An artificial intelligence-based multi-scenario intelligent question-answering management system and method

Abstract

The present invention relates to the technical field of natural language processing, and specifically to a multi-scenario intelligent question and answer management system and method based on artificial intelligence. The system includes: a semantic smoothing and optimization module that collects keywords and topic sentences in multiple question and answer scenarios, defines each keyword and topic sentence as a node, establishes semantic associations between the nodes, calculates the path lengths and association strengths between the nodes, reallocates the node weights after comparing the calculation results, and obtains optimized node weights. In the present invention, by collecting and analyzing the semantic associations of keywords and topic sentences, optimizing the node weights, and mapping user questions based on these node weights, the core node that best matches the user's query is selected, and the cumulative weights of the associated nodes are calculated. This enhances the system's ability to understand user queries and response accuracy, can more accurately identify the user's intent, and screen out the most relevant answers from the knowledge base according to the user's specific needs.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language processing, and in particular, to a multi-scenario intelligent question-answering management system and method based on artificial intelligence. Background Art

[0002] Although the prior art supports intelligent question answering in multiple scenarios, it is difficult to fully understand complex user queries in the case of involving multiple topics or requiring in-depth semantic understanding. In addition, the response of the prior art depends on a static knowledge base and a fixed answering mode, which limits its effectiveness in dealing with emerging or rare problems, resulting in a decline in user experience, leading to the problem that users feel that the answers are irrelevant or lack pertinence, thus affecting the overall service satisfaction and the practicality of the system. Summary of the Invention

[0003] The object of the present invention is to solve the deficiencies existing in the prior art, and to propose a multi-scenario intelligent question-answering management system and method based on artificial intelligence.

[0004] To achieve the above object, the present invention adopts the following technical solution: A multi-scenario intelligent question-answering management system based on artificial intelligence includes:

[0005] The semantic smoothing and optimization module collects keywords and topic sentences in multiple question-answering scenarios, defines each keyword and topic sentence as a node, establishes semantic associations between nodes, calculates the path length and association strength between nodes, reallocates node weights after comparing the calculation results, and obtains optimized node weights;

[0006] The user intention recognition module maps a user's question to a node set based on the optimized node weights, calculates the matching degree with each node, selects the core node with the highest matching degree, calculates the cumulative weight of all associated nodes according to the association degree between the core node and the remaining nodes, determines the path with the largest cumulative weight, and obtains an associated topic path;

[0007] The data fusion and adjustment module monitors the current data input of nodes based on the associated topic path, combines user interaction data, establishes a Bayesian model to calculate the conditional probability of a node as a reply under the current question, updates the connection strength and relationship of the nodes, recalculates the node state value, and generates an updated conditional probability value;

[0008] The reply path selection module obtains candidate reply content from a knowledge base or preset text based on the updated conditional probability value, calculates the matching degree between the reply content and the associated topic path, filters the reply content according to the matching degree, and generates an intelligent question-answering result.

[0009] As a further solution of the present invention, the obtaining step of reallocating node weights after comparing the calculation results is specifically as follows:

[0010] Collect keywords and topic sentences from multiple Q&A scenarios, define each keyword and topic sentence as a node, establish semantic associations between the nodes, and use the formula:

[0011] ;

[0012] Calculate the semantic association strength between node and node ;

[0013] wherein, is the node identifier, is the word frequency of node in the th Q&A scenario, is the word frequency of node in the th Q&A scenario, is the index of the Q&A scenario, is the total number of Q&A scenarios;

[0014] According to the semantic association strength between node and node , use the formula:

[0015] ;

[0016] Calculate the path length from node to node ;

[0017] According to the path length from node to node , use the formula:

[0018] ;

[0019] Calculate the updated weight of node , and obtain the optimized node weights;

[0020] wherein, is the original weight of node , is the weight adjustment factor, which is determined through experiments or the tuning process.

[0021] As a further solution of the present invention, the specific steps for obtaining the core node with the highest matching degree are as follows:

[0022] According to the node Updated weight , use the formula:

[0023] ;

[0024] Calculate the mapping score between the user's question and the node ; ;

[0025] Among them, is a binary indicator function, which takes the value of 1 when the keyword in the question belongs to the node , otherwise 0, is the total number of keywords in the question, and are adjustment coefficients, obtained from the relevance between the keywords in the user's question and the node, is the frequency weight of the node ;

[0026] According to the mapping score between the user's question and the node , use the formula: ;

[0027] ;

[0028] Calculate the matching degree of the node ; ;

[0029] Among them, is the total number of nodes, representing the number of all defined nodes, is the mapping score between the user's question and the node ;

[0030] According to the matching degree of the node , use the formula: ;

[0031] ;

[0032] Select the core node according to the matching degree ;

[0033] Among them, is used to find the node index that maximizes .

