Rail transit index question answering method and system based on large model

Abstract

The invention relates to a rail transit index question answering method and system based on a large model, and the method comprises the steps: carrying out the index scrophularization disassembly of an obtained rail transit index based on an index statistical dimension, and obtaining an original index parameter; performing marking processing of business scene labels and post role labels and association processing of the business scene labels and the post role labels on the original index parameters to obtain final index parameters to construct a basic database, and performing butt joint on the basic database and a double-track data channel to realize index data intercommunication. The method comprises the following steps: carrying out business scene identification and post role identification on a first question input by a user through an agent, carrying out dimension keyword extraction on the basis of an index statistics dimension, and calculating a query depth according to an obtained first business scene, a first post role and the dimension keyword; and querying in the basic database according to the dimension keyword and the query depth, and returning a query result to the user. Therefore, the accuracy of rail transit index question answering is improved.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a question-and-answer method and system for rail transit indicators based on a large model. Background Technology

[0002] Currently, indicator-based question-answering systems mainly focus on text interpretation and knowledge delivery, outputting content combinations based on knowledge bases by matching prompt words. However, question-answering about rail transit indicators requires not only text understanding but also statistical analysis. The question-answering results generated by traditional indicator-based question-answering systems cannot meet the needs of indicator-based question-answering in rail transit scenarios, thus affecting user experience. Summary of the Invention

[0003] The technical problem to be solved by this invention is: This invention provides a method and system for question answering rail transit indicators based on a large model. When facing question answering of indicators in rail transit scenarios, it performs statistical analysis to generate accurate question answering results, meet user needs, and optimize user experience.

[0004] To solve the above-mentioned technical problems, the technical solution adopted by the present invention is as follows:

[0005] Firstly, this invention provides a question-and-answer method for rail transit indicators based on a large model, comprising:

[0006] The process involves acquiring rail transit indicators, decomposing these indicators into raw parameters based on statistical dimensions, tagging them with business scenario labels and job role labels, as well as associating these labels to obtain final indicator parameters. A basic database is then constructed based on these final indicator parameters, and the database is connected to the dual-track data channel to achieve data exchange and obtain the connected basic database.

[0007] The system receives a first question input by the user, identifies the business scenario and job role through an intelligent agent, obtains the first business scenario and the first job role, and extracts dimension keywords from the first question based on the statistical dimensions of the indicators, obtains all dimension keywords, calculates the query depth based on the first business scenario, the first job role and all the dimension keywords, and performs a query in the integrated basic database according to the query depth based on all the dimension keywords to obtain the query results, and returns the query results to the user.

[0008] The beneficial effects of this invention are as follows: The final indicator parameters for constructing the basic database are obtained by decomposing the indicator parameters of rail transit indicators through indicator statistical dimensions. This ensures that the basic database can perform statistical analysis when answering questions about rail transit indicators. Furthermore, the basic database is connected to the dual-track data channel to ensure the comprehensiveness and integrity of the resulting basic database. When receiving the first question input by the user, the intelligent agent identifies the business scenario and job role of the first question. Based on the indicator statistical dimensions, the agent extracts dimensional keywords from the first question. Unlike traditional question-answering systems that simply match based on prompt words, this invention further divides the business scenario and job role, combining the dimensional keywords with the calculated query depth to obtain the query results. This improves the accuracy of the query results while making them more user-friendly and optimizing the user experience.

[0009] Optionally, obtaining rail transit indicators includes:

[0010] The rail transit indicators are expanded to obtain expanded rail transit indicators. The expansion process includes: synonym expansion, related word expansion, associative word expansion, and alternative name expansion.

[0011] As described above, rail transit indicators will undergo diverse expansion processing to improve their comprehensiveness and completeness, enabling the basic database to adapt to diverse scenario requirements and optimize user experience.

[0012] Optionally, the step of connecting the basic database with the dual-track data channel to achieve data exchange of indicators, resulting in a connected basic database, includes:

[0013] The basic database is connected to the dual-track data channel to achieve data exchange of indicators. The indicator data obtained from the dual-track data channel is input into the basic database according to the input parameter specification of the final indicator parameters of the basic database. At the same time, the final indicator parameters of the basic database are updated according to the indicator data to obtain the connected basic database.

[0014] The dual-track data channel includes: a structured data channel and an unstructured data channel, wherein the structured data channel includes a rail transit production system and a rail transit production support system, and the unstructured data channel includes an internet system.

[0015] As described above, the basic database's indicator data sources include both structured and unstructured data channels. The structured data channel ensures the specialization of the basic database's indicator data, while the unstructured data channel ensures the comprehensiveness of the basic database. Furthermore, when inputting data parameters, the input parameters are always performed according to the final indicator's input parameter specifications to ensure the standardization of the indicator data. The final indicator parameters are also updated based on the indicator data, further improving the real-time performance and completeness of the basic database.

[0016] Optionally, the step of identifying the business scenario and job role through the intelligent agent to obtain the first business scenario and the first job role includes:

[0017] The first question is preprocessed by an intelligent agent, including semantic analysis and text supplementation, to obtain a preprocessed first question. Business keywords and job keywords are extracted from the preprocessed first question to obtain corresponding business keywords and job keywords. A first business scenario is obtained based on the business keywords, and a first job role is obtained based on the job keywords.

