Named entity identification method and device, computer equipment and storage medium

Abstract

The invention relates to the technical field of artificial intelligence, and discloses a named entity identification method and device, computer equipment and a storage medium. The method comprises the following steps: performing named entity recognition on to-be-recognized text data, and performing word pair matrix construction on the to-be-recognized text data to obtain an initial word pair representation matrix with all word pair relationships; carrying out constraint processing on the initial word pair representation matrix by adopting a word pair relationship constraint device so as to screen out useless word pair relationships, reduce the influence caused by noise and obtain a constraint word pair representation matrix; a binary relation predictor is adopted to carry out relation prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a proximity relation probability value and a jump relation probability value, and after coding processing is carried out, a more accurate named entity recognition result is obtained, so that the recognition efficiency is improved when a nested, hierarchical or crossed entity structure is recognized, and the recognition efficiency is improved. And the named entity recognition accuracy is improved.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to a method, device, computer equipment and storage medium for identifying named entities. Background Art

[0002] With the popularization of natural language processing (NLP), the key named entity recognition (NER) task in natural language processing (NLP) is particularly important. For example, in the financial field, the accuracy of named entity recognition tasks is conducive to more accurate identification of entities such as companies and people, thereby more accurate analysis and prediction of market dynamics, public opinion trends, etc., thereby assisting financial institutions in making more accurate analysis, evaluation and management.

[0003] Named Entity Recognition (NER) helps understand important information in text, relationships between entities, and context by identifying and classifying named entities from text. Common entity categories include names of people, places, names of organizations, dates, times, currencies, percentages, product names, professional terms, etc. Named entity recognition tasks are usually completed using sequence labeling models, which assigns a single label to a specific token, but this method is difficult to capture nested, hierarchical, or cross-entity structures.

[0004] In the prior art, when dealing with complex entity structures and nested entities, named entity recognition needs to be combined with other methods or models to better capture the relationship between entities, such as capturing nested, hierarchical or intersecting entity structures by building a word pair model. However, the current word pair model cannot accurately screen out reasonable entity relationships between word pairs, that is, the existing named entity recognition task has a poor accuracy rate when recognizing nested, hierarchical or intersecting entity structures, which affects the accuracy of subsequent natural language processing and leads to excessive errors in the final prediction results. Summary of the invention

[0005] The embodiments of the present invention provide a method, apparatus, computer device and storage medium for identifying a named entity, which solves the problem of poor accuracy in existing named entity recognition tasks when identifying nested, hierarchical or cross-entity structures.

[0006] An embodiment of the present invention provides a method for identifying a named entity, comprising: Obtaining text data to be recognized; Constructing a word pair matrix for the text data to be recognized to obtain an initial word pair representation matrix; Using a word pair relationship constraint to perform constraint processing on the initial word pair representation matrix to obtain a constrained word pair representation matrix; Using a binary relationship predictor, performing relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a jump relationship probability value; The proximity relationship probability value and the jump relationship probability value are encoded to obtain a named entity recognition result.

[0007] The embodiment of the present invention further provides a device for identifying a named entity, comprising: A module for acquiring text data to be recognized, used for acquiring text data to be recognized; An initial word pair representation matrix acquisition module is used to construct a word pair matrix for the text data to be recognized and acquire an initial word pair representation matrix; A constraint word pair representation matrix acquisition module is used to use a word pair relationship constraint to perform constraint processing on the initial word pair representation matrix to obtain a constraint word pair representation matrix; A relationship probability value acquisition module, used to use a binary relationship predictor to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a neighbor relationship probability value; The named entity recognition result acquisition module is used to encode the adjacent relationship probability value and the jump relationship probability value to obtain the named entity recognition result.

[0008] An embodiment of the present invention further provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned method for recognizing named entities when executing the computer program.

[0009] An embodiment of the present invention further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method for recognizing named entities is implemented.

[0010] The above-mentioned named entity recognition method, device, computer equipment and storage medium perform named entity recognition on the text data to be recognized, and construct a word pair matrix on the text data to be recognized to obtain an initial word pair representation matrix with all word pair relationships; a word pair relationship constraint is used to constrain the initial word pair representation matrix, thereby filtering out useless word pair relationships, reducing the impact of noise, and obtaining a constrained word pair representation matrix; a binary relationship predictor is used to predict the relationship between the initial word pair representation matrix and the constrained word pair representation matrix, and obtain adjacent relationship probability values ​​and jump relationship probability values, so as to obtain more accurate named entity recognition results after encoding processing, thereby improving the accuracy of named entity recognition when recognizing nested, hierarchical or cross-entity structures. The improvement of the accuracy of named entity recognition is conducive to more accurate identification of entities such as companies and people, thereby more accurate analysis and prediction of market dynamics, public opinion trends, etc., thereby assisting financial institutions in making more accurate analysis, evaluation and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0011] 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.

[0012] Figure 1 is a schematic diagram of an application environment of a method for identifying named entities in an embodiment of the present invention; Figure 2 is a flow chart of a method for identifying named entities in one embodiment of the present invention; Figure 3 is a flowchart of step S202 of a method for identifying a named entity in one embodiment of the present invention; Figure 4 is a flowchart of step S203 of a method for identifying a named entity in one embodiment of the present invention; Figure 5 is a flowchart of step S204 of a method for identifying a named entity in one embodiment of the present invention; Figure 6 is a flowchart of step S501 of a method for identifying a named entity in one embodiment of the present invention; Figure 7 is a flowchart of step S205 of a method for identifying a named entity in one embodiment of the present invention; Figure 8 is a flowchart of a first method for identifying a named entity in another embodiment of the present invention; Fig. 9is a schematic diagram of a named entity recognition device according to an embodiment of the present invention; Fig.10 is a schematic diagram of a computer device in one embodiment of the present invention. DETAILED DESCRIPTION

[0013] 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.

