A cell type acquisition method and system

Abstract

The application relates to a cell type acquisition method and system, comprising the following steps: obtaining a cell node relationship graph according to a training cell of a sample table; performing structure conversion processing on the cell node relationship graph to obtain a cell tree structure; training an initial graph attention network according to the cell tree structure to obtain a target graph attention network; fine-tuning parameters of an initial large language model according to the target graph attention network and a preset soft prompt vector template to obtain a target large language model; and inputting a target table of a to-be-recognized cell type into the target large language model to obtain a cell type of each target cell of the target table. The application can improve the accuracy of cell type prediction in a complex table.

Description

Technical Field

[0001] This application relates to the technical field of obtaining table cell types, and in particular to a method and system for obtaining cell types. Background Technology

[0002] In existing technologies, the complex relationship between the row and column structure of tables and cell types has not been fully modeled. This results in large models lacking accurate reasoning ability when handling complex structures such as nested tables or merged cells. Consequently, these tasks often require multiple manual interventions and verifications, leading to low processing efficiency. Summary of the Invention

[0003] Therefore, the purpose of this application is to provide a method and system for obtaining cell types, which can overcome the shortcomings of the prior art.

[0004] To achieve the above objectives, the technical solution adopted in this application is as follows:

[0005] The first embodiment of this application discloses a method for obtaining cell type, including: obtaining a cell node relationship graph based on the training cells of a sample table;

[0006] The cell node relationship graph is transformed to obtain a cell tree structure;

[0007] The initial graph attention network is trained based on the cell tree structure to obtain the target graph attention network;

[0008] The parameters of the initial large language model are fine-tuned based on the target graph attention network and the preset soft cue vector template to obtain the target large language model;

[0009] The target table whose cell type is to be identified is input into the target large language model to obtain the cell type of each target cell in the target table.

[0010] As one implementation method, the step of obtaining a cell node relationship graph based on the training cells of the sample table includes:

[0011] By using the training cells of the sample table as nodes, several cell nodes are obtained;

[0012] Based on the aforementioned number of cell nodes and adjacent nodes, several node connection relationships are obtained;

[0013] The cell node relationship diagram is obtained based on the plurality of cell nodes and the plurality of node connections.

[0014] In one implementation, the plurality of cell nodes includes a plurality of header nodes;

[0015] After obtaining several cell nodes by using the training cells of the sample table as nodes, the steps include:

[0016] The first row and / or first column cells of the sample table are used as header nodes to obtain the plurality of header nodes, and the plurality of data nodes other than the header nodes.

[0017] In one implementation, the plurality of node connection relationships include a plurality of first connection relationships;

[0018] The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes:

[0019] Obtain the node relationships between the plurality of data nodes and adjacent data nodes in the same row to obtain the plurality of first connection relationships.

[0020] In one implementation, the plurality of node connection relationships include a plurality of second connection relationships;

[0021] The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes:

[0022] Based on the node relationships between the plurality of data nodes and their adjacent data nodes in the same column, the plurality of second connection relationships are obtained.

[0023] In one implementation, the plurality of node connection relationships include a plurality of third connection relationships;

[0024] The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes:

[0025] Based on the node relationships between the header nodes and adjacent data nodes, the third connection relationships are obtained.

[0026] As one implementation method, the step of performing structural transformation processing on the cell node relationship graph to obtain a cell tree structure includes:

[0027] Obtain the row relationships in the cell node relationship graph to obtain multiple definition layers;

[0028] Based on all the cell nodes in each of the defined layers, generate virtual nodes;

[0029] The cell tree structure of each virtual node is obtained based on the row relationships in the cell node relationship graph corresponding to each virtual node.