[0018] As described above, when the agent identifies the business scenario and job role in the first question, it first performs semantic analysis and text supplementation to improve the completeness of the first question, thereby improving the accuracy of the extracted business keywords and job keywords, and ensuring the accuracy of the obtained first business scenario and first job role.

[0019] Optionally, calculating the query depth based on the first business scenario, the first job role, and all the dimension keywords includes:

[0020] The first job role is matched with the job role tags in the integrated basic database to obtain the matched job role tags. All related business scenario tags are obtained based on the matched job role tags. From all related business scenario tags, business scenario tags that match the first business scenario are filtered out to obtain all filtered business scenario tags.

[0021] Obtain the process participation degree and business participation degree of the first job role in each filtered business scenario tag. Input the process participation degree and business participation degree into the first formula to calculate the business role correlation degree, and obtain the correlation degree of all business roles. The first formula is:

[0022]

[0023] in, Let i be the business role relevance of the first job role i in the j-th business scenario tag after filtering. This represents the degree of process participation of role i in the first position within the j-th business scenario tag after screening. q represents the business participation degree of the first job role i in the j-th business scenario label after screening, and q represents the basic confidence level.

[0024] The query depth is calculated based on the relevance of all business roles and the keywords of all dimensions.

[0025] As described above, by combining the first business scenario and the first job role with the integrated basic database, the process participation and business participation of the first job role in the business scenario tags associated with and corresponding to the first business scenario can be obtained. This ensures that the calculated business role correlation considers not only the business aspect but also the process aspect, thereby improving the accuracy of the business role correlation and thus improving the accuracy of the query depth calculated based on the business role correlation and dimension keywords.

[0026] Optionally, calculating the query depth based on the relevance of all business roles and all the dimension keywords includes:

[0027] Match all the aforementioned dimension keywords with the final indicator parameters in the integrated basic database to obtain the matched final indicator parameters;

[0028] Obtain the association confidence level corresponding to the association degree of each business role from the pre-set confidence level table, and obtain the number of parameters of the final indicator parameters after matching under each association confidence level. Multiply the number of parameters of each parameter by the corresponding association confidence level to obtain the scope of all business cognition.

[0029] All business knowledge ranges are sorted in descending order to obtain the sorted business knowledge ranges. The sorted business knowledge ranges are then concatenated to construct a business knowledge trapezoid. The height of the business knowledge trapezoid is calculated to obtain the query depth.

[0030] As described above, the association confidence level corresponding to each business role association degree is multiplied by the number of parameters of the final indicator parameters that match the dimension keywords under that association confidence level to ensure the accuracy and rationality of the obtained business cognition scope. The business cognition scopes after descending order are then concatenated to construct a business cognition trapezoid. The height of the business cognition trapezoid is calculated to obtain the query depth, that is, to realize the vertical mining of the business cognition scope.

[0031] Optionally, the query depth includes vertical depth and horizontal depth. The step of concatenating the business knowledge ranges after descending order to construct a business knowledge trapezoid, and calculating the trapezoidal height to obtain the query depth, includes:

[0032] The business knowledge scope that is ranked first in the descending order is concatenated with the business knowledge scope that is ranked second to construct the first business knowledge trapezoid.

[0033] The business role relevance corresponding to the first business cognition scope and the business role relevance corresponding to the second business cognition scope are input into the second formula to calculate the trapezoidal height, thus obtaining the first height of the first business cognition trapezoid. Simultaneously, the length of the first diagonal of the first business cognition trapezoid is obtained. The first diagonal length and the first height are then input into the third formula to calculate the concatenation value, thus obtaining the first concatenation value. The second formula is:

[0034]

[0035] Among them, h n This represents the first level when the scope of business knowledge is n, where n represents the scope of business knowledge. The business role relevance of the first job role i in the j-th business scenario tag after filtering;

[0036] The third formula is:

[0037]

[0038] Where L represents the first series value, h represents the first height, and c represents the first diagonal length;

[0039] Determine whether the first concatenation value is greater than the first threshold. If not, take the first height as the vertical depth. If yes, concatenate the first business cognition trapezoid with the business cognition range located in the third position to obtain a new first business cognition trapezoid and recalculate the first concatenation value to obtain a new first concatenation value. If the new first concatenation value is greater than the first threshold, concatenate the new first business cognition trapezoid with the business cognition range located in the fourth position and recalculate the first concatenation value until the new first concatenation value is less than or equal to the first threshold.

[0040] The trapezoidal volume of the first business cognition trapezoid corresponding to the vertical depth is calculated, and the trapezoidal volume is used as the horizontal depth.

[0041] As described above, the query depth includes vertical depth and horizontal depth, and the vertical depth is related to the horizontal depth. The trapezoidal volume of the first business cognition trapezoid calculated based on the vertical depth is used as the horizontal depth to ensure the rationality of the query depth. Furthermore, the vertical depth and horizontal depth are controlled by the first concatenation value calculated based on the first height and the first diagonal length of the first business cognition trapezoid. This achieves robustness optimization of the agent while improving the accuracy of the query depth.

