Methods, apparatus, equipment and media for extracting entity information from citations

Abstract

The application relates to the technical field of deep learning and information processing, and proposes a citation information entity extraction method, device, equipment and medium, wherein the method comprises the following steps: acquiring sample data set of citation data; training a preset deep learning citation block model by using the sample data set to obtain a trained target citation block model; training a preset deep learning citation classification model by using citation block data output by the target citation block model to obtain a trained target citation classification model; acquiring original citation data to be processed, and processing the original citation data to be processed by using the target citation block model and the target citation classification model to extract citation information entities from the original citation data to be processed and obtain target citation entity item results; and outputting the target citation entity item results. According to the technical scheme, the accuracy and efficiency of citation information entity extraction are improved.

Description

[Technical Field]

[0001] This application relates to the fields of deep learning and information processing technology, and in particular to a method, apparatus, device and medium for extracting citation information entities. [Background Technology]

[0002] Citation information refers to the content used in a document to cite or reference other documents. It usually includes, but is not limited to, the citation relationship between documents, co-authorship of authors, and mutual citation of journals. It is of great significance for scientific research management, academic evaluation, and discipline development.

[0003] Citation entity extraction is an important branch of information processing. It refers to the process of extracting key entities, such as document titles, authors, and journals, from citation information. These entities are crucial components of citation information and have numerous applications in academic citation analysis, academic level assessment, knowledge discovery, and literature recommendation. However, citation entity extraction currently faces many challenges, such as the diverse ways citation information is expressed, inaccurate or incomplete citation formats and content, and different styles of expression used for different document types, significantly increasing the difficulty of citation entity extraction. [Summary of the Invention]

[0004] This application provides a method, apparatus, device, and medium for extracting citation information entities, aiming to solve the technical problems such as the difficulty of extracting citation information entities in related technologies.

[0005] In a first aspect, embodiments of this application provide a method for extracting citation information entities, including:

[0006] Obtain a sample dataset of citation data, wherein the sample dataset includes at least one original citation data and a corresponding citation entity item;

[0007] The pre-defined deep learning citation segmentation model is trained using a sample dataset to obtain the trained target citation segmentation model.

[0008] The target citation classification model is trained by using the citation block data output by the target citation block model.

[0009] The raw citation data to be processed is obtained and processed through the target citation segmentation model and the target citation classification model to extract citation information entities from the raw citation data to be processed, and obtain the target citation entity item results.

[0010] Output the target citation entity item results.

[0011] In one embodiment, optionally, before training a pre-defined deep learning citation segmentation model using a sample dataset, the method further includes:

[0012] The original citation data in the sample dataset are preprocessed to determine the target location of each citation entity item in the original citation data.

[0013] In one embodiment, optionally, preprocessing is performed on each original citation data in the sample dataset to determine the target location of each citation entity item in the original citation data, including:

[0014] Obtain the first position index value of each symbol in each original citation data and record it in the position array;

[0015] Remove the symbols from each of the original citation data and citation entity items to obtain the processed original citation data and processed citation entity items;

[0016] Obtain the second position index value of the processed citation entity item in the processed original citation data and record it in the data index, wherein the second position index value is used to characterize the starting position of the processed citation entity item in the processed original citation data;

[0017] The first position index value is compared with the second position index value in the data index. If the first position index value is less than the second position index value, the second position index value is incremented by 1 to determine the actual starting position of the citation entity item in the original citation data.

[0018] Sort the first position index values ​​in the position array, and subtract the array subscript from each first position index value in the position array to determine the third position index value of each symbol in the processed original citation data;

[0019] Get the string length of the processed cited entity item;

[0020] Based on the second position index value and the string length, determine the end position of the processed citation entity item in the processed original citation data;

[0021] Determine the first number among all third position index values ​​that falls within the interval between the start position and the end position;

[0022] Based on the ending position and the first number, determine the actual ending position of the citation entity item in the original citation data;

[0023] The target position of the cited entity is determined based on the actual start position and the actual end position.

[0024] In one embodiment, optionally, the preset deep learning citation segmentation model includes a neural network layer, a Bi-GRU layer, and a CRF layer arranged sequentially.

[0025] The neural network layer extracts original citation features of different lengths from each original citation data.

