Intelligent recommendation method and device for training resources of electric power safety online learning platform

Abstract

The invention discloses a training resource intelligent recommendation method and device for an electric power safety online learning platform, and the method comprises the steps: obtaining the psychological state data and recent work content of a user, and generating a user data set according to the psychological state data and the recent work content; obtaining interaction data between the user and the learning platform, and obtaining an associated data set according to the interaction data and the user data set; acquiring training resource data, and generating a resource relation data set according to the training resource data; training the deep learning model according to the user data set, the associated data set and the resource relationship data set to obtain a recommendation model; obtaining to-be-recommended user data and an alternative resource set; the to-be-recommended user data and the alternative resource set are input into the recommendation model, and the recommendation model is used for selecting optimal resources from the alternative resource set according to the to-be-recommended user data and pushing the optimal resources, so that the online safe learning efficiency of construction personnel is effectively improved, and accurate personalized training is realized.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power teaching, and in particular to a method and device for intelligently recommending training resources for an electric power safety online learning platform. Background Art

[0002] Electric power companies have two goals when launching online power safety education. The first is long-term normalized basic education, that is, through daily learning, to help power construction personnel fully master the power safety production knowledge that they should have in their job types. The second is short-term work goal education, that is, according to the individual characteristics of specific construction personnel, to guide them to learn the knowledge related to the key work they are undertaking in the near future, so as to prepare for the recent work in terms of safety knowledge, skills and awareness.

[0003] At present, the core strategy of the recommendation algorithm is to analyze the reading interests and concerns of users through user profiling technology, and screen the training resources that fit the user's reading interests based on the in-depth exploration of the attributes of the training resources themselves, such as the existing collaborative filtering recommendation, content-based recommendation, deep learning recommendation and other general artificial intelligence algorithms. However, the existing various power resource recommendation methods mainly focus on the analysis of long-term stable user characteristics such as construction personnel's education, major, and type of work when profiling users. Although they can better meet the long-term goals of online power safety education, they also cannot meet the needs of short-term goals well because they pay less attention to users' short-term characteristics. Therefore, it is impossible to achieve dynamic and accurate recommendation of safety knowledge and skills training resources. Summary of the invention

[0004] The technical problem to be solved by the present invention is to provide a method and device for intelligently recommending training resources for an electric power safety online learning platform, so as to achieve accurate push of electric power teaching resources.

[0005] In order to solve the above technical problems, the technical solution adopted by the present invention is: A training resource intelligent recommendation method for an electric power safety online learning platform, comprising: Acquire the user's psychological state data and recent work content, and generate a user data set according to the psychological state data and recent work content; Obtaining interaction data between the user and the learning platform, and obtaining a related data set based on the interaction data and the user's short-term dynamic profile; Acquire training resource data, and generate a resource relationship data set according to the training resource data; The deep learning model is trained according to the user data set, the associated data set, and the resource relationship data set to obtain a recommendation model; Obtain the user data to be recommended and the set of candidate resources; The user data to be recommended and the candidate resource set are input into the recommendation model, and the recommendation model pushes the best resource in the candidate resource set.

[0006] In order to solve the above technical problems, another technical solution adopted by the present invention is: A training resource intelligent recommendation device for an electric power safety online learning platform comprises a memory, a processor and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, each step in the above-mentioned training resource intelligent recommendation method for an electric power safety online learning platform is implemented.

[0007] The beneficial effects of the present invention are as follows: by acquiring the recent key tasks and immediate psychological state of construction personnel to characterize the short-term dynamic characteristics of construction personnel, and combining the corresponding interactive data to obtain the associated data set, and generating the resource relationship data set according to the training resource data, so that after the deep learning model is trained by the user data set, the associated data set and the resource relationship data set, the recommendation of training resources can be tailored to the immediate psychological state of the construction personnel, helping the construction personnel to prepare safety knowledge and skills for recent on-site work tasks, effectively improving the online safety learning efficiency of the construction personnel, and realizing accurate personalized training. BRIEF DESCRIPTION OF THE DRAWINGS