[0014] The method for identifying named entities provided by the embodiment of the present invention can be applied in the following aspects: Figure 1 In the application environment shown. Figure 1 As shown, the client (computer device) communicates with the server through the network. The client, also known as the user end, refers to the program corresponding to the server that provides local services to the client. The client (computer device) includes but is not limited to various personal computers, laptops, smart phones, tablets, cameras and portable wearable devices. The server can be an independent server or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDN), and big data and artificial intelligence platforms.

[0015] The method for identifying a named entity provided by an embodiment of the present invention can be applied as follows: Figure 1 Specifically, the named entity recognition method is applied in a named entity recognition system, and the named entity recognition system includes: Figure 1 The client and server shown communicate with each other through a network to implement the method for identifying named entities provided in an embodiment of the present invention.

[0016] Understandably, in the financial field, NER can accurately build financial information retrieval systems, conduct public opinion analysis and market forecasts by marking and classifying entity information such as company names, stock codes, names, dates, etc. in financial news, announcements, and reports; NER can also be used to identify and analyze potential risk factors, such as legal proceedings, market fluctuations, economic indicators and other entity information, to assist financial institutions in risk assessment and management; at the same time, in financial transactions, NER can be used to verify customer identity information, prevent fraudulent activities, and ensure the security of transactions.

[0017] It should be noted that the named entity recognition method provided in the present invention is explained and exemplified by entity types in the financial field, and is not only used for named entity recognition in the financial field.

[0018] In one embodiment, if Figure 2 As shown, a method for identifying named entities is provided, and the method is applied in Figure 1 The server in the example is used as an example to illustrate the process, including the following steps: S201: Acquire text data to be recognized; S202: constructing a word pair matrix for the text data to be recognized, and obtaining an initial word pair representation matrix; S203: using a word pair relationship constraint to perform constraint processing on the initial word pair representation matrix to obtain a constrained word pair representation matrix; S204: using a binary relationship predictor to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a jump relationship probability value; S205: Encoding the neighboring relationship probability value and the jump relationship probability value to obtain a named entity recognition result.

[0019] As an example, in step S201, the server receives text data to be recognized input by a user or a natural language processing model, and processes the text data to be recognized.

[0020] In this example, in the financial field, when used for tasks such as financial analysis, information extraction, and risk management, in addition to personal name information, geographic location, organizational information, time and date, common text data to be identified also includes but is not limited to the following common text types: stock codes, financial indicators, market events, economic indicators, and product names.

[0021] Among them, stock codes: stock codes are often involved in financial texts; financial indicators: various indicators are often involved in the financial field, such as stock prices, market capitalization, financial indicators, etc.; market events: various events in the financial market, such as stock trading, acquisitions, financing, etc.; economic indicators: the financial field is related to economic data, such as GDP, inflation rate and other economic indicators; commodity names: commodity transactions are also common in the financial market, and NER can mark commodity names, contract types, etc. The above common text data are entities that need to be identified in NER tasks in text data in the financial field.

[0022] As an example, in step S202, after obtaining the text data to be recognized, the server constructs a word pair matrix for the representation corresponding to the text data to be recognized, and constructs a two-dimensional word pair matrix according to each word in the text data to be recognized, and obtains an initial word pair representation matrix.

[0023] In this example, the initial word pair representation matrix is ​​a word pair representation matrix that does not screen the specific associations of word pairs. The N words in the text data to be recognized are extracted two by two to form an N*N word pair representation matrix Hw, forming a two-dimensional word pair matrix as the initial word pair representation matrix.

[0024] As an example, in step S203, after obtaining the initial word pair representation matrix after matrix construction, the server inputs the initial word pair representation matrix into the word pair relationship constraint device, and performs constraint processing on the initial word pair representation matrix, that is, entity relationship screening is performed on each word pair representation in the input initial word pair representation matrix, word pair representations with entity relationships are determined, and the final constraint word pair representation matrix is ​​formed. By screening the entity relationships of word pair representations, the word pair representations without entity relationships are reduced, the noise impact on the named entity recognition task is improved, and the accuracy of the named entity recognition task is improved.

[0025] In this example, the word pair relationship constraint is used to determine whether each word pair representation has an entity relationship. It is obtained after training a binary classifier based on a multilayer perceptron (MLP). Multilayer perceptron (MLP) is an artificial neural network architecture with strong nonlinear modeling capabilities. Especially when the entity relationship recognition task handles complex relationships and abstract features, MLP can capture higher-level abstract features, thereby better modeling and screening word pair entity relationships.

[0026] As an example, in step S204, after obtaining the constraint word pair representation matrix of constraint processing, the server inputs the initial word pair representation matrix and the constraint word pair representation matrix into the binary relationship predictor, and respectively calculates the probability values ​​of the two types of relationships (adjacent relationship and jump relationship) of each word pair representation in the initial word pair representation matrix (the representation corresponding to the two matched words in the word pair representation) and the two types of relationships (adjacent relationship and jump relationship) of each corresponding word pair representation in the constraint word pair representation matrix, so as to obtain the proximity relationship probability value and jump relationship probability value corresponding to the recognized text data.

