Auxiliary response method and device based on artificial intelligence, equipment and medium

Abstract

The invention relates to the technical field of artificial intelligence, can be applied to service system platforms of medical health, financial science and technology and the like, and discloses an auxiliary response method, device and equipment based on artificial intelligence and a medium. Inputting the historical search records into the confidence rule base model to generate corresponding fuzzy confidence distribution; fusing different types of fuzzy confidence distributions in the identification framework; and constructing a rule weight and feature weight optimization model, calculating a matching degree and an activation degree of the fuzzy confidence distribution in a corresponding rule fuzzy domain, outputting a predicted intention classification according to a confidence maximum principle, and sending the predicted intention classification to the seat terminal. According to the method, confidence distributions of different categories in a framework are identified through fusion processing, and an input and output relationship between a feature space and a category space is constructed; and identification decision is carried out by adopting the principle of maximum confidence, invalid lists or low-quality lists can be filtered, the real intention of the user is identified, and the contact efficiency of the seat lists is improved.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based assisted response method, apparatus, device, and medium. Background Technology

[0002] In the financial sector, customer service personnel serve as the core bridge connecting customers and financial institutions, with their applications covering the entire process from customer consultation and business processing to risk control and marketing conversion. For example, in the insurance industry, they handle insurance claims, document review, claims progress inquiries, and policy renewal reminders. When customers encounter accidents (such as car crashes or illnesses), agents need to use empathetic communication to alleviate customer anxiety and enhance brand trust. They also identify suspicious claims through dialogue analysis (such as inquiring about accident details) to reduce payout risks.

[0003] In the medical field, customer service representatives (or medical coordinators) serve as the core communication hub between patients and medical institutions and health service providers, and their application scenarios span the entire process of pre-diagnosis consultation, in-diagnosis coordination, post-diagnosis follow-up, and health management.

[0004] However, in the financial and medical fields, agents have many customer lists to handle every day, and each list requires agent interaction. Relying solely on agents to answer customer questions consumes a lot of time and manpower. When there is a break in the operation, agents will repeatedly ask the same questions, resulting in a poor user experience. Summary of the Invention

[0005] In view of the shortcomings of the prior art, the present invention provides an artificial intelligence-based assisted response method, device, equipment and medium, which aims to solve the problems in the prior art where agents have many customer lists to deal with every day, each list requires agent contact, and relying solely on agents to answer customer inquiries consumes a lot of time and manpower costs. When there is a break in the operation, the agent will repeatedly ask the same questions, resulting in a poor user experience.

[0006] The technical solution of the present invention is as follows: The first embodiment of the present invention provides an artificial intelligence-based assisted response method, the method comprising: The fuzzy set segmentation method and membership function of the confidence rule base are set in advance to generate the confidence rule base model; Obtain the list of breakpoints, and retrieve the user's historical search records based on the breakpoint list; The historical search records are input into the confidence rule base model, and a fuzzy confidence distribution corresponding to the historical search records is generated based on the fuzzy set segmentation method and membership function. Fusion identification of fuzzy confidence distributions of different categories within the framework; Construct a rule weight and feature weight optimization model, and based on the rule weight and feature weight optimization model, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain; The confidence level is calculated based on the matching degree and activation degree, and the predicted intent classification is output according to the principle of maximizing the confidence level. The predicted intent classification is then sent to the agent terminal.

[0007] Another embodiment of the present invention provides an artificial intelligence-based assisted response device, the device comprising: The configuration module generates a confidence rule base model by pre-setting the fuzzy set segmentation method and membership function of the confidence rule base. The data acquisition module is used to obtain a list of breakpoints and to retrieve the user's historical search records based on the list of breakpoints. The data distribution generation module is used to input the historical search records into the confidence rule base model and generate the fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function. The fusion module is used to fuse fuzzy confidence distributions of different categories within the identification framework; The calculation module is used to construct a rule weight and feature weight optimization model, and based on the rule weight and feature weight optimization model, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain; The data sending module is used to calculate the confidence level based on the matching degree and activation degree, output the predicted intent classification according to the principle of maximizing the confidence level, and send the predicted intent classification to the agent terminal.

