Deep learning model numerical skill enhancement method and system for mathematical statistics

Abstract

The invention discloses a deep learning model numerical skill enhancement method and system for mathematical statistics. The method comprises the steps of data set construction, data enhancement, dimension perception enhancement, numerical reasoning optimization and model numerical skill enhancement. The invention relates to the technical field of deep learning enhancement, in particular to a deep learning model numerical skill enhancement method and a deep learning model numerical skill enhancement system for mathematical statistics. Further enhancing model training by constructing a dimension perception enhancement layer and a numerical reasoning optimization layer; carrying out data enhancement by adopting a situation learning method, and further constructing a situation background by defining situation contents of dimension perception and numerical reasoning as a knowledge base in subsequent steps; dimensionality perception enhancement is carried out by constructing a dimensionality perception enhancement knowledge base, and numerical reasoning optimization is carried out by constructing a numerical reasoning optimization layer.

Description

Technical Field

[0001] The present invention relates to the field of deep learning enhancement technology, and specifically to a method and system for enhancing the numerical skills of a deep learning model for mathematical statistics. Background Art

[0002] The method and system for enhancing the numerical skills of deep learning models in mathematical statistics aims to improve the performance of models in data analysis and inference by combining the advantages of deep learning and statistics. These methods usually include optimizing the structure and training process of the model to improve its ability to handle numerical data, for example, by expanding the training set through data enhancement technology, or using regularization methods to prevent overfitting, thereby improving the generalization ability of the model. Its role is to enhance the model's ability to recognize complex data patterns through more accurate and efficient numerical analysis, thereby providing more reliable solutions in the fields of prediction, classification, and decision support. These systems can not only help researchers and data scientists quickly gain insights, but also promote intelligent decision-making in practical applications.

[0003] However, in the existing methods for enhancing the numerical skills of deep learning models, there is a technical problem that the existing technologies for the numerical skills of deep learning models inevitably have misunderstandings about numerical units, which leads to a further decrease in the accuracy of numerical calculations and reasoning. Therefore, the numerical skills must be enhanced from both qualitative and quantitative perspectives at the same time; in the existing methods for enhancing the numerical skills of deep learning models, there is a technical problem that in addition to enhancing the basic model training method, the data itself needs to be enhanced and processed; in the existing dimensionality perception enhancement and numerical reasoning optimization methods, there is a technical problem that traditional methods focus on training multiple models to solve corresponding specific tasks, which inevitably leads to a decrease in the accuracy and usability of the superposition of multiple models, thereby reducing the numerical skills of a single deep learning model. Summary of the invention

[0004] In view of the above situation, in order to overcome the defects of the prior art, the present invention provides a method and system for enhancing the numerical skills of deep learning models for mathematical statistics. In view of the technical problem that in the existing methods for enhancing the numerical skills of deep learning models, there is an unavoidable misunderstanding of numerical units in the prior art for the numerical skills of deep learning models, which further leads to a further decrease in the accuracy of numerical calculation and reasoning, and therefore it is necessary to enhance the numerical skills from both qualitative and quantitative perspectives, this solution creatively adopts the enhancement and design in two directions of dimensional perception enhancement and numerical reasoning optimization, and further enhances model training by constructing a dimensional perception enhancement layer and a numerical reasoning optimization layer, thereby improving the dimension and depth of the deep learning model numerical skill enhancement; in view of the technical problem that in the existing methods for enhancing the numerical skills of deep learning models, in addition to the basic model training method enhancement, the data itself also needs to be enhanced and processed, this solution creatively adopts the scenario The learning method is used to enhance data, and by defining the contextual content of dimensional perception and numerical reasoning, it is further used as the knowledge base in the subsequent steps to build the contextual background, which can not only improve data quality, but also optimize the connectivity and maintainability of the scheme before and after; in the existing dimensional perception enhancement and numerical reasoning optimization methods, there are traditional methods that focus on training multiple models to solve the corresponding specific tasks, which will inevitably lead to a decrease in the accuracy and availability of multiple models, thereby reducing the technical problem of the numerical skills of a single deep learning model. This solution creatively adopts the construction of a dimensional perception enhancement knowledge base to perform dimensional perception enhancement. By constructing a unique dimensional perception enhancement knowledge base, the recognition and understanding of the qualitative part of the numerical value are enhanced, and the numerical reasoning is optimized by constructing a numerical reasoning optimization layer, which further optimizes the perception ability of the logical relationship between numerical calculations and reasoning, and improves the overall numerical recognition, understanding, reasoning and calculation performance of the deep learning model.

[0005] The technical solution adopted by the present invention is as follows: The present invention provides a method for enhancing the numerical skills of a deep learning model for mathematical statistics, the method comprising the following steps:

[0006] Step S1: Dataset construction;

[0007] Step S2: data enhancement;

[0008] Step S3: Dimensionality perception enhancement;

[0009] Step S4: numerical reasoning optimization;

[0010] Step S5: Model numerical skill enhancement.

[0011] Furthermore, in step S1, the data set construction is used to construct the original data set and the numerical skill knowledge base required for enhancing numerical skills, specifically, obtaining the original mathematical statistics data through original data collection and knowledge base construction;

[0012] The mathematical statistics raw data specifically includes a data source data set and a numerical skill knowledge base;

[0013] The data source data set specifically refers to the index value skill enhancement corpus data, including original sentence data and grammatical component triple data;

[0014] The numerical skill knowledge base specifically refers to an original knowledge base including dimension perception and numerical reasoning data sources, including a dimension perception data set and a numerical reasoning data set;

[0015] The numerical skill knowledge base is specifically constructed by performing rule-based automatic data partitioning and data screening on the data source data set in sequence to obtain a dimension-aware data set and a numerical reasoning data set, and by obtaining the dimension-aware data set and the numerical reasoning data set, the numerical skill knowledge base is constructed.

