] {candidate_solutions}

Task

[0020] Output Format Please output in JSON format, including the following fields: { "recommendations": [ { "solution_id": Solution ID, "solution_name": "Solution name", "match_score": Match score (0-10), "match_dimensions": { "grade_match": Hospital grade matching degree (0-10), "type_match": Hospital type matching degree (0-10), "region_match": Region matching score (0-10), "preference_match": Preference matching score (0-10), "content_relevance": Content relevance (0-10) }, "reason": "Reason for recommendation (within 50 words)" } ] }” The candidate solution list consists of solutions from the solution library. The {dynamic_sections} section determines the corresponding recommendation strategy based on the current hospital scenario. When it is a new hospital scenario, the focus is mainly on the hospital level, hospital type, and regional adaptability to the solution. At the same time, solutions with strong universality and ease of implementation are given priority. The following is an illustration of the recommendation strategy for a new hospital scenario: "

Hospital Characteristics Analysis

[0021] The hospital scenario determination process includes: first, determining whether the user search process provides constraints, including keyword constraints and time constraints; if constraints exist, it is determined to be a specific demand scenario; if no constraints exist, the hospital's historical usage count of the solution is compared with a preset range [n1, n2]. In this embodiment, n1=3 and n3=10. If the hospital's historical usage count of the solution is <3, it is determined to be a new hospital scenario; if the hospital's historical usage count of the solution is >10, it is determined to be an active hospital scenario; if the hospital's historical usage count of the solution is ∈ [3, 10], it is determined to be a normal hospital scenario. Through the above determination of the hospital scenario, the status of the hospital can be determined more accurately.

[0022] Step 3: Perform multi-dimensional analysis on the matched solutions based on importance weights. This multi-dimensional analysis includes: obtaining the matching degree for hospital level, hospital type, region, historical preferences, and content relevance, with matching degree scores ranging from 1 to 10. The matching degree scores for each hospital feature are weighted and summed according to their importance weights to obtain a matching degree score. Therefore, the matching degree score can make more accurate recommendations based on the current status of the hospital. After obtaining the matching degree score, it is adjusted based on business rules to obtain the final score. This process includes: obtaining a score adjustment factor according to business rules; using the sum of the matching degree score and the adjustment factor as the final score; and sorting according to the final score. The business rules include: adding points for new solutions launched less than 30 days ago to support new solutions; adding points when a solution is recommended by experts; and adding points when a solution is popular. When a solution is used more than 100 times, points are deducted to give other solutions a chance. Solutions from the same region are given points. The solutions are sorted in descending order of their final scores. The results are then adjusted for diversity to obtain the recommended solutions. The diversity adjustment process involves: starting from the second-ranked solution, judging the similarity between the current solution and the solutions before it based on a large language model. The similarity judgment is implemented based on the large language model. The similarity score is compared with a preset threshold, which is set based on empirical data. Therefore, when there is a similarity score greater than the preset threshold, it means that the two solutions are highly similar. To avoid clustering of similar solutions, the solution corresponding to the current order is excluded, and a preset number of candidate solutions are obtained as the recommended results. Therefore, the recommended results not only reflect the current state of the hospital but also provide multiple options for easy selection.

[0023] In the above scheme, the large language model can be selected from commercial APIs (such as GPT-4, Claude, and Wenxin Yiyan) and open-source models (ChatGLM and Qwen) according to the requirements, without any restrictions here.

[0024] This embodiment was evaluated on 500 test samples. The following are the comparative experimental data: The comparative experimental data above shows that, compared with traditional recommendation methods, this embodiment, through the processes of hospital feature extraction, dynamic prompt generation, multi-dimensional analysis, and secondary adjustment, can dynamically construct the information (Prompt) sent to the AI ​​based on the characteristics of different hospitals and different recommendation scenarios. This ensures that the AI ​​can obtain the most relevant and accurate contextual information for reasoning, deeply understand the semantic relationship between hospital needs and solution content, rather than simple keyword matching, accurately calculate the matching degree between hospitals and solutions, and accurately parse key information such as matching degree scores and recommendation reasons from the unstructured text returned by the AI. Based on the multi-dimensional scores, it performs reasonable sorting and generates the final recommendation list, which has the advantages of high automation, accurate matching, and high recommendation efficiency.