[0008] Another embodiment of the present invention provides a computer device, the computer device including at least one processor; and, A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, which, when executed, enable the at least one processor to perform the steps of the artificial intelligence-based assisted response method described above.

[0009] Another embodiment of the present invention provides a computer-readable storage medium storing computer-executable instructions that, when executed by one or more processors, cause the one or more processors to perform the steps of the artificial intelligence-based assisted response method described above.

[0010] Beneficial Effects: The AI-based assisted response method, apparatus, device, and medium of this invention include: pre-setting the fuzzy set segmentation method and membership function of the confidence rule base to generate a confidence rule base model; obtaining a breakpoint list and obtaining the user's historical search records based on the breakpoint list; inputting the historical search records into the confidence rule base model and generating a fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function; fusing fuzzy confidence distributions of different categories within the identification framework; constructing a rule weight and feature weight optimization model and calculating the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain based on the rule weight and feature weight optimization model; calculating the confidence degree based on the matching degree and activation degree, outputting the predicted intent classification according to the principle of maximizing confidence degree, and sending the predicted intent classification to the agent terminal. This invention, through the fusion processing of confidence distributions of different categories within the identification framework, establishes an optimization model for rule weights and feature weights, constructing the input-output relationship between the feature space and the category space. Based on this, it calculates the matching degree and activation degree of unknown intent data in the corresponding rule fuzzy domain, and uses the principle of maximizing confidence for identification decisions. This can filter invalid or low-quality lists, identify the user's true intent, and improve the contact efficiency of the agent list. Attached Figure Description

[0011] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a schematic diagram illustrating the application environment of an embodiment of an artificial intelligence-based assisted response method according to the present invention; Figure 2 This is a flowchart of a preferred embodiment of an artificial intelligence-based assisted response method of the present invention; Figure 3 This is a schematic diagram of the functional modules of a preferred embodiment of an artificial intelligence-based assisted response device of the present invention; Figure 4 This is a schematic diagram of a preferred embodiment of a computer device according to the present invention; Figure 5 This is another structural schematic diagram of a preferred embodiment of a computer device according to the present invention. Detailed Implementation

[0013] To make the objectives, technical solutions, and effects of this invention clearer and more explicit, the invention is further described in detail below. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0014] The embodiments of the present invention will be described below with reference to the accompanying drawings. Those skilled in the art will recognize that, with technological advancements and the emergence of new scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0015] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such terms are interchangeable where appropriate; this is merely a way of distinguishing objects with the same attributes in the embodiments of this application. Here, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion, so that a process, method, system, product, or apparatus that comprises a series of elements is not necessarily limited to those elements, but may include other elements not explicitly listed or inherent to those processes, methods, products, or apparatuses.

[0016] The method provided in this application can be applied to artificial intelligence (AI) scenarios. AI is the theory, method, technology, and application system that uses digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results. In other words, artificial intelligence is a branch of computer science that attempts to understand the essence of intelligence and produce a new kind of intelligent machine capable of reacting in a way similar to human intelligence. Artificial intelligence studies the design principles and implementation methods of various intelligent machines, enabling them to have the functions of perception, reasoning, and decision-making. Research in the field of artificial intelligence includes robotics, natural language processing, computer vision, decision-making and reasoning, human-computer interaction, recommendation and search, and fundamental AI theories.

[0017] The AI-based assisted response method provided in this invention can be applied to, for example... Figure 1In this application environment, the client communicates with the server via a network. The client accesses the server's network or business platform. The server can pre-set the fuzzy set segmentation method and membership function of the confidence rule base to generate a confidence rule base model; obtain a list of breakpoints, and retrieve the user's historical search records based on the breakpoint list; input the historical search records into the confidence rule base model, and generate a fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function; fuse the fuzzy confidence distributions of different categories within the identification framework; construct a rule weight and feature weight optimization model, and calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain based on the rule weight and feature weight optimization model; calculate the confidence degree based on the matching degree and activation degree, and output the predicted intent classification according to the principle of maximizing confidence degree, and send the predicted intent classification to the agent terminal. In this invention, by fusing the confidence distributions of different categories within the identification framework, an optimization model for rule weights and feature weights is established, constructing the input-output relationship between the feature space and the category space. Based on this, the matching degree and activation degree of unknown intent data in the corresponding rule fuzzy domain are calculated, and the identification decision is made using the principle of maximizing confidence. This can filter invalid or low-quality lists, identify the user's true intent, and improve the contact efficiency of the agent list. The client can be, but is not limited to, various personal computers, laptops, smartphones, tablets, and portable wearable devices. The server can be implemented using a standalone server or a server cluster consisting of multiple servers. The invention will be described in detail below through specific embodiments.