[0034] As a further solution of the present invention, the obtaining step of determining the path with the largest cumulative weight is specifically as follows:

[0035] Select the core node with reference to the matching degree , use the formula:

[0036] ;

[0037] Calculate the cumulative weight of the nodes associated with the core node ; ;

[0038] where is the degree of association between the core node and the node , is the optimized weight of the node , is the total number of all nodes;

[0039] According to the cumulative weight of the node , use the formula:

[0040] ;

[0041] Calculate the maximum path of the cumulative weight to obtain the associated topic path;

[0042] where is used to find the node index that maximizes .

[0043] As a further solution of the present invention, the step of obtaining the conditional probability that the calculation node is used as a reply under the current question is specifically as follows:

[0044] According to the current data input and user interaction data, use the formula:

[0045] ;

[0046] Calculate the data vector of the node ;

[0047] where is the current data input of the node , is the user interaction data associated with the node , is the weight for adjusting the contributions of the current data input and user interaction data;

[0048] According to the data vector of the node , referring to the maximum path of the cumulative weight, use the formula:

[0049] ;

[0050] Computing node Conditional probability as a response under the current question ;

[0051] Wherein, is the model sensitivity adjustment parameter, determined by data fitting and optimization, is the influence weight of data volatility, obtained by analyzing the historical data of the node, is the node data volatility metric, obtained through historical data analysis, is the connection strength adjustment parameter, optimized according to the influence of node connection strength on the model, is the node average connection strength with the remaining nodes, calculated based on network analysis, is the node index of the topic path , is the natural exponent, is the node data vector, represents the data volatility metric of the node , represents the average connection strength of the node with the remaining nodes.

[0052] As a further solution of the present invention, the step of obtaining the recalculated node state value is specifically as follows:

[0053] Referring to the conditional probability of the node as a response under the current question , using the formula:

[0054] ;

[0055] Calculate the connection strength between the node and the node ;

[0056] Wherein, is the balance parameter, adjusted based on historical data and response performance;

[0057] According to the connection strength between the node and the node , using the formula:

[0058] ;

[0059] Calculate the new state value of the node ;

[0060] According to the new state value of the said node adopt the formula: ;

[0061] ;

[0062] obtain the updated node conditional probability and generate the updated conditional probability value;

[0063] wherein is the node index, representing the state values of all nodes.

[0064] As a further solution of the present invention, the specific steps for obtaining the matching degree between the calculated reply content and the associated topic path are as follows:

[0065] According to the updated node conditional probability adopt the formula:

[0066] ;

[0067] obtain the relevance score of the candidate reply of the node indicating the relevance score of the candidate reply;

[0068] wherein is the vector form of the th text or data item in the knowledge base, is the cosine similarity function, used to select the maximum value;

[0069] According to the updated conditional probability vector of the said point adopt the formula:

[0070] ;

[0071] calculate the matching degree score of the reply content, and filter the reply content according to the matching degree to generate the intelligent Q&A result;

[0072] wherein is the weight coefficient, is the number of nodes in the topic path, is the node index, represents the topic path, which is a set composed of multiple nodes ;

[0073] A multi-scenario intelligent question-answering management method based on artificial intelligence, which is executed based on the above-mentioned multi-scenario intelligent question-answering management system based on artificial intelligence, and includes the following steps:

[0074] S1: Based on multi-scenario data, collect keywords and topic sentences from it, define them as nodes, construct a semantic association graph between nodes, calculate the path length and association strength between nodes, reassign node weights, and obtain optimized node weights;

[0075] S2: Based on the optimized node weights, map the user's question to the node set, calculate the matching degree with each node, select the node with the highest matching degree as the core node, calculate the cumulative weight of the node and the remaining nodes, determine the path with the largest cumulative weight, and generate an associated topic path;

[0076] S3: Utilize the associated topic path to monitor the current data input of the node, combine with user interaction data, establish a Bayesian network to calculate the conditional probability of the node under the current question, and generate an updated conditional probability value;

[0077] S4: Based on the updated conditional probability value, obtain candidate response content from the knowledge base or preset text, calculate the matching degree between the response content and the associated topic path, screen the response content with the highest matching degree, and generate candidate response content;

[0078] S5: From the candidate response content, refine the response content according to the user interaction situation to generate an intelligent question-answering result.