[0008] Figure 1 A flowchart of the steps of a method for intelligently recommending training resources for an electric power safety online learning platform in an embodiment of the present invention; Figure 2 A schematic diagram of a knowledge graph structure in a method for intelligently recommending training resources for an online learning platform for electric power safety in an embodiment of the present invention; Figure 3 A schematic diagram of a neural network architecture in a method for intelligently recommending training resources for an electric power safety online learning platform in an embodiment of the present invention; Figure 4 The present invention is a schematic diagram of the structure of a training resource intelligent recommendation device for an electric power safety online learning platform in an embodiment of the present invention. DETAILED DESCRIPTION

[0009] In order to explain the technical content, achieved objectives and effects of the present invention in detail, the following is an explanation in combination with the implementation modes and the accompanying drawings.

[0010] Please refer to Figure 1 , a training resource intelligent recommendation method for an electric power safety online learning platform, comprising: Acquire the user's psychological state data and recent work content, and generate a user data set according to the psychological state data and recent work content; Acquire the interaction data between the user and the learning platform, and obtain a related data set based on the interaction data and the user data set; Acquire training resource data, and generate a resource relationship data set according to the training resource data; The deep learning model is trained according to the user data set, the associated data set, and the resource relationship data set to obtain a recommendation model; Obtain the user data to be recommended and the set of candidate resources; The user data to be recommended and the candidate resource set are input into the recommendation model, and the recommendation model is used to select the optimal resource from the candidate resource set for push according to the user data to be recommended.

[0011] From the above description, it can be seen that the beneficial effects of the present invention are: by obtaining the recent key work and immediate psychological state of the construction personnel to characterize the short-term dynamic characteristics of the construction personnel, and combining the corresponding interactive data to obtain the associated data set, and generating a resource relationship data set based on the training resource data, so that after the deep learning model is trained by the user data set, the associated data set and the resource relationship data set, the recommendation of training resources can be in line with the immediate psychological state of the construction personnel, helping the construction personnel to prepare safety knowledge and skills for recent on-site work tasks, effectively improving the online safety learning efficiency of the construction personnel, and realizing accurate personalized training.

[0012] Furthermore, obtaining the user's psychological state data includes: Obtaining the collection time of the user's psychological state data; It is determined whether the collection time is greater than a preset collection period. If so, the user is instructed to sample psychological data according to a psychological scale to obtain the psychological state data.

[0013] From the above description, it can be seen that obtaining the user's psychological state data based on the preset collection cycle can effectively obtain the user's psychological state data in the current working cycle, so that the recommended content is more in line with the user's psychological state data in the current cycle, thereby improving the accuracy of content recommendation.

[0014] Furthermore, obtaining the user's recent work content includes: Obtain the user's work tasks and job attributes; According to the work task, all corresponding operation processes are obtained in the process system table; The job attributes are matched with all the operation procedures to obtain the job procedures corresponding to the user.

[0015] From the above description, it can be seen that by obtaining the user's work tasks and job attributes, first matching all corresponding related work processes through the work tasks, and then combining the job attributes to accurately locate the user's current job process.

[0016] Furthermore, the acquisition of interaction data between the user and the learning platform includes: The interactive data is obtained through a data collection point at a preset location on the learning platform.

[0017] From the above description, it can be seen that by setting data collection points at preset locations of the learning platform, such as login, browsing and other specific locations, when construction workers learn on the platform, their behavior data can be recorded through the collection points.

[0018] Furthermore, the interaction data includes interaction resources and resource scores corresponding to the interaction resources; The associated data sets obtained based on the interaction data and the user's short-term dynamic portrait include: Constructing related data corresponding to each user based on the psychological state data, recent work content, interactive resources, and resource scores of each user; A related data set is obtained according to all the related data.

[0019] From the above description, it can be seen that related data is generated based on the user's mental state data, recent work content, interactive resources and resource scores, that is, the user's personal mental state data, recent work content and training resources are associated, which can build the relationship between the user's mental state data, recent work content and training resources.

[0020] Furthermore, after acquiring the interactive data through the data collection point, the method further includes: Performing data cleaning on the interaction data; Interpolating the minority category samples in the cleaned interactive data to generate synthetic samples to obtain a balanced data set; The balanced data set is cleaned a second time to obtain updated interaction data.