[0018] To address the above problems, embodiments of the present invention provide an artificial intelligence-based assisted response method. Please refer to [link to relevant documentation]. Figure 2 , Figure 2 This is a flowchart illustrating a preferred embodiment of an artificial intelligence-based assisted response method according to the present invention. Figure 2 As shown, it includes: Step S100: Pre-set the fuzzy set segmentation method and membership function of the confidence rule base to generate the confidence rule base model.

[0019] This invention addresses these problems in call center operations involving breakpoint lists. In the financial and healthcare industries, a shortage of call center staff leads to a large volume of breakpoint tasks being handled by these staff, hindering the timely acquisition of user historical information and resulting in low work efficiency.

[0020] In setting up the confidence rule base model, this invention improves the connection relationship of the premise attributes in the premise part of the confidence rules, designs fuzzy set segmentation based on the statistical distribution characteristics of the dataset, and selects Cauchy distribution as the membership function, which better avoids the problem that the confidence rules cannot be effectively activated, thus leading to no effective output of the system.

[0021] The fuzzy confidence principle in this invention combines fuzzy logic and confidence theory, aiming to address data uncertainty, fuzziness, and probabilistic uncertainty. By quantifying the membership degree of fuzzy sets and the reliability of rules, it provides more flexible and realistic reasoning and decision support for complex systems. Fuzzy logic maps continuous attribute values ​​(such as temperature and user preference intensity) to the [0,1] interval through membership functions, representing the degree to which an element belongs to a certain set. For example, in social recommendation, a user's preference for a product can be represented as a fuzzy set such as "high (0.8)", "medium (0.5)", and "low (0.2)", rather than an absolute either / or choice.

[0022] Among them, the Cauchy distribution, as a membership function, is an unconventional choice in fuzzy logic. Its unique mathematical properties offer a new perspective on handling uncertainty. The Cauchy membership function can handle extreme values ​​and fuzzy boundaries. In traditional fuzzy logic, Gaussian or triangular membership functions may experience a sharp drop in membership due to extreme values, while the heavy tails of the Cauchy distribution cause membership to decay slowly away from the center, better aligning with human intuition about "fuzziness." For example, in temperature control, if the target temperature is 25℃, using the Cauchy membership function, the membership at 20℃ or 30℃ will still be relatively high (e.g., 0.5), while a Gaussian distribution might be close to 0, more reasonably reflecting the fuzziness of "moderate temperature." The Cauchy membership function is robust. It is insensitive to noise or outliers in the data, making it suitable for handling imprecise or incomplete data. For example, in social recommendation, user behavior data may contain accidental abnormal clicks; the Cauchy distribution can reduce their impact on the overall membership. By adjusting the position parameter x0 and the scale parameter γ, the center and width of the membership function can be flexibly controlled to adapt to the needs of different scenarios. For example, in fault diagnosis, increasing γ can expand the fuzzy range of the "normal state" and reduce false alarms.

[0023] The Belief Rule Base (BRB) is a hybrid intelligent model combining fuzzy logic, evidence theory, and rule systems to handle complex decision-making problems under uncertainty and incomplete information. It quantifies the uncertainty of rules by introducing a belief degree, enabling more flexible expression of fuzziness, randomness, and cognitive uncertainty in expert knowledge and real-world data. A belief rule consists of a condition part (input) and a conclusion part (output). The belief degree represents the degree of uncertainty in the rule's conclusion and can be derived from expert experience, historical data statistics, or model learning. The difference from membership degree in fuzzy logic is that membership degree describes the degree to which an input variable belongs to a fuzzy set, while belief degree describes the credibility of the output result. Rule weights represent the weights that can be assigned to each rule, wi∈[0,1], indicating the importance or reliability of the rule (usually 1 by default).