[0030] As one implementation, the step of training an initial graph attention network based on the cell tree structure to obtain a target graph attention network includes:

[0031] Based on the relationships between cell nodes in the cell tree structure, the hidden representation of each cell node in the target graph attention network is updated, including:

[0032] The hidden representation is updated using the following formula:

[0033]

[0034] in, Let N(i) be the hidden representation of cell node i at level l+1, and let N(i) be the set of neighboring nodes of cell node i. For attention weights, W (l) For a trainable weight matrix, Let σ be the hidden representation of the neighboring node v at layer l, and σ be the activation function; N(i,j) represents node v. i,j The set of neighboring nodes of v; (p,q)∈N(i,j): represents all nodes that are related to v. i,j The adjacent cell node is located at row p and column q. This indicates that in the current layer l, from neighbor v p,q to target node v i,j Attention weights; At level l, neighbor node v p,q eigenvectors.

[0035] As one implementation method, the step of fine-tuning the parameters of the initial large language model based on the target graph attention network and a preset soft cue vector template to obtain the target large language model includes:

[0036] Based on the target graph attention network, a preset soft cue vector template is obtained; the soft cue vector template includes "the previous cell type information is [Prompt], the current node embedding is [v], and the predicted node type is [MASK]"; [Prompt] is a learnable virtual vector representing the hierarchical information of the virtual node; [v] is the embedding obtained through the target graph attention network; [MASK] is the part that the large language model needs to fill in the prediction, corresponding to the cell type category set {value, valueName, index, indexName};

[0037] When training the parameters of the initial large language model using the soft cue vector template, the parameters of the initial large language model are fine-tuned according to the cross-entropy loss function to obtain the target large language model, including:

[0038] The loss function result of the cross-entropy loss function is obtained through the following formula:

[0039]

[0040] Among them, L prompt The result of the loss function; y c ∈{0,1} is the one-hot encoding of the actual label; p c ([Prompt],v) is the LLM's predicted probability for class c, which depends on the soft cue virtual vector [Prompt] and the current node embedding v generated by the GAT network.

[0041] The second embodiment of this application discloses a cell type acquisition system, including:

[0042] The node relationship graph acquisition module is used to obtain the cell node relationship graph based on the training cells of the sample table;

[0043] The tree structure acquisition module is used to perform structural transformation processing on the cell node relationship diagram to obtain the cell tree structure;

[0044] The target graph attention network acquisition module is used to train an initial graph attention network based on the cell tree structure to obtain the target graph attention network.

[0045] The target large language model acquisition module is used to fine-tune the parameters of the initial large language model based on the target graph attention network and the preset soft cue vector template to obtain the target large language model;

[0046] The cell type acquisition module is used to input the target table whose cell type is to be identified into the target large language model to obtain the cell type of each target cell in the target table.

[0047] Compared with traditional technologies, the beneficial effects of this application are:

[0048] This application obtains a cell node relationship graph based on the training cells of a sample table; performs structural transformation on the cell node relationship graph to obtain a cell tree structure; trains an initial graph attention network based on the cell tree structure to obtain a target graph attention network; and fine-tunes the parameters of the initial large language model based on the target graph attention network and a preset soft cue vector template to obtain a target large language model that accurately predicts cell types. Then, the target table of cell types to be identified is input into the target large language model to obtain the cell types of each target cell in the target table. This can improve the accuracy of cell type prediction in complex tables, while reducing the reliance on manual intervention to achieve automated processing. By efficiently combining soft cue fine-tuning technology with the large language model, the adaptability of the model is improved, enabling it to cope with diverse table structures and exhibiting strong versatility.

[0049] To better understand and implement this application, the following detailed description is provided in conjunction with the accompanying drawings. Attached Figure Description

[0050] Figure 1 This is a flowchart of a cell type acquisition method according to an embodiment of this application;

[0051] Figure 2 This is a schematic diagram of a sample table according to one embodiment of this application;

[0052] Figure 3 This is a schematic diagram of the cell node relationship diagram according to an embodiment of this application;

[0053] Figure 4 This is a schematic diagram of a cell tree structure according to an embodiment of this application;

[0054] Figure 5 This is a flowchart illustrating the process of obtaining the current node embedding according to one embodiment of this application;

[0055] Figure 6 This is a flowchart illustrating an embodiment of the present application of training an initial large language model using soft cue vector templates.