[0042] Optionally, the step of calculating the trapezoidal volume of the corresponding first business cognition trapezoid based on the vertical depth, and using the trapezoidal volume as the horizontal depth, includes:

[0043] The agent calculates the state transition probability and reward function value based on the vertical depth and the horizontal depth. It then dynamically converges the horizontal depth using the state transition probability, reward function value, and a dynamic convergence formula. During this dynamic convergence process, preset constraints are applied to obtain the dynamically converged horizontal depth. The dynamic convergence formula is as follows:

[0044] V max =argmax{V k+1 (h k+1 )};

[0045]

[0046] J k+1 =ω×V k+1 (h k+1 );

[0047] γ = O(α,β);

[0048] α=O(n×2 n );

[0049] β=O(2 n );

[0050] Among them, V max Z represents the lateral depth after dynamic convergence. k+1 J represents the state transition probability corresponding to the first height when the number of business cognition ranges is k+1; J represents the number of times the business cognition ranges are chained together when the number of business cognition ranges is k+1; J represents the total number of times the trapezoidal volume is calculated when the number of business cognition ranges is k. K+1 This represents the reward function value corresponding to the first height when the number of business knowledge scopes is k+1, where ω represents the reward weight, γ represents the discount coefficient, α represents the time dimension parameter, β represents the space dimension parameter, O() represents the complexity function, and h kV represents the first height when the number of business knowledge areas is k, where n represents the number of business knowledge areas. k (h k () indicates that the number of business knowledge scopes is k and the first height is h. k The corresponding lateral depth, h k+1 V represents the first height when the number of business knowledge areas is k+1. k+1 (h k+1 () indicates that the number of business knowledge scopes is k+1 and the first height is h. k+1 The corresponding lateral depth;

[0051] The preset constraint is as follows:

[0052] V k+1 (h k+1 -h k )+V k (h k )≤V n (h n )

[0053] Among them, V k+1 (h k+1 -h k () indicates that the number of business knowledge scopes is k+1 and the first height is h. k+1 -h k The corresponding lateral depth, V n (h n () indicates that the number of business knowledge areas is n and the first height is h. n The corresponding horizontal depth.

[0054] As described above, since the trapezoidal volume of the first business cognition trapezoid calculated based on the vertical depth is used as the horizontal depth, the horizontal depth is dynamically converged by the state transition probability and the reward function value. This avoids the trapezoidal volume being too small given the vertical depth, i.e., the business role association being too small, which would affect the accuracy of the query results.

[0055] Optionally, the query results include the current query results and follow-up prompts. The step of querying the underlying database according to the query depth based on all the stated dimension keywords to obtain the query results, and then returning the query results to the user, includes:

[0056] Obtain the initial query range input by the user, and perform a query in the connected basic database according to the query depth based on the initial query range and all the dimension keywords to obtain the current query result;

[0057] The initial query interval is enhanced based on the query depth to obtain an enhanced initial query interval. Based on the enhanced initial query interval and all the dimension keywords, a query is performed in the integrated basic database according to the query depth to obtain follow-up prompt results.

[0058] The current query results and the follow-up prompts are returned to the user.

[0059] As described above, the query results include not only the current query results but also follow-up question suggestions. Follow-up question suggestions can help users construct their next input question, improve the user's input question, assist the user in constructing the question, optimize the user experience, and the follow-up question suggestions are obtained by enhancing the initial query range according to the query depth, thereby improving the rationality and accuracy of the follow-up question suggestions.

[0060] Secondly, the present invention provides a rail transit indicator question-and-answer system based on a large model, 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, it implements the rail transit indicator question-and-answer method based on a large model as described in the first aspect.

[0061] The technical effects of the rail transit indicator question-and-answer system based on a large model provided in the second aspect are the same as those of the rail transit indicator question-and-answer method based on a large model provided in the first aspect. Attached Figure Description

[0062] Figure 1 A flowchart illustrating a large-model-based question-and-answer method for rail transit indicators provided in this embodiment;

[0063] Figure 2 This is a schematic diagram of the overall process of a rail transit indicator question-and-answer method based on a large model provided in this embodiment;

[0064] Figure 3 This is a schematic diagram illustrating the calculation of query depth in this embodiment;

[0065] Figure 4 This is a schematic diagram illustrating the construction of a business cognition trapezoid involved in this embodiment;

[0066] Figure 5 This is a schematic diagram of the structure of a rail transit indicator question-and-answer system based on a large model provided in this embodiment.

[0067] [Explanation of Labels in the Attached Image]

[0068] 1. A question-and-answer system for rail transit indicators based on a large model;

[0069] 2. Processor;

[0070] 3. Memory. Detailed Implementation

[0071] To better understand the above technical solutions, exemplary embodiments of the present invention will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of the present invention are shown in the drawings, it should be understood that the present invention can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that the present invention can be understood more clearly and thoroughly, and that the scope of the present invention can be fully conveyed to those skilled in the art.