[0024] The construction steps of a confidence rule base include defining input / output variables: determining the input variables (e.g., temperature, humidity) and output variables (e.g., failure rate, risk level). Each variable is divided into fuzzy sets (e.g., "low," "medium," "high"), which can be done through statistical distributions, expert experience, or clustering methods. Confidence rule generation methods include having domain experts directly write rules based on their knowledge and specifying the confidence level of the output results; or extracting input-output samples from historical data and generating rules using clustering algorithms (e.g., K-means) or classification algorithms (e.g., decision trees). The confidence level of the output results is calculated using statistical methods (e.g., frequency analysis). For example, if "high failure rate" occurs 30% of the time under a certain input combination, its confidence level is 0.3. Combining expert knowledge and data, the rule framework is first defined by experts, and then the confidence level is adjusted using data. Rule base optimization methods can be achieved by deleting redundant or conflicting rules (e.g., two rules with the same input but contradictory outputs). Optimization algorithms (e.g., genetic algorithms, particle swarm optimization) are used to adjust rule weights and confidence levels to minimize prediction errors (e.g., mean squared error, classification accuracy). For time-varying data, the rule base is retrained periodically to adapt to new patterns.

[0025] Step S100, which involves pre-setting the fuzzy set segmentation method and membership function of the confidence rule base to generate the confidence rule base model, includes: Step S101: Pre-set the fuzzy set segmentation method of the confidence rule base according to the statistical distribution characteristics of the dataset; Step S102: Pre-construct a membership function based on the Cauchy distribution; Step S103: Generate a confidence rule base model based on the fuzzy set segmentation method and membership function.

[0026] In the intent recognition method based on fuzzy confidence rules, setting the number of fuzzy sets is a fundamental step in realizing the fuzzy confidence rules. The model adopts a data-driven strategy to realize target intent recognition, so fuzzy set segmentation is performed according to the statistical distribution characteristics of the dataset.

[0027] Fuzzy set segmentation based on the statistical distribution characteristics of a dataset is a method combining data-driven approaches and fuzzy logic. It aims to dynamically delineate the boundaries and membership function parameters of fuzzy sets by analyzing the actual distribution characteristics of the data (such as mean, variance, skewness, kurtosis, etc.), thereby improving the adaptability and interpretability of the fuzzy system. The segmentation method specifically includes: data preprocessing and distribution analysis. First, data cleaning is performed to remove outliers and missing values ​​to ensure data quality. Next, the distribution type is identified, for example, by using histograms, QQ plots, kernel density estimation (KDE), or statistical tests (such as the Shapiro-Wilk test) to determine whether the data follows a normal, exponential, or bimodal distribution. Then, parameter estimation is performed. For example, the mean (μ), standard deviation (σ), skewness, and kurtosis of the data are calculated to provide a basis for the design of the membership function. Fuzzy set partitioning strategies include selecting partition points based on distribution characteristics, such as for unimodal distributions (such as normal distributions): dividing intervals by standard deviation with the mean as the center. For example, data can be divided into three fuzzy sets: "low," "medium," and "high," with split points μ-σ and μ+σ. For skewed distributions (e.g., right-skewed distributions): the split points can be adjusted based on the skewness. For example, if the data is right-skewed, the boundary of the "high" fuzzy set can be extended to the right to cover the long tail. For multimodal distributions (e.g., bimodal distributions): the peak positions can be identified, and each peak can be assigned to the center of a fuzzy set, with the split point located near the valley. Membership function parameters can also be dynamically adjusted: the width of the membership function (e.g., σ for a Gaussian distribution or γ for a Cauchy distribution) can be adjusted based on the dispersion of the data distribution (e.g., variance). The larger the variance, the wider the membership function, to cover more dispersed data. The symmetry of the membership function can also be adjusted based on the skewness. For example, for a right-skewed distribution, the membership function of the "high" fuzzy set can be designed as a right-skewed Gaussian or Cauchy distribution.

[0028] In fuzzy sets, membership functions need to be determined. The membership degree of an intent-identifying element belonging to a fuzzy set is objectively real. Although it may seem subjective on the surface, there is indeed an intermediate transition between the differences in intents of different targets. This objectively limits the membership degree, thus making the selection of the membership degree subject to objective laws.

