Internet financial new entity identification method and device based on big data

Abstract

The invention relates to the technical field of natural language processing, and discloses an Internet finance new entity recognition method and device based on big data, and the method comprises the steps: segmenting obtained text comment data, and obtaining a text segmentation data set; performing semantic processing on each piece of text segmentation data in the text segmentation data set to obtain a global text semantic vector set; inputting each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set; performing entity boundary classification on each text in the text distributed expression tensor set in a distributed manner to obtain a text head and tail entity boundary set; and performing entity head-tail merging processing on each text head-tail entity boundary in the text head-tail entity boundary set through a greedy matching algorithm to obtain a target entity corresponding to the text comment data, so that the target entity in the text comment data can be accurately extracted.

Description

Technical Field

[0001] The present invention relates to the technical field of data processing and natural language processing, and in particular to a method and device for identifying new Internet financial entities based on big data. Background Art

[0002] With the continuous advancement and popularization of Internet technology, the Internet financial industry has risen rapidly, various new financial formats and financial innovation products have continued to emerge, and financial information has exploded. Pushing the identified new financial entities to investors and decision makers quickly and accurately can allow them to spend less time finding the most popular new entities in finance, so that they have more time to evaluate their development prospects. The establishment of a financial entity identification solution will greatly improve the efficiency of financial information acquisition, thereby better providing information support for relevant institutions and individuals in the financial field. Therefore, how to accurately extract entities in text has become a technical problem that needs to be solved urgently. Summary of the invention

[0003] The present invention provides a method and device for identifying new Internet financial entities based on big data to solve the technical problem of how to accurately extract entities in text.

[0004] In the first aspect, a method for identifying new Internet financial entities based on big data is provided, including:

[0005] Acquire text comment data, and segment the text comment data to obtain a text segmentation data set;

[0006] Input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set;

[0007] Input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set;

[0008] Each text in the text distributed expression tensor set is input into the neural network structure for entity boundary classification to obtain the text head and tail entity boundary set;

[0009] Each text head and tail entity boundary in the text head and tail entity boundary set is merged into a head and tail entity using a greedy matching algorithm to obtain the target entity corresponding to the text comment data.

[0010] In the second aspect, a new entity identification device for Internet finance based on big data is provided, comprising:

[0011] An acquisition module is used to acquire text comment data and segment the text comment data to obtain a text segmentation data set;

[0012] A first processing module is used to input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set;

[0013] A second processing module is used to input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set;

[0014] The third processing module is used to input each text in the text distributed expression tensor set into the neural network structure for entity boundary classification to obtain the text head and tail entity boundary set;

[0015] The fourth processing module is used to merge the head and tail of each text head and tail entity boundary in the text head and tail entity boundary set through a greedy matching algorithm to obtain a target entity corresponding to the text comment data.

[0016] In a third aspect, 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 implements the steps of the above-mentioned method for identifying new Internet financial entities based on big data when executing the computer program.

[0017] In a fourth aspect, a computer-readable storage medium is provided, which stores a computer program. When the computer program is executed by a processor, the steps of the above-mentioned Internet financial new entity identification method based on big data are implemented.

[0018] In the scheme implemented by the above-mentioned Internet financial new entity recognition method and device based on big data, text comment data is obtained and segmented to obtain a text segmentation data set, and each text segmentation data in the text segmentation data set is input into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, and each global text semantic vector in the global text semantic vector set is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, so that each text in the text distributed expression tensor set can be distributed input into the neural network structure for entity boundary classification to obtain a text head and tail entity boundary set, and then each text head and tail entity boundary in the text head and tail entity boundary set can be processed by a greedy matching algorithm to merge the entity head and tail to obtain the target entity corresponding to the text comment data, which is conducive to accurately extracting the target entity in the text comment data. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings required for use in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present invention. For ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without paying creative labor.

[0020] Figure 1 It is a schematic diagram of an application environment of a method for identifying new Internet financial entities based on big data in one embodiment of the present invention;

[0021] Figure 2 It is a flow chart of a method for identifying new Internet financial entities based on big data in one embodiment of the present invention;

[0022] Figure 3 It is a structural schematic diagram of a new entity identification device for Internet finance based on big data in one embodiment of the present invention;

[0023] Figure 4 is a schematic diagram of a structure of a computer device in one embodiment of the present invention;

[0024] Figure 5 It is another structural schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0026] The Internet financial new entity identification method based on big data provided by the embodiment of the present invention can be applied in the following aspects: Figure 1In an application environment, the client communicates with the server through a network. Exemplarily, taking the natural language processing scenario as an example, a user can upload text comment data through the client. Accordingly, the server can obtain text comment data through the client, and segment the text comment data to obtain a text segmentation data set, so as to input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing, obtain a global text semantic vector set, and input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis, and obtain a text distributed expression tensor set, so that each text in the text distributed expression tensor set can be distributed input into the neural network structure for entity boundary classification, and a text head and tail entity boundary set can be obtained, and then each text head and tail entity boundary in the text head and tail entity boundary set can be processed by a greedy matching algorithm to merge the head and tail of the entity, obtain the target entity corresponding to the text comment data, and feed the target entity back to the client. Accordingly, the client can accept the target entity corresponding to the text comment data from the server, and can display the target entity corresponding to the text comment data on the client for the target user to query or browse. By adopting the big data-based Internet financial new entity recognition method provided in this application, it is helpful to accurately extract the target entity in the text comment data.