[0027] Among them, the neighbor relationship refers to the state where two entities are adjacent in the text, and the jump relationship refers to the state where two entities are non-adjacent in the text. The relationship probability value represents the probability corresponding to the type of relationship that may exist between each two entities. The neighbor relationship probability value is the probability value corresponding to the neighbor relationship, and the jump relationship probability value is the probability value corresponding to the jump relationship. For example, the probability that there is a neighbor relationship between two entities (neighbor relationship probability value) is 0.7, the probability that there is a jump relationship (jump relationship probability value) is 0.2, and the probability that there is no relationship is 0.1.

[0028] In this example, the binary relationship predictor is used to determine whether there is a proximity relationship or a jump relationship between two entities. It is built based on a multi-layer perceptron (MLP) and a dual affine linear layer. When processing the prediction of the relationship between entities, it improves the sensitivity of capturing the complex nonlinear relationship between entities. Among them, the dual affine linear layer is a combination of two affine linear transformations in the neural network, which maps the input to another space through a linear transformation. In this example, it is used to reduce the dimension of the relationship between entities and avoid overfitting by mapping it to the affine layer to achieve binary relationship classification.

[0029] As an example, in step S205, after obtaining the neighboring relationship probability value and the jump relationship probability value, the server constructs a relationship structure between entities based on the neighboring relationship probability value and the jump relationship probability value, such as constructing a sentence relationship graph, thereby realizing nested and discontinuous entity recognition and obtaining a named entity recognition result.

[0030] In this example, loss calculation can be performed based on the named entity recognition result output by the named entity recognition method and the correct label of the named entity, for example, using cross entropy as a cost function to adjust the word pair relationship constraint and the binary relationship predictor.

[0031] In this example, a word pair matrix is ​​constructed for the text data to be recognized to obtain an initial word pair representation matrix with all word pair relationships; a word pair relationship constraint is used to constrain the initial word pair representation matrix, thereby filtering out useless word pair relationships and reducing the impact of noise to obtain a constrained word pair representation matrix; a binary relationship predictor is used to predict the relationship between the initial word pair representation matrix and the constrained word pair representation matrix, and obtain neighbor relationship probability values ​​and jump relationship probability values, which are then encoded to obtain more accurate named entity recognition results, thereby improving the accuracy of named entity recognition when recognizing nested, hierarchical or cross-entity structures. The improvement in the accuracy of named entity recognition is conducive to more accurate identification of entities such as companies and people, thereby more accurate analysis and prediction of market dynamics, public opinion trends, etc., thereby assisting financial institutions in making more accurate analysis, evaluation and management.

[0032] In one embodiment, if Figure 3 As shown, in step S202, a word pair matrix is ​​constructed for the text data to be recognized, and an initial word pair representation matrix is ​​obtained, including: S301: marking the text data to be recognized to obtain a target context representation; S302: construct a word pair matrix for the target context representation to obtain an initial word pair representation matrix.

[0033] As an example, in step S301, after receiving the text data to be recognized input by the user or the natural language processing model, the server tokenizes the text data to be recognized to obtain the target context representation.

[0034] In this example, according to business needs, the text content in the financial field is used as the text data to be recognized X{X1,X2…X}, and the text data to be recognized X{X1,X2…X} is input into BERT+BiLSTM to obtain the target context representation.

[0035] As an example, in step S302, after obtaining the target context representation, the server extracts N (N is a positive integer) words in the target context representation in pairs to form a two-dimensional N*N initial word pair representation matrix to construct N*N word pair representations of N words for subsequent entity relationship prediction. In this example, each word pair representation can be expressed as (hi, hj), where hi and hj are two words in the context representation.

[0036] In one embodiment, if Figure 4 As shown, in step S203, a word pair relationship constraint is used to perform constraint processing on the initial word pair representation matrix to obtain a constrained word pair representation matrix, including: S401: extracting representations from the initial word pair representation matrix to obtain a first word pair representation; S402: performing constraint processing on the first word pair representation to obtain a second word pair representation; S403: Fusing the first word pair representation and the second word pair representation to obtain a constrained word pair representation matrix.

[0037] As an example, in step S401, after obtaining the initial word pair representation matrix, the server extracts the first word pair representation of a one-dimensional representation from the initial word pair representation matrix for processing by the word pair relationship constraint. In this example, each word pair representation in the initial word pair representation matrix is ​​extracted in order from top to bottom and from left to right, and the N*N word pair representations are concatenated.

[0038] As an example, in step S402, after obtaining the first word pair representation, the server performs constraint processing on the N*N word pair representations in the first word pair representation, that is, after identifying the entity relationship of each word pair representation, the word pair representations with entity relationships are screened out to obtain the second word pair representation. In this example, the word pair representations with entity relationships are retained, and the word pair representations without entity relationships are assigned 0), and the second word pair representation is composed according to the entity relationship recognition results (word pair representations or 0) corresponding to the order of each word pair representation in the first word pair representation.

[0039] As an example, in step S403, after obtaining the second word pair representation, the server multiplies the second word pair representation with the entity relationship recognition result with the first word pair representation to obtain a constrained word pair representation matrix. The second word pair representation is used to filter out the word pair representations with entity relationships in the first word pair representation, and the word pair representations without entity relationships are represented by 0.

[0040] In this example, after the initial word pair representation matrix is ​​processed in one dimension, the first word pair representation after one-dimensional processing is processed for entity relationship recognition. The second word pair representation is obtained according to the entity relationship recognition result, and after the first word pair representation and the second word pair representation are fused, the constrained word pair representation matrix is ​​obtained to screen out the word pair relationships that can be used for the task, that is, the constrained word pair representation matrix, to avoid using word pair representations without entity relationships for tasks. Compared with directly using all word pair representations for tasks, noise interference is reduced and task processing performance is improved.

