Education topic identification method and device, related equipment and program product

Abstract

The invention discloses an education subject identification method and device, related equipment and a program product, and relates to the field of natural language processing. Education target subjects are screened out from internet information source data through a configured education target subject classification model, and the education target subjects can be identified according to the education target subject classification model. Calculating the similarity between the education target topic and each main topic in the target topic library, if the target main topic with the similarity greater than a set similarity threshold exists, combining the education target topic with the target main topic, otherwise, adding the education target topic as a new main topic into the target topic library; and through the configured large model, according to the set education target topic label system, labeling the classification sub-labels to which the main topics in the target topic library belong. The method does not depend on manual auditing, generalization perception and recognition work of the education target themes is completed more accurately through cooperation of the large model and the small model, and the efficiency and accuracy of recognition of the education target themes are improved.

Description

Technical Field

[0001] The present application relates to the technical field of natural language processing, and more specifically, to an educational topic identification method, apparatus, related equipment and program product. Background Art

[0002] There are many types of topics in the field of education, such as academic topics, security topics, educational innovation topics, etc. Users need to discover and deal with certain types of target topics in a timely manner. However, these topic data are scattered in Internet data. The current education field lacks efficient intelligent analysis tools. The lack of intelligence leads to excessive reliance on manual operations in the information collection and processing process, which not only increases the workload, but also increases the possibility of errors, resulting in low recognition efficiency and low recognition accuracy of educational target topics. Summary of the invention

[0003] In view of the above problems, this application is proposed to provide an educational topic identification method, device, related equipment and program product to improve the efficiency and accuracy of educational target topic identification. The specific scheme is as follows:

[0004] In a first aspect, a method for identifying educational topics is provided, comprising:

[0005] Obtaining Internet information source data, and filtering out education target topics from the Internet information source data by using a configured education target topic classification model, wherein the education target topic classification model is trained using sample data labeled with education target topic classification labels; the education target topic classification model is a model with a smaller parameter scale than a large model;

[0006] For the educational target topic, calculate its similarity with each main topic in the target topic library. If there is a target main topic whose similarity is greater than the set similarity threshold, merge the educational target topic with the target main topic. Otherwise, add the educational target topic as a new main topic to the target topic library.

[0007] Through the configured large model, the main topics in the target topic library are marked with the classification sub-tags according to the set education target topic label system.

[0008] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of merging the educational target topic with the target main topic includes:

[0009] Merging the educational target topic into the sub-topic set of the target main topic;

[0010] and / or,

[0011] The educational target topic is used to complete the element information of the target main topic.

[0012] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of calculating the similarity between the educational target topic and each main topic in the target topic library includes:

[0013] Utilizing the configured educational topic semantic similarity model, respectively extract the text vector representation of the educational target topic and each main topic in the target topic library, wherein the educational topic semantic similarity model is trained by taking the identification of whether two educational target topics are the same topic as a training task;

[0014] The similarity between the educational target topic and the text vector representation of each main topic in the target topic library is calculated.

[0015] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of respectively extracting the text vector representation of the education target topic and each main topic in the target topic library using the configured education topic semantic similarity model includes:

[0016] For the educational target topic and each main topic in the target topic library:

[0017] Extract the title text and content text of the topic, convert the title text and content text of the topic into word vector sequences respectively, and send them into the education topic semantic similarity model for encoding respectively, to obtain the vector representation of the title text of the topic and the vector representation of the content text;

[0018] Average pooling is performed on the vector representation of the title text of the topic and the vector representation of the content text to obtain a first matrix, and an average operation is performed on the columns of the first matrix to obtain the text vector representation of the topic.

[0019] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of calculating the similarity between the educational target topic and each main topic in the target topic library further includes:

[0020] Comparing the classification sub-label of the educational target topic with the classification sub-label of each main topic in the target topic library to obtain a label similarity score, wherein the classification sub-label of the educational target topic is obtained by large model annotation;

[0021] Based on the similarity scores of the text vector representations of the education goal topic and each main topic and the tag similarity scores, the total similarity score of the education goal topic and each main topic is calculated.

[0022] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the large model includes a special large model obtained by fine-tuning the general large model using the target topic discrimination task data in the field of education;

[0023] The process of labeling the main topics in the target topic library with the classification sub-tags according to the set education target topic label system through the configured large model includes:

[0024] The dedicated large model is called through the first prompt instruction prompt to instruct the model to determine whether the main topic in the target topic library belongs to the target topic according to the target topic judgment rule, and to mark the classification sub-tag of the main topic belonging to the target topic with reference to the set education target topic label system.

[0025] In a possible design, in another implementation of the first aspect of the embodiments of the present application, the large model further includes a universal large model;

[0026] The process of labeling the main topics in the target topic library with the classification sub-tags thereof according to the set education target topic label system through the configured large model also includes:

[0027] For the candidate main topic determined by the special large model as belonging to the target topic, the general large model is further called through the second prompt instruction prompt to instruct the model to refer to the configured key educational target topic to determine whether the candidate main topic is the key focus topic.

