A method for distinguishing between parts of speech of special vocabulary in the field of automobile finance

Abstract

The application provides a method for distinguishing the language consciousness of special vocabulary in the field of automobile finance, comprising the following steps: S1, generating a first vocabulary chain in real time based on a target vocabulary identified in a current environment to be recognized; S2, determining a current scene category chain in real time based on the first vocabulary chain and an automobile finance scene recognition model; S3, determining a special vocabulary recognition model corresponding to the latest target scene based on the current scene category chain; and S4, distinguishing the language consciousness of the first vocabulary chain obtained in real time based on the special vocabulary recognition model until a complete language consciousness distinguishing result is obtained and output. The method can improve the accuracy of distinguishing the language consciousness of special vocabulary in the automobile finance industry and better improve the intention understanding of artificial intelligence voice in the automobile industry through the scene recognition model and the special vocabulary recognition model.

Description

Technical Field

[0001] This invention relates to the field of semantic recognition technology, and in particular to a method for semantic recognition of specialized vocabulary in the automotive finance field. Background Technology

[0002] Currently, the narrow definition of auto finance refers to the financial activity where consumers apply for loans from financial institutions to fulfill their car purchase needs. The broader definition encompasses all financial businesses related to the automotive industry, including financing methods involved in various stages of automotive development, design, production, distribution, and consumption. We usually refer to the narrow definition, specifically consumer loans for car purchases. Another common type is inventory financing, where dealers apply for financing from financial institutions when purchasing vehicles in bulk from manufacturers. Financing methods include bank acceptance bills, commercial factoring, and working capital loans.

[0003] In the field of artificial intelligence in auto finance, semantic recognition is a crucial technical component. The accuracy of semantic recognition directly affects the level of intelligence of artificial intelligence. However, because the existing semantic recognition methods in the field of artificial intelligence in auto finance do not differentiate between specific domains and semantic scenarios, the application of semantic recognition technology solutions and models in the field of artificial intelligence in auto finance is not targeted enough, and the level of intelligence and recognition accuracy need to be improved.

[0004] Therefore, this invention proposes a semantic recognition method for specialized vocabulary in the field of automotive finance. Summary of the Invention

[0005] This invention provides a semantic recognition method for specialized vocabulary in the automotive finance field. It determines the current semantic scene based on the vocabulary in the environment to be recognized, and then determines a specialized vocabulary recognition model for the automotive finance field based on the semantic scene. Based on the determined specialized vocabulary model, semantic recognition is performed on the vocabulary in the environment to be recognized, which improves the accuracy of semantic recognition of specialized vocabulary in the automotive finance industry. Through scene recognition model and specialized vocabulary recognition model, it better enhances the intent understanding of artificial intelligence speech in the automotive industry.

[0006] This invention provides a method for semantic recognition of specialized terms in the automotive finance field, including:

[0007] S1: Generate the first vocabulary chain in real time based on the target words identified in the current environment to be identified;

[0008] S2: Based on the first vocabulary chain and the auto finance scenario recognition model, the current scenario category chain is determined in real time;

[0009] S3: Determine the dedicated vocabulary recognition model corresponding to the latest target scene based on the current scene category chain;

[0010] S4: Based on a dedicated vocabulary recognition model, perform semantic recognition on the newly obtained first vocabulary chain until a complete semantic recognition result is obtained and output.

[0011] Preferably, the semantic recognition method for specialized vocabulary in the automotive finance field, S1: Based on the target vocabulary identified in the current environment to be recognized, a first vocabulary chain is generated in real time, including:

[0012] S101: Identify the target words in the current environment to be identified;

[0013] S102: Sort the identified target words according to their order of appearance to obtain the first word chain.

[0014] Preferably, in the semantic recognition method for specialized vocabulary in the automotive finance field, S2: based on the first vocabulary chain and the automotive finance scenario recognition model, the current scenario category chain is determined in real time, including:

[0015] S201: Determine the current scene category based on the first lexical chain, the lexical-scene association model, and the semantic structure-scene association model;

[0016] S202: Sort all current scene categories according to their order of appearance to obtain the current scene category chain.

