Knowledge point hybrid reasoning optimization method based on reverse deductive simulation

Abstract

The invention discloses a knowledge point hybrid reasoning optimization method based on reverse deductive simulation. The method comprises the following steps: S1, obtaining a to-be-reasoned target knowledge point set; s2, generating a standardized knowledge point set; s3, constructing a causal reasoning model, and generating an initial causal network diagram; s4, bringing the potential causal relationship identified in the derivation process into a causal network diagram; s5, generating an optimized causal network diagram; s6, generating a causal network diagram with dynamically updated weights; s7, dynamically selecting an optimal reasoning path in a reasoning process, and adjusting a reasoning sequence in combination with weights and causal relationships of knowledge points; and S8, generating a final reasoning result which comprises the reasoning path of the target knowledge point set, the reasoning logic and the optimized causal network diagram, and outputting the final reasoning result for a user to use. According to the method, powerful technical support is provided for intelligent reasoning and optimization of the knowledge network.

Description

Technical Field

[0001] The present invention relates to the technical field of knowledge point hybrid reasoning, and in particular to a knowledge point hybrid reasoning optimization method based on reverse deductive simulation. Background Art

[0002] With the rapid development of artificial intelligence and big data technology, knowledge reasoning and knowledge point analysis have gradually become important research directions in educational technology, knowledge management and intelligent systems. Knowledge point reasoning can not only simulate human logical thinking, but also help build efficient knowledge graphs and provide support for intelligent question-answering systems, adaptive learning and educational assessment scenarios. In existing technologies, knowledge point reasoning methods still face multiple challenges and are difficult to meet the diverse needs of complex knowledge networks.

[0003] At present, knowledge point reasoning mainly relies on the forward reasoning model, which derives conclusions from initial conditions in a step-by-step manner. The forward reasoning path is usually inefficient when faced with a huge knowledge network, and is particularly inefficient in scenarios that require deep logical analysis.

[0004] In addition, existing technologies also have significant deficiencies in the efficiency of knowledge point reasoning. Due to the multidimensionality and scale of knowledge networks, traditional reasoning systems need to consume a lot of computing resources to build and calculate reasoning paths. However, traditional systems usually adopt fixed reasoning strategies, which make it difficult to dynamically adjust reasoning paths and algorithm priorities according to actual problems, resulting in a waste of computing resources. At the same time, the speed of generating reasoning results is difficult to meet real-time requirements.

[0005] At the same time, the existing technology lacks an effective mechanism for the dynamic adjustment and optimization of knowledge points. In terms of dynamic adjustment of knowledge point weights and optimization of reasoning paths, traditional technologies often allocate knowledge point weights in a static manner, ignoring the dynamic change characteristics of knowledge points in the reasoning process, limiting the flexibility and applicability of the reasoning model. In addition, the existing technology usually lacks the application of hybrid reasoning strategies, and the combination of forward reasoning and reverse reasoning has not been effectively implemented, which further limits the breadth and depth of knowledge point reasoning. Summary of the invention

[0006] One purpose of the present invention is to propose a knowledge point hybrid reasoning optimization method based on reverse deductive simulation, which provides a strong technical support for the intelligent reasoning and optimization of knowledge networks.

[0007] A knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to an embodiment of the present invention comprises the following steps:

[0008] S1. Obtain a set of target knowledge points to be inferred;

[0009] S2. Preprocess each knowledge point in the target knowledge point set, clean redundant data and standardize the knowledge point expression, and generate a standardized knowledge point set;

[0010] S3. Construct a causal reasoning model, collect known causal relationships and potential causal relationships in the knowledge point set, conduct a preliminary analysis of the causal paths between the knowledge points, and generate an initial causal network diagram;

[0011] S4. Based on the reverse deductive simulation strategy, starting from the target knowledge point, reversely deduce other knowledge points that have direct or indirect causal relationships with it, mark the causal path of each knowledge point, and incorporate the potential causal relationships identified in the deduction process into the causal network diagram;

[0012] S5. Verify the causal path of the causal network diagram, use the conditional independence test method to verify the authenticity and logical consistency of the causal path, eliminate the wrong associations or weak associations in the causal network diagram, and generate an optimized causal network diagram;

[0013] S6. hierarchically process the knowledge points and their causal relationships in the knowledge point set according to the optimized causal network diagram, prioritize the knowledge points whose causal relationships are higher than the preset value, and dynamically adjust the associated weights of each knowledge point to generate a causal network diagram with dynamically updated weights;

[0014] S7. Based on the causal network diagram with dynamic weight update, a hybrid reasoning strategy is applied. Through the alternating operation of forward reasoning and reverse deduction, the optimal reasoning path is dynamically selected during the reasoning process, and the reasoning order is adjusted in combination with the weights of knowledge points and causal relationships.

[0015] S8. Generate the final reasoning result, which includes the reasoning path, reasoning logic and optimized causal network diagram of the target knowledge point set, and output it for users to use.