[0028] In a possible design, in another implementation of the first aspect of the embodiment of the present application, the process of calling the general large model through the second prompt instruction prompt to instruct the model to refer to the configured focused educational target topic to determine whether the candidate main topic is a focused topic includes:

[0029] Obtain a second prompt instruction prompt format template, the template including a second task instruction, a key focus topic slot and a candidate main topic slot, the second task instruction is used to instruct the model to refer to the key focus educational target topic configured in the key focus topic slot, and determine whether the candidate main topic in the candidate main topic slot is a key focus topic;

[0030] Fill the key focus educational target topic configured by the user into the key focus topic slot, fill the candidate main topic into the candidate main topic slot, obtain a second prompt instruction prompt, input the second prompt instruction prompt into the general big model, and obtain the result output by the general big model.

[0031] In a second aspect, a device for identifying educational topics is provided, comprising:

[0032] A topic screening unit, used to obtain Internet information source data, and screen out educational target topics from the Internet information source data through a configured educational target topic classification model, wherein the educational target topic classification model is trained using sample data labeled with educational target topic classification labels; the educational target topic classification model is a model with a smaller parameter scale than the large model;

[0033] The topic storage unit is used to calculate the similarity between the educational target topic and each main topic in the target topic library, and if there is a target main topic whose similarity is greater than a set similarity threshold, the educational target topic is merged with the target main topic; otherwise, the educational target topic is added to the target topic library as a new main topic;

[0034] The labeling unit is used to label the main topic in the target topic library with the classification sub-label according to the set education target topic label system through the configured large model.

[0035] In a third aspect, an electronic device is provided, comprising: a memory and a processor;

[0036] The memory is used to store programs;

[0037] The processor is used to execute the program to implement each step of the educational topic identification method described in any one of the first aspects of the present application.

[0038] In a fourth aspect, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the various steps of the educational topic identification method described in any one of the first aspects of the present application are implemented.

[0039] In a fifth aspect, a computer program product is provided, including a computer program, which, when executed by a processor, implements the various steps of the educational topic identification method described in any one of the first aspects of the present application.

[0040] By means of the above technical scheme, the present application pre-trains an education target topic classification model for education target topic perception. For Internet source data, the education target topic classification model can quickly screen out the education target topics and eliminate the remaining noise data. Furthermore, considering the multi-source nature of the data, the acquired Internet source data may contain different description data for the same education target topic. Therefore, for the education target topic screened by the classification model, the similarity between it and each main topic in the target topic library is further calculated. If there is a target main topic with a similarity greater than the set similarity threshold, it indicates that the education target topic and the target main topic are the same topic or a highly similar topic. Therefore, the education target topic and the target main topic can be merged to integrate multi-source information to ensure information integrity and consistency, and can effectively filter duplicate information, reduce the complexity of subsequent topic research and judgment work, and improve work efficiency. Furthermore, the present application pre-constructs a complete education target topic labeling system, which can classify and describe the education target topic in detail, which is helpful for topic evaluation, topic response and post-analysis. By leveraging the capabilities of the big model, the main topics in the target topic library are labeled with their classification sub-tags according to the education target topic label system, so as to more accurately complete the generalized perception and identification of education target topics. BRIEF DESCRIPTION OF THE DRAWINGS

[0041] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present application. Also, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0042] Figure 1 A schematic diagram of an implementation system architecture of the educational topic identification method provided in an embodiment of the present application;

[0043] Figure 2 A schematic diagram of a method for identifying educational topics provided in an embodiment of the present application;

[0044] Figure 3 A schematic diagram of another method for identifying educational topics provided in an embodiment of the present application;

[0045] Figure 4 A schematic diagram of the structure of an educational topic identification device provided in an embodiment of the present application;

[0046] Figure 5 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0047] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0048] Educational target topic identification is a type of topic analysis, that is, identifying whether the information posted on the Internet belongs to the target topic in the field of education. The current education field lacks a topic perception system with overall coordination and global judgment capabilities, which hinders the discovery and timely handling of educational target topics.

[0049] This application provides a method for identifying educational target topics, which can be applied to Figure 1 The system architecture shown in FIG. 1 may include a terminal 100 and a server 200. The server 200 may include one or more servers ( Figure 1 A server is included as an example for explanation).

[0050] The terminal 100 or the server 200 can be used alone to execute the educational topic identification method provided in the embodiment of the present application. In addition, the terminal 100 and the server 200 can also be used in collaboration to execute the educational topic identification method provided in the embodiment of the present application.

[0051] Next describe Figure 1 The product form of the mid-terminal 100;

[0052] The terminal 100 in the embodiment of the present application can be a mobile phone, a tablet computer, a desktop computer, a teaching large screen, a laptop computer, an ultra-mobile personal computer (UMPC), a netbook, a personal digital assistant (PDA), etc., and the embodiment of the present application does not impose any limitation on this.

[0053] The present application embodiment provides an educational topic identification method, taking the method applied to a computer device as an example. The computer device may be Figure 1 The terminal 100 or the system consisting of the terminal 100 and the server 200. Figure 2 , the educational topic identification method specifically includes the following steps:

[0054] Step S100: Obtain Internet information source data, and filter out educational target topics from the Internet information source data using the configured educational target topic classification model.