[0029] A fuzzy set A on the universe of discourse X is characterized by a membership function u_A(x), which can take values ​​on the closed interval [0,1]. For any x∈X, there is a unique membership function u_A(x)∈[0,1] corresponding to it. The system selects a Cauchy-type distribution as the membership function through statistical analysis of target intent recognition data, and determines parameters that are more consistent with reality based on training.

[0030] Step S200: Obtain the breakpoint list and retrieve the user's historical search records based on the breakpoint list.

[0031] A breakpoint list is a crucial tool in customer service management for tracking, managing, and optimizing incomplete service processes (i.e., "breakpoints"). By recording customer-agent interactions interrupted for various reasons, it helps teams follow up promptly, reduce customer churn, and improve service efficiency. After obtaining the breakpoint list, user information is retrieved, and their historical search records are analyzed to provide raw data for intent analysis. By analyzing user search records, the true intent of the user is determined, and this information is then used to provide prompts to agents during breakpoint handling, thereby improving agent efficiency.

[0032] Step S300: Input the historical search records into the confidence rule base model, and generate the fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function.

[0033] To achieve target intent recognition based on fuzzy confidence rules, the primary task is to construct a confidence rule base describing the input-output relationship between the feature space and the category space based on the training dataset. Historical search records are input into the confidence rule base model. In this embodiment of the invention, fuzzy set segmentation is based on data distribution characteristics. The membership function uses a Cauchy distribution. The user's historical search records are obtained; taking insurance as an example, the user's historical search history before connecting to the breakpoint job in the database is obtained, and the corresponding fuzzy confidence distribution is obtained based on the membership function. Fuzzy set segmentation refers to dividing the domain of the problem (e.g., the user's level of interest in a topic, from 0 to 100) into several semantically meaningful "fuzzy" categories, such as "low," "medium," and "high."

[0034] The membership function defines the degree to which a specific numerical value (e.g., interest value 75) belongs to a certain fuzzy set (e.g., "high"). The membership range is between [0, 1].

[0035] Historical search records: A quantifiable metric is needed to represent a user's historical behavior. A common method is to calculate search frequency, for example, the number of times a user has searched for the keyword "artificial intelligence" in the past month.

[0036] A fuzzy confidence distribution is the assignment of confidence levels across various fuzzy categories (such as "low", "medium", and "high") for a given historical search frequency. This distribution is typically calculated using a membership function.

[0037] Step S300, which involves inputting the historical search records into the confidence rule base model and generating a fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function, includes: Step S301: Obtain the training dataset for training the confidence rule base; Step S302: Construct a confidence rule base model based on the training dataset; Step S303: Input the historical search records into the trained confidence rule base model, and generate the fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function of the confidence rule base model.

[0038] In the confidence rule structure, the conclusion part adopts the confidence distribution form. Let each sample have M features, and the k-th rule be... The premise is The training sample subset is The class tag set is The training samples are Its different premises The matching degree is

[0039] The category with the highest matching degree in the training samples is used as the conclusion part of the rule.

[0040] Step S400: Fusion of fuzzy confidence distributions of different categories within the identification framework.

[0041] In fuzzy belief theory, the Frame of Discernment (FFD) is the foundation for constructing confidence functions and performing uncertainty reasoning. It defines a complete set of all possible propositions and describes the uncertainty of these propositions using fuzzy sets and confidence distributions. An FDD is a set of mutually exclusive propositions that constitute a complete set, representing all possible answers to a given question. In fuzzy belief theory, the FDD not only includes traditional deterministic propositions but also allows for fuzziness, meaning that the boundaries between propositions are not clear-cut but rather have certain transitional regions. The FDD is used to construct confidence functions in fuzzy belief. The confidence function is a core concept in fuzzy belief theory, describing the degree of confidence in the truth of a proposition. The FDD provides the foundation for constructing confidence functions because the confidence function needs to assign values ​​to each proposition within the FDD. In fuzzy belief theory, uncertainty information can be described using confidence distributions. Confidence distributions represent the distribution of the credibility of each proposition within the FDD. The identification framework provides a domain for confidence distributions, enabling us to quantify and reason about uncertain information. Fuzzy reasoning is a reasoning method based on fuzzy logic, which allows for fuzziness in the reasoning process. The identification framework provides a reasoning space for fuzzy reasoning, allowing us to reason and make decisions about fuzzy propositions within the framework.