[0016] Optionally, the S1 specifically includes:

[0017] S11. Get a set of knowledge points:

[0018] K={k1,k2,…,k n},1≤i≤n;

[0019] Among them, k i represents the i-th knowledge point, and n is the total number of knowledge points;

[0020] S12. For each knowledge point k i Extract its initial association information:

[0021] The correlation data between knowledge points R = [r ij ], where r ij Represents knowledge point ki With knowledge point k j The correlation strength satisfies 0≤r ij ≤1;

[0022] The known logical relationship of the knowledge point L = [l ij ], where l ij =1 indicates knowledge point k i With knowledge point k j There is a logical relationship, ij =0 means there is no logical relationship;

[0023] The weight distribution of knowledge points W = {w1,w2,…,w n},1≤i≤n, where w i Represents knowledge point k i The initial weight of the weight value satisfies w i >0 and

[0024] Optionally, the S2 specifically includes:

[0025] S21. Clean the redundant data of the target knowledge point set K and compare the correlation data r between the knowledge points ij and r ji , if r is satisfied ij =r ji And r ij Less than the set threshold τ r , then remove the low-correlation data and compare the logical relationship of the knowledge points. If there is l ij = 0 and the associated weight w i <τ w , then remove knowledge point k i and its related data;

[0026] S22. Performing format normalization processing on the remaining target knowledge point set and its associated information;

[0027] S23. Recalculate the weight distribution and adjust the weight value by normalization;

[0028] S24. Perform semantic consistency check on the cleaned target knowledge point set, and divide the knowledge point k i The expression of is standardized into a unified semantic format so that its identification in correlation data and logical relationships is consistent;

[0029] S25. Output the standardized target knowledge point set K' and its updated correlation data matrix R', logical relationship matrix L' and weight distribution W'.

[0030] Optionally, the S3 specifically includes:

[0031] S31. Collect the known causal relationships of the standardized target knowledge point set K' to form an initial causal relationship matrix C = [c ij ], where c ij =1 indicates knowledge point k i With knowledge point k j There is a known direct causal relationship, c ij =0 means there is no known direct causal relationship;

[0032] S32. Calculate the potential causal influence strength p from the correlation data matrix R' and the logical relationship matrix L' using the potential causal relationship mining algorithm ij :

[0033]

[0034] Among them, α and β are adjustment parameters, ∈ is the smoothing factor, and p ij Represents knowledge point k i With knowledge point k j The potential causal influence strength, c ij is the element in the initial causal relationship matrix, and m is the number of knowledge points;

[0035] S33. Constructing a comprehensive causal relationship matrix C′=[c′ ij ]:

[0036] c′ ij =c ij +p ij -c ij ·p ij ;

[0037] Among them, c′ ij Represents knowledge point k i With knowledge point k j The overall causal strength of

[0038] S34. Establish an initial causal network diagram based on the comprehensive causal relationship matrix C':

[0039] G = (V, E);

[0040] Among them, V = K' is the node set, E = {(k i ,k j )|c′ ij >τ c}, τ c is the causal relationship threshold, when c′ ij >τ c When the knowledge point k i For knowledge point k j There is a significant causal relationship;

[0041] S35. Perform causal path analysis on the initial causal network graph G and calculate the causal strength IS of each causal path ij :

[0042]

[0043] Among them, w i and w j are knowledge points k respectively i With knowledge point k j The weight of ij is knowledge point k i With knowledge point k j The topological distance between them, γ is the attenuation coefficient, represents the attenuation of causal strength due to topological distance;

[0044] S36. According to causal strength IS ij The calculation results update the edge weights in the initial causal network graph G to generate the final initial causal network graph:

[0045] G'=(V,E');

[0046] Among them, E ′ ={(k i ,k j ,IS ij )|IS ij >τ IS}, τ IS is the causal strength threshold, when IS ij >τ IS , the causal path is considered valid.

[0047] Optionally, the S4 specifically includes:

[0048] S41. Initialize the reverse deductive deduction path set P = {} starting from the target knowledge point, where P represents the path from the target knowledge point k t All causal paths that begin;

[0049] S42. Traverse the comprehensive causal relationship matrix, according to c′ jt >τ c The conditions that directly affect the target knowledge point k are identified t A collection of upstream knowledge points The path (k j ,k t ) join the path set P;

[0050] S43. Collection of upstream knowledge points Each knowledge point k in jRecursively search for its upstream knowledge point set, repeat the above S41-S43 process to construct the secondary, tertiary and higher-level causal paths in turn, until there are no more upstream knowledge points that meet the conditions. The reverse deductive inference path set is expressed as:

[0051]

[0052] Where L is the maximum number of path layers derived in reverse order;

[0053] S44. In the reverse deduction process, the cumulative causal strength CIS of each path p Perform the calculation:

[0054]

[0055] Where p represents the path;

[0056] S45. Cumulative Causal Intensity CIS p Less than the preset threshold τ cis The paths are removed from the path set P to form a valid path set P ′ ;

[0057] S46. Set the valid path set P ′ All paths in the causal network are added to G', and all potential causal relationships involved in the path are marked;

[0058] S47. Output the updated causal network diagram G", the updated causal network diagram contains the target knowledge point k t All valid reverse deduction paths and their corresponding causal relationship strengths.

