Intelligent knowledge learning method and apparatus

Abstract

The application provides a kind of intelligent knowledge learning method and device, it is related to artificial intelligence field technical field, method includes: based on pre-processing text and basic structure word library table, four-dimensional embedding vector is calculated as the input of encoding layer in transformer model to realize vocabulary enhancement;After the multiple iterations of encoding layer, the entity, attribute and relationship text classification task of pre-processing text is completed;The probability proportion of entity, attribute and relationship of pre-processing text is calculated, and the entity recognition of pre-processing text is completed;Entity, attribute and relationship are stored in graph storage structure neo4j;Iterative traversal neo4j to dynamically generate mind map.The application adopts word encoding and vocabulary encoding combination, uses relative position coding instead of actual position dimension coding, adds the dimension coding of commonly used structure word, enhances vocabulary, amplifies characteristic value, so that mind map can reflect the logical relationship between knowledge points, more suitable for learning.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence, and in particular to an intelligent knowledge learning method and apparatus. Background Technology

[0002] With the explosive growth of information today, the demands on human learning speed are increasing. Some technologies for knowledge learning propose using simple inverted indexes and BFS algorithms to generate tree nodes to display knowledge structures. This technology is essentially similar to a search engine; it can only extract key points based on word frequency and cannot truly cover all knowledge points or express the logical relationships between them. The knowledge structures displayed by these technologies are not suitable for learning a course or preparing for an exam, but are better suited for searching academic papers. Therefore, there is a need for an intelligent knowledge learning method that can automatically generate structured, logical, and colorful mind maps from input learning materials based on the learning patterns of the human brain, thereby improving learning efficiency. Summary of the Invention

[0003] In view of this, the present invention provides an intelligent knowledge learning method and apparatus to solve at least one of the aforementioned problems.

[0004] To achieve the above objectives, the present invention adopts the following solution:

[0005] According to a first aspect of the present invention, an intelligent knowledge learning method is provided, the method comprising: preprocessing a file to be learned to obtain preprocessed text; establishing a basic structure lexicon; calculating a four-dimensional embedding vector based on the preprocessed text and the basic structure lexicon as input to the encoding layer in a transformer model to achieve vocabulary enhancement, the four-dimensional embedding vector including character embedding vector, word embedding vector, structural word embedding vector, and relative position information embedding vector; completing the text classification task of entities, attributes, and relationships of the preprocessed text after multiple iterations of the encoding layer in the transformer model; calculating the probability proportions of entities, attributes, and relationships of the preprocessed text using the decoding layer in the transformer model to complete entity recognition of the preprocessed text; storing the entities, attributes, and relationships of the preprocessed text in a graph storage structure neo4j; iteratively traversing the data in the graph storage structure neo4j to dynamically generate mind map nodes, forming a knowledge learning mind map to be displayed to the user.

[0006] According to a second aspect of the present invention, an intelligent knowledge learning device is provided, the device comprising: a preprocessing unit for preprocessing a document to be learned to obtain preprocessed text; a structural lexicon building unit for building a basic structural lexicon table; a vocabulary enhancement unit for calculating a four-dimensional embedding vector based on the preprocessed text and the basic structural lexicon table as input to the encoding layer in a transformer model to achieve vocabulary enhancement, wherein the four-dimensional embedding vector includes character embedding vector, word embedding vector, structural word embedding vector, and relative position information embedding vector; a text classification unit for completing the text classification task of entities, attributes, and relationships of the preprocessed text after multiple iterations of the encoding layer in the transformer model; an entity recognition unit for calculating the probability proportions of entities, attributes, and relationships of the preprocessed text using the decoding layer in the transformer model to complete the entity recognition of the preprocessed text; a storage unit for storing the entities, attributes, and relationships of the preprocessed text in a graph storage structure neo4j; and a mind map generation unit for iteratively traversing the data in the graph storage structure neo4j to dynamically generate mind map nodes, forming a knowledge learning mind map to be displayed to the user.

[0007] According to a third aspect of the present invention, an electronic device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the above-described method.

[0008] According to a fourth aspect of the present invention, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the above-described method.