[0016] Further, in step S2, the data enhancement is used to enhance and optimize the original data set, specifically, using a contextual learning method to enhance the data source data set in the mathematical statistics original data to obtain an enhanced mathematical data set, including the following steps:

[0017] Step S21: initializing the data set, specifically performing a feature extraction operation on the data source data set in the mathematical statistics raw data, and obtaining an initialized raw data sample set by setting a feature matrix and a label vector;

[0018] Step S22: Enhanced scenario construction, specifically, designing context scenarios for dimension perception enhancement and numerical reasoning optimization, and constructing a context scenario set;

[0019] The context scenario set is used to assist the data enhancement model in understanding data characteristics, specifically to construct dimension perception enhancement scenarios and numerical reasoning optimization scenarios;

[0020] The dimension perception enhancement scenario specifically includes a unit dimension cross-language recognition perception scenario, a unit descriptive perception scenario, a unit quantity perception scenario, a unit conversion value perception scenario, and a unit occurrence frequency perception scenario;

[0021] The numerical reasoning optimization scenarios specifically include variable linear relationship reasoning scenarios, variable nonlinear relationship reasoning scenarios, proportional relationship reasoning scenarios, and constraint condition reasoning scenarios;

[0022] Step S23: data enhancement generation, specifically training a standard adversarial generative network based on the context scenario set, constructing generation conditions and performing data expansion to obtain a generated feature sample set;

[0023] Step S24: enhancing data integration, specifically integrating and merging the generated feature sample set and the initialized original data sample set to obtain an enhanced integrated data set;

[0024] Step S25: deduplication and standardization, specifically, performing data deduplication and data standardization operations on the enhanced integrated data set in sequence to obtain a comprehensive optimized data set;

[0025] The data deduplication specifically refers to removing duplicate samples in the enhanced integrated data set, specifically using a similarity detection method to automatically dedupe the data to obtain a deduplicated data set;

[0026] The data standardization specifically refers to standardizing the deduplication data set by using mean standardization to obtain a comprehensive optimized data set;

[0027] Step S26: data context enhancement, specifically, performing data context enhancement through the data set initialization, the enhanced context construction, the data enhancement generation, the enhanced data integration and the deduplication standardization to obtain an enhanced mathematical data set.

[0028] Further, in step S3, the dimension perception enhancement is used to optimize the dimension perception performance of the model, specifically, based on the enhanced mathematical data set and the numerical skill knowledge base in the mathematical statistics original data, by constructing a dimension perception enhancement knowledge base, performing dimension perception enhancement, and constructing a dimension perception enhancement layer for model training of model numerical skill enhancement;

[0029] The step of constructing a dimension perception enhancement knowledge base, performing dimension perception enhancement, and constructing a dimension perception enhancement layer includes:

[0030] Step S31: Initializing the dimension-aware enhanced knowledge base, specifically, initializing the dimension-aware enhanced knowledge base based on the dimension-aware data set in the numerical skill knowledge base and combining the context situation set in the enhanced mathematical data set to obtain the dimension-aware situation data set;

[0031] The dimension perception context data set specifically includes unit dimension cross-language recognition perception context data, unit descriptive perception context data, unit quantity perception context data, unit conversion value perception context data and unit occurrence frequency perception context data;

[0032] Step S32: constructing a dimensional knowledge architecture, specifically constructing a dimensional knowledge architecture based on the dimensional perception context data set, wherein the dimensional knowledge architecture specifically includes a unit dimension identifier, a unit Chinese name, a unit English name, a unit symbol representation, a unit second symbol representation, a unit descriptive text representation, a unit descriptive keyword, a unit dimension frequency, a unit representation quantity type, a unit dimension vector, and a unit conversion standard;

[0033] Step S33: constructing a dimensional vector, specifically defining a dimensional vector according to mathematics, and dividing the numerical dimension by defining data belonging to different numerical unit types as different dimensional vectors, to obtain a dimensional vector architecture, wherein the dimensional vector architecture specifically includes a material content dimension, a current dimension, a length dimension, a light intensity dimension, a mass dimension, a thermodynamic temperature dimension, a time dimension, and a unitless dimension;

[0034] Step S34: Calculate the dimensional knowledge frequency, specifically, calculate the occurrence frequency of each numerical unit dimension in the enhanced mathematical data set according to the dimensional vector architecture and the dimensional knowledge architecture, as the unit dimension frequency in the dimensional knowledge architecture, and obtain the dimensional knowledge frequency data set;

[0035] Step S35: constructing a dimension-awareness enhanced knowledge base, specifically, constructing the dimension-awareness enhanced knowledge base through the dimension-awareness enhanced knowledge base initialization, the dimension knowledge architecture construction, the dimension vector construction, and the dimension knowledge frequency calculation to obtain the dimension-awareness enhanced knowledge base;

[0036] Step S36: constructing a dimensional knowledge connection unit, specifically, constructing a key value candidate unit, a context coreference resolution calculation unit, and a confidence descending arrangement unit in sequence, and constructing the dimensional knowledge connection unit, including the following steps:

[0037] Step S361: construct a key value candidate unit for retrieving the corresponding unit dimension in the dimension-aware enhanced knowledge base, specifically by calculating the edit distance similarity to obtain the unit dimension candidate key value probability, and the calculation formula is:

[0038] Pr(u|M)=Levenshtein_Distance(u|M);

[0039] Where Pr(u|M) is the probability calculation function of the candidate key value per unit dimension, u is the dimension vector index, M is the set of dimension knowledge units, and Levenshtein_Distance(·) is the edit distance similarity calculation function;

[0040] Step S362: construct a context co-reference resolution calculation unit for retrieving the corresponding dimensional knowledge architecture in the dimension-aware enhanced knowledge base, specifically by calculating the context co-reference resolution through cosine similarity aggregation calculation, and the calculation formula is:

[0041] ;

[0042] Where Pr(u|C) is the context co-reference resolution calculation function, m is the total number of contexts, which is used to represent the total number of dimensional knowledge in the dimensional knowledge architecture, j is the context index, J is the dimensional knowledge unit index, and C is the context index. j is the set of context situations, k J is the contextual coreference keyword, used to represent the dimension vector in the J-th dimensional knowledge unit;

[0043] Step S363: constructing a confidence descending arrangement unit specifically includes taking the unit dimension candidate key probability and the context coreference resolution as weighted scores to calculate the dimension knowledge confidence, and constructing a confidence descending arrangement unit by sorting the dimension knowledge confidence in descending order, and performing data query and connection in the dimension-awareness enhanced knowledge base by selecting the unit dimension and dimension knowledge architecture with the highest dimension knowledge confidence;

[0044] Step S37: construct a dimensional perception enhancement layer, specifically by constructing the dimensional perception enhancement knowledge base and the dimensional knowledge connection unit to obtain the dimensional perception enhancement layer, and through the dimensional perception enhancement layer, enhance the model's perception ability of numerical quantities and numerical abstract values, and use it for model training for model numerical skill enhancement.