[0072] Example 1

[0073] Please refer to Figures 1 to 4 This invention provides a question-and-answer method for rail transit indicators based on a large model, comprising the following steps:

[0074] S1. Obtain rail transit indicators, decompose the rail transit indicators into indicator parameters based on the indicator statistical dimensions to obtain the original indicator parameters, mark the original indicator parameters with business scenario tags and job role tags, and associate the business scenario tags and job role tags to obtain the final indicator parameters, build a basic database based on the final indicator parameters, and connect the basic database with the dual-track data channel to realize the interoperability of indicator data, and obtain the connected basic database;

[0075] At this point, obtaining rail transit indicators in step S1 includes:

[0076] S11. The rail transit index is extended to obtain the extended rail transit index. The extension process includes: synonym extension, related word extension, associative word extension, and alternative name extension.

[0077] In this embodiment, as Figure 2As shown, the process involves acquiring rail transit indicators and expanding them using synonyms, related terms, and alternative names to obtain the expanded indicators. Based on statistical dimensions, these expanded indicators are then decomposed into raw indicator parameters. These statistical dimensions include indicator name, unit, time, and space. The raw indicator parameters are then tagged with business scenario labels and job role labels, and these labels are associated with each other. This means the raw indicator parameters contain business scenario labels, job role labels, and their association relationships, resulting in the final indicator parameters. The tagging process can be performed using an intelligent agent. Finally, a basic database is built based on the final indicator parameters and connected to the dual-track data channel to achieve data exchange, resulting in the connected basic database.

[0078] At this point, step S1 involves connecting the basic database with the dual-track data channel to achieve data exchange of indicators. The resulting basic database includes:

[0079] S12. Connect the basic database with the dual-track data channel to achieve data exchange of indicators. Input the indicator data obtained from the dual-track data channel according to the input parameter specification of the final indicator parameters of the basic database. At the same time, update the final indicator parameters of the basic database according to the indicator data to obtain the connected basic database.

[0080] The dual-track data channel includes: a structured data channel and an unstructured data channel, wherein the structured data channel includes a rail transit production system and a rail transit production support system, and the unstructured data channel includes an internet system.

[0081] In this embodiment, as Figure 2As shown, the basic database is connected to structured data channels, including those for rail transit production systems and rail transit production support systems, as well as to unstructured data channels, including those for the internet. Specifically, the specialized indicator data from the structured data channels and the internet indicator data from the unstructured data channels serve as the data sources for the basic database. Indicator data obtained from both data channels is input according to the final indicator parameter input specifications of the basic database. Before inputting the data, the indicator data undergoes cleaning processes, such as deleting abnormal data and filtering duplicate data. During input, indicator data from different sources is weighted; for example, the weight of indicator data from the structured data channel is set higher than that from the unstructured data channel. When different indicator data exist under the same statistical dimension, the indicator data from the structured data channel takes precedence. The cleaning and weighting processes can be adjusted according to actual conditions. Simultaneously, the final indicator parameters of the basic database are updated based on the indicator data. For example, when new synonyms, related words, or alternative names appear, the final indicator parameters are updated, resulting in the connected basic database.

[0082] S2. Receive the first question input by the user, and use the intelligent agent to identify the business scenario and job role of the first question to obtain the first business scenario and the first job role. At the same time, extract dimension keywords from the first question based on the statistical dimensions of the indicators to obtain all dimension keywords. Calculate the query depth based on the first business scenario, the first job role and all the dimension keywords. Perform a query in the connected basic database according to the query depth based on all the dimension keywords to obtain the query results, and return the query results to the user.

[0083] At this point, step S2, which involves using an agent to identify the business scenario and job role of the first question to obtain the first business scenario and the first job role, includes:

[0084] S21. The first question is preprocessed by the agent, including semantic analysis and text supplementation, to obtain the preprocessed first question. Business keywords and job keywords are extracted from the preprocessed first question to obtain the corresponding business keywords and job keywords. The first business scenario is obtained based on the business keywords, and the first job role is obtained based on the job keywords.

[0085] In this embodiment, as Figure 2As shown, the system receives the first question input by the user. An intelligent agent performs preprocessing on the first question, including semantic analysis and text supplementation. During semantic analysis and text supplementation, the agent first obtains the user's initial identity from the first question. The user's initial identity is selected autonomously when inputting the first question. If the user does not select an initial identity, it is determined based on the user's historical questions. The user's initial identity includes: an initial identity with professional knowledge of rail transit and an initial identity understanding the meaning of rail transit industry indicators. Based on the obtained initial identity, semantic analysis and text supplementation are performed on the first question to obtain the preprocessed first question. Business keywords and job keywords are extracted from the preprocessed first question to obtain the corresponding business keywords and job keywords, thereby obtaining the first business scenario and the first job role. The first job role can also be obtained by obtaining the user's registered account information. Simultaneously, dimensional keywords are extracted from the preprocessed first question based on the statistical dimensions of the indicators to obtain all dimensional keywords. The query depth is calculated based on the first business scenario, the first job role, and all dimensional keywords. The query is then performed in the integrated basic database according to the query depth based on all dimensional keywords to obtain the query results, which are then returned to the user.

[0086] At this point, the step S2, which involves calculating the query depth based on the first business scenario, the first job role, and all the aforementioned dimension keywords, includes:

[0087] S22. Match the first job role with the job role tags in the integrated basic database to obtain the matched job role tags, and obtain all related business scenario tags based on the matched job role tags. Filter out the business scenario tags that match the first business scenario from all related business scenario tags to obtain all filtered business scenario tags.