[0059] Optionally, the S5 specifically includes:

[0060] S51. For each path p in the updated causal network diagram G", the conditional independence test method is used to verify whether the causal relationship between each pair of adjacent knowledge points in the path is authentic and logically consistent, and the conditional independence test statistic T is calculated. ij :

[0061]

[0062] Among them, P(k i ,k j ) represents knowledge point k i With knowledge point k j The probability of joint occurrence, P(k i ) and P(k j ) represent knowledge point k respectively i With knowledge point k j Probability of single occurrence;

[0063] S52. For each pair of adjacent knowledge points (k i ,k j ), if T ij <τ t , then remove the edge (k i ,k j ), where τ t is the significance threshold of the conditional independence test;

[0064] S53. Recalculate the causal strength CIS of each path after the conditional independence test. p1 ;

[0065] S54. Corrected Causal Intensity CIS p1 Less than the threshold τ cis The paths are removed from the causal network graph G”, and the effective paths of causal strength are retained;

[0066] S55. Update the causal network graph G based on all path verification results to generate the optimized causal network graph G opt =(V,E opt ).

[0067] Optionally, the S6 specifically includes:

[0068] S61. Based on the optimized causal network graph G opt All knowledge points are layered according to the strength of their associated causal relationships, and the causal level L of each knowledge point is defined i :

[0069] L i =max{L j +1|(k j ,k i )∈E opt ,c′ ji >τ c};

[0070] Among them, L i Represents knowledge point k i The causal level, (k j ,k i )∈E opt Represents knowledge point k j For knowledge point k i There is a causal relationship, and the causal strength c′ ji >τ c ;

[0071] S62. Causal level L i Perform normalization to obtain the initialized causal hierarchy

[0072] S63. For each knowledge point, according to its out-degree in the causal network and in-degree Calculate the dynamic association weight w′ of the knowledge point i :

[0073]

[0074] Among them, w′ i is the initial weight, c′ ji and c′ ij The edges (k j ,k i ) and (k i ,k j ) is the causal strength, ∈1>0 is a smoothing factor to prevent the denominator from being zero;

[0075] S64. Calculate the dynamic association weight w′ i Perform normalization to obtain the updated dynamic association weight

[0076] S65. According to the dynamically adjusted weight The hierarchical order of the causal network has an impact on the causal network graph G opt Update the weights of the edges in the graph to generate a causal network diagram with dynamic weight updates:

[0077] G dyn =(V,E dyn );

[0078] in,

[0079] Optionally, the S7 specifically includes:

[0080] S71. Initialize the hybrid reasoning strategy according to the target knowledge point k t Level L t and weight Select the inference mode:

[0081] When the level of the target knowledge point L t >L avg When using reverse deductive reasoning, L avg is the average level of the causal network;

[0082] When the level of the target knowledge point L t ≤L avg When using forward reasoning, it is preferred;

[0083] S72. During the reasoning process, dynamically calculate the path priority PR p, in each step of reasoning, according to the path priority PR p Dynamically select the optimal path p * :

[0084]

[0085] in, For the current reasoning path set, select the path with the highest priority for reasoning;

[0086] S75. For the optimal path p * The node weights and edge weights in the causal network graph are updated in real time based on the inference results of each step, and the path priority PR is recalculated. p1 , adjust the subsequent reasoning order;

[0087] S76. During the reasoning process, when the path priority PR p1 Less than the preset threshold τ PR When the reasoning of the path is terminated, it is removed from the reasoning set Removed;

[0088] S77. Repeat the process of S74-S76 until all path reasoning is completed, and generate the final reasoning result set:

[0089] R 最终 ={r1,r2,…,r m};

[0090] Among them, r m Represents the mth knowledge point and its causal relationship in the reasoning result.

[0091] The beneficial effects of the present invention are:

[0092] (1) The present invention introduces a reverse deductive simulation strategy and combines it with a causal reasoning model to identify the potential causal relationship between knowledge points during the reasoning process, thereby significantly improving the accuracy and logical consistency of reasoning. It can also reversely deduce potential upstream knowledge points from the target knowledge points, and eliminate mis-associated knowledge points through causal path verification to generate an optimized causal network diagram, breaking through the limitation that traditional knowledge point reasoning is limited to correlation analysis. Through hierarchical processing and dynamic adjustment of multi-layer causal relationships, it can effectively explore implicit associations and deep logical relationships, significantly improving the construction quality of the knowledge graph and the accuracy of the reasoning results.

[0093] (2) The present invention realizes the organic combination of forward reasoning and reverse deduction through the dynamic adjustment mechanism of the hybrid reasoning strategy. In the reasoning process, the optimal reasoning path is intelligently selected according to the hierarchy and dynamic weight of the knowledge points, and the reasoning order is dynamically optimized through path priority calculation, thereby reducing the amount of calculation of redundant paths. In the construction and optimization stage of the causal network diagram, a dynamic weight update algorithm is adopted, and the utilization efficiency of computing resources is improved through weight normalization processing. In large-scale knowledge network scenarios, the reasoning efficiency is significantly improved compared with traditional methods.

[0094] (3) The present invention proposes a weight update algorithm and path priority strategy based on dynamic adjustment, which enables the reasoning system to flexibly adjust the reasoning path and weight distribution according to actual problems, thereby improving the pertinence of the reasoning results and the adaptability of the system. In the reasoning process, the weights and reasoning priorities are dynamically allocated based on the out-degree, in-degree and hierarchical relationship of the knowledge points, ensuring that the reasoning results of high-weight knowledge points can be presented first, meeting the diverse needs of different application scenarios. BRIEF DESCRIPTION OF THE DRAWINGS

[0095] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0096] Figure 1 This is a flow chart of a knowledge point hybrid reasoning optimization method based on reverse deductive simulation proposed by the present invention. DETAILED DESCRIPTION

[0097] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, which only illustrate the basic structure of the present invention in a schematic manner, and therefore only show the components related to the present invention.