[0017] Preferably, the semantic recognition method for specialized vocabulary in the automotive finance field, S201: Based on the first vocabulary chain and the vocabulary-scene association model and the semantic structure-scene association model, the current scene category is determined, including:

[0018] Determine the vocabulary size in the first vocabulary chain;

[0019] When the vocabulary size does not exceed the vocabulary size threshold, the current scene category is determined based on the first vocabulary chain and the vocabulary-scene association model.

[0020] When the vocabulary exceeds the vocabulary threshold, the current scene category is determined based on the first vocabulary chain, the vocabulary-scene association model, and the semantic structure-scene association model.

[0021] Preferably, in the aforementioned semantic recognition method for specialized vocabulary in the automotive finance field, when the vocabulary size does not exceed a vocabulary size threshold, the current scene category is determined based on a first vocabulary chain and a vocabulary-scene association model, including:

[0022] When the vocabulary size does not exceed the vocabulary size threshold, the scene category corresponding to each target word is determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model.

[0023] The scene category that appears most frequently among all the scene categories corresponding to the target words is taken as the current scene category.

[0024] Preferably, the semantic recognition method for specialized vocabulary in the automotive finance field, when the vocabulary size exceeds a vocabulary size threshold, determines the current scene category based on a first vocabulary chain, a vocabulary-scene association model, and a semantic structure-scene association model, including:

[0025] When the vocabulary exceeds the vocabulary threshold, the first scene category corresponding to each target word is determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model, and the second scene category is determined based on the target semantic structure of the first vocabulary chain and the semantic structure-scene association model.

[0026] The scene category that appears most frequently among all first and second scene categories is selected as the current scene category.

[0027] Preferably, the semantic recognition method for specialized terms in the automotive finance field, S3: determining the specialized term recognition model corresponding to the latest target scenario based on the current scenario category chain, including:

[0028] Based on the coherence analysis results of the current scene category chain, the latest target scene is determined in real time;

[0029] Based on the list of scene-specific vocabulary recognition models, the specific vocabulary recognition model corresponding to the latest target scene is determined.

[0030] Preferably, the semantic recognition method for specialized terms in the automotive finance field determines the latest target scenario in real time based on the coherence analysis results of the current scenario category chain, including:

[0031] Perform coherence analysis on the current scene category chain and determine whether there are lexical subchains in the current scene category chain that meet the coherence requirements, and obtain the coherence analysis results.

[0032] The latest target scene is determined in real time based on the results of the coherence analysis.

[0033] Preferably, the semantic recognition method for specialized vocabulary in the automotive finance field, S4: performing semantic recognition on the newly obtained first vocabulary chain based on the specialized vocabulary recognition model until a complete semantic recognition result is obtained and output, including:

[0034] Based on a dedicated vocabulary recognition model, semantic recognition is performed on the newly obtained first vocabulary chain to obtain semantic recognition results. It is then determined whether the semantic recognition results meet the recognition termination judgment condition corresponding to the latest target scene. If so, the complete semantic recognition result is determined and output based on the semantic recognition results.

[0035] Otherwise, continue to acquire new target words in the current environment to be identified until a complete semantic recognition result is determined and output.

[0036] Preferably, the semantic recognition method for specialized vocabulary in the automotive finance field continues to acquire new target words in the current environment to be recognized until a complete semantic recognition result is determined and output, including:

[0037] Continue to acquire new target words in the current environment to be identified, generate a new first word chain based on the new target words, and determine a new dedicated word recognition model based on the first word chain;

[0038] Based on the new dedicated vocabulary recognition model, semantic recognition is performed on the new first vocabulary chain to obtain new semantic recognition results. This process continues until the latest semantic recognition result meets the recognition termination judgment condition corresponding to the latest target scene. Then, the complete semantic recognition result is determined and output based on the latest semantic recognition result.

[0039] Other features and advantages of the invention will be set forth in the description which follows, and will be apparent in part from the description, or may be learned by practicing the invention. The objects and other advantages of the invention may be realized and obtained by means of the structures particularly pointed out in the written description, claims, and drawings.

[0040] The technical solution of the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. Attached Figure Description

[0041] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0042] Figure 1 This is a flowchart of a semantic recognition method for specialized terms in the field of automotive finance, as described in an embodiment of the present invention.