[0088] S23. Obtain the process participation degree and business participation degree of the first job role in each filtered business scenario tag, and input the process participation degree and business participation degree into the first formula to calculate the business role correlation degree, thereby obtaining the correlation degree of all business roles. The first formula is:

[0089]

[0090] in, Let i be the business role relevance of the first job role i in the j-th business scenario tag after filtering. This represents the degree of process participation of role i in the first position within the j-th business scenario tag after screening. q represents the business participation degree of the first job role i in the j-th business scenario label after screening, and q represents the basic confidence level.

[0091] S24. Calculate the query depth based on the relevance of all business roles and the keywords of all dimensions.

[0092] In this embodiment, as Figure 2 As shown, the first job role is matched with the job role tags in the integrated basic database to obtain the matched job role tags, thereby obtaining all related business scenario tags. From all related business scenario tags, the business scenario tags that match the first business scenario are filtered out to obtain all filtered business scenario tags. By obtaining the process participation degree and business participation degree of the first job role in each filtered business scenario tag and combining them with the first formula, the corresponding business role relevance is calculated. Since a first job role may have N filtered business scenario tags, N business role relevances will be obtained. The query depth is then calculated based on all business role relevances and all dimension keywords.

[0093] At this point, step S24 includes:

[0094] S241. Match all the aforementioned dimension keywords with the final indicator parameters in the integrated basic database to obtain the matched final indicator parameters;

[0095] S242. Obtain the association confidence level corresponding to the association degree of each business role from the pre-set confidence level table, and obtain the number of parameters of the final indicator parameters after matching under each association confidence level. Multiply the number of parameters of each parameter by the corresponding association confidence level to obtain the scope of all business cognition.

[0096] S243. Perform descending order processing on all business cognition ranges to obtain the descending order business cognition ranges. Concatenate the descending order business cognition ranges to construct a business cognition trapezoid. Calculate the trapezoid height of the business cognition trapezoid to obtain the query depth.

[0097] In this embodiment, as Figure 2As shown, all dimension keywords are matched with the final indicator parameters in the integrated basic database to obtain the matched final indicator parameters. The association confidence level corresponding to the association degree of each business role is obtained from the pre-set confidence level table, and the number of parameters of the matched final indicator parameters under each association confidence level is obtained. The number of parameters is multiplied by the corresponding association confidence level to obtain the scope of all business cognition. The larger the scope of business cognition, the stronger the association between the matched final indicator parameters and the association degree of business roles. All scopes of business cognition are sorted in descending order, that is, the earlier the sorted scopes are, the stronger the association. The scopes of business cognition are concatenated to construct a business cognition trapezoid, and the height of the business cognition trapezoid is calculated to obtain the query depth.

[0098] At this point, the query depth mentioned in step S243 includes vertical depth and horizontal depth. The step of concatenating the business knowledge ranges after descending order to construct a business knowledge trapezoid, and calculating the trapezoidal height of the business knowledge trapezoid to obtain the query depth, includes:

[0099] S2431. Prioritize connecting the first business cognition scope in the descending order with the second business cognition scope to construct the first business cognition trapezoid.

[0100] S2432. Input the business role relevance corresponding to the first business cognition scope and the business role relevance corresponding to the second business cognition scope into the second formula to calculate the trapezoidal height, thereby obtaining the first height of the first business cognition trapezoid. At the same time, obtain the length of the first diagonal of the first business cognition trapezoid. Input the first diagonal length and the first height into the third formula to calculate the concatenation value, thereby obtaining the first concatenation value. The second formula is:

[0101]

[0102] Among them, h n This represents the first level when the scope of business knowledge is n, where n represents the scope of business knowledge. The business role relevance of the first job role i in the j-th business scenario tag after filtering;

[0103] The third formula is:

[0104]

[0105] Where L represents the first series value, h represents the first height, and c represents the first diagonal length;

[0106] S2433. Determine whether the first concatenation value is greater than the first threshold. If not, take the first height as the vertical depth. If yes, concatenate the first business cognition trapezoid with the business cognition range located in the third position to obtain a new first business cognition trapezoid and recalculate the first concatenation value to obtain a new first concatenation value. If the new first concatenation value is greater than the first threshold, concatenate the new first business cognition trapezoid with the business cognition range located in the fourth position and recalculate the first concatenation value until the new first concatenation value is less than or equal to the first threshold.

[0107] S2434. Calculate the trapezoidal volume of the corresponding first business cognition trapezoid based on the vertical depth, and use the trapezoidal volume as the horizontal depth.