[0098] refer to Figure 1 , a knowledge point hybrid reasoning optimization method based on reverse deductive simulation, comprising the following steps:

[0099] S1. Obtain a set of target knowledge points to be inferred;

[0100] S2. Preprocess each knowledge point in the target knowledge point set, clean redundant data and standardize the knowledge point expression, and generate a standardized knowledge point set;

[0101] S3. Construct a causal reasoning model, collect known causal relationships and potential causal relationships in the knowledge point set, conduct a preliminary analysis of the causal paths between the knowledge points, and generate an initial causal network diagram;

[0102] S4. Based on the reverse deductive simulation strategy, starting from the target knowledge point, reversely deduce other knowledge points that have direct or indirect causal relationships with it, mark the causal path of each knowledge point, and incorporate the potential causal relationships identified in the deduction process into the causal network diagram;

[0103] S5. Verify the causal path of the causal network diagram, use the conditional independence test method to verify the authenticity and logical consistency of the causal path, eliminate the wrong associations or weak associations in the causal network diagram, and generate an optimized causal network diagram;

[0104] S6. hierarchically process the knowledge points and their causal relationships in the knowledge point set according to the optimized causal network diagram, prioritize the knowledge points whose causal relationships are higher than the preset value, and dynamically adjust the associated weights of each knowledge point to generate a causal network diagram with dynamically updated weights;

[0105] S7. Based on the causal network diagram with dynamic weight update, a hybrid reasoning strategy is applied. Through the alternating operation of forward reasoning and reverse deduction, the optimal reasoning path is dynamically selected during the reasoning process, and the reasoning order is adjusted in combination with the weights of knowledge points and causal relationships.

[0106] S8. Generate the final reasoning result, which includes the reasoning path, reasoning logic and optimized causal network diagram of the target knowledge point set, and output it for users to use.

[0107] In this implementation, S1 specifically includes:

[0108] S11. Get a set of knowledge points:

[0109] K={k1,k2,…,k n},1≤i≤n;

[0110] Among them, k i represents the i-th knowledge point, and n is the total number of knowledge points;

[0111] S12. For each knowledge point k i Extract its initial association information:

[0112] The correlation data between knowledge points R = [r ij ], where r ij Represents knowledge point k i With knowledge point k j The correlation strength satisfies 0≤r ij ≤1;

[0113] The known logical relationship of the knowledge point L = [l ij ], where l ij =1 indicates knowledge point k i With knowledge point k j There is a logical relationship,ij =0 means there is no logical relationship;

[0114] The weight distribution of knowledge points W = {w1,w2,…,w n},1≤i≤n, where w i Represents knowledge point k i The initial weight of the weight value satisfies w i >0 and

[0115] In this implementation, S2 specifically includes:

[0116] S21. Clean the redundant data of the target knowledge point set K and compare the correlation data r between the knowledge points ij and r ji , if r is satisfied ij =r ji And r ij Less than the set threshold τ r , then remove the low-correlation data and compare the logical relationship of the knowledge points. If there is l ij = 0 and the associated weight w i <τ w , then remove knowledge point k i and its related data;

[0117] S22. Performing format normalization processing on the remaining target knowledge point set and its associated information;

[0118] S23. Recalculate the weight distribution and adjust the weight value by normalization;

[0119] S24. Perform semantic consistency check on the cleaned target knowledge point set, and divide the knowledge point k i The expression of is standardized into a unified semantic format so that its identification in correlation data and logical relationships is consistent;

[0120] S25. Output the standardized target knowledge point set K' and its updated correlation data matrix R', logical relationship matrix L' and weight distribution W'.

[0121] In this implementation, S3 specifically includes:

[0122] S31. Collect the known causal relationships of the standardized target knowledge point set K' to form an initial causal relationship matrix C = [c ij ], where c ij =1 indicates knowledge point k i With knowledge point k j There is a known direct causal relationship, c ij =0 means there is no known direct causal relationship;

[0123] S32. Calculate the potential causal influence strength p from the correlation data matrix R' and the logical relationship matrix L' using the potential causal relationship mining algorithm ij :

[0124]

[0125] Among them, α and β are adjustment parameters, ∈ is the smoothing factor, and p ij Represents knowledge point k i With knowledge point k j The potential causal influence strength, c ij is the element in the initial causal relationship matrix, and m is the number of knowledge points;

[0126] S33. Constructing a comprehensive causal relationship matrix C ′ =[c ′ ij ]:

[0127] c′ ij =c ij +p ij -c ij ·p ij ;

[0128] Among them, c′ ij Represents knowledge point k i With knowledge point k j The overall causal strength of

[0129] S34. Establish an initial causal network diagram based on the comprehensive causal relationship matrix C':

[0130] G = (V, E);

[0131] Among them, V = K' is the node set, E = {(k i ,k j )|c′ ij >τ c}, τ c is the causal relationship threshold, when c′ ij >τ c When the knowledge point k i For knowledge point k j There is a significant causal relationship;

[0132] S35. Perform causal path analysis on the initial causal network graph G and calculate the causal strength IS of each causal path ij :

[0133]

[0134] Among them, w i and w j are knowledge points k respectivelyi With knowledge point k j The weight of ij is knowledge point k i With knowledge point k j The topological distance between them, γ is the attenuation coefficient, represents the attenuation of causal strength due to topological distance;

[0135] S36. According to causal strength IS ij The calculation results update the edge weights in the initial causal network graph G to generate the final initial causal network graph:

[0136] G'=(V,E');

[0137] Among them, E ′ ={(k i ,k j ,IS ij )|IS ij >τ IS}, τ IS is the causal strength threshold, when IS ij >τ IS , the causal path is considered valid.