[0055] Internet source data may include data from multiple different sources, such as data provided by different websites, applications, etc. In order to improve data processing efficiency, the Internet source data obtained in this step may be data after preliminary screening, that is, source data in the field of education may be selected based on the source channel, configured preliminary screening rules, etc. Of course, although it has been pre-screened here, the obtained Internet source data may still contain data in non-education fields, which can be filtered through subsequent processing steps.

[0056] The educational topic identification method of this embodiment can be a streaming processing process, that is, the operation of obtaining Internet source data is also performed in a streaming manner. For example, a time length is set at each interval to obtain the Internet source data collected in the previous period for further processing in subsequent steps.

[0057] In order to effectively screen out education-related target topics from massive amounts of Internet information without relying on manual review and to improve information processing efficiency, in this embodiment, an education target topic classification model can be pre-trained and configured, and the education target topic classification model can be trained using sample data labeled with education target topic classification labels. The education target topic classification model can classify input data, and classification labels may include: non-education field data, target topic data in the education field, and non-target topic data in the education field. In this way, each piece of Internet source data obtained can be classified and predicted through the education target topic classification model, and target topics in the education field can be screened out.

[0058] Among them, the educational target topic refers to a designated type of topic in the field of education, such as academic topics in the field of education, safety topics in the field of education, etc. The target topic can be further subdivided and divided into one or more sub-tags according to the tag system.

[0059] The educational goal thematic classification model can use a model with a smaller parameter scale relative to the large model, thereby speeding up the processing speed with a small amount of computing resources.

[0060] In one possible implementation, the education target topic classification model can select a pre-trained model based on the transformer architecture. Furthermore, the pre-trained model can be fine-tuned using the transfer learning method to adapt it to the education target topic classification task.

[0061] Considering that the distribution of educational target topic data in Internet data may be uneven, in this embodiment, a reasonable threshold can be determined for the last layer classification softmax activation function of the educational target topic classification model to balance the accuracy and confidence of the model. The adjustment method of the threshold Threshold can be implemented according to the following formula:

[0062]

[0063] Among them, P i represents the probability of the ith category, C represents the total number of categories, for example, C=3 (including non-educational fields, target topics in the education field, and non-target topics in the education field).

[0064] In this embodiment, a number of sample sets can be selected, and the category of each sample in the sample set can be predicted by the educational target thematic classification model, and the probability P of the sample of each category i accounting for the total number of all samples can be calculated. i , and then a reasonable threshold Threshold can be calculated according to the above formula and used as the threshold of the softmax activation function of the last layer of the education target thematic classification model.

[0065] Step S110: For the educational target topic, calculate its similarity with each main topic in the target topic library, and add the educational target topic to the target topic library according to the strategy of similarity matching.

[0066] Specifically, considering the multi-source characteristics of data, there may be multiple different data on the same education target topic in the Internet source data. In order to effectively merge multi-source topics and reduce the amount of data for subsequent target topic identification, a multi-source information fusion technology is provided in this embodiment. By calculating the similarity between the education target topic acquired and screened this time and each main topic in the target topic library, if there is a target main topic in the library with a similarity greater than the set similarity threshold, the education target topic is merged with the target main topic. If yes, it means that there is no topic in the target topic library that is the same or highly similar to the current education target topic, and the current education target topic can be added to the target topic library as a new main topic.

[0067] In a possible implementation, there are multiple ways to merge the educational goal topic with the target main topic. Several optional merging methods are provided below. Any one or more combinations thereof can be selected in actual applications:

[0068] 1) The educational goal topic can be merged into the sub-topic collection of the main goal topic.

[0069] Specifically, the target topic library can store historically confirmed educational target topics. The same educational target topic may have multiple source topics, one of which can be selected as the main topic, and the rest as sub-topics, added to the sub-topic set of the main topic. In this way, the target topic library can include several main topics, and different main topics represent different educational target topics. Each main topic corresponds to a sub-topic set, which can be empty or contain several sub-topics that belong to the same educational target topic as the main topic.

[0070] In this embodiment, when it is determined that the similarity between the current education target topic and a main topic exceeds a set similarity threshold, it can be determined that the two belong to the same topic, and thus the current education target topic can be merged into the sub-topic set of the target main topic.

[0071] Further optionally, after merging the current education target topic into the sub-topic set of the target main topic, the target main topic can be reset based on the target main topic and its sub-topic set, for example, by reselecting a topic with the highest completeness of description from the target main topic and its sub-topic set as the updated target main topic.

[0072] 2) You can use the educational goal topic to complete the element information of the main goal topic.

[0073] Specifically, different source data of the same topic may contain different element information. In order to effectively integrate multi-source information and ensure the integrity and consistency of topic information, in this embodiment, the current educational target topic can be used to complete the element information of the target main topic. For example, if the target main topic lacks time element information, and the current educational target topic contains time element information, the time element information can be used to complete the target main topic.

[0074] The above embodiments of the present application provide two optional ways of merging the educational target topic with the target main topic. In addition, other optional merging methods can also be used, which are not listed one by one in this embodiment.