[0043] Figure 2 This is a flowchart of a semantic recognition method for specialized terms in the field of automotive finance, as described in another embodiment of the present invention.

[0044] Figure 3 This is a flowchart illustrating a semantic recognition method for specialized terms in the field of automotive finance, as described in another embodiment of the present invention. Detailed Implementation

[0045] The preferred embodiments of the present invention will be described below with reference to the accompanying drawings. It should be understood that the preferred embodiments described herein are for illustration and explanation only and are not intended to limit the present invention.

[0046] Example 1:

[0047] This invention provides a method for semantic recognition of specialized terms in the automotive finance field, with reference to... Figure 1 ,include:

[0048] S1: Generate the first vocabulary chain in real time based on the target words identified in the current environment to be identified;

[0049] S2: Based on the first vocabulary chain and the auto finance scenario recognition model, the current scenario category chain is determined in real time;

[0050] S3: Determine the dedicated vocabulary recognition model corresponding to the latest target scene based on the current scene category chain;

[0051] S4: Based on a dedicated vocabulary recognition model, perform semantic recognition on the newly obtained first vocabulary chain until a complete semantic recognition result is obtained and output.

[0052] In this embodiment, the current environment to be identified is the environment in which semantic recognition methods using specialized vocabulary in the automotive finance field are needed to identify the meaning.

[0053] In this embodiment, the target vocabulary refers to the vocabulary that needs to be identified in the current environment to be identified, and whose meaning needs to be identified using a semantic recognition method that uses vocabulary specific to the automotive finance field.

[0054] In this embodiment, the first vocabulary chain is a chain structure generated based on multiple target vocabulary words.

[0055] In this embodiment, the automotive finance scenario recognition model is a model used to identify the speech scenario in the automotive finance field to which the current semantic environment belongs. The recognition results are, for example, a semantic environment of bank acceptance bill, a semantic environment of commercial factoring, or a semantic environment of working capital loan.

[0056] In this embodiment, the current scenario category chain is a chain structure consisting of multiple automotive finance semantic scenarios (including the latest target scenario) determined based on the first vocabulary chain and the automotive finance scenario recognition model.

[0057] In this embodiment, the latest target scenario is the automotive finance semantic scenario that is latest determined based on the current scenario category chain.

[0058] In this embodiment, the specialized vocabulary recognition model is a model corresponding to the automotive finance semantic scenario for recognizing specialized vocabulary in the corresponding automotive finance semantic scenario. It stores a specialized vocabulary library containing a large number of specialized vocabulary words in the corresponding semantic scenario in the automotive field and their corresponding semantic meanings. The semantic meaning of the target vocabulary word is determined by matching the target vocabulary word to be recognized with the specialized vocabulary words in the corresponding specialized vocabulary library.

[0059] In this embodiment, the complete semantic recognition result is the semantic recognition result obtained by performing semantic recognition on the newly obtained first vocabulary chain based on a dedicated vocabulary recognition model and obtaining the recognition termination judgment condition corresponding to the latest target scene.

[0060] The beneficial effects of the above technologies are as follows: the current semantic scene is determined based on the words in the environment to be identified, and then a special vocabulary recognition model for the automotive finance field is determined for the semantic scene. Based on the determined special vocabulary model, semantic recognition of words in the environment to be identified is performed, which improves the semantic recognition accuracy of special vocabulary in the automotive finance industry. Through scene recognition model and special vocabulary recognition model, the intent understanding of artificial intelligence voice in the automotive industry is better improved.

[0061] Example 2:

[0062] Based on Example 1, the semantic recognition method for specialized vocabulary in the automotive finance field, S1: Based on the target vocabulary identified in the current environment to be recognized, a first vocabulary chain is generated in real time, referencing... Figure 2 ,include:

[0063] S101: Identify the target words in the current environment to be identified;

[0064] S102: Sort the identified target words according to their order of appearance to obtain the first word chain.

[0065] In this embodiment, the order of appearance is the order in which the target words are identified.

[0066] The beneficial effect of the above technology is that by identifying and sorting the words appearing in the current environment to be identified, the corresponding word chain can be obtained.