[0021] From the above description, it can be seen that by cleaning the original data, dirty data generated by abnormal behaviors such as platform score brushing can be removed; and by generating synthetic samples through interpolation, the problem of data set imbalance can be solved, and data cleaning can be performed again to remove the noise and redundant synthetic samples introduced in the process of interpolating and generating synthetic samples, so as to prevent model performance degradation caused by unbalanced data.

[0022] Furthermore, generating a resource relationship data set according to the training resource data includes: Get the entities, entity attributes and relationships of each training resource; Obtain a resource relationship group of each training resource according to the entities, entity attributes and relationships; The resource relationship data set is obtained according to all the resource relationship groups.

[0023] From the above description, it can be seen that constructing a resource relationship dataset based on the entities, entity attributes and relationships of each training resource can effectively reveal the potential connection between the underlying knowledge characteristics of the training resources and the user's reading interests.

[0024] Further, training the deep learning model according to the user data set, the associated data set, and the resource relationship data set includes: The deep learning model is trained based on the user data set, the associated data set with the target task as the recommended resource, and the source task as the resource relationship data set embedding.

[0025] From the above description, it can be seen that while performing target task training based on the user dataset and the associated dataset, source task training is also performed through the resource relationship dataset, so that the training results obtained in the source task training can be used to improve the performance of the recommendation task.

[0026] Furthermore, the step of inputting the to-be-recommended user data and the candidate resource set into the recommendation model includes: The to-be-recommended user data and each candidate resource in the candidate resource set form a prediction group to obtain a prediction group set; The recommendation model pushes the best resource in the candidate resource set including: The recommendation model generates a prediction score for each of the prediction groups, and pushes the corresponding candidate resources in the prediction group with the highest prediction score.

[0027] From the above description, it can be seen that by forming a prediction group with the user data to be recommended and each candidate resource for score prediction, the score prediction of each candidate resource based on the user data is realized, and the candidate resource with the highest predicted score is pushed, thereby improving the accuracy of content recommendation.

[0028] Another embodiment of the present invention provides an intelligent recommendation device for training resources of an online learning platform for electric power safety, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, each step in the above-mentioned method for intelligent recommendation of training resources for an online learning platform for electric power safety is implemented.

[0029] The training resource intelligent recommendation method and device of the electric power safety online learning platform provided by the present invention can be applied to the electric power training content recommendation scenario, which is described below through specific implementation methods: Embodiment 1 Please refer to Figure 1 , a training resource intelligent recommendation method for an electric power safety online learning platform, comprising: S1. Obtain the user's psychological state data and recent work content, and generate a user data set based on the psychological state data and recent work content; the psychological state data is used to characterize the user's psychological state portrait as the user's short-term dynamic feature; the purpose of the psychological state portrait is to quantitatively evaluate the immediate psychological state of the construction personnel, so as to incorporate the consideration of the psychological state into the resource recommendation strategy to achieve psychological intervention; the main method is to use a psychological scale to conduct psychological assessments on the construction personnel at a certain period to obtain psychological ability scores, and then conduct a qualitative evaluation of their psychological state based on the psychological scores, and use them as the psychological state portraits of the construction personnel; The main steps of constructing a psychological state portrait are as follows: (1) Design Psychology Scale According to the demand characteristics of power safety learning, this embodiment designs the "Power Construction Personnel Safety Production Psychological State Scale" to evaluate the psychological state of power construction personnel in safety production, including multiple dimensions such as stress perception and emotional stability. This psychological scale is used to sample the psychological data of construction personnel.

[0030] Table 1. Psychological status scale for power construction workers in production safety

[0031] (2) Psychological data sampling Since the immediate psychological state is unstable and changes over time, it needs to be evaluated repeatedly at a certain period. Another key point is to arrange the evaluation path reasonably and simplify the sampling process. Specifically: By obtaining the collection time of the user's psychological state data, it is determined whether the collection time is greater than the preset collection period. If so, the user is instructed to sample the psychological data according to the psychological scale to obtain the psychological state data. For example, with 7 days as the evaluation period, when the construction personnel log in to the online learning platform, the platform will check whether the last evaluation result of the personnel has exceeded 7 days. If it has exceeded 7 days, the online evaluation page will pop up, requiring them to fill in the psychological scale within the preset time and re-sample the psychological data.