[0026] The original citation features of different lengths are semantically parsed using the Bi-GRU layer to obtain semantic parsing results;

[0027] The semantic parsing results are determined by context association using the CRF layer, and the original citation data is labeled with delimiters based on the context association determination results to obtain the segmented citation entity data.

[0028] In one embodiment, optionally, the pre-defined deep learning citation classification model includes: a target BERT model, a self-attention mechanism layer, and a CRF layer; the method further includes:

[0029] The pre-defined BERT model is trained using the segmented citation entity data to obtain the target BERT model.

[0030] The pre-defined deep learning citation classification model is trained using the segmented citation entity data to obtain the target citation classification model.

[0031] In one embodiment, optionally, processing is performed using the target citation chunking model and the target citation classification model to extract citation information entities from the original citation data to be processed, obtaining target citation entity item results, including:

[0032] The raw citation data to be processed is preprocessed to obtain the target raw citation data;

[0033] The target original citation data is divided into blocks using the target citation block model to generate target citation block data.

[0034] The target citation data is classified and identified using the target citation classification model to extract target citation information entities and obtain target citation entity item results.

[0035] In one embodiment, optionally, the output of the target citation entity item result includes:

[0036] The results of the target citation entity items are verified according to the preset verification rules;

[0037] For the target citation entity items that pass the validation, the results are output and stored in JSON format.

[0038] Secondly, embodiments of this application provide a citation information entity extraction apparatus, comprising:

[0039] The acquisition module is used to acquire a sample dataset of citation data, wherein the sample dataset includes at least one original citation data and a corresponding citation entity item;

[0040] The first training module is used to train a pre-defined deep learning citation segmentation model using a sample dataset to obtain the trained target citation segmentation model.

[0041] The second training module is used to train a preset deep learning citation classification model using the citation block data output by the target citation block model, so as to obtain the trained target citation classification model.

[0042] An extraction module is used to acquire the raw citation data to be processed and process it through the target citation block model and the target citation classification model to extract citation information entities from the raw citation data to be processed and obtain the target citation entity item results.

[0043] The output module is used to output the target citation entity item results.

[0044] Thirdly, a computer 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 citation information entity extraction method.

[0045] Fourthly, a computer-readable storage medium is provided, which stores a computer program that, when executed by a processor, implements the steps of the above-described citation information entity extraction method.

[0046] In the above-described scheme for extracting citation information entities, the method, apparatus, device, and medium for citation information entity extraction involve: acquiring a sample dataset of citation data, wherein the sample dataset includes at least one original citation data and corresponding citation entity items; training a preset deep learning citation segmentation model using the sample dataset to obtain a trained target citation segmentation model; training a preset deep learning citation classification model using the citation segmentation data output by the target citation segmentation model to obtain a trained target citation classification model; acquiring the original citation data to be processed, and processing it using the target citation segmentation model and the target citation classification model to extract citation information entities from the original citation data to be processed, obtaining target citation entity item results; and outputting the target citation entity item results. In this invention, a target citation segmentation model is used to segment the original citation information into blocks, and then a target citation classification model is applied to classify and identify the segmented information. The combined use of these two models achieves the extraction of citation information entities, thereby improving the accuracy and efficiency of citation information entity extraction. [Attached Image Description]

[0047] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0048] Figure 1 A schematic flowchart of a citation information entity extraction method according to an embodiment of this application is shown.

[0049] Figure 2 A schematic flowchart of step S104 in a citation information entity extraction method according to an embodiment of this application is shown.

[0050] Figure 3 A block diagram of a citation information entity extraction apparatus according to an embodiment of this application is shown.

[0051] Figure 4 A schematic diagram of the structure of a computer device according to an embodiment of this application is shown.

[0052] Figure 5 Another structural schematic diagram of a computer device according to an embodiment of this application is shown.

Detailed Implementation Methods

[0053] To better understand the technical solution of this application, the embodiments of this application will be described in detail below with reference to the accompanying drawings.

[0054] It should be understood that the described embodiments are merely some, not all, of the embodiments in this application. All other embodiments obtained by those skilled in the art based on the embodiments in this application without inventive effort are within the scope of protection of this application.