[0075] Step S120: by using the configured macro model, the main topics in the target topic library are labeled with their corresponding classification sub-tags according to the set education target topic label system.

[0076] Specifically, the educational target topic can be further subdivided into multiple sub-label types to describe the educational target topic from different dimensions. In this embodiment, an educational target topic label system can be pre-constructed to unify the target topic classification standards.

[0077] The educational goal topic label system can be set according to actual business needs. This embodiment provides an optional component structure of the educational goal topic label system, which includes at least one of the following:

[0078] Target topic subcategories, topic levels, classification of involved places, classification of involved entities, etc.

[0079] In order to label the main topic in the target topic library more accurately, this embodiment uses the ability of the large model to perform the task of labeling the classification sub-labels to which the main topic belongs. In addition, since the main topic in the target topic library is the topic selected and merged through the above steps, the labeling workload is reduced, and the classification sub-labels to which the main topic belongs can be labeled quickly and accurately.

[0080] Further optionally, after completing the labeling work, the educational goal topics judged to be of a specific level (for example, high / medium level) can be presented on a large screen, thereby achieving the goals of comprehensive aggregation of educational goal topic data, generalized situation perception, and rapid topic assessment.

[0081] In some embodiments of the present application, the above step S110 is used to illustrate the process of calculating the similarity between the educational target topic and each main topic in the target topic library.

[0082] In this embodiment, an education topic semantic similarity model can be pre-trained. The model is trained with the task of identifying whether two education target topics are the same topic (that is, whether they are the main topic or the secondary topic). The trained education topic semantic similarity model is used to extract the text vector representation of the topic.

[0083] In the process of training the education topic semantic similarity model, a training set can be collected, which contains several pairs of main and sub-topics and several pairs of non-main and sub-topics. The training set is used to train the education topic semantic similarity model, so that the model can recognize whether the input paired topic data is the main and sub-topic, and optimize the model parameters through the back propagation algorithm based on the model recognition results and the actual results. In the process, the cross entropy loss function can be used as the optimization target to improve the classification ability of the model by minimizing the loss function. At the same time, the training process can further combine learning rate scheduling, optimizer selection, batch size adjustment, automatic mixed precision training, data loading optimization and other technical means to improve the efficiency and effect of training.

[0084] The educational topic semantic similarity model can adopt a multi-layer self-attention-based Transformer structure. During encoding, each word can pay attention to the information of all input words. This structure makes the output word vector information a context-based word vector.

[0085] After the training is completed, the education topic semantic similarity model can be used to extract the text vector representation M of the current education target topic (that is, the education target topic currently obtained and to be stored in the database) new , and the text vector representation M of each main topic in the target topic library oriThen calculate the similarity between the text vector representation of the current educational target topic and each main topic in the target topic library. The similarity can be expressed by the cosine similarity of the vector representation, such as:

[0086]

[0087] After calculating the similarity score between the current educational target topic and the text vector representation of each main topic, the similarity score can be used as the similarity score of the two topics.

[0088] For the current educational target topic and each main topic in the target topic library, the process of extracting the text vector representation of the topic using the educational topic semantic similarity model can be implemented in the following way.

[0089] Each topic generally includes a title and content (the content can be the main text of the topic, or it can be an abstract extracted from the main text of the topic, with the abstract serving as the content here).

[0090] In one possible implementation, the title and content of the topic can be spliced ​​together, the spliced ​​text can be converted into a word vector sequence, and sent to the education topic semantic similarity model for encoding processing to obtain the text vector representation of the topic.

[0091] In another possible implementation, considering that the input length of the education topic semantic similarity model is limited, the length of the text after the title and content of some topics are spliced ​​may exceed the input length limit of the model, affecting the accuracy of the final encoding result. To this end, in this embodiment, the title text and content text of the topic can be extracted separately, and then the title text and content text can be converted into word vector sequences respectively, and then sent to the education topic semantic similarity model for encoding, and the vector representation H of the title text can be obtained. t , and the vector representation H of the content text c .

[0092] Furthermore, the vector representation of the title text can be expressed as t , and the vector representation H of the content text c After fusion processing, we get the text vector representation M of the topic.

[0093] An optional fusion method is to represent the vector H of the title text of the topic t , and the vector representation H of the content text c Perform average pooling to obtain the first matrix, and perform an average operation on the columns of the first matrix to obtain the text vector representation M of the topic:

[0094]

[0095] Among them, H tirepresents the vector value corresponding to the i-th position in the vector representation of the title text, H ci The vector value corresponding to the i-th position in the vector representation of the content text, where n is the dimension of the vector representation.

[0096] In some possible implementations, the similarity between the current educational target topic and each main topic in the target topic library may further consider the label similarity scores of the two topics based on the similarity scores of the text vector representations of the two topics calculated in the above embodiment. Specifically:

[0097] For the current educational target topic, the large model can also be used to annotate the classification sub-tags under the educational target topic label system. On this basis, the classification sub-tags of the current educational target topic can be compared with the classification sub-tags of each main topic in the target topic library to obtain the label similarity score.

