A method for automatically filling knowledge point information based on knowledge graph

Abstract

The present invention provides a method for automatically filling in knowledge point information based on a knowledge graph, which belongs to the field of information filling technology. It combines the structured information of textbooks to construct a knowledge point tree, and uses technical means such as textbook fragment query positioning, text summary extraction, and alias reverse query. It mainly automatically mounts the aliases, descriptions, and teaching materials and video resources of knowledge points (concepts or professional terms) in teaching scenarios, so as to improve the integrity of knowledge graph construction and accelerate the practical application of knowledge graph construction in educational and teaching scenarios.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge graph construction, and in particular to a method for automatically filling in knowledge point information based on a knowledge graph. Background Art

[0002] The industry's knowledge graph construction technology is still in its emerging stage. The triple knowledge graph constructed by pure machine learning lacks actual business value and the knowledge point's own information (alias, description, resource), which creates a considerable obstacle to the construction of the knowledge graph. The construction of the knowledge graph in related teaching scenarios mainly relies on the innovation of teacher content and cannot have universal value. The technology related to the automated construction of knowledge graphs fills in the missing knowledge point information. The construction of knowledge graphs is of great significance to education and teaching, and filling in the basic information of knowledge points is a time-consuming and labor-intensive process.

[0003] Patents CN116383413A and CN116383412A are innovations and technologies in knowledge nodes and knowledge point updates. Especially in educational and teaching scenarios, the basic information of knowledge points in the knowledge graph is of great significance to student learning and teacher lesson preparation, such as: the alias, description, and recommendation of related resources of knowledge points. The lack of basic knowledge points will reduce the application value of the knowledge graph in smart teaching.

[0004] The existing technologies for knowledge graph extraction include the following directions:

[0005] ① Existing technology patent: Patent number CN116383413A discloses a knowledge graph updating method and system based on medical data extraction. The present invention belongs to the field of data processing technology and provides a knowledge graph updating method and system based on medical data extraction. The method pre-classifies the admission and discharge record data obtained from the electronic medical record system, and performs keyword recognition and extraction on the pre-classified admission and discharge record data according to predetermined extraction rules to obtain medical keywords; according to the trained relationship extraction model, entity normalization and knowledge relationship extraction are performed to form ontology triple data to construct a medical knowledge graph; the attribute parameters of each medical entity are determined, and the formed ontology triple data is cleaned and deleted using the set cleaning rules; knowledge merging is performed based on the weight value of each determined medical entity relationship to update the constructed medical knowledge graph. The present invention can obtain a more accurate medical knowledge graph, can update the constructed medical knowledge graph, and can improve the efficiency of real data acquisition and data utilization.

[0006] ② Existing technology patent: Patent number CN116383412A provides a method and system for function point expansion based on knowledge graph. This invention belongs to the field of software analysis technology and provides a method and system for function point expansion based on knowledge graph. The method includes: extracting function point entities and knowledge relationships from existing function point analysis text to form function point triples to construct a function point knowledge graph. The function point triples include entity nodes corresponding to function point entities and unidirectional or bidirectional relationships between adjacent entity nodes. When receiving the text to be processed, the function points contained therein are identified to determine whether the function points are search keywords. When the function points are search keywords, the step of traversing the relationship path between directed entity nodes in the function point knowledge graph is executed. In the process of traversing the function point knowledge graph, a knowledge graph node queue is constructed. Function points are expanded according to the knowledge graph node queue. This invention realizes a more efficient automatic expansion process of function points and effectively avoids the problem of missing function points.

[0007] The key technology in the above-mentioned knowledge graph construction process is to construct a knowledge graph based on the relationship between knowledge points, but there is no related technology for filling in knowledge points, which requires a lot of labor costs. This addresses the pain points of existing inventions.