[0032] (3) Qualitative evaluation of psychological state Based on the results of psychological data sampling, a qualitative evaluation can be made on the psychological state of construction workers; the scores of all questions are added up to get the total score, with a full score of 75 points; the higher the total score, the better the psychological state of construction workers in terms of safe production. Based on the total score, the following qualitative evaluation can be made: Good: A score above 50 indicates a good mental state and is suitable for participating in safe production without the need for excessive psychological intervention. Therefore, when recommending training resources, emphasis should be placed on resources related to safety knowledge and skills.

[0033] Normal: The score is between 20 and 50 points. There is a certain degree of psychological pressure or fatigue, and certain psychological support is needed. When recommending training resources, appropriate resources for risk warning education in safety awareness should be recommended.

[0034] Poor: The score is less than 20 points, the mental state is poor, and immediate psychological intervention is needed. It is not suitable for participating in high-risk operations. When recommending training resources, emphasis should be placed on psychological intervention resources in safety awareness.

[0035] Among them, the recent key work profile of the user is portrayed through the recent work content as another short-term dynamic feature of the user; by incorporating the recent key work into the user profile of the construction personnel, the recommended training resources can be more in line with the user's recent work needs; the main method is to obtain the recent work arrangements of the individual construction personnel, and then map them to the work process system to describe the key work content in a refined manner. The main steps for constructing the recent key work profile are as follows: (1) Obtaining recent work arrangements According to the requirements of power work, the work task must have a corresponding work ticket, so the work ticket can be used to obtain the employee's recent work arrangements; the work ticket is a standardized written instruction in power production, which records in detail the content, location, time, safety measures, participants and other information of the work; the work ticket is generally saved in the two-ticket management system, so it can be automatically obtained from the two-ticket management system through the program interface; different types of power operations have work tickets in different formats, but all work tickets have "work shift personnel" information, so the current user can be compared with the "work shift personnel information" in the recent work ticket to obtain the work ticket related to it, and then the "work task" information can be extracted to obtain the user's recent work arrangements; in order to ensure the timeliness of the work task attributes, the screening range of work tickets is limited to one week before and after the current time point.

[0036] (2) Mapping to work processes The work arrangement obtained from the work ticket has a coarse description granularity, which only describes the overall content of the entire task. However, the construction personnel only undertake the specific work corresponding to their type of work in the entire task. Therefore, it is necessary to further refine the work arrangement in combination with the previous user's type of work in order to make more targeted training resource recommendations. This step is achieved through the mapping of type of work and process. In order to solve the problem of coarse granularity of recent work description in the existing recommendation method, the type of work and process system is introduced in this embodiment. The type of work and process describes the safety regulations and safety measures that personnel of a specific type of work in a specific profession should comply with when engaging in a certain operation process. Its typical structure is as follows: Table 2. Work process system table

[0037] The mapping method is as follows: First, find all the corresponding work processes according to the "work task"; since the "work task" generally includes the corresponding process name when describing it, this step can be completed directly using the keyword fuzzy matching algorithm; after finding the corresponding "work process", combine the "type of work" attribute of the current user to locate the corresponding type of work process; since the type of work process can accurately and uniquely identify the specific work content that a certain type of work should undertake in the power production task, the description of the current user's recent work is also more accurate.

[0038] At the same time, in addition to using psychological state data and recent work content as short-term features to characterize the user's short-term feature portrait, this embodiment also characterizes the user's long-term static feature portrait; for example: describing the user's age, education, job position and other long-term stable safety education characteristics. This part has been discussed a lot in the existing recommendation algorithm. When used, it can be selected according to specific application needs. It will not be repeated in this embodiment; for example, in this implementation example, {gender, education, major, job} 4 features are selected as portrait attributes. Based on the long-term static feature portrait constructed in the above steps and the short-term dynamic feature portrait, a user portrait is established and a user data set is obtained.