[0067] Example 3:

[0068] Based on Example 1, the semantic recognition method for specialized vocabulary in the automotive finance field, S2: Based on the first vocabulary chain and the automotive finance scenario recognition model, the current scenario category chain is determined in real time, including:

[0069] S201: Determine the current scene category based on the first lexical chain, the lexical-scene association model, and the semantic structure-scene association model;

[0070] S202: Sort all current scene categories according to their order of appearance to obtain the current scene category chain.

[0071] In this embodiment, the vocabulary-scene association model is a model that determines the corresponding automotive finance semantic scene based on vocabulary. This model is trained by a large number of specialized vocabulary words in the corresponding semantic scene in the automotive finance field. The model selects the automotive finance semantic scene corresponding to the specialized vocabulary word with the highest matching degree with the target vocabulary word from the specialized vocabulary book associated with each automotive finance semantic scene.

[0072] In this embodiment, the semantic structure-scene association model is a model that determines the corresponding automotive finance semantic scene based on the semantic structure in the lexical chain. This model is trained by a large number of semantic structures of lexical chains in the corresponding semantic scenes in the automotive finance field. The automotive finance semantic scene is selected from the semantic structure library associated with each automotive finance semantic scene by selecting the semantic structure with the greatest degree of separation between the semantic structure and the semantic structure corresponding to the lexical chain.

[0073] In this embodiment, the current scenario category is the automotive finance semantic scenario corresponding to the latest identified word chain in the environment to be identified, which is determined based on the first word chain, the word-scenario association model, and the semantic structure-scenario association model.

[0074] In this embodiment, the order of appearance is the generation order of the first word chain that determines the current scene category.

[0075] In this embodiment, the current scene category chain is a chain structure obtained by sorting all current scene categories based on their order of appearance.

[0076] The beneficial effects of the above technology are as follows: Based on the target words contained in the first word chain and the semantic structure of the first word chain, the semantic environment of the automotive finance corresponding to the currently acquired word chain is determined, and then the different semantic environments of automotive finance corresponding to different first word chains are determined. This achieves real-time determination of the semantic environment of automotive finance based on the word chain, which is more accurate than the traditional method of first dividing the word chain and then determining the semantic scene of the division result in one go.

[0077] Example 4:

[0078] Based on Example 3, the semantic recognition method for specialized vocabulary in the automotive finance field, S201: Based on the first vocabulary chain, the vocabulary-scene association model, and the semantic structure-scene association model, the current scene category is determined, including:

[0079] Determine the vocabulary size in the first vocabulary chain;

[0080] When the vocabulary size does not exceed the vocabulary size threshold, the current scene category is determined based on the first vocabulary chain and the vocabulary-scene association model.

[0081] When the vocabulary exceeds the vocabulary threshold, the current scene category is determined based on the first vocabulary chain, the vocabulary-scene association model, and the semantic structure-scene association model.

[0082] In this embodiment, the vocabulary size is the total number of target words contained in the first vocabulary chain.

[0083] In this embodiment, the vocabulary size threshold is the threshold for determining whether the semantic structure-scenario association model can be used to determine the category of auto finance scenarios. When the vocabulary size does not exceed the vocabulary size threshold, the semantic structure-scenario association model cannot be used to determine the category of auto finance scenarios. When the vocabulary size exceeds the vocabulary size threshold, the semantic structure-scenario association model can be used to determine the category of auto finance scenarios.

[0084] The beneficial effects of the above technologies are as follows: based on the vocabulary threshold, it is determined whether the semantic structure-scene association model can be used to determine the category of auto finance scenarios. Thus, while ensuring the accuracy of the results of determining the semantic scenarios of auto finance, it enriches the methods for determining the semantic scenarios of auto finance to the greatest extent and increases the accuracy of the results of determining the semantic scenarios of auto finance.

[0085] Example 5:

[0086] Based on Example 4, the semantic recognition method for specialized vocabulary in the automotive finance field, when the vocabulary size does not exceed a vocabulary size threshold, determines the current scene category based on the first vocabulary chain and the vocabulary-scene association model, including:

[0087] When the vocabulary size does not exceed the vocabulary size threshold, the scene category corresponding to each target word is determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model.