[0008] Therefore, it is necessary to provide a method for automatically filling in knowledge point information based on the knowledge graph to improve the integrity of the knowledge graph construction and accelerate the practical application of knowledge graph construction in educational and teaching scenarios. Summary of the Invention

[0009] One embodiment of this specification provides a method for automatically filling in knowledge point information based on a knowledge graph, comprising the following steps: S1: structuring the source of knowledge information in a textbook and dividing it into a multi-level directory structure; S2: constructing a knowledge tree diagram through the directory structure;

[0010] S3: Locate the smallest granularity textbook fragment through knowledge point reverse query; S4: Extract fragment description through large-scale migration of local extractive model; S5: Combine textbook fragments and model to generate knowledge point aliases describing the query; S6: Calculate similarity through sentence vector matrix and mount the reverse selected textbook and MOOC fragments.

[0011] In some embodiments, in said S1, the textbook that is the source of the knowledge information is divided into chapters, sections, and content attributes, and in said S2, the knowledge tree diagram is divided into titles from general to detailed according to the knowledge points in the textbook.

[0012] In some embodiments, in S3, based on the knowledge point name and subject restrictions, a reverse query is performed to find the textbook chapter where the knowledge point is located.

[0013] In some embodiments, the S4 also includes the following steps: S41: sending the teaching material fragment obtained by reverse query to the text generation model, wherein the text generation model references the large model LLM; S42: obtaining the training sample through fine-tuning instructions and sending it to the large model LLM to obtain a trained text generation model; S43: the text generation model outputs at least one knowledge point fragment based on the input.

[0014] In some embodiments, in S5, an alias is extracted based on a reverse query of the precise segment in S43, wherein the precise segment is derived from a teaching material.

[0015] In some embodiments, in S42, the training samples include multiple sections of textbook texts, and knowledge point information is extracted based on the textbook texts.

[0016] In some embodiments, the video information in the MOOC is converted into text based on OCR recognition.

[0017] In some embodiments, the S6 also includes the following steps: S61: split the knowledge point description into sentences to obtain a description sentence list QL with a length of m, and split the structured knowledge point data set into sentences to obtain SL with a length of n; S62: use the pre-trained model BERT to obtain the sentence embedding vector matrix A (m*L) and matrix B (n*L) of QL and SL, where L is the dimension of the vector; S63: obtain the similarity matrix M (m*n) of the two sentence lists through matrix calculation, and the similarity calculation formula is as follows: cos (θ) = A·B / |A|·|B|S64: by setting the similarity threshold, select the sentence closest to the teaching material fragment and the description; S65: use the maximum approximate sentence of the teaching material or MOOC fragment to reversely locate the inspiration position of the teaching material fragment, and obtain the resource mounting of the teaching material and the video.

[0018] Compared with the existing technology, the method of automatically filling knowledge point information based on the knowledge graph provided in this specification has at least the following beneficial effects:

[0019] 1. Ability to mount teaching materials and teaching videos during the teaching process, facilitating the expansion of knowledge points or simultaneous explanation;

[0020] 2. Accurate classification and traceability of textbook knowledge points can be constructed through multiple models, which can achieve accurate indexing of knowledge points. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] This specification will be further described in the form of exemplary embodiments, which will be described in detail with reference to the accompanying drawings. These embodiments are not limiting, and in these embodiments, like numbers represent like structures, wherein:

[0022] Figure 1It is a schematic diagram of the working principle of the present invention according to some embodiments of this specification;

[0023] Figure 2 is a schematic diagram of a localized migration model architecture according to some embodiments of this specification;

[0024] Figure 3 It is a schematic diagram of the working principle shown in some embodiments of this specification. DETAILED DESCRIPTION

[0025] To more clearly illustrate the technical solutions of the embodiments of this specification, the following briefly describes the drawings required for describing the embodiments. Obviously, the drawings described below are merely examples or embodiments of this specification. Those skilled in the art can apply this specification to other similar scenarios based on these drawings without inventive effort. Unless otherwise apparent from the context or otherwise noted, the same reference numerals in the figures represent the same structure or operation.