[0098] Based on the similarity score between the current education goal topic and the text vector representation of each main topic, and the label similarity score, the total similarity score between the current education goal topic and each main topic is calculated. The formula is as follows:

[0099] S total =S1×q1+ S2×q2

[0100] Among them, S total It represents the total similarity score of two topics, S1 represents the similarity score represented by the text vector of the two topics, S2 represents the label similarity score of the two topics, q1 and q2 are two weights used to balance the proportion of the two similarities in the total similarity, and the size can be set by the user.

[0101] In this embodiment, when considering the similarity of two topics, the text similarity of the topics and the similarity of the topic labels are considered at the same time, which can more accurately measure the similarity of the topics. Of course, it is understandable that due to the additional consideration of the similarity of the topic labels, the current education target topics to be stored need to be labeled with their classification sub-labels through a large model, which will increase the amount of calculation. In actual business applications, appropriate strategies can be selected to calculate the similarity of topics based on business accuracy and computing power.

[0102] In some embodiments of the present application, the aforementioned step S120 is described by configuring a large model and using a set education target topic labeling system to label the main topic in the target topic library with the classification sub-labels.

[0103] In order to improve the processing effect of the big model on the task of distinguishing educational target topics, in this embodiment, the education field target topic distinguishing task data can be used to perform fine-tuning SFT (Supervised Fine-Tuning) training on the general big model to obtain a fine-tuned special big model. The special big model can stably and accurately output the system labels corresponding to the educational target topics.

[0104] Then, through the configured large model, the process of labeling the classification sub-labels of the main topics in the target topic library according to the set education target topic label system may include:

[0105] The dedicated large model is called through the first prompt instruction prompt to instruct the model to determine whether the main topic in the target topic library belongs to the target topic according to the target topic judgment rules, and to mark the classification sub-label of the main topic belonging to the target topic with reference to the set education target topic label system.

[0106] Through the above-mentioned special large model, various classification sub-labels under the education goal topic label system can be stably and accurately labeled for the main topic.

[0107] Further optional, considering that in actual scenarios, users may need to dynamically adjust the target topic perception range at different times to adapt to the changing environment and enhance the ability to deal with complex topics. For example, users may set different focus topics for educational target topics every month to adjust different focuses according to recent dynamics. To this end, this embodiment provides a target topic perception range dynamic adjustment strategy, specifically:

[0108] Based on the above-mentioned use of the dedicated big model to mark the classification sub-labels to which the main topic belongs, the candidate main topics that the dedicated big model determines to belong to the target topic can be screened out. For the candidate main topics, the general big model can be called through the second prompt instruction prompt to instruct the model to refer to the configured key educational target topic to determine whether the candidate main topic is a key focus topic.

[0109] Specifically, the specialized large model after fine-tuning training is more adept at labeling the educational target topic. However, the topics that users focus on at different times may be different. In order to dynamically adjust the target topic perception range, this embodiment can use the existing capabilities of the general large model itself to design a suitable second prompt instruction prompt, so that the general large model can determine whether the candidate main topic is the topic that users focus on.

[0110] The second prompt instruction prompt can be flexibly filled with the educational target topic of focus configured by the user.

[0111] This embodiment provides a second prompt instruction prompt format template, which can be used to edit and generate the second prompt instruction prompt, and then call the general large model to execute the task. Specifically:

[0112] Get a second prompt instruction prompt format template, the template includes a second task instruction, a key focus topic slot and a candidate main topic slot, the second task instruction is used to instruct the model to refer to the key focus educational target topic configured in the key focus topic slot, and determine whether the candidate main topic in the candidate main topic slot is a key focus topic.

[0113] Further, the key focus educational target topic configured by the user is filled into the key focus topic slot, and the candidate main topic is filled into the candidate main topic slot to obtain a second prompt instruction prompt, and the second prompt instruction prompt is input into the general big model to obtain the result output by the general big model.

[0114] Further optionally, the user can also dynamically adjust the education goal topic label according to business needs, such as adding a new education goal topic label, and instructing the model to determine whether the candidate main topic belongs to the newly added education goal topic label through a second prompt instruction prompt, etc.

[0115] In this embodiment, through the parallel use of dedicated large models and general large models, on the basis of completing the basic labeling of the main topic with the classification sub-labels under the label system, it is also possible to determine whether the main topic is the topic that the user focuses on based on the educational target topics that the user focuses on at different stages, and flexibly adjust the perception range of the target topic to adapt to the ever-changing environment and enhance the ability to deal with complex topics.

[0116] Reference Figure 3 , which illustrates a process of educational objective thematic identification method.

[0117] For the acquired Internet source data, firstly, the configured education target topic classification model is used to classify and judge whether it belongs to the education field and whether it belongs to the target topic. If so, the education target topic is screened and the education target topic is merged into the target topic library.

[0118] When merging cases into the target topic library, this embodiment provides a multi-source information fusion strategy. Specifically, the text vector representation of the education target topic is extracted through the education topic semantic similarity model, and the similarity is calculated with the text vector representation of the main topic in the target topic library. If there is a target main topic whose similarity is greater than the set similarity threshold, the education target topic is merged with the target main topic.