[0079] Compared with the prior art, the advantages and positive effects of the present invention are as follows:

[0080] In the present invention, by collecting and analyzing the semantic associations of keywords and topic sentences, optimizing node weights, and mapping the user's question based on these node weights, the core node that best matches the user's query is selected, and the cumulative weight of the associated nodes is calculated. This enhances the system's ability to understand user queries and response accuracy, can more accurately identify user intentions, and screen out the most relevant answers from the knowledge base according to the specific needs of users. In addition, by dynamically monitoring the data input of nodes and updating node states, the system can adapt to the user's interaction behavior in real time and provide more personalized responses. This improves the flexibility and applicability of the question-answering system. BRIEF DESCRIPTION OF THE DRAWINGS

[0081] Figure 1 It is the system flow chart of the present invention;

[0082] Figure 2 It is the flow chart of the acquisition step of reassigning node weights after comparing and calculating the results of the present invention;

[0083] Figure 3 Flowchart of the acquisition step for selecting the core node with the highest matching degree for the present invention;

[0084] Figure 4 Flowchart of the acquisition step for determining the path with the largest cumulative weight for the present invention;

[0085] Figure 5 Flowchart of the acquisition step for calculating the conditional probability of a node as a response under the current question for the present invention;

[0086] Figure 6 Flowchart of the acquisition step for recalculating the node state value for the present invention;

[0087] Figure 7 Flowchart of the acquisition step for calculating the matching degree between the response content and the associated topic path for the present invention. Detailed implementation manners

[0088] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0089] In the description of the present invention, it should be understood that the orientation or positional relationships indicated by the terms "length", "width", "upper", "lower", "front", "rear", "left", "right", "vertical", "horizontal", "top", "bottom", "inner", "outer", etc. are based on the orientation or positional relationships shown in the 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 orientation, be constructed and operated in a specific orientation, and thus should not be construed as limiting the present invention. In addition, in the description of the present invention, "a plurality of" means two or more unless otherwise specifically defined.

[0090] Please refer to Figure 1 , the present invention provides a technical solution: A multi-scenario intelligent question and answer management system based on artificial intelligence includes.

[0091] The semantic smoothing optimization module collects keywords and topic sentences in multiple question and answer scenarios, defines each keyword and topic sentence as a node, establishes semantic associations between the nodes, calculates the path lengths and association strengths between the nodes, and redistributes the node weights after comparing the calculation results to obtain optimized node weights;

[0092] Based on the optimized node weights, the user intention recognition module maps the user's question to a set of nodes, calculates the matching degree with each node, selects the core node with the highest matching degree, calculates the cumulative weight of all associated nodes according to the association degree between the core node and the remaining nodes, determines the path with the largest cumulative weight, and obtains the associated topic path;

[0093] Based on the associated topic path, the data fusion adjustment module monitors the current data input of the nodes, combines the user interaction data, establishes a Bayesian model to calculate the conditional probability of the nodes as responses under the current question, updates the connection strength and relationship of the nodes, recalculates the node state values, and generates updated conditional probability values;

[0094] Based on the updated conditional probability values, the response path selection module obtains candidate response contents from the knowledge base or preset texts, calculates the matching degree between the response contents and the associated topic path, filters the response contents according to the matching degree, and generates intelligent question and answer results;

[0095] The optimized node weights include the reallocated node priorities, the adjusted association strength values, and the path length adjustment parameters. The associated topic path includes the core node, the set of associated nodes, the cumulative weight between the nodes, and the association degree sorting result. The updated conditional probability values include the prior probability of the nodes, the conditional dependence relationship between the nodes, and the set of values after the conditional probability is updated. The intelligent question and answer results include the best response content after screening, the corresponding matching degree score, and the priority sorting.