[0027] The client may be, but is not limited to, various personal computers, laptops, smart phones, tablet computers, and portable wearable devices. The server may be implemented by an independent server or a server cluster consisting of multiple servers. The present invention is described in detail below through specific embodiments.

[0028] See also Figure 2 As shown, Figure 2 A flowchart of a method for identifying new Internet financial entities based on big data provided by an embodiment of the present invention includes the following steps:

[0029] S10: Acquire text comment data, and segment the text comment data to obtain a text segmentation data set.

[0030] Among them, the text comment data can be understood as text comment data from various Internet financial platforms, social media, forums and other channels. The text segmentation data set may include one or more text segmentation data, which can be used to represent the data obtained after the text comment data is segmented. The server can divide the text comment data into smaller text segments by segmenting it to obtain a text segmentation data set, which can facilitate subsequent processing, such as making the size of each text segmentation data more suitable for the input requirements of the model.

[0031] S20: Input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set.

[0032] Among them, the semantic model of the bidirectional encoding structure can be a model with strong semantic understanding ability such as Bidirectional Encoder Representations from Transformers (BERT), and this application does not limit this. Through semantic processing based on the semantic model of the bidirectional encoding structure, the server can capture the contextual information of the text and generate a global text semantic vector for each text segmentation data, thereby obtaining a global text semantic vector set containing the semantic features of the text, which provides a basis for subsequent analysis.

[0033] The global text semantic vector set may include one or more global text semantic vectors, and the global text semantic vector may be used to represent a vector obtained after inputting text segmentation data into a semantic model based on a bidirectional encoding structure for semantic processing.

[0034] S30: Input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set.

[0035] Among them, the module based on the bidirectional entity boundary recognition neural network structure can be understood as a module based on BiLSTM (Bidirectional Long Short-Term Memory) and entity boundary recognition neural network (Sampling-Embedding-Feedforward Neural Network, SE-FNN). BiLSTM can process the forward and reverse information of sequence data at the same time, so as to better capture contextual information, and can be applied to modeling the temporal dependency of text in natural language processing tasks. SE-FNN can be a complex neural network structure that combines special preprocessing modules, embedding layers and feedforward neural networks, which can perform effective feature extraction and expression analysis for specific tasks (such as new entity recognition in Internet finance).

[0036] The text distributed expression tensor set may include one or more text distributed expression tensors, which can be used to represent the tensor obtained after the global text semantic vector is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis.

[0037] By performing expression analysis based on a module based on a bidirectional entity boundary recognition neural network structure, a set of text distributed expression tensors can be obtained. Each tensor in the set can be understood as a new representation of the input text. It integrates global text semantic information and higher-level features extracted by the BiLSTM and SE-FNN structures, which is beneficial to various subsequent natural language processing tasks and helps to accurately identify target entities.

[0038] S40: Input each text in the text distributed expression tensor set into the neural network structure for entity boundary classification to obtain the text head and tail entity boundary set.

[0039] The neural network structure here can be a minimalist downstream network structure, that is, a relatively simple and lightweight neural network structure, the purpose of which is to quickly and accurately process specific tasks while maintaining high efficiency. The neural network structure can include fewer fully connected layers, simple convolutional layers and activation functions, etc., and can be designed as a low-parameter, low-computation network structure according to specific task requirements, and is usually used in combination with upstream complex models (such as semantic models based on bidirectional encoding structures) to fully utilize the powerful representation capabilities of the upstream model while maintaining the simplicity and efficiency of the downstream network.

[0040] The text start and end entity boundary set may include one or more text start and end entity boundaries, and the text start and end entity boundaries may be used to represent the entity boundaries obtained after the text is distributedly input into the neural network structure for entity boundary classification.

[0041] Optionally, after obtaining the text's head and tail entity boundary sets, the server can decode the entities in the text using the head and tail pointers, where the entity decoding rules can be as follows:

[0042] entity i =sentence[s i ∶e i ]

[0043] Among them, entity i represents the i-th entity; sentence represents the original sentence; s i Indicates the start of the predicted entity; e i Indicates the end of the predicted entity.

[0044] S50: Each text head and tail entity boundary in the text head and tail entity boundary set is subjected to head and tail merging processing by a greedy matching algorithm to obtain a target entity corresponding to the text comment data.

[0045] Among them, the greedy matching algorithm usually selects the local optimal solution and gradually merges the entity boundaries in the text to determine the final target entity, that is, the new Internet financial entity identified from a large amount of text comment data. The target entity can be used to indicate the entity included in the text comment data determined by the method provided in the embodiment of the present application.

[0046] In this example, the server uses a large amount of text comment data to cover a wider range of information to improve the accuracy and comprehensiveness of entity recognition. It also uses a semantic model based on a bidirectional encoding structure, BiLSTM, a feedforward neural network structure, and a minimalist downstream network structure to effectively extract the semantic and structural features of the text to improve the performance of entity recognition. It then uses a greedy matching algorithm to merge entities from head to tail, which can quickly and effectively determine the target entity, improve the efficiency of the algorithm, and facilitate the accurate extraction of the target entity in the text comment data.

[0047] In one embodiment, the server obtains text comment data and segments the text comment data to obtain a text segmentation data set, which may include the following steps:

[0048] A1: performing punctuation segmentation processing on long text comment data that is greater than a preset length in the text comment data to obtain a first length text data set;

[0049] A2: performing dynamic step-size segmentation processing on the first-length text data in the first-length text data set to obtain a target-length text data set;

[0050] A3: Perform sequence labeling on each target length text data in the target length text data set to obtain a text labeling data set;

[0051] A4: Perform word strength conversion on the text annotation data set to obtain a text segmentation data set.

