A tree structure-based multi-text information knowledge graph construction method

Abstract

The present application relates to the technical field of intelligent teaching, and more particularly to a knowledge graph construction method of multi-text information based on tree structure, comprising: S1. obtaining multi-text information for the target field to be constructed; constructing a tree structure of each text information; S2. grafting and fusing each constructed tree structure; S3. performing pruning operation on the grafted and fused tree structure; and S4. reconstructing the tree structure knowledge graph. The present application effectively realizes the knowledge graph construction of the tree structure of multi-e-book / multi-course video through the means of grafting and fusing first, pruning and merging second, and reconstructing and combing last, so that the final knowledge graph has wide applicability, strong universality, and can be widely shared on intelligent teaching platforms of various colleges and universities, and can be effectively used to realize knowledge sharing.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent teaching, and in particular to a knowledge graph construction method for multi-text information based on a tree structure. BACKGROUND

[0002] At present, the construction of a knowledge graph in an education scene is generally based on entity recognition and extraction of a triple relationship to construct a knowledge graph. Such a graph is generally a network-shaped knowledge graph. Whether based on entity recognition or a triple relationship to construct a knowledge graph, there are problems of huge artificial cost and unguaranteed accuracy.

[0003] In an actual application scene, the starting stage of education and teaching is to construct a tree structure graph. A main node corresponds to a chapter directory, and a sub-node corresponds to a learning knowledge point. Such an existing construction method is only for a book or a course video.

[0004] However, for a course or a subject, the construction of a knowledge graph is obviously insufficient only by relying on a book or a course video. Therefore, it is particularly important to construct a tree knowledge graph of multiple books and multiple course videos for each course in an intelligent teaching scene.

[0005] However, the existing construction of a tree knowledge graph only exists in a single book and a course video. There is currently no related prior art data for constructing a comprehensive tree knowledge graph of a course from multiple books / multiple course videos. The main reason is that it is difficult to divide the tree structure of multiple books / multiple courses, and simple merging is too bloated and complex.

[0006] Therefore, in order to construct a comprehensive tree knowledge graph of a course from multiple books / multiple course videos, the present application provides a knowledge graph construction method for multi-text information based on a tree structure. SUMMARY

[0007] The present application aims to provide a knowledge graph construction method for multi-text information based on a tree structure, which is used to solve the problem of tree structure fusion of multiple books.

[0008] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0009] A knowledge graph construction method for multi-text information based on a tree structure comprises the following steps:

[0010] S1. For a target field to be constructed, multi-text information is acquired;

[0011] A tree structure of each text information is constructed;

[0012] S2. grafting and fusing the constructed tree structures;

[0013] S3. pruning the grafted and fused tree structures;

[0014] S4. reconstructing the tree structure knowledge graph.

[0015] Further, the multi-text information at least includes: text information name, directory, and text content under the directory.

[0016] Further, the multi-text information includes text information of multiple e-books and / or text information of multiple course videos.

[0017] Further, the construction of the tree structure of each text information includes:

[0018] extracting the directory information of the text information, and establishing a tree structure with the directory information as the root node;

[0019] extracting the knowledge points included in the text content under each directory.

[0020] Further, the grafting and fusing of the constructed tree structures includes:

[0021] determining the common superior of each text information as the base node;

[0022] grafting the text information on the base node, and recording the text information name as the root node.

[0023] Further, the pruning of the grafted and fused tree structures includes:

[0024] S31. comparing the similarity of each root node, setting a first similarity threshold, and merging all root nodes with a similarity greater than or equal to the first similarity threshold;

[0025] S32. comparing the directory level nodes of all merged root nodes, setting a second similarity threshold, and merging all directory level nodes with a similarity greater than or equal to the second similarity threshold.

[0026] Further, the merging of all root nodes with a similarity greater than or equal to the first similarity threshold, the root node name after merging is the root node name with the least number of characters in each merged root node, and if there is more than one root node with the least number of characters, then one of them is selected as the root node name after merging.