[0025] In addition, the algorithm in this embodiment also includes step four: real-time acquisition of user feedback information, and dynamic acquisition of attention weights for each hospital feature based on the feedback information. This process includes: acquiring user click schemes based on feedback information, sequentially judging the similarity of hospital features between any two schemes in the user click scheme based on a large language model, summing all similarities for the same hospital feature to obtain the consistency score of that hospital feature; using the consistency score ratio of hospital features as attention weights. The higher the attention to a certain hospital feature in the user click scheme, the higher the sum of all similarities for that feature, and the higher the proportion. This allows for continuous adjustment of the attention weights of different hospital features based on the content the user is interested in (user click schemes). The mean weight is calculated based on the attention weight and importance weight, and the matching score is recalculated according to the mean weight. New recommendation results are obtained based on the matching score, thereby ensuring that the recommendation results match the hospital status while dynamically adjusting the information content and weights sent to the AI ​​based on real-time user feedback, achieving continuous optimization of the recommendation strategy.

[0026] 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. An AI-based nursing teaching program recommendation algorithm, characterized in that, include: Step 1: Extract information from the current hospital to obtain its characteristics; The importance weight of each hospital's characteristics is dynamically assessed based on the current hospital scenario. Step 2: Dynamically generate a Prompt based on the current hospital scenario, call the large language model to analyze the Prompt, and obtain matching solutions; Step 3: Perform multi-dimensional analysis on the matched solutions based on importance weights to obtain a matching score. Adjust the matching score based on business rules to obtain the final score. Sort the final scores in descending order and then perform diversity adjustments on the sorting results to obtain the recommendation results.

2. The AI-based nursing teaching program recommendation algorithm according to claim 1, characterized in that, The process of dynamically generating the Prompt includes: Establish a basic Prompt template, which includes hospital information, a list of candidate solutions, tasks, and output format; Determine the corresponding recommendation strategy based on the current hospital scenario, add the recommendation strategy to the basic Prompt template, and obtain the Prompt.

3. The AI-based nursing teaching program recommendation algorithm according to claim 2, characterized in that, The hospital scenario assessment process includes: Determine whether the user search process provides constraints: If yes, then it is determined to be a specific demand scenario; If not, compare the hospital's historical usage count of the plan with the preset range [n1, n2]: If the number of times the hospital's solution has been used in the past is less than n1, then it is judged as a new hospital scenario; If the number of times the hospital's solution has been used in the past is greater than n2, then it is judged as an active hospital scenario; If the hospital's historical usage count is ∈ [n1, n2], then it is judged as a typical hospital scenario.

4. The AI-based nursing teaching program recommendation algorithm according to claim 2, characterized in that, The hospital characteristics include hospital level, hospital type, region, historical preferences, and content relevance; The process of the multidimensional analysis includes: The matching scores for hospital level, hospital type, region, historical preferences, and content relevance are obtained separately. The matching scores of hospital features are weighted and summed according to their importance to obtain the matching score.

5. The AI-based nursing teaching program recommendation algorithm according to claim 4, characterized in that, The process of adjusting the matching score based on business rules includes: The score adjustment factor is obtained according to the business rules, and the sum of the matching score and the adjustment factor is used as the final score. The scores are then sorted according to the final scores. The business rules include: Bonus points will be awarded if the new plan has been online for less than 30 days; Bonus points will be awarded if the proposal receives an expert recommendation; Deduct points if the proposal is a popular one; Bonus points for proposals from the same region.

6. The AI-based nursing teaching program recommendation algorithm according to claim 5, characterized in that, The process of adjusting for diversity includes: Starting from the second-ranked result, the similarity between the current-ranked solution and the solutions before it is determined sequentially based on the large language model. If there is a similarity score greater than a preset threshold, the solution corresponding to the current-ranked solution is excluded, and a preset number of candidate solutions are obtained as the recommendation result.

7. The AI-based nursing teaching program recommendation algorithm according to claim 6, characterized in that, The algorithm also includes: Step 4: Obtain user feedback information in real time, dynamically obtain the attention weight of each hospital feature based on the feedback information, recalculate the matching score based on the attention weight and importance weight, and obtain new recommendation results based on the matching score.

8. The AI-based nursing teaching program recommendation algorithm according to claim 7, characterized in that, The process of obtaining the attention weight includes: Based on the feedback information, user click scenarios are obtained. The similarity of hospital features between any two scenarios is judged sequentially based on the large language model. All similarities of the same hospital feature are summed to obtain the consistency score of the hospital feature. The consistency score ratio of the hospital feature is used as the attention weight.

Images