[0052] The preset length may be a pre-set length threshold for distinguishing long text from short text. The preset length may be set by the user or by the system by default, and this application does not limit this. It is understandable that if the length of the text comment data is greater than the preset length, the server may perform punctuation segmentation on it; if the length of the text comment data is less than or equal to the preset length, the server may not perform punctuation segmentation on it, and this application does not limit this.

[0053] Punctuation segmentation processing may refer to dividing a long text into smaller segments according to punctuation marks (such as periods, commas, semicolons, etc.) in the text to further obtain a first-length text data set, thereby dividing an overly long text into more easily processable parts while retaining the semantic integrity of the text.

[0054] Dynamic step segmentation can be understood as a more flexible segmentation method, which can automatically adjust the step size of segmentation according to the content and characteristics of the text to ensure that the segmented text segments are neither too long nor too short, and can better capture the local features and contextual information of the text. In other words, the target length text data set obtained by dynamic step segmentation processing has a relatively suitable length for subsequent processing and analysis.

[0055] Sequence labeling is a common natural language processing task that can assign a specific label to each element in a text (such as a word, character, etc.). For example, each character or word in a text can be labeled with its position in the text, part of speech, entity type, and other information. After sequence labeling, the resulting text labeling data set contains detailed information about the text, which is helpful for providing richer features for subsequent analysis.

[0056] Word force conversion can refer to converting text from one representation form to another representation form that is more suitable for a specific task. For example, text can be converted from character level to word level, or text can be converted from one encoding method to another encoding method. The text segmentation data set obtained after word force conversion processing, each of which can be a text segment after segmentation, annotation and conversion processing, is conducive to providing a good input for subsequent semantic processing and entity recognition tasks.

[0057] In one embodiment, performing sequence labeling on each target length text data in the target length text data set to obtain a text labeling data set may include the following steps:

[0058] Each target length text data in the target length text data set is sequence labeled using the method shown in the following formula to obtain a text labeling data set:

[0059]

[0060] Among them, f(sentence) represents a text annotation data set, which can be used to indicate the mapping relationship between sentences and the first and last labels. label Indicates the mapped tag containing the entity header; end label Indicates the mapped tag containing the end of the entity; s i Indicates the label value corresponding to the i-th position of the sentence; d(entity) represents the mapping from entity to label value; pos x Indicates the currently filled label value; pos entity_start Indicates the position of the entity head in the sentence; pos entity_endIndicates the end of the entity in the sentence.

[0061] In one embodiment, performing dynamic step-length segmentation processing on the first length text data in the first length text data set to obtain the target length text data set may include the following steps:

[0062] The first length text data in the first length text data set is subjected to dynamic step segmentation processing using the method shown in the following formula to obtain a target length text data set:

[0063] Sentence=[s0,s1,…,s c ]

[0064]

[0065] Wherein, Sentence represents a first-length text data set, s0 represents the first first-length text data in the first-length text data set, s1 represents the second first-length text data in the first-length text data set, and s c Indicates the cth first-length text data in the first-length text data set, where c represents the index in the first-length text data set; New_sentence j represents the target length text data set, j represents the index of the target length text data set; window represents the window size during the dynamic step segmentation process; step represents the step size during the dynamic step segmentation process; s j Represents the j-th target length text data.

[0066] In one embodiment, each text data in the text data set is input into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, which may include the following steps:

[0067] Each text data in the text data set is input into the semantic model based on the bidirectional encoding structure for semantic processing using the method shown in the following formula to obtain a global text semantic vector set:

[0068]

[0069] in, represents the forward masked initial vector matrix, h represents the hidden vector at the i-th position; m represents the mask vector at the i-th position; Represents the reverse masked initial vector matrix; represents the global text semantic vector set; Tanh(·) represents the activation function, which can be used here to calculate the bidirectional semantic information vector; γ(·) represents the rank of the matrix, which can be used here to calculate the rank of the matrix; Represents the transpose of the forward masked initial vector matrix; represents the transpose of the reverse masked initial vector matrix; δ represents a random bias term, which can be used to add a certain amount of randomness to the model.

[0070] Furthermore, the above h can remember the previous vector information, and the calculation rule can be as follows:

[0071]

[0072] Among them, h i The latent vector representing the i-th position can be obtained by weighted summing of the latent vectors of the previous positions, which can reflect the implicit feature representation of the current position in a specific model or algorithm; i and j are position indexes, j is a variable that changes from 0 to i to traverse all positions; (1-similarity(i,j)) i-j It can represent a weight coefficient; similarity(i,j) represents the similarity between the latent vector of the i-th position and the latent vector of the j-th position; 1-similarity(i,j) can represent the degree of dissimilarity; ij as an index can make the weight of the position farther from the current position smaller; h j represents the latent vector at the jth position, which is used as a weighted term in the summation process; x i represents the element of the latent vector at the i-th position; y j represents the element of the latent vector at the jth position.