[0108] In this embodiment, as Figure 3 As shown, the query depth includes vertical depth and horizontal depth, such as... Figure 4 As shown, Figure 4 In this context, S1 represents the business knowledge range that is ranked first after descending order, S2 represents the business knowledge range that is ranked second after descending order, S3 represents the business knowledge range that is ranked third after descending order, and S4 represents the business knowledge range that is ranked fourth after descending order. First, the business knowledge range that is ranked first (S1) is concatenated with the business knowledge range that is ranked second (S2) to construct a first business knowledge trapezoid, as shown below. Figure 3 The trapezoid enclosed by the solid lines in the diagram is used to calculate the first height of the first cognitive trapezoid by inputting the business role relevance corresponding to the first and second business cognitive scopes into the second formula. Simultaneously, the length of the first diagonal of the first cognitive trapezoid is obtained. Based on the first diagonal length, the first height, and the third formula, a first concatenation value is calculated. If the first concatenation value is not greater than a first threshold, the first height is directly used as the vertical depth. Conversely, if the first concatenation value is greater than the first threshold, the first business cognitive trapezoid is concatenated with the third business cognitive scope S3 to obtain... The new first business cognition trapezoid is recalculated for the first concatenation value and compared with the first threshold. If the recalculated first concatenation value is still greater than the first threshold, the new first business cognition trapezoid is concatenated with the business cognition range S4 located in the fourth position. This process is repeated until the new first concatenation value is not greater than the first threshold, that is, less than or equal to the first threshold. The trapezoidal volume of the corresponding first business cognition trapezoid is calculated based on the obtained vertical depth, and the trapezoidal volume is used as the horizontal depth. The first threshold is equal to the first height of the current business cognition trapezoid / 3. That is, the first threshold will change dynamically according to the different business cognition trapezoids.

[0109] At this point, step S2434 includes:

[0110] S24341. The agent calculates the state transition probability and reward function value based on the vertical depth and the horizontal depth. It then dynamically converges the horizontal depth based on the state transition probability, reward function value, and dynamic convergence formula. During the dynamic convergence process, preset constraints are applied to obtain the dynamically converged horizontal depth. The dynamic convergence formula is:

[0111] V max =argmax{V k+1 (h k+1 )};

[0112]

[0113] J k+1 =ω×V k+1 (h k+1 );

[0114] γ = O(α,β);

[0115] α=O(n×2 n );

[0116] β=O(2 n );

[0117] Among them, V max Z represents the lateral depth after dynamic convergence. k+1 J represents the state transition probability corresponding to the first height when the number of business cognition ranges is k+1; J represents the number of times the business cognition ranges are chained together when the number of business cognition ranges is k+1; J represents the total number of times the trapezoidal volume is calculated when the number of business cognition ranges is k. K+1 This represents the reward function value corresponding to the first height when the number of business knowledge scopes is k+1, where ω represents the reward weight, γ represents the discount coefficient, α represents the time dimension parameter, β represents the space dimension parameter, O() represents the complexity function, and h k V represents the first height when the number of business knowledge areas is k, where n represents the number of business knowledge areas. k (h k () indicates that the number of business knowledge scopes is k and the first height is h. k The corresponding lateral depth, h k+1 V represents the first height when the number of business knowledge areas is k+1. k+1 (h k+1 () indicates that the number of business knowledge scopes is k+1 and the first height is h. k+1 The corresponding lateral depth;

[0118] The preset constraint is as follows:

[0119] V k+1 (h k+1 -h k )+V k (h k )≤V n (h n )

[0120] Among them, V k+1 (h k+1 -h k () indicates that the number of business knowledge scopes is k+1 and the first height is h. k+1 -h k The corresponding lateral depth, V n (h n () indicates that the number of business knowledge areas is n and the first height is h. n The corresponding horizontal depth.

[0121] In this embodiment, as Figure 3 As shown, considering that the trapezoidal volume obtained based on the vertical depth may be too small, i.e., the business role association may be too small, the agent dynamically converges the trapezoidal volume, i.e., the horizontal depth. The agent calculates the state transition probability and reward function value based on the vertical depth and the horizontal depth, and combines the dynamic convergence formula with preset constraints to obtain the dynamically converged horizontal depth. The reward weight in the dynamic convergence formula depends on the calculated state transition probability, and the corresponding reward weight is obtained from the preset weight table according to different state transition probabilities.

[0122] At this point, the query results in step S2 include the current query results and follow-up prompt results. The step of querying the base database according to the query depth based on all the dimension keywords to obtain the query results, and then returning the query results to the user, includes:

[0123] S25. Obtain the initial query range input by the user, and perform a query in the connected basic database according to the query depth based on the initial query range and all the dimension keywords to obtain the current query result;

[0124] S26. Enhance the initial query interval according to the query depth to obtain the enhanced initial query interval. Query the basic database after docking according to the query depth based on the enhanced initial query interval and all the dimension keywords to obtain follow-up prompts.

[0125] S27. Return the current query result and the follow-up prompt result to the user.

[0126] In this embodiment, as Figure 2 As shown, the query results include the current query results and follow-up question suggestions. When the user enters their first question, an initial query range can be set. Based on the user-set initial query range and all dimension keywords, the database after integration is queried according to the query depth to obtain the current query results. The current query results are the answer to the user's first question. The initial query range is enhanced according to the query depth, that is, the query depth is multiplied by the initial query range to obtain the enhanced initial query range. Based on the enhanced initial query range and all dimension keywords, the database after integration is queried according to the query depth to obtain the follow-up question suggestions. The follow-up question suggestions predict the content of the user's next question, prompting and assisting the user in constructing the content of the next question. When the current query results are returned to the user, they are output according to the preset standard template. When the follow-up question suggestions are returned to the user, they are output according to the preset non-standard model, thus returning the current query results and follow-up question suggestions to the user.