[0055] The terminology used in the embodiments of this application is for the purpose of describing particular embodiments only and is not intended to be limiting of this application. The singular forms “a,” “the,” and “the” used in the embodiments of this application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0056] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0057] Please see Figure 1 , Figure 1 A schematic flowchart illustrating a citation information entity extraction method according to an embodiment of this application is shown. This citation information entity extraction method addresses the technical problem of the high difficulty in citation information entity extraction in related technologies.

[0058] like Figure 1 As shown, according to an embodiment of this application, a method for extracting citation information entities based on a short text classification task includes the following steps:

[0059] Step S101: Obtain a sample dataset of citation data, wherein the sample dataset includes at least one original citation data and a corresponding citation entity item;

[0060] Specifically, digital processing personnel extract entities from the citation information to be processed according to the processing needs. That is, they extract various citation entity information from a piece of original citation information, such as author, title of the document, source of the document, language, parent source, parent publication year, parent volume, parent publication period, parent page, parent publication place, parent publisher, edition description, thesis awarding institution, to be published, conference location, editor, and others. From this, they classify and select some data with complete structure and relatively comprehensive content as sample datasets.

[0061] In one embodiment, optionally, before training a pre-defined deep learning citation segmentation model using a sample dataset, the method further includes:

[0062] The original citation data in the sample dataset are preprocessed to determine the target location of each citation entity item in the original citation data.

[0063] Specifically, after obtaining the sample dataset, data analysis is required to match the positions of cited entity items with those in the original citation information. Unusable samples generated during processing are then filtered out, and commas, periods, colons, spaces, and other symbols present in the original citation are removed to prevent mismatches between the cited entities and the original citation. The specific algorithm is as follows:

[0064] In one embodiment, optionally, preprocessing is performed on each original citation data in the sample dataset to determine the target location of each citation entity item in the original citation data, including:

[0065] Obtain the first position index value of each symbol in each original citation data and record it in the position array; in a specific embodiment, the index value of the position of symbols such as comma, dot, colon, and space in the original citation can be obtained and recorded in the position array.

[0066] Remove symbols from each of the original citation data and citation entity items to obtain processed original citation data and processed citation entity items; in a specific embodiment, commas, dots, colons, spaces, and other symbols in the original citations and citation entity items can be removed.

[0067] The second position index value of the processed citation entity item in the processed original citation data is obtained and recorded in the data index. The second position index value is used to characterize the starting position of the processed citation entity item in the processed original citation data. In a specific embodiment, the second position index value of the citation entity item after removing symbols in the original citation after removing symbols is obtained and recorded in the variable start_index. At this time, the starting position of the citation entity item in the original citation after removing symbols is recorded. If start_index = -1, it means that the citation entity item cannot be correctly matched with the text content in the original citation and the sample should be discarded.

[0068] The first position index value is compared with the second position index value in the data index. If the first position index value is less than the second position index value, the second position index value is incremented by 1 to determine the actual starting position of the citation entity item in the original citation data. In a specific embodiment, the element value i in the position array is compared with start_index. If i is less than start_index, it means that the symbol is before start_index. In this case, start_index is incremented by 1 and moved one position to the right. At this time, the variable records the true starting position of the citation entity item in the original citation.

[0069] Sort each first position index value in the position array, and subtract the array subscript from each first position index value in the position array to determine the third position index value of each symbol in the processed original citation data; in a specific embodiment, sort the position array, subtract the element value i in the position array from its array subscript, and obtain the position index of each symbol in the original citation after removing the symbols.

[0070] Obtain the string length of the processed citation entity item;

[0071] Determine the end position of the processed citation entity item in the processed original citation data according to the second position index value and the string length; in a specific embodiment, obtain the string length of the citation entity item after removing the symbols, record it in the variable data_length, and define the variable end_index = start_index + data_length to record the end position of the citation entity item in the original citation.

[0072] Determine the first number in the range where the start position and the end position are located among all the third position index values;

[0073] Determine the actual end position of the citation entity item in the original citation data according to the end position and the first number; in a specific embodiment, sequentially determine whether the element value i in the position array is within the range (start_index, end_index). If start_index < i < end_index, then add 1 to end_index to move end_index one position backward. At this time, the variable records the actual end position of the citation entity item in the original citation.

[0074] Determine the target position of the citation entity item according to the actual start position and the actual end position. In a specific embodiment, obtain the content in the original citation whose index is within the range (start_index, end_index), which is the original appearance of the citation entity item content in the original citation content.