[0096] Please refer to Figure 2 , and the specific steps for obtaining the reallocated node weights after comparing the calculation results are as follows:

[0097] Collect keywords and topic sentences from multiple question and answer scenarios, define each keyword and topic sentence as a node, establish semantic associations between the nodes, and use the formula:

[0098] ;

[0099] Calculate the semantic association strength between node and node ;

[0100] Among them, is the node identifier, is the word frequency of node in the th question and answer scenario, is the word frequency of node in the th question and answer scenario, is the index of the question and answer scenario, ranging from 1 to , is the total number of Q&A scenarios;

[0101] According to the node and the node the semantic association strength between , use the formula:

[0102] ;

[0103] Calculate the path length from node to node ; ;

[0104] According to the path length from node to node ; , use the formula:

[0105] ;

[0106] Calculate the updated weight of node to obtain the optimized node weight; Among them,

[0107] is the original weight of node , is the weight adjustment factor, determined during the experiment or parameter tuning process, and used to control the amplitude of weight update;

[0108] The calculation process is as follows:

[0109] For the formula:

[0110] ;

[0111] The word frequency data ( and ) is obtained by counting the frequencies of keywords extracted from multiple intelligent Q&A scenarios using natural language processing (NLP) tools, is determined by the total number of different intelligent Q&A scenarios collected.

[0112] Collect the word frequency data of node from three Q&A scenarios (i.e., and node as follows:

[0113] For node :

[0114] Word frequency , , ;

[0115] For node :

[0116] Word frequency 、 、 ;

[0117] ;

[0118] Result indicates a very high semantic correlation degree between nodes and .

[0119] For the formula:

[0120] ;

[0121] Substitute the calculated into the formula:

[0122] ;

[0123] Result indicates that the path between nodes is close to the unit distance, indicating a relatively close semantic relationship between the two nodes.

[0124] For the formula:

[0125] ;

[0126] Set the original weight of the parameter (set through actual measurement or historical data), and the weight adjustment factor (set according to the results of the tuning experiment).

[0127] Substitute the aforementioned calculation result into the formula:

[0128] ;

[0129] ;

[0130] Result indicates that the weight of node has increased slightly after adjustment, reflecting the enhanced effect of the semantic relationship between the node and other associated nodes. Take the original weight as the reference value and compare it with the adjusted weight . Through comparison, it is found that the weight has increased by about 0.0994, which shows that after comprehensive consideration of the semantic association strength and path length, the relative importance or influence of node has been enhanced. This result indicates that the relevance of the node in the multi-scenario intelligent question-answering system has been optimized, reflecting the increased contribution degree of the node in the network.

[0131] Please refer to Figure 3 , the steps for obtaining the core node with the highest matching degree are specifically as follows:

[0132] According to the node updated weight , use the formula:

[0133] ;

[0134] Calculate the mapping score between the user's question and the node ;

[0135] Among them, is a binary indicator function, which takes the value of 1 when the keyword in the question belongs to the node , otherwise it is 0, is the total number of keywords in the question, and are adjustment coefficients, obtained from the relevance between specific keywords in the user's question and the node, is the frequency weight of the node , reflecting the frequency of the node appearing in all questions;

[0136] According to the mapping score between the user's question and the node , use the formula:

[0137] ;

[0138] Calculate the matching degree of the node ;

[0139] Among them, is the total number of nodes, representing the number of all defined nodes, is the mapping score between the user's question and the node ;

[0140] According to the matching degree of the node , use the formula:

[0141] ;

[0142] Select the core node according to the matching degree ;

[0143] Among them, is used to find the node index that makes the largest;

[0144] The calculation process is as follows:

[0145] For the formula:

[0146] ;

[0147] Set the parameter , extracted and statistically analyzed by NLP technology, and set to , ,

[0148] , or it can also be obtained by optimizing the response effect of the system. .

[0149] For each node, substitute the value:

[0150] Node 1:

[0151] Mapping situation: There are 3 times, and the others are 0.

[0152] Calculate the score:

[0153] ;

[0154] Node 2:

[0155] Mapping situation: There are 2 times, and the others are 0.

[0156] Calculate the score:

[0157] ;

[0158] Node 3:

[0159] Mapping situation: There is 1 time, and the others are 0.

[0160] Calculate the score:

[0161] ;

[0162] For the formula:

[0163] ;

[0164] Total score:

[0165] ;

[0166] Node 1:

[0167] ;

[0168] Node 2:

[0169] ;

[0170] Node 3:

[0171] ;

[0172] For the formula:

[0173]

[0174] Matching degree comparison:

[0175] ;

[0176] ;

[0177] ;

[0178] The node with the highest matching degree is Node 1, that is .