[0041] In one embodiment, if Figure 5 As shown, in step S204, a binary relationship predictor is used to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain neighbor relationship probability values ​​and jump relationship probability values, including: S501: using a binary relationship predictor to perform relationship prediction on an initial word pair representation matrix and a constraint word pair representation matrix, and obtaining an initial relationship score corresponding to the initial word pair representation matrix and a constraint relationship score corresponding to the constraint word pair representation matrix; S502: Add the residuals of the initial relationship score and the constraint relationship score to obtain a combined relationship score; S503: performing relationship classification on the merged relationship scores to obtain adjacent relationship scores and skip relationship scores; S504: performing vectorization processing on the neighbor relationship score and the skip relationship score to obtain a neighbor relationship vector corresponding to the neighbor relationship score and a skip relationship vector corresponding to the skip relationship score; S505: normalizing the neighbor relationship vector and the skip relationship vector to obtain a neighbor relationship probability value corresponding to the neighbor relationship vector and a skip relationship probability value corresponding to the skip relationship vector.

[0042] As an example, in step S501, after receiving the initial word pair representation matrix and the constraint word pair representation matrix, the server inputs the initial word pair representation matrix and the constraint word pair representation matrix into the binary relationship predictor respectively, performs relationship score prediction based on the initial word pair representation matrix, outputs the corresponding initial relationship score, and then performs relationship score prediction based on the constraint word pair representation matrix, and outputs the corresponding constraint relationship score.

[0043] In this example, the initial relationship score includes the jump relationship score and the neighbor relationship score corresponding to each word pair representation in the initial word pair representation matrix, and the constraint relationship score includes the jump relationship score and the neighbor relationship score corresponding to each word pair representation in the constraint word pair representation matrix. It should be noted that this process outputs the scores of the initial word pair representation matrix in two relationship dimensions, that is, the initial relationship scores; and the output constraint relationship scores are the scores of the constraint word pair representation matrix in two relationship dimensions.

[0044] As an example, in step S502, after obtaining the initial relationship score and the constraint relationship score, the server adds the residuals of the initial relationship score and the constraint relationship score to obtain a total combined relationship score. The error problem caused by the loss of context information can be avoided by adding the initial relationship score and the constraint relationship score.

[0045] In this example, the initial relationship score, the constraint relationship score, and the merged relationship score are all in the form of matrices. At this time, since there are two relationship dimensions (adjacent relationship and skip relationship), the matrix dimension corresponding to the relationship score is N*N*2 (where 2 represents adjacent relationship and skip relationship).

[0046] As an example, in step S503, after obtaining the merged relationship scores, the server classifies the merged relationship scores, that is, according to different relationships, the scores corresponding to the adjacent relationships and the skip relationships in the merged relationship scores are divided to obtain the adjacent relationship scores and the skip relationship scores.

[0047] As an example, in step S504, after obtaining the neighbor relationship score and the skip relationship score, the server vectorizes the neighbor relationship score and the skip relationship score through the MLP layer, and obtains the neighbor relationship vector corresponding to the neighbor relationship score and the skip relationship vector corresponding to the skip relationship score.

[0048] As an example, in step S505, after obtaining the neighboring relationship vector and the skip relationship vector, the server performs exponential operation on the neighboring relationship vector and the skip relationship vector to obtain the exponential result for normalization processing in the softmax layer, thereby obtaining the neighboring relationship probability value corresponding to the neighboring relationship vector and the skip relationship probability value corresponding to the skip relationship vector.

[0049] In one embodiment, if Figure 6 As shown, in step S501, a binary relationship predictor is used to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix, and an initial relationship score corresponding to the initial word pair representation matrix and a constraint relationship score corresponding to the constraint word pair representation matrix are obtained, including: S601: The initial word pair representation matrix includes an initial first vocabulary representation and an initial second vocabulary representation; the initial first vocabulary representation and the initial second vocabulary representation are respectively mapped to obtain an initial first mapping feature; S602: concatenate the initial first vocabulary representation and the initial second vocabulary representation to obtain an initial concatenated vocabulary representation; and map the concatenated vocabulary representation to obtain an initial second mapping feature; S603: Merge the initial first mapping feature and the initial second mapping feature to obtain an initial relationship score corresponding to the initial word pair representation matrix; S604: The constraint word pair representation matrix includes a constraint first vocabulary representation and a constraint second vocabulary representation; mapping processing is performed on the constraint first vocabulary representation and the constraint second vocabulary representation respectively to obtain a constraint first mapping feature; S605: concatenate the constrained first vocabulary representation and the constrained second vocabulary representation to obtain a constrained concatenated vocabulary representation; and map the concatenated vocabulary representation to obtain a constrained second mapping feature; S606: Merge the first constraint mapping feature and the second constraint mapping feature to obtain a constraint relationship score corresponding to the constraint word pair representation matrix.

[0050] Among them, the first vocabulary representation and the second vocabulary representation in the word pair representation matrix are the vocabulary representations corresponding to the two words in the word pair. Therefore, the initial first vocabulary representation and the initial second vocabulary representation are the representations corresponding to the two words in the word pair representation in the initial word pair representation matrix; the constrained first vocabulary representation and the constrained second vocabulary representation are the representations corresponding to the two words in the word pair representation in the constrained word pair representation matrix.