[0075] To better identify citation entities and avoid redundant content being classified as a particular entity, it is necessary to first identify the delimiters, find the correct delimiters, and segment the original citations into blocks. Then, the type of each entity is identified. This invention uses the PyTorch deep learning framework. The citation information entity extraction model consists of two main models: a citation segmentation model and a citation classification model. The citation segmentation model uses the Bi-GRU+CRF algorithm, and the citation classification model uses the Bert+CRF algorithm.

[0076] Before training the model, the preprocessed sample dataset can be converted into an error-free dataset, which is then divided into training, control, and test sets in an 8:1:1 ratio. The dataset is then populated using PyTorch interfaces, with the batch size and maximum text sequence truncation length set, and the dataset is packaged.

[0077] Step S102: The preset deep learning citation segmentation model is trained using the sample dataset to obtain the trained target citation segmentation model.

[0078] In one embodiment, optionally, the preset deep learning citation segmentation model includes a neural network layer, a Bi-GRU layer, and a CRF layer arranged sequentially.

[0079] The neural network layer extracts original citation features of different lengths from each original citation data.

[0080] The original citation features of different lengths are semantically parsed using the Bi-GRU layer to obtain semantic parsing results;

[0081] The semantic parsing results are determined by context association using the CRF layer, and the original citation data is labeled with delimiters based on the context association determination results to obtain the segmented citation entity data.

[0082] The target citation classification model annotates the delimiters in the error-free dataset. A pre-defined deep learning citation segmentation model is built using Bi-GRU and CRF algorithms. The model adaptively learns the delimiters between entities in the citation information. Specifically, to better enable the model to acquire citation entity information of different lengths, several convolutional neural networks (CNNs) are first established based on the feature distribution of citation information, such as six CNNs with kernel numbers of 1, 3, 5, 7, 9, and 11, and their corresponding input and output channels are set. A Bi-GRU layer follows the output layer of the CNNs. The Bi-GRU layer performs semantic parsing on the original citation features of different lengths to obtain semantic parsing results. A CRF layer follows the output layer of the Bi-GRU layer. The CRF layer performs context association determination on the semantic parsing results and annotates the delimiters on the original citation data based on the context association determination results, resulting in segmented citation entity data. In this way, the model learns the different data features of each entity.

[0083] In one specific embodiment, when training the target citation classification model, the cross-entropy loss function can be used to quantify the difference between the predicted value and the true value. The Adam optimizer is selected to iterate the model parameters a limited number of times to obtain a better model, thereby minimizing the value of the risk function. The F1 score is used as the evaluation index. After multiple iterations of training, the citation block model parameters are finally saved.

[0084] Step S103: Use the citation block data output by the target citation block model to train the preset deep learning citation classification model to obtain the trained target citation classification model.

[0085] In one embodiment, optionally, the pre-defined deep learning citation classification model includes: a target BERT model, a self-attention mechanism layer, and a CRF layer; the method further includes:

[0086] The pre-defined BERT model is trained using the segmented citation entity data to obtain the target BERT model. In one specific embodiment, the segmented citation entity data is input into the pre-defined BERT model for training. The cross-entropy loss function and Adam optimizer are used. Through multiple iterations of training, the trained target BERT model is saved, thereby enabling the BERT model to better learn the features of citation entity information.

[0087] The pre-defined deep learning citation classification model is trained using the segmented citation entity data to obtain the target citation classification model.

[0088] In one specific embodiment, a deep learning citation classification model is built by incorporating a self-attention mechanism and a CRF algorithm into a BERT pre-trained model. The trained BERT pre-trained model is then incorporating a self-attention layer and a CRF algorithm to construct the citation classification model. The model is then trained by inputting segmented citation data. Appropriate values ​​are configured for the `d_model` and `num_encoder_layers` parameters in the `attention_layer`, and the corresponding number of categories is set in the CRF. Multiple rounds of iterative training are performed, and finally, the model parameters are saved. The model adaptively learns the correspondence between citation entity items and their labels. Since the character length of each citation is different, PyTorch interfaces are used for paragraph padding. By classifying and recognizing each entity item, the purpose of citation information entity extraction is achieved.