[0042] Step S400, which involves fusing the fuzzy confidence distributions of different categories within the identification framework, includes: Obtain the current set of all categories and use the category set as the recognition framework; Obtain the confidence score for each category within the identification framework; Based on the evidence theory synthesis rules, the confidence levels of various categories are fused to generate a fused fuzzy confidence distribution.

[0043] The identification framework is based on fuzzy confidence rule-based modeling and reasoning. It involves classifying the set of categories. Considered as a framework for identification. The representative is assigned to the premise fuzzy region. The training sample set. For the training samples Its category This can be seen as support This is evidence corresponding to the conclusion of the rule. Within the framework of DS theory, this evidence can be interpreted as... Perform confidence assignment separately, and then assign the remaining confidence to the identification frame. This is used to characterize global uncertainty. This evidence can be expressed using the mass function of the following formula. to indicate

[0044] in, ≤1.

[0045] To obtain the premise The corresponding conclusion section allows for the fusion of the mass function based on Dempster's combination rule. (Rule) The confidence levels for each category in the conclusion section are:

[0046] in, This indicates the confidence level that has not been assigned to any single category.

[0047] Step S500: Construct a rule weight and feature weight optimization model, and based on the rule weight and feature weight optimization model, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain.

[0048] The core optimization process involves constructing rule weight and feature weight optimization models, addressing the question of "how to make the rule base more accurate." This is a typical data-driven parameter learning process. Optimization parameters include rule weights, which represent the importance of a rule within the entire rule base. Higher weights indicate a greater impact of the rule on the final conclusion. Feature weights represent the influence of a given attribute on the result. Higher weights indicate a more important feature, contributing more to rule matching.

[0049] An optimization model for rule weights and feature weights was established, and the input-output relationship between the feature space and the category space was constructed. Based on this, the matching degree and activation degree of unknown intent data in the corresponding rule fuzzy domain were calculated, and the identification decision was made using the principle of maximum confidence. This can filter invalid or low-quality lists, identify the user's true intent, and improve the contact efficiency of the agent list.

[0050] Step S500, namely, constructing a rule weight and feature weight optimization model, and calculating the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain based on the rule weight and feature weight optimization model, includes: Step S501: Construct an optimization model for rule weights and feature weights; Step S502: Obtain the rule weights and feature weights, and optimize the rule weights and feature weights based on the rule weight and feature weight priority model to generate target rule weights and target feature weights; Step S503: Based on the target rule weight and the target feature weight, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain.

[0051] In the confidence rule structure, the initial weight of each rule is... The number of rule weight reference values ​​to be estimated is K. The initial weight of each rule is not the rule weight used in the next step of the evidence reasoning algorithm. Feature weights This reflects the relative degree of influence of the premise attribute features on the conclusion of the rule. If the weight of the premise attribute feature is larger, its influence on the conclusion of the rule will be greater.

[0052] The output of the ER algorithm should be consistent with each confidence rule. ER stands for Evidence-Based Reasoning, and it's the engine of the Confidence Rule Base (BRB) system. It's used to integrate multiple incomplete, uncertain, or even conflicting pieces of evidence (i.e., rule conclusions in the BRB) into a comprehensive and reliable final conclusion.

[0053] The result of reasoning using a confidence rule base can be calculated using the following formula:

[0054] The mean square error can then be expressed as

[0055] in, Represents actual observation data, This represents the inference result, and T represents the number of data points in the training dataset.

[0056] By using rule weights and feature weights as decision variables in the optimization model, and mean squared error as the optimization objective, the objective function can be defined as follows:

[0057] For any training sample, if the system identifies it correctly, then =0, if the system identification result is incorrect, then =1, then the optimization objective model can be expressed as .

[0058] In step S503, based on the target rule weights and target feature weights, the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain are calculated, including: Step S531: Based on the rule weight and feature weight optimization model and the target rule weight, obtain the matching degree of the historical search record in the rule fuzzy domain; Step S532: Obtain the set of confidence rules activated for all historical search records in the confidence rule base model; Step S533: Based on the matching degree and the target rule weight, calculate the activation degree of each confidence rule in the historical search records and confidence rule set.