[0026] It should be understood that the terms "system," "device," "unit," and / or "module" used herein are a method for distinguishing different components, elements, parts, portions, or assemblies at different levels. However, if other terms can achieve the same purpose, the terms may be replaced by other expressions.

[0027] As used in this specification and claims, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not refer to the singular but also include the plural. Generally speaking, the terms "comprises" and "include" only indicate the inclusion of the steps and elements specifically identified, and these steps and elements do not constitute an exclusive list. A method or apparatus may also include other steps or elements.

[0028] Flowcharts are used throughout this specification to illustrate the operations performed by systems according to embodiments of this specification. It should be understood that preceding or following operations do not necessarily need to be performed in exact order. Instead, the steps may be processed in reverse order or simultaneously. Furthermore, other operations may be added to these processes, or one or more operations may be removed from these processes.

[0029] Example:

[0030] First, we construct a tree structure diagram for the knowledge points of the teaching scenario, as follows:

[0031] 1. Structuring the textbook information that may be the source of knowledge points, using a chapter-section-content structure, to facilitate knowledge point location and query;

[0032] 2. Use the name of the knowledge point and the subject restrictions to reversely query the textbook chapter where the knowledge point is located, and record it. For example: if the hit chapter is marked as the theme, the hit section is marked as the sub-theme, the hit segmentation title is marked as knowledge, and the hit text content is marked as an attribute; if the knowledge point list is: [PLC Overview, PLC Characteristics, PLC Application, Flexible Expansion Capabilities], according to the reverse query of the textbook, it is obtained: the theme [PLC Overview], [PCL Characteristics, PLC Applications] are sub-themes of [PLC Overview], and [Flexible Expansion Capabilities] hits the attribute marked as [PCL Characteristics] in the text.

[0033] 3. Reverse the query structure based on the textbook structure and build a tree structure diagram to accurately fill in other information. For example, based on the tree structure, it is more convenient to locate the detailed page of the text content. The knowledge point [PLC characteristics] comes from the chapter [PLC Overview], and its attributes are [high scalability] and [high reliability]. The source of PLC characteristics can be accurately located based on the upper and lower nodes;

[0034] It is worth noting that the minimum granularity textbook segment is the smallest unit knowledge point segment, that is, it covers the introduction of an entire knowledge point. The boundaries between multiple knowledge points include but are not limited to distinctions by periods, chapters, sections, titles, etc. The minimum granularity is also an aspect of the accuracy reflected in the knowledge graph. The minimum granularity of textbook segments and video segments, and video OCR text segments should be consistent. The knowledge points are the content, summary, title or generalization of textbook segments and video segments, and video OCR text segments, and the text repetition, reverse query positioning similarity, etc. are the same.

[0035] It is worth noting that in the above reverse query structure, the retrieved teaching material fragments often contain a large amount of content, and the content is too long to be intuitively used as a description of the knowledge point. For this reason, this embodiment constructs a text generation model with a deep learning transform structure. By referring to the latest large model LLM, it obtains training samples (extractive samples, that is, the output is likely to come from the original data) by adjusting instructions and fine-tuning models. The input is the knowledge point itself and the positioned large paragraph of teaching material text, and the output is the output of the large model. The training obtains a localized extractive text generation model, which extracts key information from the long text fragment of the teaching material in the above figure, such as [high reliability, simple programming] and other characteristics (due to the innovation of the large model, the extraction result may be a rewrite of the original text). The definition of the text generation model is as follows: The generation model is a machine learning model that generates new data by learning the probability distribution of data. The localized migration model architecture is as follows: Figure 2After extracting the description of the knowledge point textbook fragment, since the current description is most likely derived from the precise fragment of the textbook itself, the alias is extracted through the reverse query rule, for example: XXX(), XXX is also known as, XXX is called, etc. to complete the alias information. At this point, the alias and description of the knowledge point are constructed; finally, the textbook resources and MOOC video resources of the resource library are reversely queried based on the knowledge point alias and description, and the textbook fragments and video playback fragments are selected for the recommended resources, where the MOOC video information is structured text data after OCR recognition. Since the description may not be derived from the textbook, there will be a certain amount of rewriting. In order to accurately select the textbook and video summary, the present invention proposes a matrix calculation reverse query method, the process is as follows:

[0036] 1. Split the knowledge point description into sentences to obtain a list of description sentences, query_list, with a length of m, and a sentence_list of the structured teaching material (video) dataset, with a length of n;

[0037] 2. Use the pre-trained model bert to obtain the sentence embedding vector matrix A(m*l) and matrix B(n*l) of query_list and sentence_list (the vectorized representation of the sentence output by the model), where l is the dimension of the vector;

[0038] 3. The similarity matrix of the two sentence lists is obtained through matrix calculation, M(m*n). The similarity calculation formula is as follows:

[0039] cos(θ)=A·B / |A|·|B|

[0040] 4. By setting a similarity threshold, select the sentence in the textbook fragment that is closest to the description, for example: threshold = 0.9, that is, the sentence with the matrix element greater than the threshold and the maximum similarity to sentence_list, as shown in Table 1 below: query1 sentence corresponds to textbook fragment sentence3, and query2 corresponds to sentence2.

[0041] Table 1 Similarity calculation between textbooks / video clips and knowledge point lists

[0042] cos sentence1 sentence2 sentence3 sentence4 query1 0.6 0.9 1.0 0.1 query2 0.1 0.9 0.3 0.2

[0043] 5. Use the most similar sentences where the textbook (MOOC) fragment is located: sentence3 and sentence3 to reversely locate the starting position of the textbook fragment to achieve the purpose of selecting textbook resources.

[0044] It is worth noting that the video teaching materials in this embodiment are not limited to MOOC videos. All existing public learning and education videos for public use can be processed in the above manner.

[0045] In summary, please refer to Figure 3 , which is a crucial link in the commercial application stage of knowledge graph construction. Based on structured textbook data and MOOC video data, it adopts information retrieval technology, tree diagram construction method, large model fine-tuning localization model (extractive description), matrix calculation selection of textbooks and rule query alias and other technical means to fill in the basic information of knowledge points in the teaching process, and constructs a mechanism for automatically filling in knowledge point aliases, descriptions, textbook clip mounting, and video clip mounting, creating conditions for the actual application of knowledge graphs in the field of student learning.

[0046] It should be noted that the MOOC videos are provided for illustrative purposes only and are not intended to limit the scope of this specification. A person of ordinary skill in the art would be able to make various modifications or variations based on the description of this specification, such as the publicly available professional technical videos. However, such modifications and variations would not deviate from the scope of this specification.

[0047] While the basic concepts have been described above, it will be apparent to those skilled in the art that the detailed disclosure is merely illustrative and does not limit this specification. Although not explicitly stated herein, various modifications, improvements, and revisions to this specification may be made by those skilled in the art. Such modifications, improvements, and revisions are suggested in this specification and remain within the spirit and scope of the exemplary embodiments of this specification.

[0048] This specification also uses specific terms to describe the embodiments of this specification. For example, "one embodiment," "an embodiment," and / or "some embodiments" refer to a feature, structure, or characteristic associated with at least one embodiment of this specification. Therefore, it should be emphasized and noted that references to "one embodiment," "an embodiment," or "an alternative embodiment" two or more times in different locations in this specification do not necessarily refer to the same embodiment. Furthermore, certain features, structures, or characteristics of one or more embodiments of this specification may be appropriately combined.