[0045] Further, in step S4, the numerical reasoning optimization is used to optimize the numerical reasoning performance of the model, specifically, based on the enhanced mathematical data set and the dimension-aware enhanced knowledge base, the numerical reasoning optimization is performed, and a numerical reasoning optimization layer is constructed for model training for model numerical skill enhancement;

[0046] The steps of performing numerical reasoning optimization and constructing a numerical reasoning optimization layer include:

[0047] Step S41: initializing the numerical reasoning optimization knowledge base, specifically initializing the numerical reasoning optimization knowledge base based on the numerical reasoning data set in the numerical skill knowledge base and combining the context scenario set in the enhanced mathematical data set to obtain the numerical reasoning optimization scenario data set;

[0048] The numerical reasoning optimization scenario data set specifically includes variable linear relationship reasoning scenario data, variable nonlinear relationship reasoning scenario data, proportional relationship reasoning scenario data and constraint condition reasoning scenario data;

[0049] Step S42: constructing a numerical reasoning knowledge framework, specifically, constructing a numerical reasoning knowledge framework based on the numerical reasoning optimization scenario data set, wherein the numerical reasoning knowledge framework specifically includes a variable identifier, a variable name, a variable type, a numerical reasoning formula, a reasoning formula input, a reasoning formula output, a reasoning step representation, a unit dimension knowledge representation used in the reasoning process, a reasoning process scenario representation, a variable relationship type, and a variable constraint condition;

[0050] Step S43: constructing a numerical reasoning vector, specifically using the word2vec method, performing numerical reasoning vector transformation according to the numerical reasoning optimization scenario data set and the numerical reasoning knowledge architecture, and obtaining a numerical reasoning vector set;

[0051] Step S44: constructing a numerical reasoning optimization layer, specifically, constructing the numerical reasoning optimization layer through the initialization of the numerical reasoning optimization knowledge base, the construction of the numerical reasoning knowledge architecture and the construction of the numerical reasoning vector, and through the numerical reasoning optimization layer, enhancing the model's perception of the logical relationship between calculations and conversions between numerical values, and used for model training for enhancing the model's numerical skills.

[0052] Furthermore, in step S5, the model numerical skill enhancement is used to implement model training and realize the numerical skill enhancement of the deep learning model. Specifically, through the dimension-aware enhancement knowledge base, the dimension-aware enhancement layer and the numerical reasoning optimization layer, a standard transformer is used in combination with a seq2seq method to perform model training to obtain a generative deep learning numerical skill enhancement model, and mathematical and statistical calculations are performed by applying the generative deep learning numerical skill enhancement model.

[0053] The present invention provides a deep learning model numerical skill enhancement system for mathematical statistics, comprising a data set construction module, a data enhancement module, a dimension perception enhancement module, a numerical reasoning optimization module and a model numerical skill enhancement module;

[0054] The data set construction module is used for data set construction, and obtains mathematical statistics raw data through the data set construction module, and sends the mathematical statistics raw data to the data enhancement module;

[0055] The data enhancement module is used for data enhancement, and obtains an enhanced mathematical data set through data enhancement, and sends the enhanced mathematical data set to the dimension perception enhancement module and the numerical reasoning optimization module;

[0056] The dimension perception enhancement module is used for dimension perception enhancement, and a dimension perception enhancement layer is obtained through dimension perception enhancement, and the dimension perception enhancement layer is used for model training in the model numerical skill enhancement module;

[0057] The numerical reasoning optimization module is used for numerical reasoning optimization, and obtains a numerical reasoning optimization layer through numerical reasoning optimization, and uses the numerical reasoning optimization layer for model training in the model numerical skill enhancement module;

[0058] The model numerical skill enhancement module is used for model numerical skill enhancement. Through the model numerical skill enhancement, a generative deep learning numerical skill enhancement model is obtained, and mathematical statistical calculations are performed through the generative deep learning numerical skill enhancement model.

[0059] The beneficial effects achieved by the present invention using the above scheme are as follows:

[0060] (1) In the existing methods for enhancing the numerical skills of deep learning models, there is an unavoidable misunderstanding of numerical units in the existing technologies for the numerical skills of deep learning models, which further reduces the accuracy of numerical calculations and reasoning. Therefore, it is necessary to enhance the numerical skills from both qualitative and quantitative perspectives. This solution creatively adopts the enhancement and design in two directions: dimensionality perception enhancement and numerical reasoning optimization. It further enhances model training by constructing a dimensionality perception enhancement layer and a numerical reasoning optimization layer, thereby improving the dimension and depth of the numerical skill enhancement of deep learning models.

[0061] (2) In view of the technical problem that in addition to the basic model training method enhancement, the data itself needs to be enhanced and processed in the existing deep learning model numerical skill enhancement methods, this solution creatively adopts the context learning method for data enhancement and further constructs the context background as the knowledge base in the subsequent steps by defining the context content of dimension perception and numerical reasoning. This can not only improve the data quality, but also optimize the connectivity and maintainability of the solution before and after.

[0062] (3) In the existing dimensionality perception enhancement and numerical reasoning optimization methods, there is a technical problem that traditional methods focus on training multiple models to solve corresponding specific tasks, which inevitably leads to a decrease in the accuracy and availability of multiple models, thereby reducing the numerical skills of a single deep learning model. This solution creatively adopts the construction of a dimensionality perception enhancement knowledge base to perform dimensionality perception enhancement. By constructing a unique dimensionality perception enhancement knowledge base, it enhances the recognition and understanding of the qualitative part of numerical values, and optimizes numerical reasoning by constructing a numerical reasoning optimization layer, further optimizing the perception ability of the logical relationship between numerical calculations and reasoning, and improving the overall performance of the deep learning model in numerical recognition, understanding, reasoning and calculation. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1A schematic diagram of a flow chart of a method for enhancing numerical skills of a deep learning model for mathematical statistics provided by the present invention;

[0064] Figure 2 A schematic diagram of a numerical skill enhancement system for a deep learning model for mathematical statistics provided by the present invention;

[0065] Figure 3 Schematic diagram of the process of data enhancement in step S2;

[0066] Figure 4 This is a flow chart of the dimensional perception enhancement process in step S3;

[0067] Figure 5 Schematic diagram of the process flow for optimizing the numerical reasoning in step S4.