[0009] As can be seen from the above technical solution, the intelligent knowledge learning method and device provided by the present invention have improved the artificial intelligence algorithm, combined character encoding and vocabulary encoding, used relative position encoding to replace actual position dimension encoding, added dimension encoding of commonly used structural words, enhanced vocabulary, amplified feature values, so that the structure of automatically identified knowledge points can more accurately reflect the logical relationship between knowledge points, and is more suitable for users to learn. Attached Figure Description

[0010] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort. In the drawings:

[0011] Figure 1This is a flowchart illustrating an intelligent knowledge learning method provided in an embodiment of this application;

[0012] Figure 2 This is a schematic diagram of a mind map provided in an embodiment of this application;

[0013] Figure 3 This is a schematic diagram of another mind map provided in an embodiment of this application;

[0014] Figure 4 This is a schematic diagram of the structure of an intelligent knowledge learning device provided in an embodiment of this application;

[0015] Figure 5 This is a schematic block diagram of the system configuration of the electronic device provided in the embodiments of this application. Detailed Implementation

[0016] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the embodiments of the present invention will be further described in detail below with reference to the accompanying drawings. Here, the illustrative embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0017] Currently, knowledge learning methods only use simple inverted indexes and BFS algorithms to generate tree nodes to display knowledge structures. This technology can only extract key points based on word frequency and cannot truly cover all knowledge points or express the logical relationships between them. Therefore, this application proposes an intelligent knowledge learning method and device that can automatically generate structured and logical mind maps from input learning materials based on the learning patterns of the human brain, thereby improving learning efficiency.

[0018] like Figure 1 The diagram shown is a flowchart of an intelligent knowledge learning method provided in an embodiment of this application. The method includes the following steps:

[0019] Step S101: Preprocess the file to be studied to obtain preprocessed text.

[0020] The files to be studied can be, for example, question banks for exam review or files containing relevant knowledge obtained from the internet. The file format can be Word documents, PDF files, or images.

[0021] Preferably, the preprocessing in this step can be to convert the document to be studied into plain text using text extraction technology. For text files such as Word or PDF, text extraction technology in natural language processing can be used for text extraction, while for image files, image recognition technology such as Optical Character Recognition (OCR) can be used for text extraction.

[0022] Step S102: Establish the basic structural lexicon table.

[0023] The basic structural words here refer to commonly used terms related to knowledge organization, such as those that describe something as composed of, divided into layers, the first layer, containing, or having components, advantages, and disadvantages. These words can be given higher weights, leading to better entity recognition results through the subsequent attention mechanism of the transformer.

[0024] In this embodiment, the basic structural lexicon can be established by filtering existing vocabulary lists, or by manually annotating commonly used structural words to establish the basic structural lexicon. The basic structural lexicon can be supplemented later based on new structural words.

[0025] Step S103: Based on the preprocessed text and the basic structure lexicon, calculate a four-dimensional embedding vector as the input to the encoding layer in the transformer model to achieve vocabulary enhancement. The four-dimensional embedding vector includes character embedding vector, word embedding vector, structural word embedding vector, and relative position information embedding vector.

[0026] In this embodiment, entity recognition of preprocessed text is accomplished using the transformer model. The transformer model is a neural network architecture based on the attention mechanism, which is widely used in the field of natural language processing. It includes a word embedding layer, an encoding layer, and a decoding layer.

[0027] Due to the significantly more complex word segmentation characteristics in Chinese compared to English, and the substantial impact of segmentation on word meaning, neither character encoding nor word encoding alone can adequately represent semantic features. Therefore, this embodiment employs a four-dimensional embedding vector, comprising character embedding vectors, word embedding vectors, structural word embedding vectors, and relative position information embedding vectors, as input to the encoding layer to achieve lexical enhancement.

[0028] When calculating word embedding vectors, the preprocessed text can first be segmented into Chinese words using a tokenizer. Word vectors are then embedded using word2vec, and character vectors and structural word vectors are also embedded.

[0029] Currently, the positional information used in the transformer model is mostly absolute positional information. However, since the absolute positional information of characters or words is expressed the same in sentences of different lengths, it is severely limited by the length of the sentence. Therefore, this embodiment uses relative positional information for encoding.

[0030] Preferably, the relative position encoding method in this embodiment may include one of the following two methods: one is to use the head-tail pattern to represent the relative position, and the other is to represent the relative position by multiplying the transposed tail matrix and the head matrix by dot product.

[0031] Currently, when representing location information, the frequency of trigonometric function transformations is commonly used: PE represents location embedding encoding, which uses linear transformations of sin and cosine functions to process location information, using sin for odd numbers and cosine for even numbers. Choosing to alternate between odd and even values ​​for the trigonometric functions sin and cosine ensures that the values ​​are distributed between (-1, 1).