[0049] In addition, unless expressly stated in the claims, the order of the processing elements and sequences, the use of alphanumeric characters, or the use of other names described in this specification are not intended to limit the order of the processes and methods of this specification. Although the above disclosure discusses some of the invention embodiments currently considered useful through various examples, it should be understood that such details are for illustrative purposes only, and the appended claims are not limited to the disclosed embodiments. On the contrary, the claims are intended to cover all modifications and equivalent combinations that are consistent with the spirit and scope of the embodiments of this specification. For example, although the system components described above can be implemented by hardware devices, they can also be implemented only by software solutions, such as installing the described system on an existing server or mobile device.

[0050] Similarly, it should be noted that, in order to simplify the presentation of this specification and thus facilitate understanding of one or more embodiments of the invention, the foregoing descriptions of the embodiments of this specification sometimes combine multiple features into a single embodiment, figure, or description thereof. However, this disclosure method does not imply that the subject matter of this specification requires more features than those recited in the claims. In fact, an embodiment may have fewer features than all of the features of a single disclosed embodiment.

[0051] Finally, it should be understood that the embodiments described in this specification are intended only to illustrate the principles of the embodiments of this specification. Other variations may also fall within the scope of this specification. Therefore, by way of example and not limitation, alternative configurations of the embodiments of this specification may be considered consistent with the teachings of this specification. Accordingly, the embodiments of this specification are not limited to the embodiments explicitly described and illustrated in this specification.

Claims

1. A method for automatically filling in knowledge point information based on a knowledge graph, characterized in that: The following steps are involved: S1: Structuralize the sources of knowledge information in textbooks and divide them into multi-level directory structures; S2: Construct a knowledge tree through the directory structure; S3: Locate the smallest granularity teaching material segment through reverse query of knowledge points; S4: Extracting snippet descriptions by large-scale transfer of local extractive models; S5: Combine the textbook fragment and the model to generate knowledge point aliases describing the query; S6: Calculate similarity through sentence vector matrix and mount reverse selected textbooks and MOOC segments; In S3, based on the knowledge point name and subject restrictions, the textbook chapter where the knowledge point is located is searched in reverse; The S4 further includes the following steps: S41: Send the reverse-queried textbook fragment to the text generation model, where the text generation model references the large model LLM; S42: Obtain training samples through fine-tuning instructions and send them to the large model LLM to obtain a trained text generation model; S43: The text generation model outputs at least one knowledge point segment based on the input; The S6 further comprises the following steps: S61: Split the knowledge point description into sentences to obtain a description sentence list QL with a length of m. Split the structured knowledge point dataset into sentences to obtain SL with a length of n. S62: Use the pre-trained model BERT to obtain the sentence embedding vector matrix A (m*L) and matrix B (n*L) of QL and SL, where L is the dimension of the vector; S63: The similarity matrix M (m*n) of the two sentence lists is obtained through matrix calculation. The similarity calculation formula is as follows: cos(θ) = A·B / |A|·|B| S64: by setting a similarity threshold, selecting the sentence closest to the teaching material segment and the description; S65: Use the maximum approximate sentence of the textbook or MOOC segment to reversely locate the inspiration position of the textbook segment, and obtain the resource mounting of the textbook and video.

2. The method for automatically filling in knowledge point information based on a knowledge graph according to claim 1, characterized in that: In said S1, the textbook serving as the source of the knowledge information is divided into chapters, sections, and contents by attributes. In said S2, the knowledge tree diagram is divided into titles from general to detailed according to the knowledge points in the textbook.

3. The method for automatically filling in knowledge point information based on a knowledge graph according to claim 1, characterized in that: In S5, an alias is extracted based on a reverse query of the precise segment in S43, wherein the precise segment comes from a teaching material.

4. The method for automatically filling in knowledge point information based on a knowledge graph according to claim 3, characterized in that: In the above-mentioned S42, the training samples include multiple sections of textbook texts, and knowledge point information is extracted based on the textbook texts.

5. The method for automatically filling in knowledge point information based on a knowledge graph according to claim 4, characterized in that: The video information in MOOCs is converted into text based on OCR recognition.

Images