[0056] Figure 7 This is a schematic diagram of the module connections of a cell type acquisition system according to an embodiment of this application.

[0057] 100. Cell type acquisition system; 101. Node relationship graph acquisition module; 102. Tree structure acquisition module; 103. Target graph attention network acquisition module; 104. Target large language model acquisition module; 105. Cell type acquisition module. Detailed Implementation

[0058] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings.

[0059] It should be understood that the described embodiments are merely some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of the embodiments of this application.

[0060] In the following description, when referring to the accompanying drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. In the description of this application, it should be understood that the terms "first," "second," "third," etc., are used only to distinguish similar objects and are not necessarily used to describe a specific order or sequence, nor should they be construed as indicating or implying relative importance. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances. The singular forms "a," "the," and "the" used in this application and the appended claims are also intended to include the plural forms, unless the context clearly indicates otherwise. The word "if" as used herein can be interpreted as "when," "when," or "in response to determination."

[0061] Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, or B alone. The character " / " generally indicates that the preceding and following related objects have an "or" relationship.

[0062] Please see Figure 1-6 The first embodiment of this application discloses a method for obtaining cell type, including:

[0063] S1: Obtain the cell node relationship graph based on the training cells of the sample table.

[0064] Please see Figure 2 The sample table and the individual training cells included in the sample table, such as Figure 2 As shown, the content of each training cell determines its cell type. For example, in the sample table, the cell type for "Year" is index name (indexName); the cell type for "Product A Sales Amount" is value name (valueName); and the cell type for "1 million" is value (value).

[0065] The cell type defined above is as follows:

[0066] value: Represents the actual data content in the cell;

[0067] valueName (value name): A descriptive name that summarizes the values ​​in the same column of data;

[0068] index: An identifier used to locate or sort data;

[0069] indexName(index_name): The name that describes the index.

[0070] By labeling each training cell with its cell type, the internal structure and semantic relationships of the table can be effectively captured, enabling subsequent data extraction, analysis, and information retrieval tasks to be performed automatically, efficiently, and accurately.

[0071] S2: Perform structural transformation on the cell node relationship diagram to obtain a cell tree structure;

[0072] S3: Train the initial graph attention network based on the cell tree structure to obtain the target graph attention network;

[0073] S4: Fine-tune the parameters of the initial large language model based on the target graph attention network and the preset soft cue vector template to obtain the target large language model;

[0074] S5: Input the target table of cell types to be identified into the target large language model to obtain the cell types of each target cell in the target table.

[0075] In a feasible embodiment, step S1: obtaining the cell node relationship graph based on the training cells of the sample table, includes:

[0076] S11: Using the training cells of the sample table as nodes, a number of cell nodes are obtained;

[0077] S12: Based on the aforementioned number of cell nodes and adjacent nodes, obtain several node connection relationships;

[0078] S13: Based on the plurality of cell nodes and the plurality of node connection relationships, the cell node relationship diagram is obtained.

[0079] In one feasible embodiment, the plurality of cell nodes includes a plurality of header nodes;

[0080] S11: After the step of obtaining several cell nodes by taking the training cells of the sample table as nodes, the following steps are included:

[0081] The first row and / or first column cells of the sample table are used as header nodes to obtain the plurality of header nodes, and the plurality of data nodes other than the header nodes.

[0082] In one feasible embodiment, the plurality of node connection relationships include a plurality of first connection relationships;

[0083] S12: The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes:

[0084] Obtain the node relationships between the plurality of data nodes and adjacent data nodes in the same row to obtain the plurality of first connection relationships.

[0085] The first connection relationship can be represented as:

[0086] For adjacent cells v in the same row ij With v i,j+1 Establish the edge (i,j)→(i,j+1).