[0119] For the main topic in the target topic library, label grading and classification are performed through the big model. Here, this embodiment adopts a dynamic adjustment strategy for the target topic perception range. Specifically, by fine-tuning the general big model with the special big model trained by the target topic discrimination task data in the education field, the main topic is marked with various sub-labels under the education target topic label system, including but not limited to: education target topic subcategory classification, topic level classification, involved place classification, involved subject classification, etc. Further, through the general big model, it is judged whether the main topic is the topic that the user has focused on recently. It also supports dynamic adjustment of the education target topic label, for example, adding an education target topic label, so as to judge whether the main topic is the dynamically adjusted education target topic label through the general big model.

[0120] For topics that are judged by large models to be of medium or high level, they can be presented on the system's large screen for staff to review and process.

[0121] The method provided in this embodiment selects educational target topics from massive Internet texts by training an educational target topic classification model, without relying on manual review, thereby improving information processing efficiency. Afterwards, different source topics are merged to complete the topic information, and a complete educational target topic label system is constructed to conduct topic analysis and evaluation by dynamically adjusting the perception range, thereby comprehensively improving the accuracy of generalized perception and recognition of educational target topics, while reducing dependence on manual work and improving recognition efficiency.

[0122] Multi-source information fusion technology determines the similarities between topics and makes judgments. It merges the topic information determined to be the same topic, effectively filters out duplicate information, reduces the complexity of subsequent classification and analysis, effectively integrates multi-source information, and ensures information integrity and consistency.

[0123] The technology of dynamic adjustment of perception range realizes dynamic adjustment of the perception range of target topics by using both dedicated and general large models, adapts to the ever-changing environment, enhances the ability to deal with complex topics, and better generalizes the perception and identification of complex situations of educational target topics in reality.

[0124] Finally, by constructing system labels for educational goal topics and combining them with topic level classification, the educational goal topics judged to be at the high school level are presented on a large screen in a systematic manner, providing a highly aggregated, intelligent and efficient educational goal topic situation awareness system platform, achieving the goals of comprehensive aggregation of educational goal topic data, generalized situation perception, and rapid topic assessment.

[0125] The following is a description of an educational topic identification device provided in an embodiment of the present application. The educational topic identification device described below and the educational topic identification method described above can be referenced to each other.

[0126] See also Figure 4, Figure 4 This is a schematic diagram of the structure of an educational topic identification device disclosed in an embodiment of the present application.

[0127] like Figure 4 As shown, the device may include:

[0128] The topic screening unit 11 is used to obtain Internet information source data, and screen out educational target topics from the Internet information source data through a configured educational target topic classification model, wherein the educational target topic classification model is trained using sample data labeled with educational target topic classification labels; the educational target topic classification model is a model with a smaller parameter scale than the large model;

[0129] The topic storage unit 12 is used to calculate the similarity between the educational target topic and each main topic in the target topic library, and if there is a target main topic whose similarity is greater than a set similarity threshold, merge the educational target topic with the target main topic, otherwise, add the educational target topic as a new main topic to the target topic library;

[0130] The labeling unit 13 is used to label the main topics in the target topic library with their corresponding classification sub-tags according to the set education target topic labeling system through the configured macro model.

[0131] In a possible implementation, the process of the topic storage unit merging the educational target topic with the target main topic includes:

[0132] Merging the educational target topic into the sub-topic set of the target main topic;

[0133] and / or,

[0134] The educational target topic is used to complete the element information of the target main topic.

[0135] In a possible implementation, the process of the topic storage unit calculating the similarity between the educational target topic and each main topic in the target topic database includes:

[0136] Utilizing the configured educational topic semantic similarity model, respectively extract the text vector representation of the educational target topic and each main topic in the target topic library, wherein the educational topic semantic similarity model is trained by taking the identification of whether two educational target topics are the same topic as a training task;

[0137] The similarity between the educational target topic and the text vector representation of each main topic in the target topic library is calculated.

[0138] In a possible implementation, the process of extracting the text vector representation of the education target topic and each main topic in the target topic library by the topic storage unit using the configured education topic semantic similarity model includes:

[0139] For the educational target topic and each main topic in the target topic library:

[0140] Extract the title text and content text of the topic, convert the title text and content text of the topic into word vector sequences respectively, and send them into the education topic semantic similarity model for encoding respectively, to obtain the vector representation of the title text of the topic and the vector representation of the content text;

[0141] Average pooling is performed on the vector representation of the title text of the topic and the vector representation of the content text to obtain a first matrix, and an average operation is performed on the columns of the first matrix to obtain the text vector representation of the topic.

[0142] In a possible implementation, the process of the topic storage unit calculating the similarity between the educational target topic and each main topic in the target topic library further includes:

[0143] Comparing the classification sub-label of the educational target topic with the classification sub-label of each main topic in the target topic library to obtain a label similarity score, wherein the classification sub-label of the educational target topic is obtained by large model annotation;

[0144] Based on the similarity scores of the text vector representations of the education goal topic and each main topic and the tag similarity scores, the total similarity score of the education goal topic and each main topic is calculated.