[0027] Further, the merging of all the catalog level nodes with the similarity greater than or equal to the second similarity threshold value is performed, and in the merging, a catalog level node with the maximum number of node levels is selected as a base branch, and if there is more than one catalog level node with the maximum number of node levels, an optional catalog level node with the maximum number of node levels is selected as the base branch;

[0028] The other catalog level nodes are filled into the base branch.

[0029] Further, the reconstructed tree structure knowledge graph comprises:

[0030] S41. The root nodes and the catalog level nodes are modified.

[0031] S42. For the knowledge points under the merged base branch, it is determined again whether the knowledge points are related knowledge points of the base branch.

[0032] S43. The tail nodes are cleaned.

[0033] S44. The de-duplication operation is performed again.

[0034] Further, the determination again whether the knowledge points under the merged base branch are related knowledge points of the base branch further comprises:

[0035] If the knowledge points under the merged base branch are not necessary knowledge points of the base branch, the knowledge points are removed; otherwise, the knowledge points are retained.

[0036] The present application has at least the following advantages:

[0037] The present application effectively realizes the construction of the tree structure knowledge graph of multiple electronic books / multiple course videos through the means of grafting and fusing first, pruning and merging second, and reconstruction and carding last.

[0038] The present application uses the idea of "addition first, pruning second, and reconstruction last" to "trim" the complex tree structure knowledge graph, adopts a simple similarity comparison method to prune and merge the nodes of the tree one by one, and performs multiple reconstruction and modification, so that the final knowledge graph has wide applicability, strong universality, and can be widely shared on intelligent teaching platforms of various colleges and universities, and can be effectively used to realize knowledge sharing and break the independent barriers of course and knowledge point learning among colleges and universities. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings needed in the embodiment description will be briefly introduced. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor.

[0040] Figure 1 A schematic diagram of constructing a separate tree structure knowledge graph for a single book / video;

[0041] Figure 2 A schematic diagram of a certain embodiment of the grafting result;

[0042] Figure 3 A schematic diagram of the redundant branches of the tree structure knowledge graph and the pruning step;

[0043] Figure 4 A schematic diagram of a certain embodiment of the result after reconstruction of the tree structure knowledge graph. DETAILED DESCRIPTION

[0044] In order to make the purpose, technical scheme and advantages of the present application clearer, further detailed description will be made below in combination with the drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and do not limit the present application.

[0045] The main purpose of the present application is to prune and reconstruct the redundant knowledge point branches, so that the complex tree structure knowledge map constructed by multiple books can be more widely shared and applied.

[0046] The process of constructing a knowledge map available for multiple books or courses is as follows:

[0047] S1. Construct a tree structure knowledge map of a book or a course video:

[0048] In order to construct a complex "big tree", a "small sapling" must first be grown, so it is particularly important to construct a tree structure knowledge map. The process of constructing a tree structure knowledge map is described in detail in Figure 1 , which is as follows:

[0049] S11. Obtain text information of an e-book or text information of a course video;

[0050] The text information described herein includes at least directory information and text content under each directory.

[0051] S12. Extract the directory information and directory structure of the text information as the root node of the tree structure;

[0052] S13. Extract the knowledge points included in the text content under each directory.

[0053] As for how to extract knowledge points from text content, this technology is a relatively mature existing technology, and specific reference can be made to the following existing literature:

[0054] CN108334493B discloses an automatic knowledge point extraction method based on neural networks; CN104820724B discloses a method for obtaining a knowledge point prediction model for text-based educational resources and a method for applying the model. This invention does not target specific knowledge point extraction technologies; in implementing this invention, existing knowledge point extraction methods can be selectively used.

[0055] S2. Multiple knowledge maps are directly grafted together from the root node to form a complex tree-structured knowledge map:

[0056] For cases involving multiple books or courses, this invention first constructs separate tree knowledge maps for each book (the construction process is described in S1 above), and then performs simple grafting and fusion.