[0059] set up This represents the unknown intent data to be identified, which allows us to obtain the unknown intent within the rules. Fuzzy domain The matching degree on is

[0060] in, ( ) is a fuzzy set with premise The membership function. Let... This represents the set of rules in the confidence rule base that are activated by all input samples, i.e.

[0061] In the confidence conclusion section, according to The value is used to determine whether a rule is activated, which mainly depends on two factors: matching degree. and rule weight . It is a representation of the degree of similarity between the unknown intent and the premise of the confidence rule, while This represents the stability of the confidence rule. Let This indicates that the input sample y corresponds to the rule activation level, for activation level Defined as

[0062] Essentially, activation can characterize the effectiveness of classifying the input sample y in the conclusion part of the activated rule.

[0063] Step S600: Calculate the confidence level based on the matching degree and activation degree, and output the predicted intent classification according to the principle of maximizing confidence level, and send the predicted intent classification to the agent terminal.

[0064] Step S600, which involves calculating the confidence level based on the matching degree and activation degree, and outputting the predicted intent classification according to the principle of maximizing confidence, includes: Step S601: Generate the corresponding confidence level based on the activation level of each confidence rule; Step S602: Perform discount operations on the activated confidence rules using the discount operator to generate the target confidence level; Step S603: Utilize the principle of maximizing confidence and the target confidence level to make an identification decision and output the predicted intent classification; Step S604: The predicted intent is classified and sent to the agent terminal.

[0065] Within the framework of fuzzy confidence rules, the Shafer discount operator is used to perform discount operations on the activated confidence rules. The discount must consider the influence of activation degree on the conclusion of the activated rule, thus obtaining...

[0066] The mass function is fused using Dempster's combination rule, except... and (k=1,2,..., ) Apart from that, the basic probability of all other elements is assigned to zero. The analytical expression for a combination rule can be written as:

[0067]

[0068]

[0069] The identification decision is made using the principle of maximizing confidence, i.e., the identification result is represented as follows:

[0070] By integrating the confidence distributions of different categories within the identification framework, a rule weight and feature weight optimization model is established, and the input-output relationship between the feature space and the category space is constructed. Based on this, the matching degree and activation degree of unknown intent data in the corresponding rule fuzzy domain are calculated, and the identification decision is made using the principle of maximizing confidence. This can filter invalid or low-quality lists, identify the user's true intent, and improve the contact efficiency of the agent list.

[0071] The agent obtains the breakpoint list issued by the system. At the same time, the agent also obtains the user's online search history. Then, through model training, the agent obtains the user's true intent classification and provides real-time prompts to the agent through bubble messages. In this way, the agent knows the user's true intent in advance before communicating with the user, which improves the agent's work efficiency.

[0072] Bubble messages, providing real-time prompts to agents, are an efficient and intuitive interactive design method that significantly improves customer service efficiency and customer experience. The underlying support for bubble messages includes real-time communication technology, front-end interaction design, and back-end logic processing. Real-time communication technology includes the WebSocket protocol to establish a full-duplex communication channel, ensuring low-latency message transmission (e.g., customer input is displayed in the bubble on the agent's end in real time). Long polling is used as an alternative to simulate real-time effects when WebSocket is unavailable, suitable for scenarios with high compatibility requirements. The MQTT protocol is a lightweight push notification protocol, suitable for mobile devices or resource-constrained environments, reducing bandwidth consumption. Front-end interaction design includes bubble component development: using HTML / CSS / JavaScript to implement dynamic bubble effects (e.g., pop-up position, animation transitions, disappearance time). Framework integration: encapsulating bubble components based on frameworks such as Vue / React, supporting custom styles and behaviors (e.g., triggering a quick reply when clicking a bubble). Accessibility design: adding ARIA tags to the bubbles to ensure visually impaired users can access information through screen readers. Backend logic processing includes message queues: decoupling message production and consumption through middleware such as RabbitMQ / Kafka to avoid system crashes under high concurrency. Intelligent routing: automatically assigning message bubble notifications based on customer issue type and agent skill tags (e.g., routing "return / exchange" issues to after-sales service). Status synchronization: recording the read / unread status of message bubble messages to avoid duplicate notifications (e.g., the relevant bubble automatically disappears after the customer replies "okay").