[0145] In a possible implementation, the big model includes a special big model that is fine-tuned on a general big model using target topic discrimination task data in the field of education; the labeling unit labels the main topic in the target topic library with the classification sub-labels according to the set education target topic label system through the configured big model, including:

[0146] The dedicated large model is called through the first prompt instruction prompt to instruct the model to determine whether the main topic in the target topic library belongs to the target topic according to the target topic judgment rule, and to mark the classification sub-tag of the main topic belonging to the target topic with reference to the set education target topic label system.

[0147] In a possible implementation, the macro model further includes a general macro model; the process in which the labeling unit labels the classification sub-labels belonging to the main theme in the target theme library according to the set education target theme label system through the configured macro model also includes:

[0148] For the candidate main topic determined by the special large model as belonging to the target topic, the general large model is further called through the second prompt instruction prompt to instruct the model to refer to the configured key educational target topic to determine whether the candidate main topic is the key focus topic.

[0149] In a possible implementation, the labeling unit calls the general large model through the second prompt instruction prompt to instruct the model to refer to the configured key focus educational target topic and determine whether the candidate main topic is a key focus topic, including:

[0150] Obtain a second prompt instruction prompt format template, the template including a second task instruction, a key focus topic slot and a candidate main topic slot, the second task instruction is used to instruct the model to refer to the key focus educational target topic configured in the key focus topic slot, and determine whether the candidate main topic in the candidate main topic slot is a key focus topic;

[0151] Fill the key focus educational target topic configured by the user into the key focus topic slot, fill the candidate main topic into the candidate main topic slot, obtain a second prompt instruction prompt, input the second prompt instruction prompt into the general big model, and obtain the result output by the general big model.

[0152] The present application also provides an electronic device in an embodiment. Figure 5 As shown, it shows a schematic diagram of the structure of an electronic device suitable for implementing the embodiment of the present application. The electronic device in the embodiment of the present application may include but is not limited to fixed terminals such as mobile phones, tablet computers, desktop computers, teaching large screens, etc. Figure 5 The electronic device shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0153] like Figure 5 As shown, the electronic device may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage device 608 into a random access memory (RAM) 603 to implement the educational topic identification method of the aforementioned embodiment of the present application. When the electronic device is powered on, various programs and data required for the operation of the electronic device are also stored in the RAM 603. The processing device 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0154] Typically, the following devices may be connected to the I / O interface 605: an input device 606 including, for example, a touch screen, a touch pad, a keyboard, a mouse, a camera, a microphone, an accelerometer, a gyroscope, etc.; an output device 607 including, for example, a liquid crystal display (LCD), a speaker, a vibrator, etc.; a storage device 608 including, for example, a memory card, a hard disk, etc.; and a communication device 609. The communication device 609 may allow the electronic device to communicate with other devices wirelessly or by wire to exchange data. Although Figure 5 An electronic device having various devices is shown, but it should be understood that it is not required to implement or possess all the devices shown. More or fewer devices may be implemented or possessed instead.

[0155] An embodiment of the present application also provides a computer program product including computer-readable instructions. When the computer-readable instructions are executed on an electronic device, the electronic device implements any educational topic identification method provided in the embodiment of the present application.

[0156] A computer-readable storage medium is also provided in an embodiment of the present application. The storage medium carries one or more computer programs. When the one or more computer programs are executed by an electronic device, the electronic device can implement any educational topic identification method provided in the embodiment of the present application.

[0157] It should also be noted that the device embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed over multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. In addition, in the drawings of the device embodiments provided by the present application, the connection relationship between the modules indicates that there is a communication connection between them, which may be specifically implemented as one or more communication buses or signal lines.

[0158] Through the description of the above implementation mode, the technicians in the field can clearly understand that the present application can be implemented by means of software plus necessary general hardware, and of course, it can also be implemented by special hardware including special integrated circuits, special CPUs, special memories, special components, etc. In general, all functions completed by computer programs can be easily implemented by corresponding hardware, and the specific hardware structure used to implement the same function can also be various, such as analog circuits, digital circuits or special circuits. However, for the present application, software program implementation is a better implementation mode in more cases. Based on such an understanding, the technical solution of the present application is essentially or the part that contributes to the prior art can be embodied in the form of a software product, which is stored in a readable storage medium, such as a computer floppy disk, a U disk, a mobile hard disk, a ROM, a RAM, a disk or an optical disk, etc., including a number of instructions to enable a computer device (which can be a personal computer, a training device, or a network device, etc.) to execute the methods described in each embodiment of the present application.

[0159] In the above embodiments, all or part of the embodiments may be implemented by software, hardware, firmware or any combination thereof. When implemented by software, all or part of the embodiments may be implemented in the form of a computer program product.

[0160] The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable devices. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from a website site, a computer, a training device, or a data center by wired (e.g., coaxial cable, optical fiber, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) mode to another website site, computer, training device, or data center. The computer-readable storage medium may be any available medium that a computer can store or a data storage device such as a training device, a data center, etc. that includes one or more available media integrations. The available medium may be a magnetic medium, (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium (e.g., a solid-state drive (SSD)), etc.

[0161] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The various embodiments can be combined as needed, and the same or similar parts can refer to each other.