[0032]

[0033]

[0034] In the above formula, pos refers to the position of a word in the sentence, with a value range of [0, max_sequence_length), where max_sequence_length is the maximum length of the processed sequence; i refers to the dimension index of the word vector, with a value range of [0, embedding_dimension / 2), where embedding_dimension is the dimension size of the embedding vector, and d model This refers to the value of embedding_dimension.

[0035] The two methods of representing relative positions are described below:

[0036] (1) Head-tail pattern indicates relative position

[0037] Position is represented using the relative distance of dynamic word boxes, not absolute position; a dynamically adaptive word box encoded with relative position is used. Because the lengths of the dynamically adaptive word boxes vary, relative position encoding is used. For two input word boxes, X... i X j There are three types of relationships: overlapping, inclusion, and non-overlapping. These three relationships are determined by their beginning and end.

[0038] For this relationship, instead of using the ordinary transformer direct encoding mode, we learn image recognition and model it using dense vector matrices. We calculate the transformation of head and tail information, representing not only the relationship between two tags but also the distance between a character and a word. Let head[i] and tail[i] represent the positions of the head and tail of the dynamic word box. Four different relationship distances can represent X. i X j The relationship. See matrix r below for details. ij .

[0039]

[0040] Each element is a matrix, and it is a symmetric matrix.

[0041] The nonlinear transformation transformer is represented by combining the ReLU activation function with a weighting method that cascades relative position information:

[0042] r ij =ReLU(w r (concatenation(PE(r ij 1 ),PE(r ij 2 ),PE(r ij 3 ),PE(r ij 4 ))

[0043] Where concatenation means concatenation, linking the encoded relative position information together, w r Indicates the weight.

[0044] (2) The dot product of the transposed tail matrix and the head matrix represents the relative position.

[0045] Based on the above representation of trigonometric functions for positional information, the following formula is embedded at the character position:

[0046]

[0047]

[0048] The position embedding of the t-th character is:

[0049]

[0050] Where d is d model abbreviation, c i It is a constant determined by i, and equal to 1 / 10000. 2i / d .

[0051] According to the formula cos(xy)=sin(x)sin(y)+cos(x)cos(y), we can obtain:

[0052]

[0053]

[0054]

[0055] Preferably, in this embodiment, the character embedding vector and word embedding vector can be concatenated and then added to the structure word embedding vector and the relative position information embedding vector. The result is then used as the input to the encoding layer in the transformer model to achieve lexical enhancement. Thus, this embodiment processes the encoding using three dimensions: semantics (character embedding and word embedding), relative position, and structure word. Then, the semantic encoding, relative position encoding, and structure word encoding are added together, transforming the three dimensions into a single dimension as the input to the encoding layer.

[0056] Since many individual Chinese characters have their own meanings, combining character embedding and word embedding, and using multiple dimensions to describe the features of sentences, can enhance the vocabulary by using positional information to improve the relationship between words, helping machines to understand semantics more accurately. At the same time, it transforms the information from three dimensions into a single dimension, reducing workload and improving efficiency.

[0057] Step S104: After multiple iterations of the encoding layer in the transformer model, the text classification task of entities, attributes and relationships of the preprocessed text is completed.

[0058] In the Transformer model, the encoding layer consists of multiple identical encoders. The output of each encoder is used as the input of the next encoder for iteration. The specific number of iterations can be designed as needed, such as 10 iterations. The Transformer's encoder includes a multi-head attention mechanism sublayer and a fully connected feedforward sublayer. First, the multi-head attention mechanism is used to calculate the score, then it is passed through an FFN feedforward neural network, and finally the subsequent decoding layer performs entity recognition based on the score.

[0059] Preferably, this step may further include: after adding the weight calculation of the four-dimensional embedding vector input to the multi-head attention mechanism, performing a layer regularization; then, after processing by the feedforward neural network, adding it to the input that has not been processed by the feedforward neural network, then performing residual calculation, and then performing layer regularization again; after 10 iterations, the text classification task is completed.

[0060] Step S105: Calculate the probability proportions of entities, attributes, and relationships in the preprocessed text using the decoding layer in the transformer model, and complete the entity recognition of the preprocessed text.

[0061] Preferably, in the decoding layer, the softmax activation function can be used to calculate the probability proportions of entities, attributes, and relationships in the preprocessed text. By calculating the probability proportions, it can be determined whether the word belongs to an entity, an attribute, or a relationship. For example, if the probability of an entity is 70%, the probability of a relationship is 5%, and the probability of an attribute is 10%, then the word is determined to be an entity, thereby completing the entity recognition of the preprocessed text.