[0051] As an example, in step S601, after obtaining the initial word pair representation matrix and the constraint word pair representation matrix, the server maps the reduced dimensional initial first vocabulary representation through the bi-affine linear layer in the binary relationship predictor, and performs transposition processing to obtain the initial first vocabulary representation; and maps the reduced dimensional initial second vocabulary representation to obtain the corresponding initial first mapping features between the initial second vocabulary representations.

[0052] In this example, the mapping process includes inputting the dimension-reduced initial first vocabulary representation and the dimension-reduced initial second vocabulary representation into a pre-selected trained first linear change matrix, thereby outputting a first mapping feature.

[0053] As an example, in step S602, after acquiring the initial first vocabulary representation and the initial second vocabulary representation, the server concatenates the initial first vocabulary representation and the initial second vocabulary representation, and performs mapping processing on the concatenated initial concatenated vocabulary representation to obtain an initial second mapping feature.

[0054] In this example, the mapping process includes inputting the concatenated initial concatenated vocabulary representation into a pre-selected trained second linear transformation matrix, thereby outputting a second mapping feature.

[0055] As an example, in step S603, after obtaining the initial first mapping feature and the initial second mapping feature, the server also obtains the bias value obtained after pre-training, and merges the bias value, the initial first mapping feature and the initial second mapping feature obtained after pre-training to obtain the initial relationship score corresponding to the initial word pair representation matrix.

[0056] As an example, in step S604, after obtaining the constraint word pair representation matrix and the constraint word pair representation matrix, the server maps the reduced-dimensional constraint first vocabulary representation through the bi-affine linear layer in the binary relationship predictor, and performs transposition processing to obtain the constraint first vocabulary representation; and maps the reduced-dimensional constraint second vocabulary representation to obtain corresponding constraint first mapping features between the constraint second vocabulary representations.

[0057] In this example, the mapping process includes inputting the dimension-reduced constrained first vocabulary representation and the dimension-reduced constrained second vocabulary representation into a pre-selected trained first linear change matrix, thereby outputting a first mapping feature.

[0058] As an example, in step S604, after acquiring the constrained first vocabulary representation and the constrained second vocabulary representation, the server concatenates the constrained first vocabulary representation and the constrained second vocabulary representation, and performs mapping processing on the concatenated constrained concatenated vocabulary representation to obtain the constrained second mapping feature.

[0059] In this example, the mapping process includes inputting the concatenated constrained concatenated vocabulary representation into a pre-selected trained second linear change matrix, thereby outputting a second mapping feature.

[0060] As an example, in step S605, after obtaining the first constrained mapping feature and the second constrained mapping feature, the server also obtains the bias value obtained after pre-training, and merges the bias value, the first constrained mapping feature and the second constrained mapping feature obtained after pre-training to obtain the constraint relationship score corresponding to the constraint word pair representation matrix.

[0061] In this example, a binary relationship predictor based on MLP and dual affine linear layers is used to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix. The data is dimensionality reduced and mapped to an affine layer to avoid overfitting, thereby obtaining the initial relationship scores corresponding to the initial word pair representation matrix and the constraint relationship scores corresponding to the constraint word pair representation matrix.

[0062] In one embodiment, if Figure 7As shown, in step S205, a word pair relationship constraint is used to perform constraint processing on the initial word pair representation matrix to obtain a constrained word pair representation matrix, including: S701: Analyze the neighboring relationship probability value and the jump relationship probability value to obtain the sentence relationship result; S702: Perform entity path prediction processing according to the vocabulary proximity relationship in the sentence relationship result to obtain a first prediction result; S703: Perform entity type and boundary prediction processing according to the vocabulary jump relationship in the sentence relationship result to obtain a second prediction result; S704: Obtain a named entity recognition result according to the first prediction result and the second prediction result.

[0063] As an example, in step S701, after obtaining the proximity relationship probability value and the jump relationship probability value, the server analyzes the proximity relationship probability value and the jump relationship probability value to obtain a sentence relationship result. In this example, the sentence relationship result is the relationship structure in each vocabulary representation. For example, in the text data to be recognized as "Shanghai is a scenic spot worth visiting", the sentence relationship result is that the two words "Shanghai" are in a proximity relationship, and "Shanghai" is in a jump relationship with "attractions". The relationship trend between each word can also be shown through a sentence relationship chart.

[0064] As an example, in step S702, after obtaining the sentence relationship result, the server performs entity path prediction processing according to the word proximity relationship in the sentence relationship result, thereby obtaining a first prediction result. In this example, the entity path is predicted according to the proximity relationship, that is, the possible relationship path between words is predicted by analyzing the proximity relationship between them, and the predicted result is used as the first prediction result.

[0065] As an example, in step S703, after obtaining the sentence relationship result, the server performs entity type and boundary prediction processing according to the word jump relationship in the sentence relationship result, thereby obtaining the second prediction result. In this example, the entity type and boundary are determined according to the jump relationship, that is, the jump relationship between words is analyzed to determine their types and boundaries in the text.

[0066] As an example, in step S704, after obtaining the first prediction result corresponding to the entity path prediction and the second prediction result corresponding to the entity type and boundary prediction, the server predicts the entity path based on the proximity relationship, and determines the entity type and boundary through the jump relationship, thereby realizing nested and discontinuous entity recognition and obtaining a named entity recognition result.

[0067] In this example, accurate recognition of nested and discontinuous entities is achieved by predicting entity paths based on proximity relationships and determining entity types and boundaries through jump relationships, thereby improving the accuracy of named entity recognition when recognizing nested, hierarchical, or intersecting entity structures.