[0179] The highest matching degree obtained through calculation is Node 1( ), indicating that the user's question best matches the semantic range of Node 1. This result reflects that Node 1 is the node that can best represent the user's needs, which helps to improve the response accuracy of the system and the user experience.

[0180] Please refer to Figure 4 to determine that the specific steps for obtaining the path with the largest cumulative weight are as follows:

[0181] Select the core node with reference to the matching degree , and adopt the formula:

[0182] ;

[0183] Calculate the cumulative weight of the node associated with the core node ;

[0184] Among them, is the association degree between the core node and the node , is the optimization weight of the node , is the total number of all nodes;

[0185] According to the cumulative weight of the node , adopt the formula:

[0186] ;

[0187] Calculate the maximum path of the cumulative weight , and obtain the associated topic path;

[0188] Among them, is used to find the node index that makes the largest ;

[0189] The calculation process is as follows:

[0190] For the formula:

[0191] ;

[0192] The association degree between nodes is obtained from the analysis of historical interaction data, such as by calculating semantic similarity or interaction frequency.

[0193] Set the parameter ;

[0194] Set three nodes, and the weights and association degrees are as follows:

[0195] Core node , and its association degree with other nodes is: , , and the node weight is: (core node), , .

[0196] The cumulative weight of node 2:

[0197] ;

[0198] The cumulative weight of node 3:

[0199] ;

[0200] The cumulative weight of node 2 ;

[0201] The cumulative weight of node 3 .

[0202] For the formula:

[0203]

[0204] Compare the cumulative weights of each node:

[0205] ;

[0206] ;

[0207] The maximum cumulative weight is node 2, that is:

[0208] 。

[0209] The calculation results show that Node 2 is the path with the largest cumulative association weight with the core Node 1, and its cumulative weight is 。Compared with the benchmark value (such as setting the cumulative weight to 1.0 as a reference), this value is significantly higher than the benchmark, indicating that Node 2 plays an important role in the semantic network with the core node. This shows that in the semantic network of the intelligent question-answering system, Node 2 may represent a key theme or semantic direction, worthy of key attention or further optimization. The calculation results help to identify the most influential semantic paths in the system and optimize the response effect.

[0210] Please refer to Figure 5 , the specific steps for obtaining the conditional probability of a node as a reply under the current question are as follows:

[0211] According to the current data input and user interaction data, use the formula:

[0212] ;

[0213] Calculate the data vector of node ; ;

[0214] Among them, is the current data input of node , representing the latest data collected from this node, is the user interaction data associated with node , including user queries and feedback, is the weight for adjusting the contributions of the current data input and user interaction data;

[0215] According to the data vector of node , referring to the path with the maximum cumulative weight , use the formula:

[0216] ;

[0217] Calculate the conditional probability of node as a reply under the current question;

[0218] Among them, is the model sensitivity adjustment parameter, determined through data fitting and optimization, is the influence weight of data volatility, obtained by analyzing the historical data of the node, is node The data volatility measure, usually the standard deviation, is obtained through historical data analysis. is the connection strength adjustment parameter, which is optimized according to the impact of node connection strength on the model. is the node The average connection strength with other nodes, calculated based on network analysis. is the topic path The node index of, used to determine which nodes to consider in the summation process. is the natural exponent, used to exponentiate the data vector , is the node The data vector of represents the data volatility measure of the node represents the node The average connection strength with the remaining nodes;

[0219] The calculation process is as follows:

[0220] For the formula:

[0221] ;

[0222] Similarly, for the node , according to the formula:

[0223] ;

[0224] Set the parameters and Obtained by monitoring the real-time data of node and user interactions. For example, the current data input can be statistical data such as the latest response time and access times of the node, while the user interaction data can include click-through rate, user feedback score, etc. , adjust the balance weight according to historical data analysis to highlight the contribution of the current data input.

[0225] Set the data input and interaction data of the node as follows:

[0226] Node 1: , ;

[0227] Node 2: , ;

[0228] Node 3: , ;

[0229] The comprehensive data vector of Node 1:

[0230] ;

[0231] Composite data vector of Node 2:

[0232]

[0233] Composite data vector of Node 3:

[0234]

[0235] For the formula:

[0236] ;

[0237] Set the parameter Determined by data fitting and optimization, it is 0.8. Obtained by analyzing the historical data of the node, it is 0.5. It is 0.6. Optimized according to the influence of the node connection strength on the model, and the standard deviation is , , , and the node connection strength is , , .