[0087] In one feasible embodiment, the plurality of node connection relationships include a plurality of second connection relationships;

[0088] S12: The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes:

[0089] Based on the node relationships between the plurality of data nodes and their adjacent data nodes in the same column, the plurality of second connection relationships are obtained.

[0090] The second connection relationship can be represented as:

[0091] For adjacent cells v in the same column ij With v i+1,j Establish an edge (i,j)→(i+1,j).

[0092] In one feasible embodiment, the plurality of node connection relationships include a plurality of third connection relationships;

[0093] S12: The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes:

[0094] Based on the node relationships between the header nodes and adjacent data nodes, the third connection relationships are obtained.

[0095] The third connection can be represented as v ij v ij This indicates the connection between the header cell and the data cell.

[0096] In summary, the cell node relationship graph G = (V, E) includes the set of nodes. The set of connection relations E = {(v ij ,v i,j+1 )}∪{(v ij ,v i+1,j )}∪E header .

[0097] In a feasible embodiment, step S2: performing structural transformation processing on the cell node relationship graph to obtain a cell tree structure includes:

[0098] S21: Obtain the row relationships in the cell node relationship graph to obtain multiple definition layers.

[0099] Obtain the structural hierarchy (including logical levels such as rows or columns) in the cell node relationship graph, and group nodes with similar structures into the same definition layer.

[0100] S22: Generate virtual nodes based on all the cell nodes in each of the defined layers.

[0101] For each cell node {v ij ∈L k Construct virtual nodes satisfy:

[0102]

[0103] Among them, L k The k-th definition layer contains all cell nodes belonging to that layer.

[0104] v i,j ∈L k The cell node in row i and column j belongs to the k-th level.

[0105] x i,j Node v i,j The initial input features (which may include text embedding, position encoding, etc.)

[0106] The feature vector of the virtual node at layer k.

[0107] S23: Based on the row relationships in the cell node relationship diagram corresponding to each virtual node, obtain the cell tree structure of each virtual node.

[0108] In a feasible embodiment, the step of training an initial graph attention network based on the cell tree structure to obtain a target graph attention network includes:

[0109] Based on the relationships between cell nodes in the cell tree structure, the hidden representation of each cell node in the target graph attention network is updated, including:

[0110] The hidden representation is updated using the following formula:

[0111]

[0112] in, Let N(i) be the hidden representation of cell node i at level l+1, and let N(i) be the set of neighboring nodes of cell node i. For attention weights, W (l) For a trainable weight matrix, Let σ be the hidden representation of the neighboring node v at layer l, and σ be the activation function; N(i,j) represents node v. i,j The set of neighboring nodes of v; (p,q)∈N(i,j): represents all nodes that are related to v. i,j The adjacent cell node is located at row p and column q. This indicates that in the current layer l, from neighbor v p,q to target node v i,j Attention weights; At level l, neighbor node v p,q eigenvectors.

[0113] In a feasible embodiment, the step of fine-tuning the parameters of the initial large language model based on the target graph attention network and a preset soft cue vector template to obtain the target large language model includes:

[0114] Based on the target graph attention network, a preset soft cue vector template is obtained; the soft cue vector template includes "the previous cell type information is [Prompt], the current node embedding is [v], and the predicted node type is [MASK]"; [Prompt] is a learnable virtual vector representing the hierarchical information of the virtual node; [v] is the embedding obtained through the target graph attention network; [MASK] is the part that the large language model needs to fill in the prediction, corresponding to the cell type category set {value, valueName, index, indexName};

[0115] When training the parameters of the initial large language model using the soft cue vector template, the parameters of the initial large language model are fine-tuned according to the cross-entropy loss function to obtain the target large language model.

[0116] In a feasible embodiment, the step of fine-tuning the parameters of the initial large language model according to the cross-entropy loss function to obtain the target large language model when training the parameters of the initial large language model using the soft cue vector template includes:

[0117] The loss function result of the cross-entropy loss function is obtained through the following formula:

[0118]

[0119] Among them, L prompt The result of the loss function; y c ∈{0,1} is the one-hot encoding of the actual label; p c ([Prompt],v) is the LLM's predicted probability for class c, which depends on the soft cue virtual vector [Prompt] and the current node embedding v generated by the GAT network.