[0068] In another embodiment, if Figure 8 As shown, in step S203, a word pair relationship constraint is used to perform constraint processing on the initial word pair representation matrix. Before obtaining the constraint word pair representation matrix, the named entity recognition method further includes: S801: Obtain sample data formed based on entity prior relations; S802: Perform feature extraction processing on the sample data to obtain sample context representation; S803: construct a word pair matrix for the sample context representation to obtain a sample word pair representation matrix; S804: Input the sample word pair representation matrix into the multi-layer perceptron for classification processing to obtain the sample recognition result corresponding to the sample data; S805: performing loss calculation on the sample recognition result corresponding to the sample data and the real label data corresponding to the sample data to obtain a relationship prediction loss value; S806: Based on the relationship prediction loss value, the multi-layer perceptron is updated to obtain a word pair relationship constraint.

[0069] As an example, in step S801, the server receives sample data for training input by a user or a natural language processing model, where the sample data is formed based on entity prior relationships, that is, real sample data obtained by labeling or recognition.

[0070] As an example, in step S802, after receiving the sample data input by the user or the natural language processing model, the server tokenizes the sample text data to obtain the sample context representation.

[0071] In this example, according to business needs, sample content in the financial field is used as sample text data X{X1, X2…X}, and the sample text data X{X1, X2…X} is input into BERT+BiLSTM to obtain sample context representation.

[0072] As an example, in step S803, after obtaining the sample context representation, the server extracts N (N is a positive integer) words in the sample context representation in pairs to form a two-dimensional N*N sample word pair representation matrix, so as to construct N*N word pair representations of N words for subsequent entity relationship prediction.

[0073] As an example, in step S804, after obtaining the sample word pair representation matrix, the server inputs the sample word pair representation matrix into the multi-layer perceptron for prediction and classification processing to obtain the sample recognition result corresponding to the sample data. In this example, classification prediction is performed for each word pair representation in the sample word pair representation matrix. When the word pair representation has an entity relationship, the word pair representation with the entity relationship is marked as 1, and the word pair representation without the entity relationship is marked as 0. When all the word pair representations in the sample word pair representation matrix are classified, the sample recognition result is obtained.

[0074] As an example, in step S805, after obtaining the sample recognition result corresponding to the sample data, the server performs loss calculation according to the sample recognition result corresponding to the sample data and the real label data corresponding to the sample data to obtain the relationship prediction loss value. In this example, the cross entropy of the sample recognition result corresponding to the sample data and the real label data corresponding to the sample data is used to obtain the corresponding relationship prediction loss value for training the word pair relationship constraint.

[0075] As an example, in step S806, the server updates the multilayer perceptron based on the relationship prediction loss value to obtain the word pair relationship constraint. In this example, the multilayer perceptron is updated according to the preset conditions. When the relationship prediction loss value reaches the preset conditions, the multilayer perceptron is used as the word pair relationship constraint and optimized in subsequent use.

[0076] In this example, the binary classifier of the Multilayer Perceptron (MLP) is trained with sample data to obtain a word pair relationship constraint, thereby taking advantage of the strong nonlinear modeling ability to determine whether each word pair representation has an entity relationship, improving the screening of word pair relationships, thereby screening out useless word pair relationships and reducing the impact of noise.

[0077] 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.

[0078] In one embodiment, a device for identifying a named entity is provided, and the device for identifying a named entity corresponds one-to-one to the method for identifying a named entity in the above embodiment. Fig. 9 As shown, the named entity recognition device includes a to-be-recognized text data acquisition module 901, an initial word pair representation matrix acquisition module 902, a constraint word pair representation matrix acquisition module 903, a relationship probability value acquisition module 904, and a named entity recognition result acquisition module 905. Each functional module is described in detail as follows: The module 901 for acquiring text data to be recognized is used to acquire text data to be recognized; An initial word pair representation matrix acquisition module 902 is used to construct a word pair matrix for the text data to be recognized and acquire an initial word pair representation matrix; A constraint word pair representation matrix acquisition module 903 is used to use a word pair relationship constraint to perform constraint processing on the initial word pair representation matrix to obtain a constraint word pair representation matrix; The relationship probability value acquisition module 904 is used to use a binary relationship predictor to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a neighbor relationship probability value; The named entity recognition result acquisition module 905 is used to encode the adjacent relationship probability value and the jump relationship probability value to obtain the named entity recognition result.

[0079] In one embodiment, the initial word pair representation matrix acquisition module 902 includes: A target context representation acquisition unit, used for marking the text data to be recognized and acquiring the target context representation; The initial word pair representation matrix acquisition unit is used to construct a word pair matrix for the target context representation and acquire an initial word pair representation matrix.

[0080] In one embodiment, the constraint word pair representation matrix acquisition module 903 includes: A first word pair representation acquisition unit, used to extract the representation from the initial word pair representation matrix to obtain a first word pair representation; A second word pair representation acquisition unit, used for performing constraint processing on the first word pair representation to acquire a second word pair representation; The constraint word pair representation matrix acquisition unit is used to fuse the first word pair representation and the second word pair representation to obtain the constraint word pair representation matrix.