[0062] It should be noted that the relationships in this step can include relationships between entities and attributes, as well as relationships between entities.

[0063] Preferably, regarding the relationship between entities, this embodiment may further include the following steps:

[0064] In the decoding layer, the TransE translation model is used to calculate the relationship between entities and obtain a relationship score. The size of the relationship score can represent the distance between two entities in the mind map.

[0065] The translation model TransE maps entities and relationships to the same low-dimensional vector space, and represents the relationships between entities as translation operations between entity vectors.

[0066] The model defines the following scoring function for the triple (h,r,t):

[0067]

[0068] Where h and t are entities, and r is a relation.

[0069] L1 is the Manhattan distance, L1 = |X1-X2|+|Y1-Y2|;

[0070] L2 is the Euclidean distance.

[0071] Therefore, in the mind map generated later, entities with close relationships can be represented close together or connected in the middle, while important entities are represented in more important positions on the mind map.

[0072] Step S106: Store the entities, attributes, and relationships of the preprocessed text in the graph storage structure neo4j.

[0073] In this embodiment, entities or attributes are represented by being stored in graph nodes, and relationships are represented by connecting lines. The same entity is stored in the same graph node, and the same graph node is connected to other graph nodes by lines. Multiple connecting lines indicate that the entity has multiple relationships, can have multiple attributes, or can be related to multiple other entities.

[0074] Step S107: Iteratively traverse the data in the graph storage structure neo4j to dynamically generate mind map nodes, forming a knowledge learning mind map to be displayed to the user.

[0075] Preferably, when generating a knowledge learning mind map, structural words in the basic structural word library can be highlighted, for example, by using color display or bolding / blackening. When using color display, the color can be randomized or displayed according to the template corresponding to the structural word and color.

[0076] like Figure 2 and Figure 3 As shown, this is the mind map generated according to the method described above in this embodiment, wherein... Figure 2 Different shades of gray can represent different colors.

[0077] As can be seen from the above technical solution, the intelligent knowledge learning method provided by the present invention improves the artificial intelligence algorithm by combining character encoding and vocabulary encoding, using relative position encoding to replace actual position dimension encoding, and adding dimension encoding of commonly used structural words, thereby enhancing vocabulary and amplifying feature values. This makes the structure of automatically identified knowledge points more accurately reflect the logical relationship between knowledge points, making it more suitable for users to learn.

[0078] like Figure 4 The diagram shown is a structural schematic of an intelligent knowledge learning device provided in an embodiment of this application. The device includes: a preprocessing unit 410, a structural lexicon building unit 420, a vocabulary enhancement unit 430, a text classification unit 440, an entity recognition unit 450, a storage unit 460, and a mind map generation unit 470, which are connected in sequence.

[0079] in:

[0080] The preprocessing unit 410 is used to preprocess the file to be studied to obtain preprocessed text;

[0081] The structural lexicon building unit 420 is used to build the basic structural lexicon table;

[0082] The vocabulary enhancement unit 430 is used to calculate a four-dimensional embedding vector based on the preprocessed text and the basic structure lexicon table, which is used as the input of the encoding layer in the transformer model to achieve vocabulary enhancement. The four-dimensional embedding vector includes character embedding vector, word embedding vector, structural word embedding vector and relative position information embedding vector.

[0083] The text classification unit 440 is used to complete the text classification task of entities, attributes and relationships of the preprocessed text after multiple iterations of the encoding layer in the transformer model.

[0084] The entity recognition unit 450 is used to calculate the probability ratio of entities, attributes and relationships in the preprocessed text using the decoding layer in the transformer model, and to complete the entity recognition of the preprocessed text.

[0085] Storage unit 460 is used to store the entities, attributes and relationships of the preprocessed text in the graph storage structure neo4j;

[0086] The mind map generation unit 470 is used to iteratively traverse the data in the graph storage structure neo4j to dynamically generate mind map nodes, forming a knowledge learning mind map to be displayed to the user.

[0087] Preferably, the preprocessing unit 410 can be used to convert the document to be studied into preprocessed text in plain text form using text extraction technology.

[0088] Preferably, the vocabulary enhancement unit 430 can be specifically used to: concatenate the character embedding vector and the word embedding vector, add them to the structured word embedding vector and the relative position information embedding vector, and then use the calculation result as the input of the encoding layer in the transformer model to achieve vocabulary enhancement.