[0138] In this implementation, S4 specifically includes:

[0139] S41. Initialize the reverse deductive deduction path set P = {} starting from the target knowledge point, where P represents the path from the target knowledge point k t All causal paths that begin;

[0140] S42. Traverse the comprehensive causal relationship matrix, according to c′ jt >τ c The conditions that directly affect the target knowledge point k are identified t A collection of upstream knowledge points The path (k j ,k t ) join the path set P;

[0141] S43. Collection of upstream knowledge points Each knowledge point k in j Recursively search for its upstream knowledge point set, repeat the above S41-S43 process to construct the secondary, tertiary and higher-level causal paths in turn, until there are no more upstream knowledge points that meet the conditions. The reverse deductive inference path set is expressed as:

[0142]

[0143] Where L is the maximum number of path layers derived in reverse order;

[0144] S44. In the reverse deduction process, the cumulative causal strength CIS of each path p Perform the calculation:

[0145]

[0146] Where p represents the path;

[0147] S45. Cumulative Causal Intensity CIS p Less than the preset threshold τ cis The paths are removed from the path set P to form a valid path set P ′ ;

[0148] S46. Set the valid path set P ′ All paths in the causal network are added to G', and all potential causal relationships involved in the path are marked;

[0149] S47. Output the updated causal network diagram G", the updated causal network diagram contains the target knowledge point k t All valid reverse deduction paths and their corresponding causal relationship strengths.

[0150] In this implementation, S5 specifically includes:

[0151] S51. For each path p in the updated causal network diagram G", the conditional independence test method is used to verify whether the causal relationship between each pair of adjacent knowledge points in the path is authentic and logically consistent, and the conditional independence test statistic T is calculated. ij :

[0152]

[0153] Among them, P(k i ,k j ) represents knowledge point k i With knowledge point k j The probability of joint occurrence, P(k i ) and P(k j ) represent knowledge point k respectively i With knowledge point k j Probability of single occurrence;

[0154] S52. For each pair of adjacent knowledge points (k i ,k j ), if T ij <τ t , then remove the edge (k i ,k j ), where τ t is the significance threshold of the conditional independence test;

[0155] S53. Recalculate the causal strength CIS of each path after the conditional independence test. p1 ;

[0156] S54. Corrected Causal Intensity CIS p1 Less than the threshold τ cis The paths are removed from the causal network graph G”, and the effective paths of causal strength are retained;

[0157] S55. Update the causal network graph G based on all path verification results to generate the optimized causal network graph G opt =(V,E opt ).

[0158] In this implementation, S6 specifically includes:

[0159] S61. Based on the optimized causal network graph G opt All knowledge points are layered according to the strength of their associated causal relationships, and the causal level L of each knowledge point is defined i :

[0160] L i =max{L j +1|(k j ,k i )∈E opt ,c′ ji >τ c};

[0161] Among them, L i Represents knowledge point k i The causal level, (k j ,k i )∈E opt Represents knowledge point k j For knowledge point k i There is a causal relationship, and the causal strength c′ ji >τ c ;

[0162] S62. Causal level L i Perform normalization to obtain the initialized causal hierarchy

[0163] S63. For each knowledge point, according to its out-degree in the causal network and in-degree Calculate the dynamic association weight w′ of the knowledge point i :

[0164]

[0165] Among them, w′ iis the initial weight, c′ ji and c′ ij The edges (k j ,k i ) and (k i ,k j ) is the causal strength, ∈1>0 is a smoothing factor to prevent the denominator from being zero;

[0166] S64. Calculate the dynamic association weight w′ i Perform normalization to obtain the updated dynamic association weight

[0167] S65. According to the dynamically adjusted weight The hierarchical order of the causal network has an impact on the causal network graph G opt Update the weights of the edges in the graph to generate a causal network diagram with dynamic weight updates:

[0168] G dyn =(V,E dyn );

[0169] in,

[0170] In this implementation, S7 specifically includes:

[0171] S71. Initialize the hybrid reasoning strategy according to the target knowledge point k t Level L t and weight Select the inference mode:

[0172] When the level of the target knowledge point L t >L avg When using reverse deductive reasoning, L avg is the average level of the causal network;

[0173] When the level of the target knowledge point L t ≤L avg When using forward reasoning, it is preferred;

[0174] S72. During the reasoning process, dynamically calculate the path priority PR p , in each step of reasoning, according to the path priority PR p Dynamically select the optimal path p * :

[0175]

[0176] in, For the current reasoning path set, select the path with the highest priority for reasoning;

[0177] S75. For the optimal path p * The node weights and edge weights in the causal network graph are updated in real time based on the result of each step of reasoning, and the path priority PR is recalculated. p1 , adjust the subsequent reasoning order;

[0178] S76. During the reasoning process, when the path priority PR p1 Less than the preset threshold τ PR When the reasoning of the path is terminated, it is removed from the reasoning set Removed;

[0179] S77. Repeat the process of S74-S76 until all path reasoning is completed, and generate the final reasoning result set:

[0180] R 最终 ={r1,r2,…,r m};

[0181] Among them, r m Represents the mth knowledge point and its causal relationship in the reasoning result.