[0068] The accompanying drawings are used to provide further understanding of the present invention and constitute a part of the specification. They are used to explain the present invention together with the embodiments of the present invention and do not constitute a limitation of the present invention. DETAILED DESCRIPTION

[0069] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments; based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0070] In the description of the present invention, it should be understood that terms such as “upper”, “lower”, “front”, “back”, “left”, “right”, “top”, “bottom”, “inside” and “outside” indicating directions or positional relationships are based on the directions or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as limiting the present invention.

[0071] Example 1, see Figure 1 The present invention provides a method for enhancing the numerical skills of a deep learning model for mathematical statistics, the method comprising the following steps:

[0072] Step S1: Dataset construction;

[0073] Step S2: data enhancement;

[0074] Step S3: Dimensionality perception enhancement;

[0075] Step S4: numerical reasoning optimization;

[0076] Step S5: Model numerical skill enhancement.

[0077] By performing the above operations, in order to solve the technical problem that in the existing methods for enhancing the numerical skills of deep learning models, the existing technologies inevitably have misunderstandings about numerical units for the numerical skills of deep learning models, which further reduces the accuracy of numerical calculations and reasoning. Therefore, it is necessary to enhance the numerical skills from both qualitative and quantitative perspectives. This solution creatively adopts enhancement and design combining dimensionality perception enhancement and numerical reasoning optimization, and further enhances model training by constructing dimensionality perception enhancement layer and numerical reasoning optimization layer, thereby improving the dimension and depth of the enhancement of the numerical skills of deep learning models.

[0078] Example 2, see Figure 1 and Figure 2 In step S1, the data set is constructed to construct the original data set and the numerical skill knowledge base required for enhancing numerical skills, specifically, by collecting original data and constructing the knowledge base to obtain the original mathematical statistics data;

[0079] The mathematical statistics raw data specifically includes a data source data set and a numerical skill knowledge base;

[0080] The data source data set specifically refers to the index value skill enhancement corpus data, including original sentence data and grammatical component triple data;

[0081] The original sentence data specifically refers to Chinese sentence data in the original corpus extracted from mathematics test websites, electronic information forums, industrial knowledge graphs, and general domain graphs;

[0082] The grammatical component triple data specifically refers to semi-structured triple data obtained by annotating, labeling and extracting information from the original corpus;

[0083] The semi-structured triples specifically include a subject, a predicate and an object;

[0084] The numerical skill knowledge base specifically refers to an original knowledge base including dimension perception and numerical reasoning data sources, including a dimension perception data set and a numerical reasoning data set;

[0085] The dimensional perception specifically refers to the recognition and perception of numerical units;

[0086] The numerical reasoning specifically refers to quantitative calculation and reasoning about quantity;

[0087] The numerical skill knowledge base is specifically constructed by performing rule-based automatic data partitioning and data screening on the data source data set in sequence to obtain a dimension-aware data set and a numerical reasoning data set, and by obtaining the dimension-aware data set and the numerical reasoning data set, the numerical skill knowledge base is constructed.

[0088] Example 3, see Figure 1 , Figure 2 and Figure 3 This embodiment is based on the above embodiment. In step S2, the data enhancement is used to enhance and optimize the original data set. Specifically, the data source data set in the original mathematical statistics data is enhanced by using a contextual learning method to obtain an enhanced mathematical data set, including the following steps:

[0089] Step S21: Data set initialization, specifically, performing feature extraction operation on the data source data set in the mathematical statistics raw data, and obtaining the initialized raw data sample set by setting the feature matrix and label vector, the calculation formula is:

[0090] ;

[0091] Where D ori is the initialization original data sample set, x i is the i-th original data feature matrix, i is the original data index, y i is the i-th original data label vector, n is the total number of original data;

[0092] Step S22: Enhanced scenario construction, specifically, designing context scenarios for dimension perception enhancement and numerical reasoning optimization, and constructing a context scenario set;

[0093] The context scenario set is used to assist the data enhancement model in understanding data characteristics, specifically to construct dimension perception enhancement scenarios and numerical reasoning optimization scenarios;

[0094] The dimension perception enhancement scenario specifically includes a unit dimension cross-language recognition perception scenario, a unit descriptive perception scenario, a unit quantity perception scenario, a unit conversion value perception scenario, and a unit occurrence frequency perception scenario;

[0095] The numerical reasoning optimization scenarios specifically include variable linear relationship reasoning scenarios, variable nonlinear relationship reasoning scenarios, proportional relationship reasoning scenarios, and constraint condition reasoning scenarios;

[0096] Step S23: data enhancement generation, specifically training a standard adversarial generative network based on the context scenario set, constructing generation conditions and performing data expansion to obtain a generated feature sample set;

[0097] The calculation formula of the generation condition is:

[0098] ;

[0099] Where D aug,jis the generated feature sample set, f(·) is the generated data function, which is specifically used to represent the standard adversarial generation network function, x aug,i is the extended data feature matrix corresponding to the i-th original data, i is the original data index, y i is the i-th original data label vector, C j is a set of context situations, where j is the context situation index, and D ori It is the initialization of the original data sample set;

[0100] The calculation formula for the data expansion is:

[0101] ;

[0102] In the formula, x aug,i is the extended data feature matrix corresponding to the i-th original data, g(·) is the context feature transformation function, x i is the i-th original data feature matrix, C j is a collection of context situations;

[0103] Step S24: Enhanced data integration, specifically, integrating and merging the generated feature sample set and the initialized original data sample set to obtain an enhanced integrated data set, and the calculation formula is:

[0104] ;

[0105] Where D aug is the enhanced integrated dataset, m is the total number of context situations, j is the context situation index, and D aug,j It is to generate feature sample set;

[0106] Step S25: deduplication and standardization, specifically, performing data deduplication and data standardization operations on the enhanced integrated data set in sequence to obtain a comprehensive optimized data set;

[0107] The data deduplication specifically refers to removing duplicate samples in the enhanced integrated data set, specifically using a similarity detection method to automatically dedupe the data to obtain a deduplicated data set;

[0108] The data standardization specifically refers to standardizing the deduplication data set by using mean standardization to obtain a comprehensive optimized data set;

[0109] Step S26: data context enhancement, specifically, performing data context enhancement through the data set initialization, the enhanced context construction, the data enhancement generation, the enhanced data integration and the deduplication standardization to obtain an enhanced mathematical data set.