[0073] In one embodiment, each global text semantic vector in the global text semantic vector set is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, which may include the following steps:

[0074] B1: Input each global text semantic vector in the global text semantic vector set into the mixed data denoising layer to obtain a text enhanced semantic vector set;

[0075] B2: Input each text enhancement semantic vector in the text enhancement semantic vector set into the bidirectional long short-term memory layer to obtain a text feature representation set;

[0076] B3: Embed each text feature representation in the text feature representation set into the entity boundary recognition neural network layer to perform entity boundary recognition, and obtain a text entity boundary recognition information set;

[0077] B4: Each text entity boundary recognition information in the text entity boundary recognition information set is calculated and processed through an efficient hybrid self-distillation model to obtain a set of text distributed expression tensors.

[0078] Among them, the server can add noise to the data to obtain a set of text-enhanced semantic vectors. First, the server inputs each global text semantic vector in the global text semantic vector set into the mixed data denoising layer (MixupNoiseBlock) to increase the diversity and robustness of the data by randomly perturbing the original vector and mixing the features of different vectors, thereby obtaining a set of text-enhanced semantic vectors that can better capture the complex semantic information in the text and have better adaptability to noise and changes.

[0079] Next, the server inputs each text enhancement semantic vector in the text enhancement semantic vector set into the bidirectional long short-term memory (BiLSTM) layer, which can simultaneously consider the previous and next contextual information of the text and effectively capture the long-term dependencies in the sequence data, so that each text enhancement semantic vector is converted into a more expressive text feature representation, thereby obtaining a text feature representation set that contains the dynamic information and semantic features of the text in the time dimension.

[0080] Then, the server embeds each text feature representation in the text feature representation set into the entity boundary identification neural network (SE-FNN) layer for entity boundary identification, so as to determine the boundary position of the entity in the text by performing complex nonlinear transformation and feature extraction on the input text feature representation, so as to further obtain a text entity boundary identification information set. Among them, each element in the text entity boundary identification information set can be used to represent possible entity boundary information in a text. It can be understood that the SE-FNN layer can be a neural network structure specially designed for entity boundary identification, and the present application does not limit this.

[0081] Finally, the server calculates and processes each text entity boundary identification information in the text entity boundary identification information set through a Highly Efficient Mixed Self-Distillation Attention (HMSDA) model, and introduces a self-distillation mechanism and utilizes knowledge transfer between models to further optimize the text entity boundary identification information, thereby obtaining a more accurate and representative set of text distributed expression tensors, which is beneficial to various subsequent natural language processing tasks and helps to accurately identify target entities in text comment data.

[0082] In one embodiment, the server inputs each global text semantic vector in the global text semantic vector set into the mixed data noise enhancement layer to obtain the text enhancement semantic vector set, which may include the following steps:

[0083] Each global text semantic vector in the global text semantic vector set is input into the mixed data denoising layer using the method shown in the following formula to obtain a text enhanced semantic vector set:

[0084]

[0085] in, Represents a text enhancement semantic sample in a text enhancement semantic vector set; represents the label of the text enhancement semantic sample in the text enhancement semantic vector set; σ represents the bias term, which can be used to control the proportion of the two samples; x i represents any global text semantic vector in the global text semantic vector set, such as the ith global text semantic vector; x j Indicates the global text semantic vector set except x i Any global text semantic vector other than y is called the jth global text semantic vector; i Represents the label of the i-th global text semantic vector; y j represents the label of the jth global text semantic vector; e represents the natural logarithm; α represents a random factor, which can be used for noise addition here.

[0086] In one embodiment, each text entity boundary recognition information in the text entity boundary recognition information set is calculated and processed by an efficient hybrid self-distillation model to obtain a text distributed expression tensor set, which may include the following steps:

[0087] Each text entity boundary recognition information in the text entity boundary recognition information set is calculated and processed by an efficient hybrid self-distillation model using the method shown in the following formula to obtain a text distributed expression tensor set:

[0088] L θ =CE loss (MixupNoiseBlock(BERT(x,θ)),y)+L θ '

[0089]

[0090] Among them, L θ Represents a set of distributed expression tensors of text, i.e., the result of self-distillation; CE loss (·) represents cross entropy, which can be used here to calculate the relationship between the model and the original label; MixupNoiseBlock(·) represents processing through the mixed data noise layer; BERT(·) represents processing through the first bidirectional encoder; x represents text entity boundary recognition information; y represents the true label; θ represents any sentence vector; L θ ′ represents the mean square error loss value; α represents the coefficient factor; MSE loss(·) denotes mean squared error; BERT′(·) denotes processing through the second bidirectional encoder; K denotes the number of past time steps; k denotes the index of the past time step; θ t-k Represents the sentence vector at time step tk, that is, the sentence vector of the kth time step in the past.

[0091] It should be understood that the Internet financial new entity recognition method based on big data provided by the server in this application can be implemented based on the Internet financial new entity recognition model based on big data. The Internet financial new entity recognition model based on big data can be used to execute the Internet financial new entity recognition method based on big data provided by the above embodiment. In the training process of the Internet financial new entity recognition model based on big data, the server can use the above method to iterate the initial model for multiple rounds, and introduce an adaptive model to calculate the hierarchical weights, and at the same time add a reward mechanism to increase the error penalty, thereby saving better model parameters and obtaining an Internet financial new entity recognition model based on big data.