[0182] Embodiment 1:

[0183] Example 1 In the online education system of a key middle school on September 1, 2024, 5,000 senior high school students in the key middle school are taking a comprehensive mathematics test. The test involves 120 knowledge points. The system records and analyzes the students' answers in real time. The students' test time is 120 minutes. The answering process of each student is broken down into binary data of the mastery of the knowledge points (mastery is represented by 1, and non-mastery is represented by 0). Among them, a student, Xiao Ming, has a high error rate in questions related to the knowledge point of "monotonicity of function". The system quickly locates the root cause of the problem and recommends a personalized learning path through the hybrid reasoning optimization method of the present invention.

[0184] After Xiao Ming finished the test, the system first extracted Xiao Ming's answer data and found that his mastery of the three knowledge points of "monotonicity of function", "graph change of logarithmic function" and "solving inequalities" was 0, and the error rate of his related knowledge points exceeded 50%. The system took the target knowledge point "monotonicity of function" as the starting point, and used the reverse deductive simulation strategy of the present invention to dynamically derive potential weak related knowledge points from the knowledge point causal network diagram.

[0185] In the reverse deductive reasoning process, the system starts from the target knowledge point "monotonicity of function", identifies the upstream knowledge point "basic properties of logarithmic function" directly related to it, and calculates the path priority. The reasoning results show that the path "basic properties of logarithmic function → image change of logarithmic function → monotonicity of function" has the highest priority, with a causal association strength of 0.87 and a weight-adjusted priority of 0.92.

[0186] Subsequently, the system used the conditional independence test to verify the authenticity of the path. The statistical calculation results showed that the conditional independence test values ​​between all nodes were greater than the preset threshold of 0.7. Therefore, the path was confirmed as a valid reasoning path. The system eliminated other lower priority paths and removed them from the reasoning set.

[0187] Next, the system recalculated the dynamic weights of Xiao Ming's knowledge point set by combining the optimized causal network diagram and the dynamic weight update mechanism. The results showed that the weights of the two knowledge points, "Basic Properties of Logarithmic Functions" and "Solving Inequalities", were significantly increased to 0.76 and 0.71 respectively, which were much higher than the weights of other related knowledge points. Based on these weight adjustment results, the system generated a personalized learning path for Xiao Ming.

[0188] In the learning path recommendation, the system suggests that Xiao Ming first review the "basic properties of logarithmic functions", then learn the "graph changes of logarithmic functions" and "monotonicity of functions" in turn, and finally consolidate the knowledge point of "solving inequalities". Each learning module is equipped with system-generated high-frequency wrong question analysis and targeted exercises. The learning path is pushed to Xiao Ming through the personalized learning function of the online education platform, and his learning behavior and completion status are recorded in real time.

[0189] After the test, the system further analyzed the data of 5,000 students as a whole, and generated the following data in Table 1 by comparing the method of the present invention with the traditional forward reasoning method:

[0190] Table 1 Comparative data analysis of the method of the present invention and the traditional forward reasoning method

[0191]

[0192] It can be seen from the data in Table 1 that the method of the present invention is significantly superior to the traditional method in terms of reasoning efficiency, error location accuracy and personalized learning recommendation satisfaction. In the subsequent test, after Xiao Ming studied according to the recommended path for 2 weeks, the system tested his knowledge point mastery again. The results showed that Xiao Ming's accuracy in questions related to "monotonicity of functions" increased from 40% to 85%, and his accuracy in questions related to "basic properties of logarithmic functions" increased from 50% to 92%, and he reached a mastery level of 80% in questions related to "solving inequalities".

[0193] This embodiment shows that the method of the present invention has extremely high feasibility and practicality in real online education scenarios, can effectively improve the accuracy of knowledge point reasoning and learning path recommendation, and optimize students' learning effects.

[0194] The present invention introduces a reverse deductive simulation strategy and combines it with a causal reasoning model to identify potential causal relationships between knowledge points during the reasoning process, thereby significantly improving the accuracy and logical consistency of reasoning. It can also reversely deduce potential upstream knowledge points from target knowledge points, and eliminate mis-associated knowledge points through causal path verification to generate an optimized causal network diagram, breaking through the limitation that traditional knowledge point reasoning is limited to correlation analysis. Through hierarchical processing and dynamic adjustment of multi-layer causal relationships, it can effectively mine implicit associations and deep logical relationships, significantly improving the construction quality of the knowledge graph and the accuracy of the reasoning results.

[0195] The present invention realizes the organic combination of forward reasoning and reverse deduction through the dynamic adjustment mechanism of hybrid reasoning strategy. In the reasoning process, the optimal reasoning path is intelligently selected according to the hierarchy and dynamic weight of the knowledge points, and the reasoning order is dynamically optimized through path priority calculation, thereby reducing the amount of calculation of redundant paths. A dynamic weight update algorithm is adopted in the construction and optimization stage of the causal network diagram, and the utilization efficiency of computing resources is improved through weight normalization processing. In large-scale knowledge network scenarios, the reasoning efficiency is significantly improved compared with the traditional method. Experiments show that the algorithm of the present invention can shorten the reasoning time to about 60% of the traditional method while reducing the consumption of computing resources by more than 30%, thereby meeting the real-time reasoning requirements.