[0110] By performing the above operations, in view of the technical problem that in the existing deep learning model numerical skill enhancement methods, in addition to the basic model training method enhancement, the data itself needs to be enhanced and processed, this solution creatively adopts the situational learning method for data enhancement, and by defining the situational content of dimensional perception and numerical reasoning, further constructs the situational background as the knowledge base in the subsequent steps, which can not only improve the data quality, but also optimize the connectivity and maintainability before and after the solution.

[0111] Example 4, see Figure 1 , Figure 2 and Figure 4 This embodiment is based on the above embodiment. In step S3, the dimension perception enhancement is used to optimize the dimension perception performance of the model. Specifically, according to the enhanced mathematical data set and the numerical skill knowledge base in the mathematical statistics original data, the dimension perception enhancement is performed by constructing a dimension perception enhancement knowledge base, and a dimension perception enhancement layer is constructed for model training of model numerical skill enhancement;

[0112] The step of constructing a dimension perception enhancement knowledge base, performing dimension perception enhancement, and constructing a dimension perception enhancement layer includes:

[0113] Step S31: Initializing the dimension-aware enhanced knowledge base, specifically, initializing the dimension-aware enhanced knowledge base based on the dimension-aware data set in the numerical skill knowledge base and combining the context situation set in the enhanced mathematical data set to obtain the dimension-aware situation data set;

[0114] The dimension perception context data set specifically includes unit dimension cross-language recognition perception context data, unit descriptive perception context data, unit quantity perception context data, unit conversion value perception context data and unit occurrence frequency perception context data;

[0115] Step S32: constructing a dimensional knowledge architecture, specifically constructing a dimensional knowledge architecture based on the dimensional perception context data set, wherein the dimensional knowledge architecture specifically includes a unit dimension identifier, a unit Chinese name, a unit English name, a unit symbol representation, a unit second symbol representation, a unit descriptive text representation, a unit descriptive keyword, a unit dimension frequency, a unit representation quantity type, a unit dimension vector, and a unit conversion standard;

[0116] Step S33: constructing a dimensional vector, specifically defining a dimensional vector according to mathematics, and dividing the numerical dimension by defining data belonging to different numerical unit types as different dimensional vectors, to obtain a dimensional vector architecture, wherein the dimensional vector architecture specifically includes a material content dimension, a current dimension, a length dimension, a light intensity dimension, a mass dimension, a thermodynamic temperature dimension, a time dimension, and a unitless dimension;

[0117] Step S34: Calculate the dimensional knowledge frequency, specifically, according to the dimensional vector architecture and the dimensional knowledge architecture, calculate the occurrence frequency of each numerical unit dimension in the enhanced mathematical data set as the unit dimension frequency in the dimensional knowledge architecture, and obtain the dimensional knowledge frequency data set. The calculation formula is:

[0118] ;

[0119] In the formula, Score(·) is a unit dimension frequency score calculation function, which is used to represent the weighted score value of dimensional knowledge and assist in the calculation of dimensional knowledge frequency data set. M is a set of dimensional knowledge units, and its specific value range is {G, H, C}. J is a dimensional knowledge unit index, where when J is G, it is used to represent the usage rate of a unit dimension in the enhanced mathematical data set. When J is H, it is used to represent the artificial commonality evaluation rate of a unit dimension in the enhanced mathematical data set. When J is C, it is used to represent the occurrence frequency of a unit dimension in the enhanced mathematical data set. is the frequency score weight of the unit dimension, Freq J (·) is the unit dimension frequency value of the J-th dimension knowledge unit, and u is the dimension vector index, which is used to represent the unit dimension vector in the dimension vector architecture;

[0120] Freq(u) is the dimensional knowledge frequency calculation function corresponding to the u-th dimensional vector in the dimensional knowledge frequency data set. is the frequency weight of dimension knowledge, min is the function for finding the minimum value, and max is the function for finding the maximum value;

[0121] in, When J is G, the specific value is 0.25. When J is H, the specific value is 0.25. When J is C, the specific value is 0.5. The value of is 0.1;

[0122] Step S35: constructing a dimension-awareness enhanced knowledge base, specifically, constructing the dimension-awareness enhanced knowledge base through the dimension-awareness enhanced knowledge base initialization, the dimension knowledge architecture construction, the dimension vector construction, and the dimension knowledge frequency calculation to obtain the dimension-awareness enhanced knowledge base;

[0123] Step S36: constructing a dimensional knowledge connection unit, specifically, constructing a key value candidate unit, a context coreference resolution calculation unit, and a confidence descending arrangement unit in sequence, and constructing the dimensional knowledge connection unit, including the following steps:

[0124] Step S361: construct a key value candidate unit for retrieving the corresponding unit dimension in the dimension-aware enhanced knowledge base, specifically by calculating the edit distance similarity to obtain the unit dimension candidate key value probability, and the calculation formula is:

[0125] Pr(u|M)=Levenshtein_Distance(u|M);

[0126] Where Pr(u|M) is the probability calculation function of the candidate key value per unit dimension, u is the dimension vector index, M is the set of dimension knowledge units, and Levenshtein_Distance(·) is the edit distance similarity calculation function;

[0127] Step S362: construct a context co-reference resolution calculation unit for retrieving the corresponding dimensional knowledge architecture in the dimension-aware enhanced knowledge base, specifically by calculating the context co-reference resolution through cosine similarity aggregation calculation, and the calculation formula is:

[0128] ;