[0127] In this embodiment, the process also includes handling the user's next input question after receiving the current query result and follow-up prompt result. Based on the user's next input question, the accuracy of the previously generated current query result and follow-up prompt result is judged to optimize the agent. Furthermore, based on the next input question, the user's primary role and primary business scenario are further refined to construct a personal profile of the user. When the user asks questions later, the personal profile is directly called for question and answer processing, thereby improving the efficiency of question and answer processing.

[0128] Example 2

[0129] Please refer to Figure 5 The present invention provides a rail transit indicator question and answer system 1 based on a large model, including a memory 3, a processor 2, and a computer program stored on the memory 3 and run on the processor 2. When the processor 2 executes the computer program, it implements the steps in Embodiment 1.

[0130] Since the systems / devices described in the above embodiments of the present invention are systems / devices used to implement the methods of the above embodiments of the present invention, those skilled in the art can understand the specific structure and modifications of the systems / devices based on the methods described in the above embodiments of the present invention, and therefore will not be repeated here. All systems / devices used in the methods of the above embodiments of the present invention fall within the scope of protection of the present invention.

[0131] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of a computer program product embodied on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0132] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, as well as combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions.

[0133] It should be noted that any reference numerals placed between parentheses in the claims should not be construed as limiting the claims. The word "comprising" does not exclude the presence of components or steps not listed in the claims. The word "a" or "an" preceding a component does not exclude the presence of a plurality of such components. The invention can be implemented by means of hardware comprising several different components and by means of a suitably programmed computer. In claims that enumerate several means, several of these means may be embodied by the same hardware. The use of the terms first, second, third, etc., is merely for convenience of expression and does not indicate any order. These terms can be understood as part of the component names.

[0134] Furthermore, it should be noted that in the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Furthermore, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0135] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the claims should be interpreted to include both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0136] 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, then this invention should also include these modifications and variations.

Claims

1. A question-and-answer method for rail transit indicators based on a large model, characterized in that, include: The process involves acquiring rail transit indicators, decomposing these indicators into raw parameters based on statistical dimensions, tagging them with business scenario labels and job role labels, as well as associating these labels to obtain final indicator parameters. A basic database is then constructed based on these final indicator parameters, and the database is connected to the dual-track data channel to achieve data exchange and obtain the connected basic database. The system receives a first question input by the user, identifies the business scenario and job role through an intelligent agent, obtains the first business scenario and the first job role, and extracts dimension keywords from the first question based on the statistical dimensions of the indicators, obtains all dimension keywords, calculates the query depth based on the first business scenario, the first job role and all the dimension keywords, and performs a query in the integrated basic database according to the query depth based on all the dimension keywords to obtain the query results, and returns the query results to the user.

2. The rail transit indicator question-answering method based on a large model as described in claim 1, characterized in that, The acquisition of rail transit indicators includes: The rail transit indicators are expanded to obtain expanded rail transit indicators. The expansion process includes: synonym expansion, related word expansion, associative word expansion, and alternative name expansion.

3. The rail transit indicator question-answering method based on a large model as described in claim 1, characterized in that, The process of connecting the basic database with the dual-track data channel to achieve data exchange of indicators results in a basic database after the connection, which includes: The basic database is connected to the dual-track data channel to achieve data exchange of indicators. The indicator data obtained from the dual-track data channel is input into the basic database according to the input parameter specification of the final indicator parameters of the basic database. At the same time, the final indicator parameters of the basic database are updated according to the indicator data to obtain the connected basic database. The dual-track data channel includes: a structured data channel and an unstructured data channel, wherein the structured data channel includes a rail transit production system and a rail transit production support system, and the unstructured data channel includes an internet system.

4. The rail transit indicator question-answering method based on a large model as described in claim 1, characterized in that, The step of identifying the business scenario and job role through the intelligent agent in the first question to obtain the first business scenario and the first job role includes: The first question is preprocessed by an intelligent agent, including semantic analysis and text supplementation, to obtain a preprocessed first question. Business keywords and job keywords are extracted from the preprocessed first question to obtain corresponding business keywords and job keywords. A first business scenario is obtained based on the business keywords, and a first job role is obtained based on the job keywords.

5. The rail transit indicator question-answering method based on a large model as described in claim 1, characterized in that, The step of calculating the query depth based on the first business scenario, the first job role, and all the dimension keywords includes: The first job role is matched with the job role tags in the integrated basic database to obtain the matched job role tags. All related business scenario tags are obtained based on the matched job role tags. From all related business scenario tags, business scenario tags that match the first business scenario are filtered out to obtain all filtered business scenario tags. Obtain the process participation degree and business participation degree of the first job role in each filtered business scenario tag. Input the process participation degree and business participation degree into the first formula to calculate the business role correlation degree, and obtain the correlation degree of all business roles. The first formula is: in, Let i be the business role relevance of the first job role i in the j-th business scenario tag after filtering. This represents the degree of process participation of role i in the first position within the j-th business scenario tag after screening. q represents the business participation degree of the first job role i in the j-th business scenario label after screening, and q represents the basic confidence level. The query depth is calculated based on the relevance of all business roles and the keywords of all dimensions.