[0092] In one embodiment, the server introduces an adaptive model into the initial model to calculate the layer weights, which can be seen in the following formula:

[0093]

[0094]

[0095] in, Indicates the weights corresponding to each layer in the initial model, which can be used to adjust the contribution of different layers in the final output; Dense unit Represents a fully connected layer or similar processing unit; SoftMax(·) represents a normalization operation, which can be used here to obtain the weights of each layer of the initial model to ensure that the weights of each layer are within a reasonable range and the relative importance is reflected; representation i The layers of the model, that is, the feature representations at different levels in the initial model, each layer performs specific processing and transformation on the input data; output represents the final output, which is the result obtained by weighted summation of the representations of each layer of the initial model, representing the processing result of the model on the input data; d model The hidden layer dimensions of the initial model can reflect the size or complexity of the model's internal representation when processing data.

[0096] In one embodiment, the server adds a reward mechanism, the main purpose of which is to guide the model to better learn and identify entity boundaries by adjusting the loss function, such as giving different feedback to the correct and incorrect situations of the model in entity boundary identification, so as to improve the accuracy and generalization ability of the model.

[0097] When calculating the loss during model training, the server can use the mask to control the weight of the loss, increase the loss corresponding to the entity boundary that should have been recognized but was not recognized, so as to increase the penalty for the model; reduce the loss of the entity boundary that was recognized but should not be recognized, so as to increase its reward. The calculation rules can be as follows:

[0098]

[0099] Among them, Loss reward represents the reward loss function, which adjusts the loss of model training by considering the entity boundary recognition results of different situations, so as to guide the model to better identify new entities in Internet finance; δ represents the penalty factor, which can be used to control the intensity of the increased penalty for the loss corresponding to the entity boundary that should have been recognized but was not recognized. It is understandable that the larger the δ value, the stronger the penalty; β represents the penalty factor, which can be used to control the reward intensity when the loss of the entity boundary that was recognized but should not be recognized increases. It is understandable that the larger the β value, the stronger the reward intensity; P a represents the penalty fragments that need to be enlarged, that is, the fragments corresponding to the entity boundaries that the model should have recognized but did not recognize; Q a represents the fragments that need to be rewarded, that is, the fragments corresponding to the entity boundaries that the model recognizes but should not actually recognize; s α (i,j) represents the entity segment predicted by the model. (i,j) can represent information such as the position index of the entity segment in the text; e represents the natural logarithm.

[0100] This application uses a sliding window to segment small fragments of data and splices them in a fixed window sliding step to increase the amount of trainable data; the text data uses a pre-trained model based on a bidirectional encoding structure to capture the global semantic information of the text, and uses a masking method to mask the text forward and backward to ensure that the subsequent information will not be leaked during forward encoding, and vice versa; the BiLSTM-SE-FNN module is introduced to prevent the text from forgetting the forward or backward information due to being too long, and MixupNoiseBlock is used before to add noise to the data to achieve the effect of strengthening the data; then a minimalist two-layer feedforward neural network is used to extract the head and tail boundaries of the entity, and the entity is extracted by greedily combining the head and tail of the entity, and finally an efficient hybrid self-distillation method is used to reduce the number of model parameters, making training faster and more efficient; combined with financial business scenarios, a dynamic step size is proposed to segment the data, adaptive learning is introduced, and a reward mechanism is added to increase the reward or loss of the model predicting the entity, increase the penalty for not predicting the entity, and use efficient hybrid self-distillation to reduce the number of model parameters; in addition, the entire training process uses distributed expression tensors and multi-batch training, which is faster.

[0101] It can be seen that in the above scheme, the server obtains text comment data and segments the text comment data to obtain a text segmentation data set, and inputs each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, and each global text semantic vector in the global text semantic vector set is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, so that each text in the text distributed expression tensor set can be distributed input into the neural network structure for entity boundary classification to obtain a text head and tail entity boundary set, and then each text head and tail entity boundary in the text head and tail entity boundary set can be processed by a greedy matching algorithm to merge the entity head and tail to obtain the target entity corresponding to the text comment data, which is conducive to accurately extracting the target entity in the text comment data.

[0102] It should be understood that the order of execution of the steps in the above embodiment does not necessarily mean 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 embodiment of the present invention.

[0103] In one embodiment, a device for identifying new Internet financial entities based on big data is provided, and the device for identifying new Internet financial entities based on big data corresponds to the method for identifying new Internet financial entities based on big data in the above embodiment. Figure 3 As shown, the Internet financial new entity identification device based on big data includes an acquisition module 101, a first processing module 102, a second processing module 103, a third processing module 104 and a fourth processing module 105. The functional modules are described in detail as follows:

[0104] The acquisition module 101 is used to acquire text comment data and segment the text comment data to obtain a text segmentation data set;

[0105] The first processing module 102 is used to input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set;

[0106] The second processing module 103 is used to input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set;

[0107] The third processing module 104 is used to input each text in the text distributed expression tensor set into the neural network structure for entity boundary classification to obtain the text head and tail entity boundary set;

[0108] The fourth processing module 105 is used to perform a head-to-tail entity merging process on each text head-to-tail entity boundary in the text head-to-tail entity boundary set through a greedy matching algorithm to obtain a target entity corresponding to the text comment data.

[0109] In one embodiment, the acquisition module 101 is used to acquire text comment data and segment the text comment data to obtain a text segmentation data set, which is specifically used to:

[0110] Performing punctuation segmentation processing on long text comment data greater than a preset length in the text comment data to obtain a first length text data set;

[0111] Performing dynamic step-length segmentation processing on the first-length text data in the first-length text data set to obtain a target-length text data set;

[0112] Perform sequence labeling on each target length text data in the target length text data set to obtain a text labeling data set;

[0113] The text annotation data set is processed by word intensity conversion to obtain a text segmentation data set.