[0129] Where Pr(u|C) is the context co-reference resolution calculation function, m is the total number of contexts, which is used to represent the total number of dimensional knowledge in the dimensional knowledge architecture, j is the context index, J is the dimensional knowledge unit index, and C is the context index. j is the set of context situations, k J is the contextual coreference keyword, used to represent the dimension vector in the J-th dimensional knowledge unit;

[0130] Step S363: constructing a confidence descending arrangement unit specifically includes taking the unit dimension candidate key probability and the context coreference resolution as weighted scores to calculate the dimension knowledge confidence, and constructing a confidence descending arrangement unit by sorting the dimension knowledge confidence in descending order, and performing data query and connection in the dimension-awareness enhanced knowledge base by selecting the unit dimension and dimension knowledge architecture with the highest dimension knowledge confidence;

[0131] Step S37: construct a dimensional perception enhancement layer, specifically by constructing the dimensional perception enhancement knowledge base and the dimensional knowledge connection unit to obtain the dimensional perception enhancement layer, and through the dimensional perception enhancement layer, enhance the model's perception ability of numerical quantities and numerical abstract values, and use it for model training for model numerical skill enhancement.

[0132] Example 5, see Figure 1 and Figure 2This embodiment is based on the above embodiment. In step S4, the numerical reasoning optimization is used to optimize the numerical reasoning performance of the model. Specifically, the numerical reasoning optimization is performed based on the enhanced mathematical data set and the dimension-aware enhanced knowledge base, and a numerical reasoning optimization layer is constructed to be used for model training for model numerical skill enhancement.

[0133] The steps of performing numerical reasoning optimization and constructing a numerical reasoning optimization layer include:

[0134] Step S41: initializing the numerical reasoning optimization knowledge base, specifically initializing the numerical reasoning optimization knowledge base based on the numerical reasoning data set in the numerical skill knowledge base and combining the context scenario set in the enhanced mathematical data set to obtain the numerical reasoning optimization scenario data set;

[0135] The numerical reasoning optimization scenario data set specifically includes variable linear relationship reasoning scenario data, variable nonlinear relationship reasoning scenario data, proportional relationship reasoning scenario data and constraint condition reasoning scenario data;

[0136] Step S42: constructing a numerical reasoning knowledge framework, specifically, constructing a numerical reasoning knowledge framework based on the numerical reasoning optimization scenario data set, wherein the numerical reasoning knowledge framework specifically includes a variable identifier, a variable name, a variable type, a numerical reasoning formula, a reasoning formula input, a reasoning formula output, a reasoning step representation, a unit dimension knowledge representation used in the reasoning process, a reasoning process scenario representation, a variable relationship type, and a variable constraint condition;

[0137] Step S43: constructing a numerical reasoning vector, specifically using the word2vec method, performing numerical reasoning vector transformation according to the numerical reasoning optimization scenario data set and the numerical reasoning knowledge architecture, and obtaining a numerical reasoning vector set;

[0138] Step S44: constructing a numerical reasoning optimization layer, specifically, constructing the numerical reasoning optimization layer through the initialization of the numerical reasoning optimization knowledge base, the construction of the numerical reasoning knowledge architecture and the construction of the numerical reasoning vector, and through the numerical reasoning optimization layer, enhancing the model's perception of the logical relationship between calculations and conversions between numerical values, and used for model training for enhancing the model's numerical skills.

[0139] By performing the above operations, in view of the technical problem that in the existing dimensionality perception enhancement and numerical reasoning optimization methods, the traditional methods focus on training multiple models to solve the corresponding specific tasks, which will inevitably lead to the decrease in the accuracy and availability of multiple models superimposed, thereby reducing the numerical skills of a single deep learning model, this solution creatively adopts the construction of a dimensionality perception enhancement knowledge base to perform dimensionality perception enhancement. By constructing a unique dimensionality perception enhancement knowledge base, the recognition and understanding of the qualitative part of the numerical values ​​are enhanced, and the numerical reasoning is optimized by constructing a numerical reasoning optimization layer, which further optimizes the perception ability of the logical relationship between numerical calculations and reasoning, and improves the overall performance of the deep learning model in numerical recognition, understanding, reasoning and calculation.

[0140] Example 6, see Figure 1 and Figure 2 This embodiment is based on the above embodiment. In step S5, the model numerical skill enhancement is used to implement model training and realize the numerical skill enhancement of the deep learning model. Specifically, the model is trained through the dimension-aware enhancement knowledge base, the dimension-aware enhancement layer and the numerical reasoning optimization layer, using a standard transformer combined with a seq2seq method to obtain a generative deep learning numerical skill enhancement model, and mathematical and statistical calculations are performed by applying the generative deep learning numerical skill enhancement model.

[0141] Embodiment 8, see Figure 1 and Figure 2 , this embodiment is based on the above embodiment, and the present invention provides a deep learning model numerical skill enhancement system for mathematical statistics, including a data set construction module, a data enhancement module, a dimension perception enhancement module, a numerical reasoning optimization module and a model numerical skill enhancement module;

[0142] The data set construction module is used for data set construction, and obtains mathematical statistics raw data through the data set construction module, and sends the mathematical statistics raw data to the data enhancement module;

[0143] The data enhancement module is used for data enhancement, and obtains an enhanced mathematical data set through data enhancement, and sends the enhanced mathematical data set to the dimension perception enhancement module and the numerical reasoning optimization module;

[0144] The dimension perception enhancement module is used for dimension perception enhancement, and a dimension perception enhancement layer is obtained through dimension perception enhancement, and the dimension perception enhancement layer is used for model training in the model numerical skill enhancement module;

[0145] The numerical reasoning optimization module is used for numerical reasoning optimization, and obtains a numerical reasoning optimization layer through numerical reasoning optimization, and uses the numerical reasoning optimization layer for model training in the model numerical skill enhancement module;

[0146] The model numerical skill enhancement module is used for model numerical skill enhancement. Through the model numerical skill enhancement, a generative deep learning numerical skill enhancement model is obtained, and mathematical statistical calculations are performed through the generative deep learning numerical skill enhancement model.

[0147] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0148] While the embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that many changes, modifications, substitutions and variations can be made to the embodiments without departing from the principles and spirit of the invention.