[0238] Calculate the conditional probability of Node 1:

[0239] ;

[0240] ;

[0241] ;

[0242] Calculate the conditional probability of Node 2:

[0243] ;

[0244] ;

[0245] ;

[0246] Calculate the conditional probability of Node 3:

[0247] ;

[0248] ;

[0249] ;

[0250] As a result, Node 2 has the highest conditional probability , indicating that it is most likely to be a response node under the current user question. The conditional probabilities of Node 1 and Node 3 are and , showing their secondary influence in the semantic network. This result helps the system accurately identify the most suitable response path in complex semantic situations, optimizing the user interaction experience.

[0251] Please refer to Figure 6 , and the specific steps to recalculate the node state value are as follows:

[0252] Refer to the conditional probability of node as a response under the current question , and use the formula:

[0253] ;

[0254] Calculate the connection strength between node and node ;

[0255] Among them, is the balance parameter, used to adjust the influence ratio of the original connection strength and the newly calculated probability value, adjusted based on historical data and response performance;

[0256] According to the connection strength between node and node , use the formula:

[0257] ;

[0258] Calculate the new state value of node ;

[0259] According to the new state value of node , use the formula:

[0260] ;

[0261] Obtain the updated node conditional probability , and generate the updated conditional probability value;

[0262] Among them, is the node index, indicating the state values of all nodes;

[0263] The calculation process is as follows:

[0264] For the formula:

[0265] ; ​​​​

[0266] Set parameters Based on historical data and response performance adjustment, for , conditional probability , , .

[0267] The initial connection strength is set to:

[0268] , ;

[0269] , ;

[0270] , ;

[0271] Update the connection strength :

[0272] ;

[0273] Update the connection strength :

[0274] ;

[0275] Update the connection strength :

[0276] ;

[0277] Update the connection strength :

[0278] ;

[0279] Update the connection strength :

[0280] ;

[0281] Update the connection strength :

[0282] ;

[0283] For the formula:

[0284] ;

[0285] Calculate the new state value of node 1 :

[0286] ;

[0287] The new state value of computing node 2 :

[0288] ;

[0289] The new state value of computing node 3 :

[0290] ;

[0291] The updated conditional probability of computing node 1 :

[0292] ;

[0293] The updated conditional probability of computing node 2 :

[0294] ;

[0295] The updated conditional probability of computing node 3 :

[0296] ;

[0297] The updated node state values and conditional probabilities reflect the adjustment of the dynamic relationships between nodes in the system. By introducing new connection strengths and probability values, the system can adapt to the changing user interaction environment in real time. The updated conditional probabilities show the redistribution of the relative importance of each node in the network, which helps to optimize the response accuracy and decision-making process of the system.

[0298] Please refer to Figure 7 , the specific steps for obtaining the matching degree between the computed reply content and the associated topic path are as follows:

[0299] According to the updated node conditional probabilities , use the formula:

[0300] ;

[0301] Get the relevance score of the candidate reply for node , indicating the relevance score of the candidate reply;

[0302] wherein, is the vector form of the th text or data item in the knowledge base, is the cosine similarity function, used to select the maximum value;

[0303] According to the updated conditional probability vector of point ​ , use the formula:

[0304] ;

[0305] Calculate the matching degree score of the reply content , filter the reply content according to the matching degree, and generate the intelligent Q&A result;

[0306] Among them, is the weight coefficient, obtained through system optimization, used to balance the influence of different matching degrees, is the number of nodes in the topic path, is the node index, used to consider the new state values of all nodes during the summation process, represents the topic path, which is a set composed of multiple nodes ;

[0307] The calculation process is as follows:

[0308] For the formula:

[0309] ;

[0310] The definition of cosine similarity is:

[0311] ;

[0312] The value returned by this function is between -1 and 1, where 1 represents complete similarity.

[0313] Set the following vectors:

[0314] ;

[0315] ;

[0316] Calculate the dot product:

[0317] ;

[0318] Calculate the modulus length of the vector:

[0319] ;

[0320] ;

[0321] Calculate the cosine similarity:

[0322] ;

[0323] Therefore, the correlation score is approximately 0.915.