[0114] In one embodiment, the acquisition module 101 is used to perform sequence labeling on each target length text data in the target length text data set to obtain a text labeling data set, specifically for:

[0115] Each target length text data in the target length text data set is sequence labeled using the method shown in the following formula to obtain a text labeling data set:

[0116]

[0117] Among them, f(sentence) represents the text annotation data set; start label Indicates the mapped tag containing the entity header; end label Indicates the mapped tag containing the end of the entity; s i Indicates the label value corresponding to the i-th position of the sentence; d(entity) represents the mapping from entity to label value; pos x Indicates the currently filled label value; pos entity_start Indicates the position of the entity head in the sentence; pos entity_end Indicates the end of the entity in the sentence.

[0118] In one embodiment, the acquisition module 101 is used to perform dynamic step segmentation processing on the first length text data in the first length text data set to obtain a target length text data set, specifically for:

[0119] The first length text data in the first length text data set is subjected to dynamic step segmentation processing using the method shown in the following formula to obtain a target length text data set:

[0120] Sentence=[s0,s1,…,s c ]

[0121]

[0122] Wherein, Sentence represents a first-length text data set, s0 represents the first first-length text data in the first-length text data set, s1 represents the second first-length text data in the first-length text data set, and s c Indicates the cth first-length text data in the first-length text data set, where c represents the index in the first-length text data set; New_sentence j represents the target length text data set, j represents the index of the target length text data set; window represents the window size during the dynamic step segmentation process; step represents the step size during the dynamic step segmentation process; s j Represents the j-th target length text data.

[0123] In one embodiment, the first processing module 102 is used to input each text data in the text data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, specifically for:

[0124] Each text data in the text data set is input into the semantic model based on the bidirectional encoding structure for semantic processing using the method shown in the following formula to obtain a global text semantic vector set:

[0125]

[0126] in, represents the forward masked initial vector matrix, h represents the hidden vector of the i-th character; m represents the mask vector of the i-th character; Represents the reverse masked initial vector matrix; represents the global text semantic vector set; Tanh(·) represents the activation function; γ(·) represents the rank of the matrix; Represents the transpose of the forward masked initial vector matrix; represents the transpose of the inverse masked initial vector matrix; δ represents a random bias term.

[0127] In one embodiment, the second processing module 103 is used to input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, specifically for:

[0128] Input each global text semantic vector in the global text semantic vector set into the mixed data denoising layer to obtain a text enhanced semantic vector set;

[0129] Input each text enhancement semantic vector in the text enhancement semantic vector set into the bidirectional long short-term memory layer to obtain a text feature representation set;

[0130] Each text feature representation in the text feature representation set is embedded into the entity boundary recognition neural network layer to perform entity boundary recognition, thereby obtaining a text entity boundary recognition information set;

[0131] Each text entity boundary recognition information in the text entity boundary recognition information set is calculated and processed through an efficient hybrid self-distillation model to obtain a set of text distributed expression tensors.

[0132] In one embodiment, the second processing module 103 is used to input each global text semantic vector in the global text semantic vector set into the mixed data noise enhancement layer to obtain a text enhancement semantic vector set, specifically for:

[0133] Each global text semantic vector in the global text semantic vector set is input into the mixed data denoising layer using the method shown in the following formula to obtain a text enhanced semantic vector set:

[0134]

[0135]

[0136] in, Represents a text enhancement semantic sample in a text enhancement semantic vector set; represents the label of the text enhancement semantic sample in the text enhancement semantic vector set; σ represents the bias term; x i represents any global text semantic vector in the global text semantic vector set, such as the ith global text semantic vector; x j Indicates the global text semantic vector set except x i Any global text semantic vector other than y is called the jth global text semantic vector; i Represents the label of the i-th global text semantic vector; y j represents the label of the jth global text semantic vector; e represents the natural logarithm; α represents a random factor.

[0137] In one embodiment, the second processing module 103 is used to calculate and process each text entity boundary recognition information in the text entity boundary recognition information set through an efficient hybrid self-distillation model to obtain a text distributed expression tensor set, specifically for:

[0138] Each text entity boundary recognition information in the text entity boundary recognition information set is calculated and processed by an efficient hybrid self-distillation model using the method shown in the following formula to obtain a text distributed expression tensor set:

[0139] L θ =CE loss (MixupNoiseBlock(BERT(x,θ)),y)+L θ '

[0140]

[0141] Among them, L θ Represents a set of distributed text expression tensors; CE loss (·) represents cross entropy; MixupNoiseBlock(·) represents processing through the mixed data noise layer; BERT(·) represents processing through the first bidirectional encoder; x represents text entity boundary recognition information; y represents the true label; θ represents an arbitrary sentence vector; L θ ′ represents the mean square error loss value; α represents the coefficient factor; MSE loss (·) denotes mean squared error; BERT′(·) denotes processing through the second bidirectional encoder; K denotes the number of past time steps; k denotes the index of the past time step; θ t-k Represents the sentence vector at time step tk, that is, the sentence vector of the kth time step in the past.

[0142] The present invention provides a new entity recognition device for Internet finance based on big data. The device obtains text comment data and segments the text comment data to obtain a text segmentation data set, and inputs each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, and each global text semantic vector in the global text semantic vector set is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, so that each text in the text distributed expression tensor set can be distributedly input into a neural network structure for entity boundary classification to obtain a text head and tail entity boundary set, and then each text head and tail entity boundary in the text head and tail entity boundary set can be processed by a greedy matching algorithm to perform head and tail merging of entities to obtain a target entity corresponding to the text comment data, which is conducive to accurately extracting the target entity in the text comment data.