[0149] The present invention and its embodiments are described above, and such description is not restrictive. The drawings show only one embodiment of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A method for enhancing the numerical skills of deep learning models for mathematical statistics, characterized by: The method comprises the following steps: Step S1: Data set construction, through raw data collection and knowledge base construction, to obtain mathematical statistics raw data; the mathematical statistics raw data specifically includes a data source data set and a numerical skill knowledge base; Step S2: data enhancement, which is used to enhance and optimize the original data set, specifically, using a contextual learning method to enhance the data source data set in the mathematical statistics original data to obtain an enhanced mathematical data set, including the following steps: Step S21: data set initialization; Step S22: Enhanced scenario construction, specifically, for dimension perception enhancement and numerical reasoning optimization, context scenario design is performed, and a context scenario set is constructed; the context scenario set is used to assist the data enhancement model in understanding data characteristics, specifically to construct dimension perception enhancement scenarios and numerical reasoning optimization scenarios; the dimension perception enhancement scenarios specifically include unit dimension cross-language recognition perception scenarios, unit descriptive perception scenarios, unit quantity perception scenarios, unit conversion value perception scenarios, and unit occurrence frequency perception scenarios; the numerical reasoning optimization scenarios specifically include variable linear relationship reasoning scenarios, variable nonlinear relationship reasoning scenarios, proportional relationship reasoning scenarios, and constraint condition reasoning scenarios; Step S23: data enhancement generation; Step S24: enhanced data integration; Step S25: deduplication standardization; Step S26: data context enhancement; Step S3: Dimension perception enhancement, based on the enhanced mathematical data set and the numerical skill knowledge base in the mathematical statistics original data, by constructing a dimension perception enhancement knowledge base, dimension perception enhancement is performed, and a dimension perception enhancement layer is constructed for model training of model numerical skill enhancement; Step S4: numerical reasoning optimization, performing numerical reasoning optimization based on the enhanced mathematical data set and the dimension-aware enhanced knowledge base, and constructing a numerical reasoning optimization layer for model training for model numerical skill enhancement; Step S5: Model numerical skill enhancement, perform model training to obtain a generative deep learning numerical skill enhancement model, and apply the generative deep learning numerical skill enhancement model to perform mathematical and statistical calculations.

2. The method for enhancing numerical skills of a deep learning model for mathematical statistics according to claim 1, characterized in that: In step S1, the data set is constructed to construct the original data set and the numerical skill knowledge base required for enhancing numerical skills, and obtain the original mathematical statistics data; The mathematical statistics raw data specifically includes a data source data set and a numerical skill knowledge base; The data source data set specifically refers to the index value skill enhancement corpus data, including original sentence data and grammatical component triple data; The numerical skill knowledge base specifically refers to an original knowledge base including dimension perception and numerical reasoning data sources, including a dimension perception data set and a numerical reasoning data set.

3. The method for enhancing numerical skills of a deep learning model for mathematical statistics according to claim 2, characterized in that: In step S2, the data enhancement is used to enhance and optimize the original data set, including the following steps: Step S21: initializing the data set, specifically performing a feature extraction operation on the data source data set in the mathematical statistics raw data, and obtaining an initialized raw data sample set by setting a feature matrix and a label vector; Step S22: Enhanced scenario construction, specifically, designing context scenarios for dimension perception enhancement and numerical reasoning optimization, and constructing a context scenario set; Step S23: data enhancement generation, specifically training a standard adversarial generative network based on the context scenario set, constructing generation conditions and performing data expansion to obtain a generated feature sample set; Step S24: enhancing data integration, specifically integrating and merging the generated feature sample set and the initialized original data sample set to obtain an enhanced integrated data set; Step S25: deduplication and standardization, specifically, performing data deduplication and data standardization operations on the enhanced integrated data set in sequence to obtain a comprehensive optimized data set; Step S26: data context enhancement, specifically, performing data context enhancement through the data set initialization, the enhanced context construction, the data enhancement generation, the enhanced data integration and the deduplication standardization to obtain an enhanced mathematical data set.

4. The method for enhancing numerical skills of a deep learning model for mathematical statistics according to claim 3, characterized in that: The dimension perception enhancement is used to optimize the dimension perception performance of the model. The step of constructing a dimension perception enhancement knowledge base, performing dimension perception enhancement, and constructing a dimension perception enhancement layer includes: Step S31: Initializing the dimension-aware enhanced knowledge base, specifically, initializing the dimension-aware enhanced knowledge base based on the dimension-aware data set in the numerical skill knowledge base and combining the context situation set in the enhanced mathematical data set to obtain the dimension-aware situation data set; The dimension perception context data set specifically includes unit dimension cross-language recognition perception context data, unit descriptive perception context data, unit quantity perception context data, unit conversion value perception context data and unit occurrence frequency perception context data; Step S32: constructing a dimensional knowledge architecture, specifically constructing a dimensional knowledge architecture based on the dimensional perception context data set, wherein the dimensional knowledge architecture specifically includes a unit dimension identifier, a unit Chinese name, a unit English name, a unit symbol representation, a unit second symbol representation, a unit descriptive text representation, a unit descriptive keyword, a unit dimension frequency, a unit representation quantity type, a unit dimension vector, and a unit conversion standard; Step S33: constructing a dimensional vector, specifically defining a dimensional vector according to mathematics, and dividing the numerical dimension by defining data belonging to different numerical unit types as different dimensional vectors, to obtain a dimensional vector architecture, wherein the dimensional vector architecture specifically includes a material content dimension, a current dimension, a length dimension, a light intensity dimension, a mass dimension, a thermodynamic temperature dimension, a time dimension, and a unitless dimension; Step S34: Calculate the dimensional knowledge frequency, specifically, calculate the occurrence frequency of each numerical unit dimension in the enhanced mathematical data set according to the dimensional vector architecture and the dimensional knowledge architecture, as the unit dimension frequency in the dimensional knowledge architecture, and obtain the dimensional knowledge frequency data set; Step S35: constructing a dimension-awareness enhanced knowledge base, specifically, constructing the dimension-awareness enhanced knowledge base through the dimension-awareness enhanced knowledge base initialization, the dimension knowledge architecture construction, the dimension vector construction, and the dimension knowledge frequency calculation to obtain the dimension-awareness enhanced knowledge base; Step S36: constructing a dimensional knowledge connection unit, specifically, constructing a key value candidate unit, a context coreference resolution calculation unit, and a confidence descending arrangement unit in sequence to construct the dimensional knowledge connection unit; Step S37: construct a dimensional perception enhancement layer, specifically by constructing the dimensional perception enhancement knowledge base and the dimensional knowledge connection unit to obtain the dimensional perception enhancement layer, and through the dimensional perception enhancement layer, enhance the model's perception ability of numerical quantities and numerical abstract values, and use it for model training for model numerical skill enhancement.