[0088] The scene category that appears most frequently among all the scene categories corresponding to the target words is taken as the current scene category.

[0089] In this embodiment, the scenario category refers to different automotive finance semantic scenarios.

[0090] The beneficial effects of the above technology are: to determine the current scene category based on the target words in the word chain when the vocabulary size of the first word chain does not exceed the vocabulary size threshold.

[0091] Example 6:

[0092] Based on Example 4, the semantic recognition method for specialized vocabulary in the automotive finance field, when the vocabulary size exceeds a vocabulary size threshold, determines the current scene category based on the first vocabulary chain, the vocabulary-scene association model, and the semantic structure-scene association model, including:

[0093] When the vocabulary exceeds the vocabulary threshold, the first scene category corresponding to each target word is determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model, and the second scene category is determined based on the target semantic structure of the first vocabulary chain and the semantic structure-scene association model.

[0094] The scene category that appears most frequently among all first and second scene categories is selected as the current scene category.

[0095] In this embodiment, the first scene category is the scene category corresponding to each target word determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model when the vocabulary size exceeds the vocabulary size threshold.

[0096] In this embodiment, the second scene category is determined based on the target semantic structure of the first lexical chain and the semantic structure-scene association model, including:

[0097] Extract the target semantic structure of the first lexical chain, and determine the corresponding scene category based on the target semantic structure and the semantic structure-scene association model;

[0098] The target semantic structure of the first lexical chain is extracted, including:

[0099] Based on the word attribute association rules, the association rules corresponding to each word attribute are determined;

[0100] Identify each word attribute corresponding to each target word in the first word chain, and based on the association rules corresponding to each part-of-speech attribute, determine the association rules between all target words corresponding to the corresponding word attribute;

[0101] Based on the association rules among all target words in the first vocabulary chain, the association network of the first vocabulary chain is determined;

[0102] The association relationships between all target words in the first vocabulary chain were determined based on the association network.

[0103] Assign a corresponding personalized vocabulary attribute representation value to each vocabulary attribute of each target vocabulary (i.e., the representation value of each vocabulary attribute of each target vocabulary is different for each target vocabulary).

[0104] The sum of the personalized lexical attribute representation values ​​of the two target words corresponding to the association relationship is used as the association relationship representation value of the corresponding association relationship.

[0105] Based on the association network and the association representation value of each association contained in the association network, an association representation matrix is ​​constructed.

[0106] In the first vocabulary chain, the key target words are identified, and the association relationship representation values ​​corresponding to the association relationships containing the key target words in the association relationship representation matrix are set to 0 to obtain the association iteration matrix;

[0107] The association relationship representation matrix and the association iterative relationship matrix are multiplied by a preset number of iterations to obtain the iterative relationship matrix corresponding to each iteration process.

[0108] Identify the target association relationships in the iterative relationship matrix whose association relationship representation value is 0;

[0109] After deleting all target associations determined by the iterative process in the association network, the iterative association network is obtained;

[0110] Based on all remaining associations contained in the iterative association network, the target words contained in the first lexical chain are associated and divided to obtain the target semantic structure of the first lexical chain.

[0111] By identifying the associations of target words in the first vocabulary chain under different vocabulary attributes, an association network is constructed. Different representation values ​​are assigned based on the different vocabulary attributes of different target words, and the representation value of each association is determined. Then, an association representation matrix is ​​constructed to represent the associations between all target words in the first vocabulary chain. Based on the determination of keywords in the first vocabulary chain, the iteration center and iteration matrix are determined. The association representation matrix is ​​iterated by cumulative multiplication. Then, the associations that become 0 after each iteration are deleted. After a preset number of iterations, the stable associations between all target words in the first vocabulary chain are determined as the dividing criteria to divide the first vocabulary chain and obtain the corresponding target semantic structure.

[0112] The iterative relation matrix obtained in each iteration is:

[0113]

[0114] In the formula, D n Let A be the iterative relation matrix obtained in the nth iteration, B be the correlation iterative relation matrix, and i be the order number of the current iteration process.