[0196] The present invention proposes a weight updating algorithm and path priority strategy based on dynamic adjustment, which enables the reasoning system to flexibly adjust the reasoning path and weight distribution according to actual problems, thereby improving the pertinence of the reasoning results and the adaptability of the system. In the reasoning process, the weights and reasoning priorities are dynamically allocated in combination with the out-degree, in-degree and hierarchical relationship of the knowledge points, ensuring that the reasoning results of high-weight knowledge points can be presented first, meeting the diverse needs of different application scenarios.

[0197] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A knowledge point hybrid reasoning optimization method based on reverse deductive simulation, characterized in that: The steps include: S1. Obtain a set of target knowledge points to be inferred; S2. Preprocess each knowledge point in the target knowledge point set, clean redundant data and standardize the knowledge point expression, and generate a standardized knowledge point set; S3. Construct a causal reasoning model, collect known causal relationships and potential causal relationships in the knowledge point set, conduct a preliminary analysis of the causal paths between the knowledge points, and generate an initial causal network diagram; S4. Based on the reverse deductive simulation strategy, starting from the target knowledge point, reversely deduce other knowledge points that have direct or indirect causal relationships with it, mark the causal path of each knowledge point, and incorporate the potential causal relationships identified in the deduction process into the causal network diagram; S5. Verify the causal path of the causal network diagram, use the conditional independence test method to verify the authenticity and logical consistency of the causal path, eliminate the wrong associations or weak associations in the causal network diagram, and generate an optimized causal network diagram; S6. hierarchically process the knowledge points and their causal relationships in the knowledge point set according to the optimized causal network diagram, prioritize the knowledge points whose causal relationships are higher than the preset value, and dynamically adjust the associated weights of each knowledge point to generate a causal network diagram with dynamically updated weights; S7. Based on the causal network diagram with dynamic weight update, a hybrid reasoning strategy is applied. Through the alternating operation of forward reasoning and reverse deduction, the optimal reasoning path is dynamically selected during the reasoning process, and the reasoning order is adjusted in combination with the weights of knowledge points and causal relationships. S8. Generate the final reasoning result, which includes the reasoning path, reasoning logic and optimized causal network diagram of the target knowledge point set, and output it for users to use.

2. According to claim 1, a knowledge point hybrid reasoning optimization method based on reverse deductive simulation is characterized in that: The S1 specifically includes: S11. Get a set of knowledge points: K={k1,k2,…,k n },1≤i≤n; Among them, k i represents the i-th knowledge point, and n is the total number of knowledge points; S12. For each knowledge point k i Extract its initial association information: The correlation data between knowledge points R = [r ij ], where r ij Represents knowledge point k i With knowledge point k j The correlation strength satisfies 0≤r ij ≤1; The known logical relationship of the knowledge point L = [l ij ], where l ij =1 indicates knowledge point k i With knowledge point k j There is a logical relationship, ij =0 means there is no logical relationship; The weight distribution of knowledge points W = {w1,w2,…,w n },1≤i≤n, where w i Represents knowledge point k i The initial weight of the weight value satisfies w i >0 and 3. The knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to claim 1 is characterized in that: The S2 specifically includes: S21. Clean the redundant data of the target knowledge point set K and compare the correlation data r between the knowledge points ij and r ji , if r is satisfied ij =r ji And r ij Less than the set threshold τ r , then remove the low-correlation data and compare the logical relationship of the knowledge points. If there is l ij = 0 and the associated weight w i <τ w , then remove knowledge point k i and its related data; S22. Performing format normalization processing on the remaining target knowledge point set and its associated information; S23. Recalculate the weight distribution and adjust the weight value by normalization; S24. Perform semantic consistency check on the cleaned target knowledge point set, and divide the knowledge point k i The expression of is standardized into a unified semantic format so that its identification in correlation data and logical relationships is consistent; S25. Output the standardized target knowledge point set K' and its updated correlation data matrix R', logical relationship matrix L' and weight distribution W'.

4. The knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to claim 1 is characterized in that: The S3 specifically includes: S31. Collect the known causal relationships of the standardized target knowledge point set K' to form an initial causal relationship matrix C = [c ij ], where c ij =1 indicates knowledge point k i With knowledge point k j There is a known direct causal relationship, c ij =0 means there is no known direct causal relationship; S32. Calculate the potential causal influence strength p from the correlation data matrix R' and the logical relationship matrix L' using the potential causal relationship mining algorithm ij : Among them, α and β are adjustment parameters, ∈ is the smoothing factor, and p ij Represents knowledge point k i With knowledge point k j The potential causal influence strength, c ij is the element in the initial causal relationship matrix, and m is the number of knowledge points; S33. Constructing a comprehensive causal relationship matrix C′=[c′ ij ]: c′ ij =c ij +p ij -c ij ·p ij ; Among them, c′ ij Represents knowledge point k i With knowledge point k j The overall causal strength of S34. Establish an initial causal network diagram based on the comprehensive causal relationship matrix C': G = (V, E); Among them, V = K' is the node set, E = {(k i ,k j )|c′ ij >τ c }, τ c is the causal relationship threshold, when c′ ij >τ c When the knowledge point k i For knowledge point k j There is a significant causal relationship; S35. Perform causal path analysis on the initial causal network graph G and calculate the causal strength IS of each causal path ij : Among them, w i and w j are knowledge points k respectively i With knowledge point k j The weight of ij is knowledge point k i With knowledge point k j The topological distance between them, γ is the attenuation coefficient, e -γdij represents the attenuation of causal strength due to topological distance; S36. According to causal strength IS ij The calculation results update the edge weights in the initial causal network graph G to generate the final initial causal network graph: G'=(V,E'); Where E′={(k i ,k j ,IS ij )|IS ij >τ IS }, τ IS is the causal strength threshold, when IS ij >τ IS , the causal path is considered valid.