5. The method for enhancing numerical skills of a deep learning model for mathematical statistics according to claim 4, characterized in that: In step S36, the step of constructing a dimensional knowledge connection unit includes: Step S361: construct a key value candidate unit for retrieving the corresponding unit dimension in the dimension-aware enhanced knowledge base, specifically by calculating the edit distance similarity to obtain the unit dimension candidate key value probability, and the calculation formula is: Pr(u|M)=Levenshtein_Distance(u|M); Where Pr(u|M) is the probability calculation function of the candidate key value per unit dimension, u is the dimension vector index, M is the set of dimension knowledge units, and Levenshtein_Distance(·) is the edit distance similarity calculation function; Step S362: construct a context co-reference resolution calculation unit for retrieving the corresponding dimensional knowledge architecture in the dimension-aware enhanced knowledge base, specifically by calculating the context co-reference resolution through cosine similarity aggregation calculation, and the calculation formula is: ； Where Pr(u|C) is the context co-reference resolution calculation function, m is the total number of contexts, which is used to represent the total number of dimensional knowledge in the dimensional knowledge architecture, j is the context index, J is the dimensional knowledge unit index, and C is the context index. j is the set of context situations, k J is the contextual coreference keyword, used to represent the dimension vector in the J-th dimensional knowledge unit; Step S363: Constructing a confidence descending arrangement unit specifically comprises taking the unit dimension candidate key probability and the context co-reference resolution as weighted scores to calculate the dimension knowledge confidence, and constructing a confidence descending arrangement unit by sorting the dimension knowledge confidence in descending order, and performing data query and connection in the dimension-aware enhanced knowledge base by selecting the unit dimension and dimension knowledge architecture with the highest dimension knowledge confidence.

6. The method for enhancing numerical skills of a deep learning model for mathematical statistics according to claim 5, characterized in that: In step S4, the numerical reasoning optimization is used to optimize the numerical reasoning performance of the model. The steps of performing the numerical reasoning optimization and constructing the numerical reasoning optimization layer include: Step S41: initializing the numerical reasoning optimization knowledge base, specifically initializing the numerical reasoning optimization knowledge base based on the numerical reasoning data set in the numerical skill knowledge base and combining the context scenario set in the enhanced mathematical data set to obtain the numerical reasoning optimization scenario data set; The numerical reasoning optimization scenario data set specifically includes variable linear relationship reasoning scenario data, variable nonlinear relationship reasoning scenario data, proportional relationship reasoning scenario data and constraint condition reasoning scenario data; Step S42: constructing a numerical reasoning knowledge framework, specifically, constructing a numerical reasoning knowledge framework based on the numerical reasoning optimization scenario data set, wherein the numerical reasoning knowledge framework specifically includes a variable identifier, a variable name, a variable type, a numerical reasoning formula, a reasoning formula input, a reasoning formula output, a reasoning step representation, a unit dimension knowledge representation used in the reasoning process, a reasoning process scenario representation, a variable relationship type, and a variable constraint condition; Step S43: constructing a numerical reasoning vector, specifically using the word2vec method, performing numerical reasoning vector transformation according to the numerical reasoning optimization scenario data set and the numerical reasoning knowledge architecture, and obtaining a numerical reasoning vector set; Step S44: constructing a numerical reasoning optimization layer, specifically, constructing the numerical reasoning optimization layer through the initialization of the numerical reasoning optimization knowledge base, the construction of the numerical reasoning knowledge architecture and the construction of the numerical reasoning vector, and through the numerical reasoning optimization layer, enhancing the model's perception of the logical relationship between calculations and conversions between numerical values, and used for model training for enhancing the model's numerical skills.

7. The method for enhancing numerical skills of a deep learning model for mathematical statistics according to claim 6, characterized in that: In step S5, the model numerical skill enhancement is used to implement model training and realize the numerical skill enhancement of the deep learning model. Specifically, the model is trained by using the dimension-aware enhancement knowledge base, the dimension-aware enhancement layer and the numerical reasoning optimization layer, using a standard transformer combined with a seq2seq method to obtain a generative deep learning numerical skill enhancement model, and mathematical and statistical calculations are performed by applying the generative deep learning numerical skill enhancement model.

8. A system for enhancing numerical skills of a deep learning model for mathematical statistics, for implementing a method for enhancing numerical skills of a deep learning model for mathematical statistics as claimed in any one of claims 1 to 7, characterized in that: It includes data set construction module, data enhancement module, dimension awareness enhancement module, numerical reasoning optimization module and model numerical skill enhancement module.

9. The deep learning model numerical skill enhancement system for mathematical statistics according to claim 8, characterized in that: The data set construction module is used for data set construction, and obtains mathematical statistics raw data through data set construction, and sends the mathematical statistics raw data to the data enhancement module; The data enhancement module is used for data enhancement, and obtains an enhanced mathematical data set through data enhancement, and sends the enhanced mathematical data set to the dimension perception enhancement module and the numerical reasoning optimization module; The dimension perception enhancement module is used for dimension perception enhancement, and a dimension perception enhancement layer is obtained through dimension perception enhancement, and the dimension perception enhancement layer is used for model training in the model numerical skill enhancement module; The numerical reasoning optimization module is used for numerical reasoning optimization, and a numerical reasoning optimization layer is obtained through numerical reasoning optimization, and the numerical reasoning optimization layer is used for model training in the model numerical skill enhancement module; The model numerical skill enhancement module is used for model numerical skill enhancement. Through the model numerical skill enhancement, a generative deep learning numerical skill enhancement model is obtained, and mathematical statistical calculations are performed through the generative deep learning numerical skill enhancement model.

Images