[0115] The beneficial effects of the above technology are: when the vocabulary size of the first vocabulary chain exceeds the vocabulary size threshold, the current scene category can be determined based on the target vocabulary and semantic structure in the vocabulary chain.

[0116] Example 7:

[0117] Based on Example 1, the semantic recognition method for specialized vocabulary in the automotive finance field, S3: determining the specialized vocabulary recognition model corresponding to the latest target scenario based on the current scenario category chain, including:

[0118] Based on the coherence analysis results of the current scene category chain, the latest target scene is determined in real time;

[0119] Based on the list of scene-specific vocabulary recognition models, the specific vocabulary recognition model corresponding to the latest target scene is determined.

[0120] In this embodiment, the coherence analysis result is the result obtained after performing a coherence analysis on the current scene chain.

[0121] In this embodiment, the latest target scenario is the automotive finance semantic scenario that is most recently determined based on the coherence analysis results of the current scenario category chain.

[0122] In this embodiment, the scenario-specific vocabulary recognition model list is a list that includes automotive finance semantic scenarios and corresponding specific vocabulary recognition models.

[0123] The beneficial effects of the above technologies are: to accurately determine the automotive finance semantic scenario corresponding to the environment to be identified based on the coherence analysis results of the current scenario category chain, and thus to determine the professional vocabulary recognition model for the automotive finance semantic scenario corresponding to the environment to be identified.

[0124] Example 8:

[0125] Based on Example 7, the semantic recognition method for specialized vocabulary in the automotive finance field, based on the coherence analysis results of the current scene category chain, determines the latest target scene in real time, including:

[0126] Perform coherence analysis on the current scene category chain and determine whether there are lexical subchains in the current scene category chain that meet the coherence requirements, and obtain the coherence analysis results.

[0127] The latest target scene is determined in real time based on the results of the coherence analysis.

[0128] In this embodiment, the vocabulary subchain is the partial chain structure that meets the coherence requirement contained in the current scene chain.

[0129] In this embodiment, a coherence analysis is performed on the current scene category chain, and it is determined whether there is a lexical subchain in the current scene category chain that meets the coherence requirement, including:

[0130] Identify consecutive consistent scene category subchains in the current scene chain, and determine the ratio of the total number of consecutive consistent scene categories in the current scene category subchain to the total number of scene categories in the current scene category chain as the consecutive proportion of the corresponding scene category subchain, i.e.:

[0131]

[0132] In the formula, α is the continuous proportion of the corresponding scene category subchain, S is the ordinal number of the first scene category in the scene category subchain corresponding to the continuous consistent scene category in the current scene category chain (i.e., the first scene category in the scene category subchain corresponding to the continuous consistent scene category is located at the Sth position in the current scene category chain), E is the ordinal number of the last scene category in the scene category subchain corresponding to the continuous consistent scene category in the current scene category chain (i.e., the last scene category in the scene category subchain corresponding to the continuous consistent scene category is located at the Eth position in the current scene category chain), M is the total number of continuous consistent scene categories contained in the scene category subchain, and N is the total number of scene categories contained in the current scene category chain.

[0133] For example, if S is 1, E is 10, M is 10, and N is 100, then α is 0.155.

[0134] Determine whether the continuous proportion is not less than the continuous proportion threshold. If so, determine that the corresponding scenario category subchain is a word subchain that meets the coherence requirement; otherwise, determine that the corresponding scenario category subchain is not a word subchain that meets the coherence requirement.

[0135] In this embodiment, the latest target scene is determined in real time based on the coherence analysis results, including:

[0136] Based on the coherence analysis results, determine whether there is a word subchain that meets the coherence requirement in the current scene category chain. If so, determine whether there is only one word subchain that meets the coherence requirement. If so, take the continuous and consistent scene category corresponding to the corresponding word subchain as the latest target scene. Otherwise, take the continuous and consistent scene category corresponding to the word subchain with the largest continuous proportion as the latest target scene.

[0137] If there is no word subchain in the current scene category chain that meets the coherence requirement, then the scene category that appears most frequently in the current scene category chain will be taken as the latest target scene.

[0138] The beneficial effects of the above technology are: to determine the latest target scene by judging whether the current scene chain contains a word subchain that meets the coherence requirement, so that the scene category can be accurately determined based on the coherence of the scene category determination result in the current scene category chain.