[0089] Preferably, the relative position information embedding vector adopts relative position encoding, and the relative position encoding method includes: using a head-tail pattern to represent the relative position, or representing the relative position by multiplying the transposed tail matrix and the head matrix by a dot product.

[0090] Preferably, the entity recognition unit 450 can be used to: calculate the probability proportions of entities, attributes and relationships in the preprocessed text using the activation function softmax.

[0091] Preferably, the text classification unit 440 can be used to: after the four-dimensional embedded vector input is fed into the weight calculation of the multi-head attention mechanism, perform a layer regularization; then after processing by the feedforward neural network, add it to the input that has not been processed by the feedforward neural network, then perform residual calculation, and then perform layer regularization again; after 10 iterations, the text classification task is completed.

[0092] Preferably, the aforementioned relationships include relationships between entities and attributes, as well as relationships between entities.

[0093] Preferably, the above-mentioned device may further include a calculation unit for: calculating the relationship between entities using the translation model TransE to obtain a relationship score, wherein the relationship score represents the distance between the two entities in the mind map.

[0094] For detailed descriptions of each of the above units, please refer to the corresponding descriptions in the foregoing method embodiments, which will not be repeated here.

[0095] As can be seen from the above technical solution, the intelligent knowledge learning device provided by the present invention has improved the artificial intelligence algorithm, combined character encoding and vocabulary encoding, used relative position encoding to replace actual position dimension encoding, added dimension encoding of commonly used structural words, enhanced vocabulary, amplified feature values, so that the structure of automatically identified knowledge points can more accurately reflect the logical relationship between knowledge points, and is more suitable for users to learn.

[0096] This invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the above-described method.

[0097] This invention also provides a computer-readable storage medium storing a computer program for performing the above-described methods.

[0098] like Figure 5 As shown, the electronic device 600 may also include: a communication module 110, an input unit 120, an audio processor 130, a display 160, and a power supply 170. It is worth noting that the electronic device 600 does not necessarily need to include these components. Figure 5 All components shown; in addition, the electronic device 600 may also include Figure 5 For components not shown, please refer to existing technologies.

[0099] like Figure 5 As shown, the central processing unit 100, sometimes also referred to as a controller or operating control, may include a microprocessor or other processor device and / or logic device. The central processing unit 100 receives inputs and controls the operation of various components of the electronic device 600.

[0100] The memory 140 may be, for example, one or more of a cache, flash memory, hard drive, removable media, volatile memory, non-volatile memory, or other suitable devices. It may store the aforementioned failure-related information, and also store a program for executing that information. The central processing unit 100 may execute the program stored in the memory 140 to perform information storage or processing, etc.

[0101] Input unit 120 provides input to central processing unit 100. Input unit 120 may be, for example, a keypad or touch input device. Power supply 170 provides power to electronic device 600. Display 160 displays images and text. Display may be, for example, an LCD display, but is not limited thereto.

[0102] The memory 140 can be a solid-state memory, such as a read-only memory (ROM), random access memory (RAM), a SIM card, etc. It can also be a memory that retains information even when power is off, can be selectively erased, and contains more data; examples of this type of memory are sometimes referred to as EPROMs. The memory 140 can also be some other type of device. The memory 140 includes a buffer memory 141 (sometimes referred to as a buffer). The memory 140 may include an application / function storage unit 142 for storing application programs and function programs or processes for executing the operation of the electronic device 600 via the central processing unit 100.

[0103] The memory 140 may also include a data storage unit 143 for storing data, such as contacts, digital data, pictures, sounds, and / or any other data used by the electronic device. The driver storage unit 144 of the memory 140 may include various drivers for the electronic device's communication functions and / or for performing other functions of the electronic device (such as messaging applications, address book applications, etc.).

[0104] The communication module 110 is a transmitter / receiver 110 that transmits and receives signals via antenna 111. The communication module (transmitter / receiver) 110 is coupled to the central processing unit 100 to provide input signals and receive output signals, which can be the same as in a conventional mobile communication terminal.

[0105] Based on different communication technologies, multiple communication modules 110 can be configured in the same electronic device, such as cellular network modules, Bluetooth modules, and / or wireless LAN modules. The communication module (transmitter / receiver) 110 is also coupled to a speaker 131 and a microphone 132 via an audio processor 130 to provide audio output via the speaker 131 and receive audio input from the microphone 132, thereby enabling typical telecommunications functions. The audio processor 130 may include any suitable buffer, decoder, amplifier, etc. Additionally, the audio processor 130 is coupled to a central processing unit 100, enabling on-device recording via the microphone 132 and on-device playback of stored audio via the speaker 131.