[0143] For the specific definition of the Internet financial new entity identification device based on big data, please refer to the definition of the Internet financial new entity identification method based on big data above, which will not be repeated here. Each module in the above-mentioned Internet financial new entity identification device based on big data can be implemented in whole or in part by software, hardware and a combination thereof. The above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to the above modules.

[0144] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Figure 4 As shown. The computer device includes a processor, a memory, a network interface and a database connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client through a network connection. When the computer program is executed by the processor, it realizes the functions or steps on the service side of a new entity identification method for Internet finance based on big data.

[0145] In one embodiment, a computer device is provided. The computer device may be a client, and its internal structure diagram may be as follows: Figure 5As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server through a network connection. When the computer program is executed by the processor, the functions or steps on the client side of a new entity identification method for Internet finance based on big data are realized.

[0146] 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 implements the following steps when executing the computer program:

[0147] Acquire text comment data, and segment the text comment data to obtain a text segmentation data set;

[0148] Input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set;

[0149] Input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set;

[0150] Each text in the text distributed expression tensor set is input into the neural network structure for entity boundary classification to obtain the text head and tail entity boundary set;

[0151] Each text head and tail entity boundary in the text head and tail entity boundary set is merged into a head and tail entity using a greedy matching algorithm to obtain the target entity corresponding to the text comment data.

[0152] The present invention provides a computer device, which obtains text comment data and segments the text comment data to obtain a text segmentation data set, so that each text segmentation data in the text segmentation data set is input into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, and each global text semantic vector in the global text semantic vector set is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, so that each text in the text distributed expression tensor set can be distributedly input into a neural network structure for entity boundary classification to obtain a text head and tail entity boundary set, and then each text head and tail entity boundary in the text head and tail entity boundary set can be processed by a greedy matching algorithm to perform head and tail merging of entities to obtain a target entity corresponding to the text comment data, which is conducive to accurately extracting the target entity in the text comment data.

[0153] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the following steps are implemented:

[0154] Acquire text comment data, and segment the text comment data to obtain a text segmentation data set;

[0155] Input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set;

[0156] Input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set;

[0157] Each text in the text distributed expression tensor set is input into the neural network structure for entity boundary classification to obtain the text head and tail entity boundary set;

[0158] Each text head and tail entity boundary in the text head and tail entity boundary set is merged into a head and tail entity using a greedy matching algorithm to obtain the target entity corresponding to the text comment data.

[0159] The present invention provides a computer-readable storage medium, which obtains text comment data and segments the text comment data to obtain a text segmentation data set, so that each text segmentation data in the text segmentation data set is input into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set, and each global text semantic vector in the global text semantic vector set is input into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, so that each text in the text distributed expression tensor set can be distributedly input into a neural network structure for entity boundary classification to obtain a text head and tail entity boundary set, and then each text head and tail entity boundary in the text head and tail entity boundary set can be processed by a greedy matching algorithm to perform head and tail merging of entities to obtain a target entity corresponding to the text comment data, which is conducive to accurately extracting the target entity in the text comment data.

[0160] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant descriptions on the server side and the client side in the aforementioned method embodiment. To avoid repetition, they will not be described one by one here.

[0161] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database 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. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0162] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional units and modules is used as an example. In actual applications, the above-mentioned functions can be distributed and completed by different functional units and modules as needed, that is, the internal structure of the device can be divided into different functional units or modules to complete all or part of the functions described above.

[0163] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. A method for identifying new Internet financial entities based on big data, characterized in that: The Internet financial new entity identification method based on big data includes: Acquire text comment data, and segment the text comment data to obtain a text segmentation data set; Inputting each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set; Inputting each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set; Input each text in the text distributed expression tensor set into a neural network structure for entity boundary classification to obtain a text head and tail entity boundary set; Each text head and tail entity boundary in the text head and tail entity boundary set is subjected to head and tail merging processing by a greedy matching algorithm to obtain a target entity corresponding to the text comment data.

2. According to the big data-based Internet financial new entity identification method of claim 1, it is characterized in that: The acquiring of text comment data and segmentation of the text comment data to obtain a text segmentation data set includes: Performing punctuation segmentation processing on long text comment data greater than a preset length in the text comment data to obtain a first length text data set; Performing dynamic step-length segmentation processing on the first-length text data in the first-length text data set to obtain a target-length text data set; Performing sequence labeling processing on each target length text data in the target length text data set to obtain a text labeling data set; The text annotation data set is subjected to character strength conversion processing to obtain a text segmentation data set.

3. The method for identifying new Internet financial entities based on big data according to claim 2 is characterized in that: The step of performing sequence labeling on each target length text data in the target length text data set to obtain a text labeling data set includes: Each target length text data in the target length text data set is sequence labeled using the method shown in the following formula to obtain a text labeling data set: Among them, f(sentence) represents the text annotation data set; start label Indicates the mapped tag containing the entity header; end label Indicates the mapped tag containing the end of the entity; s i Indicates the label value corresponding to the i-th position of the sentence; d(entity) represents the mapping from entity to label value; pos x Indicates the currently filled label value; pos entity_start Indicates the position of the entity head in the sentence; pos entity_end Indicates the end of the entity in the sentence.