Claims

1. A method for identifying educational topics, characterized in that: include: Obtaining Internet information source data, and filtering out education target topics from the Internet information source data by using a configured education target topic classification model, wherein the education target topic classification model is trained using sample data labeled with education target topic classification labels; the education target topic classification model is a model with a smaller parameter scale than a large model; For the educational target topic, calculate its similarity with each main topic in the target topic library. If there is a target main topic whose similarity is greater than the set similarity threshold, merge the educational target topic with the target main topic. Otherwise, add the educational target topic as a new main topic to the target topic library. Through the configured large model, the main topics in the target topic library are marked with the classification sub-tags according to the set education target topic label system.

2. The method according to claim 1, characterized in that The process of merging the educational objective topic with the objective main topic comprises: Merging the educational target topic into the sub-topic set of the target main topic; and / or, The educational target topic is used to complete the element information of the target main topic.

3. The method according to claim 1, characterized in that The process of calculating the similarity between the educational target topic and each main topic in the target topic library includes: Utilizing the configured educational topic semantic similarity model, respectively extract the text vector representation of the educational target topic and each main topic in the target topic library, wherein the educational topic semantic similarity model is trained by taking the identification of whether two educational target topics are the same topic as a training task; The similarity between the educational target topic and the text vector representation of each main topic in the target topic library is calculated.

4. The method according to claim 3, characterized in that The process of respectively extracting the text vector representation of the educational target topic and each main topic in the target topic library by using the configured educational topic semantic similarity model includes: For the educational target topic and each main topic in the target topic library: Extract the title text and content text of the topic, convert the title text and content text of the topic into word vector sequences respectively, and send them into the education topic semantic similarity model for encoding respectively, to obtain the vector representation of the title text of the topic and the vector representation of the content text; Average pooling is performed on the vector representation of the title text of the topic and the vector representation of the content text to obtain a first matrix, and an average operation is performed on the columns of the first matrix to obtain the text vector representation of the topic.

5. The method according to claim 3, characterized in that: The process of calculating the similarity between the educational target topic and each main topic in the target topic library also includes: Comparing the classification sub-label of the educational target topic with the classification sub-label of each main topic in the target topic library to obtain a label similarity score, wherein the classification sub-label of the educational target topic is obtained by large model annotation; Based on the similarity scores of the text vector representations of the education goal topic and each main topic and the tag similarity scores, the total similarity score of the education goal topic and each main topic is calculated.

6. The method according to claim 1, characterized in that The large model includes a special large model obtained by fine-tuning the general large model using target topic discrimination task data in the field of education; The process of labeling the main topics in the target topic library with the classification sub-tags according to the set education target topic label system through the configured large model includes: The dedicated large model is called through the first prompt instruction prompt to instruct the model to determine whether the main topic in the target topic library belongs to the target topic according to the target topic judgment rule, and to mark the classification sub-tag of the main topic belonging to the target topic with reference to the set education target topic label system.

7. The method according to claim 6, characterized in that The macromodel also includes a universal macromodel; The process of labeling the main topics in the target topic library with the classification sub-tags thereof according to the set education target topic label system through the configured large model also includes: For the candidate main topic determined by the special large model as belonging to the target topic, the general large model is further called through the second prompt instruction prompt to instruct the model to refer to the configured key educational target topic to determine whether the candidate main topic is the key focus topic.

8. The method according to claim 7, characterized in that The process of calling the general large model through the second prompt instruction prompt to instruct the model to refer to the configured key educational target topic to determine whether the candidate main topic is the key educational topic includes: Obtain a second prompt instruction prompt format template, the template including a second task instruction, a key focus topic slot and a candidate main topic slot, the second task instruction is used to instruct the model to refer to the key focus educational target topic configured in the key focus topic slot, and determine whether the candidate main topic in the candidate main topic slot is a key focus topic; Fill the key focus educational target topic configured by the user into the key focus topic slot, fill the candidate main topic into the candidate main topic slot, obtain a second prompt instruction prompt, input the second prompt instruction prompt into the general big model, and obtain the result output by the general big model.

9. An educational topic identification device, characterized in that: include: A topic screening unit, used to obtain Internet information source data, and screen out educational target topics from the Internet information source data through a configured educational target topic classification model, wherein the educational target topic classification model is trained using sample data labeled with educational target topic classification labels; the educational target topic classification model is a model with a smaller parameter scale than the large model; The topic storage unit is used to calculate the similarity between the educational target topic and each main topic in the target topic library, and if there is a target main topic whose similarity is greater than a set similarity threshold, the educational target topic is merged with the target main topic; otherwise, the educational target topic is added to the target topic library as a new main topic; The labeling unit is used to label the main topic in the target topic library with the classification sub-label according to the set education target topic label system through the configured large model.

10. An electronic device, characterized in that: include: Memory and processor; The memory is used to store programs; The processor is used to execute the program to implement the various steps of the educational topic identification method as described in any one of claims 1 to 8.

11. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the educational topic identification method as described in any one of claims 1 to 8 is implemented.

12. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, each step of the educational topic identification method as described in any one of claims 1 to 8 is implemented.

Images