5. The knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to claim 1 is characterized in that: The S4 specifically includes: S41. Initialize the reverse deductive deduction path set P = {} starting from the target knowledge point, where P represents the path from the target knowledge point k t All causal paths that begin; S42. Traverse the comprehensive causal relationship matrix, according to c′ jt >τ c The conditions that directly affect the target knowledge point k are identified t A collection of upstream knowledge points The path (k j ,k t ) join the path set P; S43. Collection of upstream knowledge points Each knowledge point k in j Recursively search for its upstream knowledge point set, repeat the above S41-S43 process to construct the secondary, tertiary and higher-level causal paths in turn, until there are no more upstream knowledge points that meet the conditions. The reverse deductive inference path set is expressed as: Where L is the maximum number of path layers derived in reverse order; S44. In the reverse deduction process, the cumulative causal strength CIS of each path p Perform the calculation: Among them, p represents the path; S45. Cumulative Causal Intensity CIS p Less than the preset threshold τ cis The paths are removed from the path set P to form a valid path set P′; S46. Add all paths in the valid path set P′ to the causal network graph G′, and mark all potential causal relationships involved in the paths; S47. Output the updated causal network diagram G", the updated causal network diagram contains the target knowledge point k t All valid reverse deduction paths and their corresponding causal relationship strengths.

6. The knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to claim 1 is characterized in that: The S5 specifically includes: S51. For each path p in the updated causal network diagram G", the conditional independence test method is used to verify whether the causal relationship between each pair of adjacent knowledge points in the path is authentic and logically consistent, and the conditional independence test statistic T is calculated. ij : Among them, P(k i ,k j ) represents knowledge point k i With knowledge point k j The probability of joint occurrence, P(k i ) and P(k j ) represent knowledge point k respectively i With knowledge point k j Probability of single occurrence; S52. For each pair of adjacent knowledge points (k i ,k j ), if T ij <τ t , then remove the edge (k i ,k j ), where τ t is the significance threshold of the conditional independence test; S53. Recalculate the causal strength CIS of each path after the conditional independence test. p1 ; S54. Corrected Causal Intensity CIS p1 Less than the threshold τ cis The paths are removed from the causal network graph G”, and the effective paths of causal strength are retained; S55. Update the causal network graph G based on all path verification results to generate the optimized causal network graph G opt =(V,E opt ).

7. The knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to claim 1 is characterized in that: The S6 specifically includes: S61. Based on the optimized causal network graph G opt All knowledge points are layered according to the strength of their associated causal relationships, and the causal level L of each knowledge point is defined i : L i =max{L j +1∣(k j ,k i )∈E opt ,c′ ji >t c }; Among them, L i Represents knowledge point k i The causal level, (k j ,k i )∈E opt Represents knowledge point k j For knowledge point k i There is a causal relationship, and the causal strength c′ ji >τ c ; S62. Causal level L i Perform normalization to obtain the initialized causal hierarchy S63. For each knowledge point, according to its out-degree in the causal network and in-degree Calculate the dynamic association weight w′ of the knowledge point i : Among them, w′ i is the initial weight, c′ ji and c′ ij The edges (k j ,k i ) and (k i ,k j ) is the causal strength, ∈1>0 is a smoothing factor to prevent the denominator from being zero; S64. Calculate the dynamic association weight w′ i Perform normalization to obtain the updated dynamic association weight S65. According to the dynamically adjusted weight The hierarchical order of the causal network has an impact on the causal network graph G opt Update the weights of the edges in the graph to generate a causal network diagram with dynamically updated weights: G dyn =(V,E dyn ); in, 8. The knowledge point hybrid reasoning optimization method based on reverse deductive simulation according to claim 1 is characterized in that: The S7 specifically includes: S71. Initialize the hybrid reasoning strategy according to the target knowledge point k t Level L t and weight Select the inference mode: When the level of the target knowledge point L t >L avg When using reverse deductive reasoning, L avg is the average level of the causal network; When the level of the target knowledge point L t ≤L avg When using forward reasoning, it is preferred; S72. During the reasoning process, dynamically calculate the path priority PR p , in each step of reasoning, according to the path priority PR p Dynamically select the optimal path p * : in, For the current reasoning path set, select the path with the highest priority for reasoning; S75. For the optimal path p * The node weights and edge weights in the causal network graph are updated in real time based on the result of each step of reasoning, and the path priority PR is recalculated. p1 , adjust the subsequent reasoning order; S76. During the reasoning process, when the path priority PR p1 Less than the preset threshold τ PR When the reasoning of the path is terminated, it is removed from the reasoning set Removed; S77. Repeat the process of S74-S76 until all path reasoning is completed, and generate the final reasoning result set: R 最终 ＝{r1,r2,…,r m}； Among them, r m Represents the mth knowledge point and its causal relationship in the reasoning result.

Images