[0081] In one embodiment, the relationship probability value acquisition module 904 includes: An initial relationship score and constraint relationship score acquisition unit, used to use a binary relationship predictor to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix, and acquire an initial relationship score corresponding to the initial word pair representation matrix and a constraint relationship score corresponding to the constraint word pair representation matrix; A merged relationship score acquisition unit, used for adding the residuals of the initial relationship score and the constraint relationship score to acquire the merged relationship score; A neighboring relationship score and a skip relationship score obtaining unit, used for classifying the combined relationship scores and obtaining neighboring relationship scores and a skip relationship score; A neighbor relationship vector and a skip relationship vector acquisition unit, used to perform vectorization processing on the neighbor relationship score and the skip relationship score, and acquire a neighbor relationship vector corresponding to the neighbor relationship score and a skip relationship vector corresponding to the skip relationship score; The neighbor relationship probability value and jump relationship probability value acquisition unit is used to normalize the neighbor relationship vector and the jump relationship vector to obtain the neighbor relationship probability value corresponding to the neighbor relationship vector and the jump relationship probability value corresponding to the jump relationship vector.

[0082] In one embodiment, the initial relationship score and constraint relationship score acquisition subunit includes: An initial first mapping feature acquisition subunit is used for an initial word pair representation matrix including an initial first vocabulary representation and an initial second vocabulary representation; the initial first vocabulary representation and the initial second vocabulary representation are respectively mapped to obtain an initial first mapping feature; The initial second mapping feature acquisition subunit is used to perform concatenation processing on the initial first vocabulary representation and the initial second vocabulary representation to obtain an initial concatenated vocabulary representation; and perform mapping processing on the concatenated vocabulary representation to obtain an initial second mapping feature; An initial relationship score acquisition subunit is used to merge the initial first mapping feature and the initial second mapping feature to obtain an initial relationship score corresponding to the initial word pair representation matrix; A constrained first mapping feature acquisition subunit is used to constrain the word pair representation matrix to include a constrained first vocabulary representation and a constrained second vocabulary representation; the constrained first vocabulary representation and the constrained second vocabulary representation are respectively mapped to obtain a constrained first mapping feature; The constrained second mapping feature acquisition subunit is used to perform concatenation processing on the constrained first vocabulary representation and the constrained second vocabulary representation to obtain the constrained concatenated vocabulary representation; perform mapping processing on the concatenated vocabulary representation to obtain the constrained second mapping feature; The constraint relationship score acquisition subunit is used to merge the constraint first mapping feature and the constraint second mapping feature to obtain the constraint relationship score corresponding to the constraint word pair representation matrix.

[0083] In one embodiment, the named entity recognition result acquisition module 905 includes: A sentence relationship result acquisition unit, used to analyze the neighbor relationship probability value and the jump relationship probability value to acquire the sentence relationship result; A first prediction result acquisition unit, configured to perform entity path prediction processing according to the vocabulary proximity relationship in the sentence relationship result to acquire a first prediction result; A second prediction result acquisition unit, configured to perform entity type and boundary prediction processing according to the vocabulary jump relationship in the sentence relationship result to acquire a second prediction result; The named entity recognition result acquisition unit is used to acquire the named entity recognition result according to the first prediction result and the second prediction result.

[0084] In another embodiment, the apparatus for identifying a named entity further includes: A sample data acquisition module is used to acquire sample data formed based on entity prior relations; A sample context representation acquisition module is used to perform feature extraction processing on sample data to obtain sample context representation; A sample word pair representation matrix acquisition module is used to construct a word pair matrix for sample context representation and obtain a sample word pair representation matrix; The sample recognition result acquisition module is used to input the sample word pair representation matrix into the multi-layer perceptron for classification processing to obtain the sample recognition result corresponding to the sample data; A relationship prediction loss value acquisition module is used to calculate the loss of the sample recognition result corresponding to the sample data and the real label data corresponding to the sample data to obtain the relationship prediction loss value; The word pair relationship constraint acquisition module is used to update the multi-layer perceptron based on the relationship prediction loss value to obtain the word pair relationship constraint.

[0085] For the specific definition of the named entity recognition device, please refer to the definition of the named entity recognition method above, which will not be repeated here. The various modules in the above-mentioned named entity recognition device 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.

[0086] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.10 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 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 database of the computer device is used to execute the data used or generated in the process of the method for identifying named entities. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a method for identifying named entities is implemented.

[0087] 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. When the processor executes the computer program, the method for recognizing named entities in the above embodiment is implemented, for example Figure 2 S201-S205 as shown, or Figures 3 to 8 Alternatively, when the processor executes the computer program, the functions of each module / unit in the embodiment of the named entity recognition device are realized, for example Fig. 9 The functions of the to-be-recognized text data acquisition module 901, the initial word pair representation matrix acquisition module 902, the constraint word pair representation matrix acquisition module 903, the relationship probability value acquisition module 904 and the named entity recognition result acquisition module 905 are not described here in detail to avoid repetition.

[0088] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the method for identifying named entities in the above embodiment is implemented, for example Figure 2 S201-S205 as shown, or Figures 3 to 8 Alternatively, when the computer program is executed by a processor, the functions of each module / unit in the embodiment of the apparatus for identifying named entities are realized, for example, Fig. 9 The functions of the to-be-recognized text data acquisition module 901, the initial word pair representation matrix acquisition module 902, the constraint word pair representation matrix acquisition module 903, the relationship probability value acquisition module 904 and the named entity recognition result acquisition module 905 are not described here in detail to avoid repetition.

[0089] A person of ordinary skill in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be implemented 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 by the present invention may include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of 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), etc.

[0090] 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.

[0091] The above embodiments 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 be included in the protection scope of the present invention.