[0057] When grafting, it is necessary to determine the common superior features of the multiple books / courses, and denote them as base nodes.

[0058] Please see Figure 2 This diagram illustrates a simple example of grafting and merging. Among the various books on advanced mathematics shown are titles such as *Linear Algebra*, *Calculus*, *Discrete Functions*, ..., *Linear Algebra Second Edition*, etc. Each of these books is constructed using the tree structure described in S1; these tree structures are then grafted onto advanced mathematics. Advanced mathematics serves as the base node, while *Linear Algebra*, *Calculus*, *Discrete Functions*, ..., *Linear Algebra Second Edition*, etc., are all root nodes (first-level nodes).

[0059] S3. Locate and prune any excess branches;

[0060] For the knowledge map integrating multiple books / courses mentioned above, there are varying degrees of overlap in knowledge points across different versions and sub-fields of courses. Furthermore, the importance of different knowledge points varies across different versions. Therefore, identifying the similarities and differences, as well as redundant branches, of each sub-branch is crucial. This invention designs a method for deduplication that compares root nodes and child nodes, and also compares different root nodes. For details, please refer to [link / reference]. Figure 3 ,as follows:

[0061] S31. First, natural language processing technology is used to compare the similarity of each root node (first-level node), a first similarity threshold is set, and all root nodes with similarity greater than or equal to the first similarity threshold are merged.

[0062] For example, the books "Linear Algebra" and "Linear Algebra, Second Edition" can be merged.

[0063] The name of the merged root node can be the root node with the least number of words in the merged root nodes. For example, the names of Linear Algebra, Linear Algebra, Second Edition, and Linear Algebra, Third Edition are merged into Linear Algebra.

[0064] It is known that the similarity of comparison text using natural language processing technology is also a relatively mature existing technology, which can be referred to CN106776559B disclosed a method for calculating text semantic similarity; CN106776503B discloses a method for determining text semantic similarity; CN107562824B discloses a text similarity detection method; and the like. The innovation of the present application is not directed to the related technology of the present application. In the implementation of the present application, the existing technology can be selected.

[0065] In addition, the setting of the first similarity threshold can be adjusted according to the accuracy requirement in the specific implementation by those skilled in the art, and the present application does not make specific limitation, and it is known that any range of the similarity threshold should be within the protection scope of the present application.

[0066] S32. Compare all directory level nodes (including all second level nodes, third level nodes to the last level nodes) of each root node (first level node) that needs to be merged, and set a second similarity threshold. For all directory level nodes with a similarity greater than or equal to the second similarity threshold, merge them.

[0067] Here, when merging, the directory level node with the highest branch and depth (node level) is selected as the base branch, and other directory level nodes are filled into the base branch. If there are two or more directory level nodes with the maximum node level that need to be merged, one can be selected as the base branch.

[0068] S4. Reconstruct the tree structure knowledge graph.

[0069] After the above-mentioned merging branch operation of S3, the useful branches are merged into the associated base branch, and the redundant branches have been pruned, and the basic knowledge map is basically completed.

[0070] However, the directory node and the like need to be cleaned and reconstructed into a general name before being used in actual business, so the final reconstructed node is a crucial step, and the main work is as follows:

[0071] S41. Use Chinese text processing technology to clean and correct all directory name or section name nodes (i.e. directory level nodes). That is, according to the order of chapters, sections, etc., the nodes are renamed in order.

[0072] For example:

Chapter I

Chapter II

[0073]

Section I

Section II

[0074]

Principle

Overview

[0075] S42. For the newly merged knowledge points, determine again whether they are related knowledge points of the base branch, if not, clear, if yes, keep.

[0076] For example:

Matrix

Linear Equations

[0077] S43. The tail nodes of the tree structure map are cleaned.

[0078] For example:

Solution of Linear Equations

Linear Equations

[0079] S44. The modified tree structure is de-duplicated again.