[0039] S2. Obtaining the interaction data between the user and the learning platform, and obtaining a related data set based on the interaction data and the user data set, specifically including the following steps: (1) Interactive dataset format: Considering the impact of users’ short-term dynamic features on reading interest, the collected interactive dataset format is set as<ui,pp,pw,cj,s> , that is, the triple of <user i, immediate mental state, recent key tasks, resource j, score>, where “immediate mental state” and “recent key tasks” are both derived from the current user’s short-term dynamic feature user portrait, describing the reading interest level of user i for training resource j when he is in p mental states and undertakes w tasks.

[0040] (2) Interest scoring rules: The score quantitatively reflects the user's interest in reading the training resource, and the value range is [0,1]. 0 is a negative example, that is, the user opened the training resource but closed it without reading it effectively, indicating that the user has no interest in it; 1 is a positive example, that is, the user has had effective interactive behaviors on the training resource: effective reading, commenting, liking, and scoring.

[0041] (3) Data sampling method: This embodiment adopts the buried point method for data sampling, and sets data collection points at specific locations such as login and browsing of the learning platform. When the construction personnel learn on the platform, the collection points will record their behavior data; the deep learning model will use these data to analyze the user's learning behavior pattern, preference and demand, so as to establish a recommendation strategy; the specific implementation method of data sampling in this implementation can be customized according to factors such as the technology selection and programming language of the learning platform. A general implementation example is as follows: Front-end tracking technology: Use JavaScript, HTML5 and other technologies to embed tracking codes in web pages or HTML5 applications; Back-end tracking technology: record user request and response data by setting up an interceptor on the server side; Tracking locations: Tracking is performed at four locations: when the user clicks the comment button, clicks the like button, clicks the score button, and when the user reads the training resource for 3 minutes (triggered by a timer); Storage of tracking data: Save it to an unstructured database, such as MongoDB, according to the "interactive data set format" defined above.

[0042] (4) Interactive data collation method: In order to improve the quality of the sampled interactive data and meet the probabilistic distribution requirements of deep learning model training, the data needs to be sorted; the data sorting implementation method can be customized according to the specific application scenario. A general implementation example is as follows: first, the original data needs to be cleaned to remove the dirty data generated by abnormal behaviors such as platform score brushing; then the problem of data set imbalance needs to be dealt with. Due to the characteristics of the employee composition of the power grid company, for example, the proportion of front-line personnel is large while the proportion of personnel in some positions is extremely small, resulting in data imbalance. The imbalance of the data set will cause the model to overfit most categories. Therefore, in this embodiment, oversampling is used for processing, and the specific method is as follows: First, SMOTE is used to interpolate between minority category samples to generate synthetic samples to increase the number of minority category samples. Then the Tomek Links algorithm is applied to perform secondary cleaning on the data to remove the noise and redundant synthetic samples introduced in the SMOTE process to prevent the model performance degradation caused by unbalanced data, while reflecting the user's real reading interest habits.

[0043] S3. Obtain training resource data and generate a resource relationship dataset based on the training resource data. In order to reveal the potential connection between the underlying knowledge features of training resources and user reading interests, it is necessary to establish a knowledge graph that can describe the knowledge features of training resources and their internal logical relationships as a training dataset for the knowledge graph embedding subtask in training the deep learning model. The specific method is as follows: (1) Constructing ontology model First, we need to formally represent the knowledge in the training resources by building an ontology model, where the ontology model can be customized according to the characteristics of the application scenario. A typical ontology model example is shown in the following table: Table 3. Entity model

[0044] Table 4. Relationship model

[0045] (2) Constructing HRT triples According to the established ontology model, it is necessary to analyze the resources one by one according to the ontology model, and construct the corresponding <head entity h, relationship r, tail entity t> triple (hereinafter referred to as HRT triple) for each training resource. Each training resource can be mapped to a unique entity in the ontology model. When constructing, the entities, entity attributes, and relationships in Tables 3 and 4 are directly extracted from the corresponding fields in the database. At the same time, since the values ​​of each field have been selected from the alternative set according to the unified specification when the resource is released, it is unique, so there is no need to perform entity disambiguation and coreference resolution.