[0139] Example 9:

[0140] Based on Example 1, the semantic recognition method for specialized vocabulary in the automotive finance field, S4: performing semantic recognition on the newly obtained first vocabulary chain based on the specialized vocabulary recognition model until a complete semantic recognition result is obtained and output, including:

[0141] Based on a dedicated vocabulary recognition model, semantic recognition is performed on the newly obtained first vocabulary chain to obtain semantic recognition results. It is then determined whether the semantic recognition results meet the recognition termination judgment condition corresponding to the latest target scene. If so, the complete semantic recognition result is determined and output based on the semantic recognition results.

[0142] Otherwise, continue to acquire new target words in the current environment to be identified until a complete semantic recognition result is determined and output.

[0143] In this embodiment, the semantic recognition result is the result obtained after performing semantic recognition on the newly obtained first vocabulary chain based on a dedicated vocabulary recognition model.

[0144] In this embodiment, the identification termination determination condition is the discrimination condition for determining whether to stop identification corresponding to different scene categories. For example, if the latest obtained semantic recognition result contains the termination word contained in the corresponding scene category, then the identification termination determination condition is the termination word of the corresponding scene category.

[0145] In this embodiment, the complete semantic recognition result is the semantic recognition result when the latest obtained semantic recognition result satisfies the recognition termination judgment condition corresponding to the latest target scene.

[0146] The beneficial effects of the above technology are as follows: Based on the determined latest target scenario, the corresponding recognition termination judgment condition is determined, thereby enabling different recognition termination judgment conditions to be used for different auto finance scenario categories, making the semantic recognition termination judgment more accurate, and also making the semantic recognition result more accurate.

[0147] Example 10:

[0148] Based on Example 9, the semantic recognition method for specialized vocabulary in the automotive finance field continues to acquire new target words in the current environment to be recognized until a complete semantic recognition result is determined and output, including:

[0149] Continue to acquire new target words in the current environment to be identified, generate a new first word chain based on the new target words, and determine a new dedicated word recognition model based on the first word chain;

[0150] Based on the new dedicated vocabulary recognition model, semantic recognition is performed on the new first vocabulary chain to obtain new semantic recognition results. This process continues until the latest semantic recognition result meets the recognition termination judgment condition corresponding to the latest target scene. Then, the complete semantic recognition result is determined and output based on the latest semantic recognition result.

[0151] The beneficial effects of the above technology are: when the latest semantic recognition result does not meet the recognition termination judgment condition corresponding to the latest target scene, the target words in the recognition environment continue to be recognized until a complete semantic recognition result is determined, thus achieving complete semantic recognition of the target words in the recognition environment.

[0152] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A method for semantic recognition of specialized terminology in the field of auto finance, characterized in that, include: S1: Generate the first vocabulary chain in real time based on the target words identified in the current environment to be identified; S2: Based on the first vocabulary chain and the auto finance scenario recognition model, the current scenario category chain is determined in real time, including: S201: Determine the current scene category based on the first lexical chain, the lexical-scene association model, and the semantic structure-scene association model; S202: Sort all current scene categories according to their order of appearance to obtain the current scene category chain; S3: Determine the dedicated vocabulary recognition model corresponding to the latest target scene based on the current scene category chain; S4: Perform semantic recognition on the newly obtained first vocabulary chain based on a dedicated vocabulary recognition model until a complete semantic recognition result is obtained and output; This also includes: Based on all remaining associations contained in the iterative association network, the target words contained in the first lexical chain are associated and divided to obtain the target semantic structure of the first lexical chain. By identifying the relationships between target words in the first lexical chain under different lexical attributes, a relational network is constructed. Different representation values ​​are assigned based on the different lexical attributes of different target words, and the representation value of each relation is determined. Then, a relational representation matrix is ​​constructed to represent the relationships between all target words in the first lexical chain. Based on the identification of keywords in the first lexical chain, the iteration center and iteration matrix are determined. The relational representation matrix is ​​iterated by cumulative multiplication. Relationships that become 0 after each iteration are deleted. After a preset number of iterations, the remaining relationships between all target words in the first lexical chain are used as the dividing criterion to divide the first lexical chain and obtain the corresponding target semantic structure.