[0120] Compared with traditional technologies, the beneficial effects of this application are:

[0121] This application obtains a cell node relationship graph based on the training cells of a sample table; performs structural transformation on the cell node relationship graph to obtain a cell tree structure; trains an initial graph attention network based on the cell tree structure to obtain a target graph attention network; and fine-tunes the parameters of the initial large language model based on the target graph attention network and a preset soft cue vector template to obtain a target large language model that accurately predicts cell types. Then, the target table of cell types to be identified is input into the target large language model to obtain the cell types of each target cell in the target table. This can improve the accuracy of cell type prediction in complex tables, while reducing the reliance on manual intervention to achieve automated processing. By efficiently combining soft cue fine-tuning technology with the large language model, the adaptability of the model is improved, enabling it to cope with diverse table structures and exhibiting strong versatility.

[0122] Please see Figure 7 The second embodiment of this application discloses a cell type acquisition system 100, including:

[0123] The node relationship graph acquisition module 101 is used to obtain the cell node relationship graph based on the training cells of the sample table;

[0124] The tree structure acquisition module 102 is used to perform structural transformation processing on the cell node relationship diagram to obtain a cell tree structure.

[0125] The target graph attention network acquisition module 103 is used to train an initial graph attention network based on the cell tree structure to obtain the target graph attention network.

[0126] The target large language model acquisition module 104 is used to fine-tune the parameters of the initial large language model according to the target graph attention network and the preset soft cue vector template to obtain the target large language model;

[0127] The cell type acquisition module 105 is used to input the target table of cell types to be identified into the target large language model to obtain the cell type of each target cell of the target table.

[0128] It should be noted that the cell type acquisition system 100 provided in the second embodiment of this application is only illustrated by the above-described division of functional modules when executing the cell type acquisition method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above. In addition, the cell type acquisition system 100 provided in the second embodiment of this application and the cell type acquisition method in the first embodiment of this application belong to the same concept, and its implementation process is detailed in the method embodiment, which will not be repeated here.

[0129] The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this application according to actual needs. Those skilled in the art can understand and implement this without any inventive effort.

[0130] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application 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.

[0131] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. 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... Figure 1 One or more processes and / or boxes Figure 1 The 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 selected in one or more boxes.

[0132] 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 selected in one or more boxes.

[0133] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0134] Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0135] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0136] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0137] The above are merely embodiments of this application and are not intended to limit the scope of this application. Various modifications and variations can be made to this application by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the scope of the claims of this application.

Claims

1. A method for obtaining cell type, characterized in that, include: Based on the training cells of the sample table, a cell node relationship graph is obtained; The cell node relationship graph is transformed to obtain a cell tree structure; The initial graph attention network is trained based on the cell tree structure to obtain the target graph attention network; The parameters of the initial large language model are fine-tuned based on the target graph attention network and the preset soft cue vector template to obtain the target large language model; The target table whose cell type is to be identified is input into the target large language model to obtain the cell type of each target cell in the target table; The step of performing structural transformation processing on the cell node relationship graph to obtain a cell tree structure includes: Obtain the row relationships in the cell node relationship graph to obtain multiple definition layers; Obtain the structural hierarchy of the cell node relationship graph, including the logical hierarchy of rows or columns, and group nodes with similar structures into the same definition level; Based on all the cell nodes in each of the defined layers, generate virtual nodes; For each cell node in each layer, construct a virtual node. satisfy: ； in, The k-th definition layer contains all cell nodes belonging to that layer; The cell node in row i and column j belongs to the k-th level; :node The initial input features include text embedding and position encoding; : Feature vector of the k-th layer virtual node; Based on the row relationships in the cell node relationship diagram corresponding to each virtual node, the cell tree structure of each virtual node is obtained; The step of fine-tuning the parameters of the initial large language model based on the target graph attention network and the preset soft cue vector template to obtain the target large language model includes: Based on the target graph attention network, a preset soft cue vector template is obtained; the soft cue vector template includes "the previous cell type information is [Prompt], the current node embedding is [v], and the predicted node type is [MASK]"; [Prompt] is a learnable virtual vector representing the hierarchical information of the virtual node; [v] is the embedding obtained through the target graph attention network; [MASK] is the part that the large language model needs to fill in the prediction, corresponding to the cell type category set {value, valueName, index, indexName}.