4. The method for identifying new Internet financial entities based on big data according to claim 3 is characterized in that: The step of performing dynamic step-length segmentation processing on the first-length text data in the first-length text data set to obtain a target-length text data set includes: The first length text data in the first length text data set is subjected to dynamic step length segmentation processing by the method shown in the following formula to obtain a target length text data set: Sentence=[s0,s1,…,s c ] Wherein, Sentence represents a first-length text data set, s0 represents the first first-length text data in the first-length text data set, s1 represents the second first-length text data in the first-length text data set, and s c Indicates the cth first-length text data in the first-length text data set, where c represents the index in the first-length text data set; New_sentence j represents the target length text data set, j represents the index of the target length text data set; window represents the window size during the dynamic step segmentation process; step represents the step size during the dynamic step segmentation process; s j Represents the j-th target length text data.

5. The method for identifying new Internet financial entities based on big data according to claim 4 is characterized in that: The step of inputting each text data in the text data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set includes: Each text data in the text data set is input into a semantic model based on a bidirectional encoding structure for semantic processing using the method shown in the following formula to obtain a global text semantic vector set: in, represents the forward masked initial vector matrix, h represents the hidden vector of the i-th character; m represents the mask vector of the i-th character; Represents the reverse masked initial vector matrix; represents the global text semantic vector set; Tanh(·) represents the activation function; γ(·) represents the rank of the matrix; Represents the transpose of the forward masked initial vector matrix; represents the transpose of the inverse masked initial vector matrix; δ represents a random bias term.

6. The method for identifying new Internet financial entities based on big data according to claim 5, characterized in that: The step of inputting each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set, including: Input each global text semantic vector in the global text semantic vector set into the mixed data denoising layer to obtain a text enhanced semantic vector set; Inputting each text enhancement semantic vector in the text enhancement semantic vector set into a bidirectional long short-term memory layer to obtain a text feature representation set; Embed each text feature representation in the text feature representation set into an entity boundary recognition neural network layer to perform entity boundary recognition, thereby obtaining a text entity boundary recognition information set; Each text entity boundary identification information in the text entity boundary identification information set is calculated and processed through an efficient hybrid self-distillation model to obtain a text distributed expression tensor set.

7. The method for identifying new Internet financial entities based on big data according to claim 6 is characterized in that: The step of inputting each global text semantic vector in the global text semantic vector set into the mixed data denoising layer to obtain a text enhanced semantic vector set includes: Each global text semantic vector in the global text semantic vector set is input into the mixed data denoising layer using the method shown in the following formula to obtain a text enhanced semantic vector set: in, Represents a text enhancement semantic sample in a text enhancement semantic vector set; represents the label of the text enhancement semantic sample in the text enhancement semantic vector set; σ represents the bias term; x i represents any global text semantic vector in the global text semantic vector set, such as the ith global text semantic vector; x j Indicates the global text semantic vector set except x i Any global text semantic vector other than y is called the jth global text semantic vector; i Represents the label of the i-th global text semantic vector; y j represents the label of the jth global text semantic vector; e represents the natural logarithm; α represents a random factor.

8. The method for identifying new Internet financial entities based on big data according to claim 7 is characterized in that: The step of calculating and processing each text entity boundary recognition information in the text entity boundary recognition information set through an efficient hybrid self-distillation model to obtain a text distributed expression tensor set includes: Each text entity boundary recognition information in the text entity boundary recognition information set is calculated and processed by an efficient hybrid self-distillation model using the method shown in the following formula to obtain a text distributed expression tensor set: L θ =CE loss (MixupNoiseBlock(BERT(x,θ)),y)+L θ ′ Among them, L θ Represents a set of distributed text expression tensors; CE loss (·) represents cross entropy; MixupNoiseBlock(·) represents processing through the mixed data noise layer; BERT(·) represents processing through the first bidirectional encoder; x represents text entity boundary recognition information; y represents the true label; θ represents an arbitrary sentence vector; L θ ′ represents the mean square error loss value; α represents the coefficient factor; MSE loss (·) denotes mean squared error; BERT′(·) denotes processing through the second bidirectional encoder; K denotes the number of past time steps; k denotes the index of the past time step; θ t-k Represents the sentence vector at time step tk, that is, the sentence vector of the kth time step in the past.

9. A new entity identification device for Internet finance based on big data, characterized in that: The Internet financial new entity identification device based on big data includes: An acquisition module, used for acquiring text comment data and segmenting the text comment data to obtain a text segmentation data set; A first processing module is used to input each text segmentation data in the text segmentation data set into a semantic model based on a bidirectional encoding structure for semantic processing to obtain a global text semantic vector set; A second processing module is used to input each global text semantic vector in the global text semantic vector set into a module based on a bidirectional entity boundary recognition neural network structure for expression analysis to obtain a text distributed expression tensor set; A third processing module is used to distribute each text in the text distributed expression tensor set into a neural network structure for entity boundary classification to obtain a text head and tail entity boundary set; The fourth processing module is used to perform head-to-tail merging processing on each text head-to-tail entity boundary in the text head-to-tail entity boundary set through a greedy matching algorithm to obtain a target entity corresponding to the text comment data.

10. A terminal, characterized in that: The method comprises a processor, an input device, an output device and a memory, wherein the processor, the input device, the output device and the memory are interconnected, wherein the memory is used to store a computer program, the computer program comprises program instructions, and the processor is configured to call the program instructions to execute the method according to any one of claims 1 to 8.

Images