Claims

1. A method for identifying named entities, characterized in that: include: Obtaining text data to be recognized; Constructing a word pair matrix for the text data to be recognized to obtain an initial word pair representation matrix; Using a word pair relationship constraint to perform constraint processing on the initial word pair representation matrix to obtain a constrained word pair representation matrix; Using a binary relationship predictor, performing relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a jump relationship probability value; The proximity relationship probability value and the jump relationship probability value are encoded to obtain a named entity recognition result.

2. The method for identifying named entities according to claim 1, characterized in that: The step of constructing a word pair matrix for the text data to be recognized to obtain an initial word pair representation matrix includes: Marking the text data to be recognized to obtain a target context representation; A word pair matrix is ​​constructed for the target context representation to obtain an initial word pair representation matrix.

3. The method for identifying named entities according to claim 1, characterized in that: The constraining the initial word pair representation matrix to obtain the constrained word pair representation matrix includes: Performing representation extraction on the initial word pair representation matrix to obtain a first word pair representation; Performing constraint processing on the first word pair representation to obtain a second word pair representation; The first word pair representation and the second word pair representation are fused to obtain a constrained word pair representation matrix.

4. The method for identifying named entities according to claim 1, characterized in that: The method of using a binary relationship predictor to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a jump relationship probability value includes: Using a binary relationship predictor, perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix, and obtain an initial relationship score corresponding to the initial word pair representation matrix and a constraint relationship score corresponding to the constraint word pair representation matrix; Adding residuals of the initial relationship score and the constraint relationship score to obtain a combined relationship score; Performing relationship classification on the combined relationship scores to obtain adjacent relationship scores and skip relationship scores; Performing vectorization processing on the neighbor relationship score and the skip relationship score to obtain a neighbor relationship vector corresponding to the neighbor relationship score and a skip relationship vector corresponding to the skip relationship score; The neighbor relationship vector and the skip relationship vector are normalized to obtain a neighbor relationship probability value corresponding to the neighbor relationship vector and a skip relationship probability value corresponding to the skip relationship vector.

5. The method for identifying named entities according to claim 4, characterized in that: The performing relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain an initial relationship score corresponding to the initial word pair representation matrix and a constraint relationship score corresponding to the constraint word pair representation matrix includes: The initial word pair representation matrix includes an initial first vocabulary representation and an initial second vocabulary representation; the initial first vocabulary representation and the initial second vocabulary representation are respectively mapped to obtain an initial first mapping feature; Performing concatenation processing on the initial first vocabulary representation and the initial second vocabulary representation to obtain an initial concatenated vocabulary representation; performing mapping processing on the concatenated vocabulary representation to obtain an initial second mapping feature; Merging the initial first mapping feature and the initial second mapping feature to obtain an initial relationship score corresponding to the initial word pair representation matrix; The constraint word pair representation matrix includes a constraint first vocabulary representation and a constraint second vocabulary representation; mapping processing is performed on the constraint first vocabulary representation and the constraint second vocabulary representation respectively to obtain a constraint first mapping feature; The first constrained vocabulary representation and the second constrained vocabulary representation are concatenated to obtain a constrained concatenated vocabulary representation; the concatenated vocabulary representation is mapped to obtain a constrained second mapping feature; The first constraint mapping feature and the second constraint mapping feature are combined to obtain a constraint relationship score corresponding to the constraint word pair representation matrix.

6. The method for identifying named entities according to claim 1, characterized in that: The encoding process of the proximity relationship probability value and the jump relationship probability value to obtain the named entity recognition result includes: Analyze the neighboring relationship probability value and the jump relationship probability value to obtain a sentence relationship result; Perform entity path prediction processing according to the vocabulary proximity relationship in the sentence relationship result to obtain a first prediction result; According to the vocabulary jump relationship in the sentence relationship result, perform entity type and boundary prediction processing to obtain a second prediction result; Obtain a named entity recognition result based on the first prediction result and the second prediction result.

7. The method for identifying named entities according to claim 1, characterized in that: Before the word pair relationship constraint is adopted to perform constraint processing on the initial word pair representation matrix to obtain the constraint word pair representation matrix, the named entity recognition method further includes: Obtain sample data based on entity prior relationships; Performing feature extraction processing on the sample data to obtain sample context representation; Constructing a word pair matrix for the sample context representation to obtain a sample word pair representation matrix; Inputting the sample word pair representation matrix into a multi-layer perceptron for classification processing to obtain a sample recognition result corresponding to the sample data; Performing loss calculation on the sample recognition result corresponding to the sample data and the real label data corresponding to the sample data to obtain a relationship prediction loss value; Based on the relationship prediction loss value, the multi-layer perceptron is updated to obtain a word pair relationship constraint.

8. A named entity recognition device, characterized in that: include: A module for acquiring text data to be recognized, used for acquiring text data to be recognized; An initial word pair representation matrix acquisition module is used to construct a word pair matrix for the text data to be recognized and acquire an initial word pair representation matrix; A constraint word pair representation matrix acquisition module is used to use a word pair relationship constraint to perform constraint processing on the initial word pair representation matrix to obtain a constraint word pair representation matrix; A relationship probability value acquisition module, used to use a binary relationship predictor to perform relationship prediction on the initial word pair representation matrix and the constraint word pair representation matrix to obtain a neighbor relationship probability value and a neighbor relationship probability value; The named entity recognition result acquisition module is used to encode the adjacent relationship probability value and the jump relationship probability value to obtain the named entity recognition result.

9. A computer device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method for recognizing a named entity according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for recognizing a named entity as claimed in any one of claims 1 to 7 is implemented.

Images