[0080] For example:

IP Datagram

IP Multicast

IP

IP

[0081] Therefore, after the above steps, the tree structure knowledge map of the fusion of multiple books is basically completed, which meets the requirements of completeness, clarity, simplicity, generalization and the like, and can be shared as a resource and learning by different regions and students.

[0082] The final effect of a specific embodiment after completion is shown in Figure 4 .

[0083] As can be seen from the above:

[0084] In the field of education and teaching resource sharing, the tree structure knowledge map constructed by one book is often too simple and not widely applicable, and the tree structure knowledge map constructed by the fusion of multiple books and video resources is too complex and not convenient for application in actual scenarios. The present application ingeniously avoids the fusion idea in the map construction stage, and instead adopts a “first addition-pruning-then reconstruction” idea, which “prunes” the complex tree structure knowledge map into a clear and concise one. At the same time, a simple natural language processing technology (similarity comparison method) is adopted to compare and prune the nodes of the tree one by one, and the tree is reconstructed and modified in multiple ways. The final knowledge map is widely applicable and has strong knowledge point universality, and can be widely shared on intelligent teaching platforms in various colleges and universities, realizing true knowledge sharing and breaking down the barriers of independent courses and knowledge points learning among colleges and universities, and truly realizing “knowledge sharing”.

[0085] The above shows and describes the basic principles, main features and advantages of the present application. Those skilled in the art should understand that the present application is not limited to the above-mentioned embodiments, and the above-mentioned embodiments and descriptions in the specification are only the principles of the present application. Various changes and improvements can be made without departing from the spirit and scope of the present application, and these changes and improvements all fall within the scope of the claimed present application. The scope of protection of the present application is defined by the appended claims and their equivalents.

Claims

1.A method for constructing a knowledge graph of multi-text information based on a tree structure, characterized in that, The method comprises the following steps: S1. obtaining multi-text information for a target field to be constructed; constructing a tree structure for each of the text information; S2. grafting and fusing the constructed tree structures; S3. pruning the grafted and fused tree structure; S4. reconstructing a tree structure knowledge graph; The multi-text information at least comprises: text information name, directory, and text content under the directory; The multi-text information comprises text information of multiple e-books and / or text information of multiple course videos; The constructing of the tree structure for each of the text information comprises: extracting directory information of the text information and establishing a tree structure with the directory information as a root node; extracting knowledge points included in the text content under each directory; The grafting and fusing of the constructed tree structures comprises: determining a common superior of each text information as a base node; grafting the text information on the base node, with the text information name as a root node; The pruning of the grafted and fused tree structure comprises: S31. comparing the similarity of each root node, setting a first similarity threshold, and merging all root nodes with a similarity greater than or equal to the first similarity threshold; S32. comparing the directory level nodes of all merged root nodes, setting a second similarity threshold, and merging all directory level nodes with a similarity greater than or equal to the second similarity threshold. 2.The method of claim 1, wherein, The merging of all root nodes with a similarity greater than or equal to the first similarity threshold, the name of the merged root node is the name of the root node with the least number of characters in each merged root node, and if there is more than one root node with the least number of characters, one of them is selected as the name of the merged root node. 3.The method of claim 1, wherein, The merging of all directory level nodes with a similarity greater than or equal to the second similarity threshold, when merging, the directory level node with the maximum node level is selected as the base branch, and if there is more than one directory level node with the maximum node level, one of them is selected as the base branch; filling other directory level nodes to the base branch. 4.The method of claim 1, wherein, The reconstructing of the tree structure knowledge graph comprises: S41. correcting all root nodes and directory level nodes; S42. determining again whether the knowledge points under the merged base branch are related knowledge points of the base branch; S43. cleaning the tail nodes; S44. performing a de-duplication operation again. 5.The method of claim 4, wherein, The determining again whether the knowledge points under the merged base branch are related knowledge points of the base branch further comprises: if the knowledge points under the merged base branch are not necessary knowledge points of the base branch, they are removed; otherwise, they are retained.

Images