[0089] Step S104: Obtain the raw citation data to be processed, and process it through the target citation block model and the target citation classification model to extract citation information entities from the raw citation data to be processed, and obtain the target citation entity item result;

[0090] like Figure 2 As shown, in one embodiment, optionally, step S104 includes:

[0091] Step S201: Preprocess the raw citation data to be processed to obtain the target raw citation data.

[0092] Specifically, the original citation can be simply formatted and its content identified using regular expressions, and some special characters can be removed.

[0093] Step S202: The target original citation data is divided into blocks using the target citation block model to generate target citation block data;

[0094] Step S203: Classify and identify the target citation block data using the target citation classification model to extract target citation information entities and obtain target citation entity item results.

[0095] Step S105: Output the target citation entity item result.

[0096] In one embodiment, optionally, step S105 includes:

[0097] The results of the target citation entity items are verified according to the preset verification rules;

[0098] For the target citation entity items that pass the validation, the results are output and stored in JSON format.

[0099] In the above embodiments, deep learning technology is used to extract citation information by segmenting and classifying it. Specifically, a citation segmentation model is constructed using a convolutional neural network combined with Bi-GRU and CRF algorithms to segment the original citation information. Then, a citation classification model is constructed using the BERT deep learning pre-trained model for semantic understanding, combined with attention mechanisms and CRF algorithms, to classify and identify the segmented information. By using the two models in tandem, citation information extraction is achieved, thereby improving the accuracy and efficiency of citation information entity extraction.

[0100] It should be understood that the sequence number of each step in the above embodiments does not imply the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present invention.

[0101] Figure 3 A block diagram of a citation information entity extraction apparatus according to an embodiment of this application is shown.

[0102] like Figure 3 As shown, in a second aspect, embodiments of this application provide a citation information entity extraction apparatus 30, comprising:

[0103] The acquisition module 31 is used to acquire a sample dataset of citation data, wherein the sample dataset includes at least one original citation data and a corresponding citation entity item;

[0104] The first training module 32 is used to train a preset deep learning citation segmentation model using a sample dataset to obtain the trained target citation segmentation model.

[0105] The second training module 33 is used to train a preset deep learning citation classification model using the citation block data output by the target citation block model, so as to obtain the trained target citation classification model.

[0106] Extraction module 34 is used to acquire the raw citation data to be processed, and process it through the target citation block model and the target citation classification model to extract citation information entities from the raw citation data to be processed, and obtain the target citation entity item result;

[0107] Output module 35 is used to output the target citation entity item result.

[0108] In one embodiment, optionally, before training a pre-defined deep learning citation segmentation model using a sample dataset, the apparatus further includes:

[0109] The preprocessing module is used to preprocess each original citation data in the sample dataset to determine the target position of each citation entity item in the original citation data.

[0110] In one embodiment, optionally, the preprocessing module includes:

[0111] The first processing unit is used to obtain the first position index value of each symbol in each original citation data and record it in the position array;

[0112] The second processing unit is used to remove symbols from each of the original citation data and citation entity items to obtain the processed original citation data and processed citation entity items.

[0113] The first acquisition unit is used to acquire the second position index value of the processed citation entity item in the processed original citation data and record it in the data index, wherein the second position index value is used to characterize the starting position of the processed citation entity item in the processed original citation data.

[0114] The comparison unit is used to compare the first position index value with the second position index value in the data index respectively. In response to the first position index value being less than the second position index value, the second position index value is incremented by 1 to determine the actual starting position of the citation entity item in the original citation data.

[0115] The sorting unit is used to sort the first position index values ​​in the position array and subtract the array subscript from each first position index value in the position array to determine the third position index value of each symbol in the processed original citation data.

[0116] The second acquisition unit is used to obtain the string length of the processed citation entity item;

[0117] The first determining unit is configured to determine the end position of the processed citation entity item in the processed original citation data based on the second position index value and the string length.

[0118] The second determining unit is used to determine the first number among all third position index values ​​that is in the interval between the starting position and the ending position;

[0119] The third determining unit is used to determine the actual end position of the citation entity item in the original citation data based on the end position and the first number;

[0120] The fourth determining unit is used to determine the target position of the cited entity item based on the actual start position and the actual end position.

[0121] In one embodiment, optionally, the preset deep learning citation segmentation model includes a neural network layer, a Bi-GRU layer, and a CRF layer arranged sequentially.

[0122] The neural network layer extracts original citation features of different lengths from each original citation data.