2. The semantic recognition method for specialized terms in the automotive finance field according to claim 1, characterized in that, S1: Based on the target words identified in the current environment to be identified, generate the first word chain in real time, including: S101: Identify the target words in the current environment to be identified; S102: Sort the identified target words according to their order of appearance to obtain the first word chain.

3. The semantic recognition method for specialized terms in the automotive finance field according to claim 1, characterized in that, S201: Based on the first lexical chain, the lexical-scene association model, and the semantic structure-scene association model, the current scene category is determined, including: Determine the vocabulary size in the first vocabulary chain; When the vocabulary size does not exceed the vocabulary size threshold, the current scene category is determined based on the first vocabulary chain and the vocabulary-scene association model. When the vocabulary size exceeds the vocabulary size threshold, the current scene category is determined based on the first vocabulary chain, the vocabulary-scene association model, and the semantic structure-scene association model.

4. The semantic recognition method for specialized terms in the automotive finance field according to claim 3, characterized in that, When the vocabulary size does not exceed the vocabulary threshold, the current scene category is determined based on the first vocabulary chain and the vocabulary-scene association model, including: When the vocabulary size does not exceed the vocabulary size threshold, the scene category corresponding to each target word is determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model. The scene category that appears most frequently among all the scene categories corresponding to the target words is taken as the current scene category.

5. The semantic recognition method for specialized terms in the automotive finance field according to claim 3, characterized in that, When the vocabulary size exceeds the vocabulary threshold, the current scene category is determined based on the first vocabulary chain, the vocabulary-scene association model, and the semantic structure-scene association model, including: When the vocabulary exceeds the vocabulary threshold, the first scene category corresponding to each target word is determined based on the target words contained in the first vocabulary chain and the vocabulary-scene association model, and the second scene category is determined based on the target semantic structure of the first vocabulary chain and the semantic structure-scene association model. The scene category that appears most frequently among all first and second scene categories is selected as the current scene category.

6. The semantic recognition method for specialized terms in the field of automotive finance according to claim 1, characterized in that, S3: Based on the current scene category chain, determine the dedicated vocabulary recognition model corresponding to the latest target scene, including: Based on the coherence analysis results of the current scene category chain, the latest target scene is determined in real time; Based on the list of scene-specific vocabulary recognition models, the specific vocabulary recognition model corresponding to the latest target scene is determined.

7. The semantic recognition method for specialized terms in the automotive finance field according to claim 6, characterized in that, Based on the coherence analysis results of the current scene category chain, the latest target scene is determined in real time, including: Perform coherence analysis on the current scene category chain and determine whether there are lexical subchains in the current scene category chain that meet the coherence requirements, and obtain the coherence analysis results. The latest target scene is determined in real time based on the results of the coherence analysis.

8. The semantic recognition method for specialized terms in the field of automotive finance according to claim 1, characterized in that, S4: Perform semantic recognition on the newly obtained first vocabulary chain based on a dedicated vocabulary recognition model until a complete semantic recognition result is obtained and output, including: Based on a dedicated vocabulary recognition model, semantic recognition is performed on the newly obtained first vocabulary chain to obtain semantic recognition results. It is then determined whether the semantic recognition results meet the recognition termination judgment condition corresponding to the latest target scene. If so, the complete semantic recognition result is determined and output based on the semantic recognition results. Otherwise, continue to acquire new target words in the current environment to be identified until a complete semantic recognition result is determined and output.

9. The semantic recognition method for specialized terms in the field of automotive finance according to claim 8, characterized in that, Continue acquiring new target words in the current environment to be identified until a complete semantic recognition result is determined and output, including: Continue to acquire new target words in the current environment to be identified, generate a new first word chain based on the new target words, and determine a new dedicated word recognition model based on the first word chain; Based on the new dedicated vocabulary recognition model, semantic recognition is performed on the new first vocabulary chain to obtain new semantic recognition results. This process continues until the latest semantic recognition result meets the recognition termination judgment condition corresponding to the latest target scene. Then, the complete semantic recognition result is determined and output based on the latest semantic recognition result.

Images