[0106] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0107] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0108] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0109] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0110] Specific embodiments have been used to illustrate the principles and implementation methods of this invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of this invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this invention. Therefore, the content of this specification should not be construed as a limitation of this invention.

Claims

1. An intelligent knowledge learning method, characterized in that, The method includes: The learning files are preprocessed to obtain preprocessed text; Establish a basic structural lexicon table; Based on the preprocessed text and the basic structure lexicon, a four-dimensional embedding vector is calculated as the input to the encoding layer in the transformer model to achieve vocabulary enhancement. The four-dimensional embedding vector includes character embedding vector, word embedding vector, structural word embedding vector, and relative position information embedding vector. After multiple iterations of the encoding layer in the transformer model, the text classification task of entities, attributes and relationships of the preprocessed text is completed; The probability proportions of entities, attributes, and relationships in the preprocessed text are calculated using the decoding layer in the transformer model to complete entity recognition of the preprocessed text. The entities, attributes, and relationships of the preprocessed text are stored in the graph storage structure neo4j; The data in the Neo4j graph storage structure is iteratively traversed to dynamically generate mind map nodes, forming a knowledge learning mind map to be displayed to the user.

2. The intelligent knowledge learning method as described in claim 1, characterized in that, The preprocessing of the file to be studied to obtain preprocessed text includes: using text extraction technology to convert the file to be studied into preprocessed text in plain text form.

3. The intelligent knowledge learning method as described in claim 1, characterized in that, The calculation of the four-dimensional embedding vector as input to the encoding layer in the transformer model to achieve vocabulary enhancement includes: concatenating the character embedding vector and the word embedding vector, adding them to the structured word embedding vector and the relative position information embedding vector, and then using the calculation result as input to the encoding layer in the transformer model to achieve vocabulary enhancement.

4. The intelligent knowledge learning method as described in claim 1, characterized in that, The relative position information embedding vector adopts relative position encoding, and the relative position encoding method includes: using a head-tail pattern to represent the relative position, or representing the relative position by multiplying the transposed tail matrix and the head matrix by a dot product.

5. The intelligent knowledge learning method as described in claim 1, characterized in that, The step of calculating the probability proportions of entities, attributes, and relationships in the preprocessed text using the decoding layer in the transformer model includes: using the activation function softmax to calculate the probability proportions of entities, attributes, and relationships in the preprocessed text.

6. The intelligent knowledge learning method as described in claim 1, characterized in that, The text classification task of entities, attributes, and relationships of the preprocessed text, completed after multiple iterations through the encoding layer in the transformer model, includes: After the four-dimensional embedded vector input is weighted by a multi-head attention mechanism, it undergoes a layer regularization. Then, after processing by a feedforward neural network, it is added to the input that has not undergone feedforward neural network processing. Next, residual calculation is performed, followed by another layer regularization. After 10 iterations, the text classification task is completed.

7. The intelligent knowledge learning method as described in claim 1, characterized in that, The relationships include those between entities and attributes, as well as those between entities.

8. The intelligent knowledge learning method as described in claim 7, characterized in that, The method further includes: using the translation model TransE to calculate the relationship between entities and obtain a relationship score, wherein the relationship score represents the distance between the two entities in the mind map.

9. An intelligent knowledge learning device, characterized in that, The device includes: The preprocessing unit is used to preprocess the files to be studied to obtain preprocessed text; The structural lexicon building unit is used to build the basic structural lexicon table; The vocabulary enhancement unit is used to calculate a four-dimensional embedding vector based on the preprocessed text and the basic structure lexicon table, which serves as the input to the encoding layer in the transformer model to achieve vocabulary enhancement. The four-dimensional embedding vector includes character embedding vector, word embedding vector, structural word embedding vector, and relative position information embedding vector. The text classification unit is used to complete the text classification task of entities, attributes and relationships of the preprocessed text after multiple iterations of the encoding layer in the transformer model. The entity recognition unit is used to calculate the probability proportions of entities, attributes, and relationships in the preprocessed text using the decoding layer in the transformer model, thereby completing the entity recognition of the preprocessed text. A storage unit is used to store the entities, attributes, and relationships of the preprocessed text in a graph storage structure neo4j; The mind map generation unit is used to iteratively traverse the data in the neo4j graph storage structure to dynamically generate mind map nodes, forming a knowledge learning mind map to be displayed to the user.

10. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 8.

11. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 8.

Images