[0123] The original citation features of different lengths are semantically parsed using the Bi-GRU layer to obtain semantic parsing results;

[0124] The semantic parsing results are determined by context association using the CRF layer, and the original citation data is labeled with delimiters based on the context association determination results to obtain the segmented citation entity data.

[0125] In one embodiment, optionally, the pre-defined deep learning citation classification model includes: a target BERT model, a self-attention mechanism layer, and a CRF layer; the second training module includes:

[0126] The first training unit is used to train the preset BERT model using the segmented citation entity data to obtain the target BERT model.

[0127] The second training unit is used to train the preset deep learning citation classification model using the segmented citation entity data to obtain the target citation classification model.

[0128] In one embodiment, optionally, the extraction module includes:

[0129] The preprocessing unit is used to preprocess the raw citation data to be processed to obtain the target raw citation data;

[0130] The segmentation unit is used to segment the target original citation data into blocks using the target citation segmentation model to generate target citation segmentation data.

[0131] The classification unit is used to classify and identify the target citation block data through the target citation classification model, so as to extract the target citation information entities and obtain the target citation entity item results.

[0132] In one embodiment, optionally, the output module includes:

[0133] The verification unit is used to verify the result of the target citation entity item according to the preset verification rules;

[0134] The output unit is used to output and store the results of the validated target citation entity items in JSON format.

[0135] Specific limitations regarding the citation entity extraction device can be found in the limitations of the citation entity extraction method described above, and will not be repeated here. Each module in the aforementioned citation entity extraction device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the operations corresponding to each module.

[0136] In one embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a citation information entity extraction method on the server side.

[0137] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5 As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a citation information entity extraction method on the client side.

[0138] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0139] The computer device in this application embodiment exists in various forms, including but not limited to:

[0140] (1) Mobile communication devices: These devices are characterized by their mobile communication capabilities and primarily aim to provide voice and data communication. These terminals include: smartphones (e.g., iPhones), multimedia phones, feature phones, and low-end phones, etc.

[0141] (2) Ultra-mobile personal computer devices: These devices fall under the category of personal computers, possessing computing and processing capabilities, and generally also have mobile internet access features. These terminals include PDAs, MIDs, and UMPCs, such as the iPad.

[0142] (3) Portable entertainment devices: These devices can display and play multimedia content. This category includes audio and video players (such as iPods), handheld game consoles, e-book readers, as well as smart toys and portable car navigation devices.

[0143] (4) Server: A device that provides computing services. The components of a server include a processor, hard disk, memory, system bus, etc. Servers are similar to general computer architectures, but because they need to provide highly reliable services, they have higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0144] (5) Other electronic devices with data interaction functions.

[0145] In one embodiment, a computer 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 perform the following steps:

[0146] Obtain a sample dataset of citation data, wherein the sample dataset includes at least one original citation data and a corresponding citation entity item;

[0147] The pre-defined deep learning citation segmentation model is trained using a sample dataset to obtain the trained target citation segmentation model.

[0148] The target citation classification model is trained by using the citation block data output by the target citation block model.

[0149] The raw citation data to be processed is obtained and processed through the target citation segmentation model and the target citation classification model to extract citation information entities from the raw citation data to be processed, and obtain the target citation entity item results.

[0150] Output the target citation entity item results.

[0151] In one embodiment, optionally, before training a pre-defined deep learning citation segmentation model using a sample dataset, the method further includes:

[0152] The original citation data in the sample dataset are preprocessed to determine the target location of each citation entity item in the original citation data.

[0153] In one embodiment, optionally, preprocessing is performed on each original citation data in the sample dataset to determine the target location of each citation entity item in the original citation data, including:

[0154] Obtain the first position index value of each symbol in each original citation data and record it in the position array;

[0155] Remove the symbols from each of the original citation data and citation entity items to obtain the processed original citation data and processed citation entity items;

[0156] Obtain the second position index value of the processed citation entity item in the processed original citation data and record it in the data index, wherein the second position index value is used to characterize the starting position of the processed citation entity item in the processed original citation data;

[0157] The first position index value is compared with the second position index value in the data index. If the first position index value is less than the second position index value, the second position index value is incremented by 1 to determine the actual starting position of the citation entity item in the original citation data.