[0046] For example, for the training resource library named "Resource A", the entity attributes are: substation major, safety awareness class, video resource, and the following HRT triples can be constructed based on the ontology model: <Header entity = Resource A, Relationship = Related Majors, Tail entity = Substation Major> <Header entity = resource A, relationship = representation method, tail entity = video resource> <Header entity = Resource A, Relationship = Subject Type, Tail entity = Security Awareness> According to the above method, each resource in the training resource library can construct 3 corresponding HRT triples. Assuming the number of resources in the resource library is n, 3*n HRT triples will be generated in the end.

[0047] (3) Generate knowledge graph Through the above steps, we will eventually obtain a set of HRT triples containing all training resource entities. This set constitutes the knowledge graph of the entire resource library, which can fully describe the inherent knowledge structure between all resources. At the same time, it can be directly imported into graph databases such as Neo4J for storage, or displayed through knowledge graph visualization tools such as SmartKG. Figure 2 The following is an example of the knowledge graph that is finally generated.

[0048] S4. The deep learning model is trained according to the user data set, the associated data set and the resource relationship data set to obtain a recommendation model; that is, three data sets can be obtained through steps S1-S3: a user portrait data set for describing user characteristics, a training resource knowledge graph data set for describing training resource characteristics, and an interaction data set between users and training resources for describing reading interests; in this embodiment, the following deep learning model is designed based on the MKR algorithm, a general recommendation algorithm based on the knowledge graph; Please refer to Figure 3 The model constructed by this algorithm is a multi-task transfer learning neural network architecture. The source task is the knowledge graph embedding (hereinafter referred to as KGE) subtask, and the target task is the recommendation subtask. The knowledge gained from the KGE task training can be used to improve the performance of the recommendation task. For details, please refer to the MKR algorithm specification.

[0049] Based on the user portrait dataset, the training resource knowledge graph dataset, and the user-training resource interaction dataset, four types of vectors are generated during model training: resource vector c, user vector u, head entity vector h, and relationship vector r. Since each training resource can be mapped to a unique entity in the knowledge graph and has co-reference, the two can be placed in the same vector space for training, thereby integrating the entity feature information in the knowledge graph into the resource vector, thereby strengthening the influence of the knowledge structure of training resources on search results.

[0050] The recommendation subtask uses a collaborative filtering algorithm, with u' and resource vector c as input. u' contains aggregated information about user features and entity relationships. At the same time, the training resource vector c and entity vector e are in the same vector space, so the knowledge structure information in the knowledge graph can effectively enhance the prediction results. For details on model training rounds, positive and negative example construction, etc., please refer to the MKR algorithm specification.

[0051] S5. Obtain the user data to be recommended and the candidate resource set; that is, construct the user profile to be recommended through the method in step S1 to obtain the corresponding user data set ui; the candidate resource set is represented as Ci={c1,c2,...cn}.

[0052] S6. Input the user data to be recommended and the candidate resource set into the recommendation model. The recommendation model is used to select the best resource from the candidate resource set according to the user data to be recommended for push. Specifically: S61, the user data to be recommended and each candidate resource in the candidate resource set are combined into a prediction group to obtain a prediction group set; that is, each candidate resource cn is extracted one by one from the candidate resource set Ci, and combined with the user data set ui to form a prediction group set.<ui,cn> Yes, as a parameter input to the prediction subtask part; S62: The recommendation model generates a prediction score for each prediction group, and pushes the corresponding candidate resources in the prediction group with the highest prediction score; that is, the recommendation model generates a prediction score for each prediction group.<ui,cn> Calculate the reading interest prediction score y; finally select the prediction group with the highest reading interest prediction score y<ui,cn> , where the resource corresponding to cn is the training resource that user ui is currently most interested in, and it is recommended to the user.

[0053] Embodiment 2 Please refer to Figure 4 , a training resource intelligent recommendation device for an electric power safety online learning platform, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, the processor implements the various steps in the above-mentioned training resource intelligent recommendation method for an electric power safety online learning platform.