[0324] For the formula:

[0325] ;

[0326] Set parameters , (i.e., the path contains 3 nodes), and the relevance scores of other nodes ;

[0327] Calculate the average cosine similarity of the path:

[0328] ;

[0329] Calculate :

[0330] ;

[0331] The final matching degree score is approximately 0.8955. This value represents the matching degree between the candidate response and the topic path. The higher the score, the higher the matching degree. In practical applications, a value close to 1 means that the candidate response is highly relevant and closely related to the topic path. A lower value may indicate that the response is less relevant or unrelated to the topic. By calculating these values, the system can preferentially select more accurate and contextually relevant responses.

[0332] A multi-scenario intelligent Q&A management method based on artificial intelligence. The multi-scenario intelligent Q&A management method based on artificial intelligence is executed based on the above-mentioned multi-scenario intelligent Q&A management system based on artificial intelligence, and includes the following steps:

[0333] S1: Based on multi-scenario data, collect keywords and topic sentences from it and define them as nodes, construct a semantic association graph between the nodes, calculate the path length and association strength between the nodes, reassign the node weights, and obtain optimized node weights;

[0334] S2: Based on the optimized node weights, map the user's question to the node set, calculate the matching degree with each node, select the node with the highest matching degree as the core node, calculate the cumulative weight of the node with the remaining nodes, determine the path with the largest cumulative weight, and generate an associated topic path;

[0335] S3: Utilize the associated topic path, monitor the current data input of the node, combine with the user interaction data, establish a Bayesian network to calculate the conditional probability of the node under the current question, and generate an updated conditional probability value;

[0336] S4: Based on the updated conditional probability value, obtain candidate response content from the knowledge base or preset text, calculate the matching degree between the response content and the associated topic path, screen the response content with the highest matching degree, and generate candidate response content;

[0337] S5: Refine the response content from the candidate response content according to the user interaction situation to generate an intelligent Q&A result.

[0338] The above are only the preferred embodiments of the present invention, and do not limit the present invention in other forms. Any person skilled in the art may use the technical content disclosed above to make changes or modifications into equivalent embodiments with equivalent changes and apply them to other fields. However, as long as it does not depart from the technical solution content of the present invention, any simple modification, equivalent change, and modification made to the above embodiments based on the technical essence of the present invention still fall within the protection scope of the technical solution of the present invention.

Claims

1. A multi-scenario intelligent question-answering management system based on artificial intelligence, characterized in that: The system comprises: The semantic smoothing optimization module collects keywords and topic sentences in multiple question-answering scenarios, defines each keyword and topic sentence as a node, establishes semantic associations between nodes, calculates the path length and association strength between nodes, and redistributes node weights after comparing the calculation results to obtain optimized node weights. The steps for obtaining the node weights after comparing the calculation results are specifically as follows: Collect keywords and topic sentences from multiple question-answering scenarios, define each keyword and topic sentence as a node, establish semantic associations between nodes, and use the formula: ； Compute Node With Node The semantic relationship strength between ; in, is the node identifier, Is a node In the The frequency of words in the question-answering scenario, Is a node In the The frequency of words in the question-answering scenario, is the index of the question-answering scenario, is the total number of question-answering scenarios; According to the node With Node The semantic relationship strength between , using the formula: ； Compute Node To Node The path length ; According to the node To Node The path length , using the formula: ； Compute Node Updated weights , get the optimized node weight; in, Is a node The original weight of is the weight adjustment factor, which is determined through experiments or parameter adjustment; The user intention recognition module maps the user's question to a node set based on the optimized node weights, calculates the matching degree with each node, selects the core node with the highest matching degree, calculates the cumulative weight of all associated nodes according to the correlation between the core node and the remaining nodes, determines the path with the largest cumulative weight, and obtains the associated topic path; The data fusion adjustment module monitors the current data input of the node based on the associated topic path, combines the user interaction data, establishes a Bayesian model to calculate the conditional probability of the node as a response under the current question, updates the connection strength and relationship of the node, recalculates the node state value, and generates an updated conditional probability value; The reply path selection module obtains candidate reply content from the knowledge base or preset text based on the updated conditional probability value, calculates the matching degree between the reply content and the associated subject path, screens the candidate reply content according to the matching degree, and generates an intelligent question and answer result; The steps for obtaining the degree of matching between the reply content and the associated subject path are specifically as follows: According to the updated node conditional probability , using the formula: ； Get Node The relevance score of the candidate answer , represents the relevance score of the candidate response; in, The knowledge base A vector of text or data items, is the cosine similarity function, Used to select the maximum value; According to the node The relevance score of the candidate answer ； Calculate the matching score of the reply content , filter candidate responses based on matching degree and generate intelligent question-answering results; in, is the weight coefficient, is the number of nodes in the topic path, is the node index, Represents the topic path, which is composed of multiple nodes Composed of a collection.