[0158] Sort the first position index values ​​in the position array, and subtract the array subscript from each first position index value in the position array to determine the third position index value of each symbol in the processed original citation data;

[0159] Get the string length of the processed cited entity item;

[0160] Based on the second position index value and the string length, determine the end position of the processed citation entity item in the processed original citation data;

[0161] Determine the first number among all third position index values ​​that falls within the interval between the start position and the end position;

[0162] Based on the ending position and the first number, determine the actual ending position of the citation entity item in the original citation data;

[0163] The target position of the cited entity is determined based on the actual start position and the actual end position.

[0164] In one embodiment, optionally, the preset deep learning citation segmentation model includes a neural network layer, a Bi-GRU layer, and a CRF layer arranged sequentially.

[0165] The neural network layer extracts original citation features of different lengths from each original citation data.

[0166] The original citation features of different lengths are semantically parsed using the Bi-GRU layer to obtain semantic parsing results;

[0167] The semantic parsing results are determined by context association using the CRF layer, and the original citation data is labeled with delimiters based on the context association determination results to obtain the segmented citation entity data.

[0168] In one embodiment, optionally, the pre-defined deep learning citation classification model includes: a target BERT model, a self-attention mechanism layer, and a CRF layer; the method further includes:

[0169] The pre-defined BERT model is trained using the segmented citation entity data to obtain the target BERT model.

[0170] The pre-defined deep learning citation classification model is trained using the segmented citation entity data to obtain the target citation classification model.

[0171] In one embodiment, optionally, processing is performed using the target citation chunking model and the target citation classification model to extract citation information entities from the original citation data to be processed, obtaining target citation entity item results, including:

[0172] The raw citation data to be processed is preprocessed to obtain the target raw citation data;

[0173] The target original citation data is divided into blocks using the target citation block model to generate target citation block data.

[0174] The target citation data is classified and identified using the target citation classification model to extract target citation information entities and obtain target citation entity item results.

[0175] In one embodiment, optionally, the output of the target citation entity item result includes:

[0176] The results of the target citation entity items are verified according to the preset verification rules;

[0177] For the target citation entity items that pass the validation, the results are output and stored in JSON format.

[0178] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or electronic device described above can be referred to the relevant descriptions in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0179] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0180] It should be understood that although the terms "first," "second," etc., may be used to describe the setting units in the embodiments of this application, these setting units should not be limited to these terms. These terms are only used to distinguish the setting units from each other. For example, without departing from the scope of the embodiments of this application, the first setting unit may also be referred to as the second setting unit, and similarly, the second setting unit may also be referred to as the first setting unit.

[0181] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0182] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the apparatus embodiments described above are merely illustrative; for instance, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection between apparatuses or units may be electrical, mechanical, or other forms.

[0183] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or in a combination of hardware and software functional units.

[0184] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in various forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0185] The above-described embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included within the protection scope of the present invention.

Claims

1. A method of citation information entity extraction, characterized by, The method comprises the following steps: obtaining a sample data set of citation data, wherein the sample data set comprises at least one original citation data and a corresponding citation entity item; training a preset deep learning citation segmentation model using the sample data set to obtain a trained target citation segmentation model; training a preset deep learning citation classification model using citation segmentation data output by the target citation segmentation model to obtain a trained target citation classification model; obtaining original citation data to be processed and processing the original citation data to be processed by the target citation segmentation model and the target citation classification model to extract citation information entities from the original citation data to be processed to obtain a target citation entity item result; outputting the target citation entity item result; before training the preset deep learning citation segmentation model using the sample data set, the method further comprises the following steps: preprocessing each original citation data in the sample data set to determine the target position of each citation entity item in the original citation data; the preprocessing each original citation data in the sample data set to determine the target position of each citation entity item in the original citation data comprises the following steps: obtaining a first position index value of each symbol in each original citation data and recording it in a position array; removing the symbols in the original citation data and the citation entity item to obtain processed original citation data and processed citation entity item; obtaining a second position index value of the processed citation entity item in the processed original citation data and recording it in a data index, wherein the second position index value is used to represent the starting position of the processed citation entity item in the processed original citation data; comparing the first position index value and the second position index value in the data index respectively, and in response to the first position index value being less than the second position index value, adding 1 to the second position index value to determine the actual starting position of the citation entity item in the original citation data; sorting the first position index values in the position array and subtracting the array subscript of each first position index value in the position array from the first position index value to determine the third position index value of each symbol in the processed original citation data; obtaining the string length of the processed citation entity item; determining the end position of the processed citation entity item in the processed original citation data according to the second position index value and the string length; determining the first number in the interval of the starting position and the end position among all third position index values; determining the actual end position of the citation entity item in the original citation data according to the end position and the first number; determining the target position of the citation entity item according to the actual starting position and the actual end position.