6. The rail transit indicator question-answering method based on a large model as described in claim 5, characterized in that, The calculation of query depth based on the relevance of all business roles and all dimension keywords includes: Match all the aforementioned dimension keywords with the final indicator parameters in the integrated basic database to obtain the matched final indicator parameters; Obtain the association confidence level corresponding to the association degree of each business role from the pre-set confidence level table, and obtain the number of parameters of the final indicator parameters after matching under each association confidence level. Multiply the number of parameters of each parameter by the corresponding association confidence level to obtain the scope of all business cognition. All business knowledge ranges are sorted in descending order to obtain the sorted business knowledge ranges. The sorted business knowledge ranges are then concatenated to construct a business knowledge trapezoid. The height of the business knowledge trapezoid is calculated to obtain the query depth.

7. The rail transit indicator question-answering method based on a large model as described in claim 6, characterized in that, The query depth includes vertical depth and horizontal depth. The process of concatenating the business knowledge ranges after descending order to construct a business knowledge trapezoid, and calculating the trapezoidal height to obtain the query depth, includes: The business knowledge scope that is ranked first in the descending order is concatenated with the business knowledge scope that is ranked second to construct the first business knowledge trapezoid. The business role relevance corresponding to the first business cognition scope and the business role relevance corresponding to the second business cognition scope are input into the second formula to calculate the trapezoidal height, thus obtaining the first height of the first business cognition trapezoid. Simultaneously, the length of the first diagonal of the first business cognition trapezoid is obtained. The first diagonal length and the first height are then input into the third formula to calculate the concatenation value, thus obtaining the first concatenation value. The second formula is: Among them, h n This represents the first level when the scope of business knowledge is n, where n represents the scope of business knowledge. The business role relevance of the first job role i in the j-th business scenario tag after filtering; The third formula is: Where L represents the first series value, h represents the first height, and c represents the first diagonal length; Determine whether the first concatenation value is greater than the first threshold. If not, take the first height as the vertical depth. If yes, concatenate the first business cognition trapezoid with the business cognition range located in the third position to obtain a new first business cognition trapezoid and recalculate the first concatenation value to obtain a new first concatenation value. If the new first concatenation value is greater than the first threshold, concatenate the new first business cognition trapezoid with the business cognition range located in the fourth position and recalculate the first concatenation value until the new first concatenation value is less than or equal to the first threshold. The trapezoidal volume of the first business cognition trapezoid corresponding to the vertical depth is calculated, and the trapezoidal volume is used as the horizontal depth.

8. The rail transit indicator question-answering method based on a large model as described in claim 7, characterized in that, The step of calculating the trapezoidal volume of the corresponding first business cognition trapezoid based on the vertical depth, and using the trapezoidal volume as the horizontal depth, includes: The agent calculates the state transition probability and reward function value based on the vertical depth and the horizontal depth. It then dynamically converges the horizontal depth using the state transition probability, reward function value, and a dynamic convergence formula. During this dynamic convergence process, preset constraints are applied to obtain the dynamically converged horizontal depth. The dynamic convergence formula is as follows: V max =argmax{V k+1 (h k+1 )}; J k+1 =ω×V k+1 (h k+1 ); γ = O(α,β); α=O(n×2 n ); β＝O(2 n )； Among them, V max Z represents the lateral depth after dynamic convergence. k+1 J represents the state transition probability corresponding to the first height when the number of business cognition ranges is k+1; J represents the number of times the business cognition ranges are chained together when the number of business cognition ranges is k+1; J represents the total number of times the trapezoidal volume is calculated when the number of business cognition ranges is k. K+1 This represents the reward function value corresponding to the first height when the number of business knowledge scopes is k+1, where ω represents the reward weight, γ represents the discount coefficient, α represents the time dimension parameter, β represents the space dimension parameter, O() represents the complexity function, and h k V represents the first height when the number of business knowledge areas is k, where n represents the number of business knowledge areas. k (h k () indicates that the number of business knowledge scopes is k and the first height is h. k The corresponding lateral depth, h k+1 V represents the first height when the number of business knowledge areas is k+1. k+1 (h k+1 () indicates that the number of business knowledge scopes is k+1 and the first height is h. k+1 The corresponding lateral depth; The preset constraint is as follows: V k+1 (h k+1 -h k )+V k (h k )≤V n (h n ) Among them, V k+1 (h k+1 -h k () indicates that the number of business knowledge scopes is k+1 and the first height is h. k+1 -h k The corresponding lateral depth, V n (h n () indicates that the number of business knowledge areas is n and the first height is h. n The corresponding horizontal depth.

9. The rail transit indicator question-answering method based on a large model as described in claim 1, characterized in that, The query results include the current query results and follow-up prompts. The process of querying the base database according to the query depth based on all the specified dimension keywords to obtain the query results, and then returning the query results to the user, includes: Obtain the initial query range input by the user, and perform a query in the connected basic database according to the query depth based on the initial query range and all the dimension keywords to obtain the current query result; The initial query interval is enhanced based on the query depth to obtain an enhanced initial query interval. Based on the enhanced initial query interval and all the dimension keywords, a query is performed in the integrated basic database according to the query depth to obtain follow-up prompt results. The current query results and the follow-up prompts are returned to the user.

10. A rail transit indicator question-and-answer system based on a large model, 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, it implements the method as described in any one of claims 1 to 9.

Images