2. The cell type acquisition method according to claim 1, characterized in that, The step of obtaining the cell node relationship graph based on the training cells of the sample table includes: By using the training cells of the sample table as nodes, several cell nodes are obtained; Based on the aforementioned number of cell nodes and adjacent nodes, several node connection relationships are obtained; The cell node relationship diagram is obtained based on the plurality of cell nodes and the plurality of node connections.

3. The cell type acquisition method according to claim 2, characterized in that, The plurality of cell nodes includes a plurality of header nodes; After obtaining several cell nodes by using the training cells of the sample table as nodes, the steps include: The first row and / or first column cells of the sample table are used as header nodes to obtain the plurality of header nodes, and the plurality of data nodes other than the header nodes.

4. The cell type acquisition method according to claim 3, characterized in that, The plurality of node connection relationships include a plurality of first connection relationships; The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes: Obtain the node relationships between the plurality of data nodes and adjacent data nodes in the same row to obtain the plurality of first connection relationships.

5. The cell type acquisition method according to claim 3, characterized in that, The plurality of node connection relationships include a plurality of second connection relationships; The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes: Based on the node relationships between the plurality of data nodes and their adjacent data nodes in the same column, the plurality of second connection relationships are obtained.

6. The cell type acquisition method according to claim 3, characterized in that, The node connection relationships include several third connection relationships; The step of obtaining the connection relationships of several nodes based on the several cell nodes and adjacent nodes includes: Based on the node relationships between the header nodes and adjacent data nodes, the third connection relationships are obtained.

7. The cell type acquisition method according to claim 1, characterized in that, The step of training an initial graph attention network based on the cell tree structure to obtain a target graph attention network includes: Based on the relationships between cell nodes in the cell tree structure, the hidden representation of each cell node in the target graph attention network is updated, including: The hidden representation is updated using the following formula: ； in, For a trainable weight matrix, For activation functions; : Represents a node The set of neighboring nodes; : indicates all with The adjacent cell node is located at row p and column q. : indicates the current number l In the floor, from the neighbor To the target node Attention weights; : No. l Layer-time neighbor nodes eigenvectors.

8. The cell type acquisition method according to claim 1, characterized in that, When training the parameters of the initial large language model using the soft cue vector template, the parameters of the initial large language model are fine-tuned according to the cross-entropy loss function to obtain the target large language model, including: The loss function result of the cross-entropy loss function is obtained through the following formula: ； in, The result of the loss function; , is the one-hot encoding of the actual label; The LLM predicts the probability of class c, which depends on the soft cue virtual vector. , and the current node embedding v generated by the GAT network.

9. A cell type retrieval system, characterized in that, include: The node relationship graph acquisition module is used to obtain the cell node relationship graph based on the training cells of the sample table; The tree structure acquisition module is used to perform structural transformation processing on the cell node relationship diagram to obtain the cell tree structure; The target graph attention network acquisition module is used to train an initial graph attention network based on the cell tree structure to obtain the target graph attention network. The target large language model acquisition module is used to fine-tune the parameters of the initial large language model based on the target graph attention network and the preset soft cue vector template to obtain the target large language model; The cell type acquisition module is used to input the target table whose cell type is to be identified into the target large language model to obtain the cell type of each target cell in the target table; The system is implemented using the cell type acquisition method as described in any one of claims 1-8.

Images