2. The multi-scenario intelligent question-answering management system based on artificial intelligence according to claim 1 is characterized in that: The steps of obtaining the core node with the highest matching degree are specifically as follows: According to the node Updated weights , using the formula: ； Calculate user questions and nodes The mapping score ; in, is a binary indicator function. When the key words in the question Belongs to Node The value is 1 when , otherwise it is 0. is the total number of keywords in the question, and is the adjustment coefficient, which is obtained by the correlation between the keywords in the user's question and the node. Is a node The frequency weight of According to the user's question and node The mapping score , using the formula: ； Compute Node The matching degree ; in, is the total number of nodes, indicating the number of all defined nodes, It is the user's question and node The mapping score of According to the node The matching degree , using the formula: ； Select core nodes based on matching degree ; in, Used to find out Maximum node index .

3. The multi-scenario intelligent question and answer management system based on artificial intelligence according to claim 2 is characterized in that: The step of obtaining the path with the largest cumulative weight is specifically as follows: Select the core node based on the matching degree , using the formula: ； Compute and core nodes Associated Nodes The cumulative weight of ; in, It is a core node With Node The correlation between Is a node The optimization weight of is the total number of nodes; According to the node The cumulative weight of , using the formula: ； Calculate the maximum path with cumulative weight , get the associated topic path; in, Used to find out Maximum node index .

4. The multi-scenario intelligent question-answering management system based on artificial intelligence according to claim 3 is characterized in that: The steps for obtaining the conditional probability of the computing node as a response to the current question are specifically as follows: According to the current data input and user interaction data, the formula is adopted: ； Compute Node The data vector ; in, Is a node The current data input, Is with the node associated user interaction data, It is to adjust the weight of the current data input and the contribution of user interaction data; According to the node The data vector , refer to the maximum path with cumulative weight , using the formula: ； Compute Node The conditional probability of being the answer under the current question ; in, is the model sensitivity adjustment parameter, which is determined by data fitting and optimization. is the impact weight of data volatility, obtained by analyzing the historical data of the node. Is a node The data volatility measure is obtained through historical data analysis. is the connection strength adjustment parameter, which is optimized according to the influence of node connection strength on the model. Is a node The average connection strength with the rest of the nodes, calculated based on network analysis, Is the subject path The node index of is the natural index, Is a node The data vector, Representation Node The data volatility measure is Representative Node The average connection strength with the rest of the nodes.

5. The multi-scenario intelligent question-answering management system based on artificial intelligence according to claim 4 is characterized in that: The steps for recalculating the node status value are specifically as follows: Reference to the node The conditional probability of being the answer under the current question , using the formula: ； Compute Node With Node The connection strength between ; in, is a balancing parameter, adjusted based on historical data and response performance; According to the node With Node The connection strength between , using the formula: ； Compute Node The new status value ; According to the node The new status value , using the formula: ； Get the updated node conditional probability , generate updated conditional probability values; in, It is the node index, which indicates the status value of all nodes.

6. A multi-scenario intelligent question-answering management method based on artificial intelligence, characterized in that: The method according to any one of claims 1 to 5 is implemented by the multi-scenario intelligent question and answer management system based on artificial intelligence, comprising the following steps: Based on multi-scenario data, keywords and topic sentences are collected to define nodes, a semantic association graph between nodes is constructed, the path length and association strength between nodes are calculated, and node weights are redistributed to obtain optimized node weights; Based on the optimized node weights, the user's questions are mapped to a node set, the matching degree with each node is calculated, the node with the highest matching degree is selected as the core node, the cumulative weight of the node and the remaining nodes is calculated, the path with the largest cumulative weight is determined, and the associated topic path is generated; Using the associated topic path, monitoring the current data input of the node, combining the user interaction data, establishing a Bayesian network to calculate the conditional probability of the node under the current question, and generating an updated conditional probability value; Based on the updated conditional probability value, candidate reply content is obtained from the knowledge base or preset text, the matching degree between the reply content and the associated subject path is calculated, the reply content with the highest matching degree is screened, and the candidate reply content is generated; From the candidate answer contents, the answer contents are refined according to the user interaction situation to generate intelligent question and answer results.

Images