2. The method of claim 1, wherein, The preset deep learning citation segmentation model comprises a neural network layer, a Bi-GRU layer and a CRF layer arranged in sequence; obtaining original citation features of different lengths from each original citation data through the neural network layer; The Bi-GRU layer is configured to perform semantic analysis on the original citation features of different lengths to obtain a semantic analysis result. The CRF layer is configured to determine context association based on the semantic analysis result, and mark a segmentation point in the original citation data based on a context association determination result to obtain citation entity data after segmentation.

3. The method of claim 2, wherein, The preset deep learning citation classification model includes a target Bert model, a self-attention mechanism layer, and a CRF layer, and the method further includes: The target Bert model is obtained by training the preset Bert model using the citation entity data after segmentation. The target citation classification model is obtained by training the preset deep learning citation classification model using the citation entity data after segmentation.

4. The method of claim 1, wherein, The target citation entity item result is obtained by processing the target citation segmentation model and the target citation classification model to extract citation information entities from the original citation data to be processed. The target original citation data is obtained by preprocessing the original citation data to be processed. The target citation segmentation data is generated by segmenting the target original citation data using the target citation segmentation model. The target citation entity item result is obtained by classifying and identifying the target citation segmentation data using the target citation classification model to extract target citation information entities.

5. The method of claim 1, wherein, The target citation entity item result is output, including: The target citation entity item result is verified according to a preset verification rule. The target citation entity item result that passes the verification is output and stored in JSON format.

6. A citation information entity extraction apparatus characterized by comprising: The device includes: The acquisition module is configured to acquire a sample data set of citation data, wherein the sample data set includes at least one original citation data and a corresponding citation entity item. The first training module is configured to train a preset deep learning citation segmentation model using the sample data set to obtain a trained target citation segmentation model. The second training module is configured to train a preset deep learning citation classification model using citation segmentation data output by the target citation segmentation model to obtain a trained target citation classification model. The extraction module is configured to acquire original citation data to be processed and process the original citation data to be processed using the target citation segmentation model and the target citation classification model to extract citation information entities from the original citation data to be processed to obtain a target citation entity item result. The output module is configured to output the target citation entity item result. Before the sample data set is used to train the preset deep learning citation segmentation model, the device further includes: The preprocessing module is configured to preprocess each original citation data in the sample data set to determine a target position of each citation entity item in the original citation data. The preprocessing module includes: The first processing unit is configured to acquire a first position index value of each symbol in each original citation data and record the first position index value in a position array. The second processing unit is configured to remove symbols in the original citation data and the citation entity item to obtain processed original citation data and processed citation entity items. The first acquisition unit is configured to acquire a second position index value of the processed citation entity item in the processed original citation data and record the second position index value in a data index, wherein the second position index value is used to represent a starting position of the processed citation entity item in the processed original citation data. The comparison unit is configured to compare the first position index value with the second position index value in the data index, and in response to the first position index value being smaller than the second position index value, add 1 to the second position index value to determine an actual starting position of the citation entity item in the original citation data. The sorting unit is configured to sort the first position index values in the position array, and subtract the array index of each first position index value from the first position index value to determine a third position index value of each symbol in the processed original citation data. The second acquisition unit is configured to acquire a string length of the processed citation entity item. The first determination unit is configured to determine an ending position of the processed citation entity item in the processed original citation data according to the second position index value and the string length. The second determination unit is configured to determine a first number in all third position index values that is in an interval between the starting position and the ending position. The third determination unit is configured to determine an actual ending position of the citation entity item in the original citation data according to the ending position and the first number. The fourth determination unit is configured to determine a target position of the citation entity item according to the actual starting position and the actual ending position.

7. A computer device, comprising: The apparatus comprises: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are configured to perform the method of any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, The computer executable instructions are stored in the computer readable medium, and the computer executable instructions are used to perform the method of any one of claims 1 to 5.

Images