[0054] In summary, the method and device for intelligently recommending training resources for the online learning platform for electric power safety provided by the present invention not only consider the long-term stable characteristics of construction personnel when constructing user portraits, but also consider the short-term dynamic characteristics of immediate psychological state and recent key work components, so as to describe the dynamic safety training demand characteristics of construction personnel; at the same time, the knowledge ontology structure of training resources is modeled using a knowledge graph containing triples of work types and processes, so that safety knowledge and safety skill resources can be recommended based on the specific work details undertaken by construction personnel in construction tasks, further refining the recommendation granularity; and on this basis, a deep learning model is trained to construct a deep learning model that can automatically mine and dynamically update the inherent deep logical relationship between psychological state factors and key work factors and recommended needs, and based on this model, accurate recommendation of training resources is achieved.

[0055] The above descriptions are merely embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent transformations made using the contents of the present invention's specification and drawings, or directly or indirectly applied in related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A training resource intelligent recommendation method for an electric power safety online learning platform, characterized in that: include: Acquire the user's psychological state data and recent work content, and generate a user data set according to the psychological state data and recent work content; Acquire the interaction data between the user and the learning platform, and obtain a related data set based on the interaction data and the user data set; Acquire training resource data, and generate a resource relationship data set according to the training resource data; The deep learning model is trained according to the user data set, the associated data set, and the resource relationship data set to obtain a recommendation model; Obtain the user data to be recommended and the set of candidate resources; The user data to be recommended and the candidate resource set are input into the recommendation model, and the recommendation model is used to select the optimal resource from the candidate resource set for push according to the user data to be recommended.

2. According to claim 1, a training resource intelligent recommendation method for an electric power safety online learning platform is characterized in that: Obtaining the user's psychological state data includes: Obtaining the collection time of the user's psychological state data; It is determined whether the collection time is greater than a preset collection period. If so, the user is instructed to sample psychological data according to a psychological scale to obtain the psychological state data.

3. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 1 is characterized in that: Recent work on acquiring users includes: Obtain the user's work tasks and job attributes; According to the work task, all corresponding operation processes are obtained in the process system table; The job attributes are matched with all the operation procedures to obtain the job procedures corresponding to the user.

4. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 1 is characterized in that: The acquisition of interaction data between the user and the learning platform includes: The interactive data is obtained through a data collection point at a preset location on the learning platform.

5. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 4 is characterized in that: The interaction data includes interaction resources and resource scores corresponding to the interaction resources; The associated data set obtained according to the interaction data and the user data set includes: Constructing related data corresponding to each user based on the psychological state data, recent work content, interactive resources, and resource scores of each user; A related data set is obtained according to all the related data.

6. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 4, characterized in that: After acquiring the interactive data through the data collection point, the method further includes: Performing data cleaning on the interaction data; Classifying the cleaned interactive data, and interpolating minority category samples whose number of classified samples is less than a preset number to generate synthetic samples, to obtain a balanced data set; The balanced data set is cleaned a second time to obtain updated interaction data.

7. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 1, characterized in that: Generating a resource relationship data set according to the training resource data includes: Get the entities, entity attributes and relationships of each training resource; Obtain a resource relationship group of each training resource according to the entities, entity attributes and relationships; The resource relationship data set is obtained according to all the resource relationship groups.

8. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 1, characterized in that: Training the deep learning model according to the user data set, the associated data set, and the resource relationship data set includes: The deep learning model is trained based on the user data set, the associated data set with the target task as the recommended resource, and the source task as the resource relationship data set embedding.

9. The method for intelligently recommending training resources for an electric power safety online learning platform according to claim 1, characterized in that: The step of inputting the to-be-recommended user data and the candidate resource set into the recommendation model comprises: The to-be-recommended user data and each candidate resource in the candidate resource set form a prediction group to obtain a prediction group set; The recommendation model is used to select the best resource from the candidate resource set for push according to the user data to be recommended, including: The recommendation model generates a prediction score for each of the prediction groups, and pushes the corresponding candidate resources in the prediction group with the highest prediction score.

10. An intelligent training resource recommendation device for an electric power safety online learning platform, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, each step of the method for intelligently recommending training resources for an electric power safety online learning platform as described in any one of claims 1 to 9 is implemented.

Images