[0073] Bubble messages allow for real-time input preview; as customers type in the input box, the agent's terminal displays the content in a bubble (e.g., "Hello, I would like to inquire about insurance premiums"), helping agents prepare responses in advance. Sensitive word filtering: Automatically blocks abusive, advertising, and other prohibited content; the bubble displays "[Sensitive information filtered]".

[0074] When providing quick reply suggestions, high-frequency reply templates are recommended in the bubble based on historical conversations and the knowledge base. Dynamic adjustments are made: suggestions are dynamically updated based on customer question keywords (e.g., recommending "invoice issuance process" when "invoice" is entered). For multi-task reminders, when handling multiple conversations simultaneously, the bubble displays the number of unread messages in a tag format (e.g., "Conversation A (3 unread messages)") to avoid omissions. For priority sorting, the bubble display order can be adjusted based on customer level (e.g., VIP) and issue urgency (e.g., complaint). For emotion recognition and warnings, customer tone can be analyzed using NLP (e.g., anger, anxiety), and the bubble displays an emotion tag (e.g., "Customer is emotionally agitated") to remind agents to adjust their communication strategies. During automatic transfer, when the emotion score exceeds a threshold, the bubble prompts "Recommend transferring to senior customer service."

[0075] Compared to existing technologies, this invention first improves the connection relationship of the premise attributes in the premise part of the confidence rule. It designs fuzzy set segmentation based on the statistical distribution characteristics of the dataset and selects a Cauchy-type distribution as the membership function, effectively avoiding the problem of confidence rules not being effectively activated, thus leading to no effective output from the system. Secondly, it integrates the confidence distributions of different categories within the identification framework, establishes an optimization model for rule weights and feature weights, and constructs the input-output relationship between the feature space and the category space. Based on this, it calculates the matching degree and activation degree of unknown intent data in the corresponding rule fuzzy domain, and uses the principle of maximizing confidence for identification decisions. This can filter invalid or low-quality lists, identify the user's true intent, and improve the contact efficiency of the agent list.

[0076] It should be noted that there is no necessary order between the above steps. Those skilled in the art will understand from the description of the embodiments of the present invention that the above steps may have different execution orders in different embodiments, that is, they may be executed in parallel or in turn, etc.

[0077] Another embodiment of the present invention provides an artificial intelligence-based assisted response device, which corresponds one-to-one with the artificial intelligence-based assisted response method described in the above embodiments. For example... Figure 3 As shown, device 1 includes: Module 100 is configured to generate a confidence rule base model by pre-setting the fuzzy set segmentation method and membership function of the confidence rule base. The data acquisition module 200 is used to acquire a list of breakpoints and to acquire the user's historical search records based on the list of breakpoints. The data distribution generation module 300 is used to input the historical search records into the confidence rule base model and generate the fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and membership function. The fusion module 400 is used to fuse fuzzy confidence distributions of different categories within the identification framework; The calculation module 500 is used to construct a rule weight and feature weight optimization model, and based on the rule weight and feature weight optimization model, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain; The data sending module 600 is used to calculate the confidence level based on the matching degree and activation degree, and output the predicted intent classification according to the principle of maximizing the confidence level, and send the predicted intent classification to the agent terminal.

[0078] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0079] In one embodiment, the setting module 100 is specifically used for: The fuzzy set segmentation method of the confidence rule base is set in advance according to the statistical distribution characteristics of the dataset; Pre-construct a membership function based on the Cauchy distribution; Based on the fuzzy set segmentation method and membership function, a confidence rule base model is generated.

[0080] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0081] In one embodiment, the data distribution generation module 300 is specifically used for: Obtain the training dataset for training the confidence rule base; A confidence rule base model is constructed based on the training dataset; The historical search records are input into the trained confidence rule base model, and the fuzzy confidence distribution corresponding to the historical search records is generated based on the fuzzy set segmentation method and membership function of the confidence rule base model.

[0082] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0083] In one embodiment, the fusion module 400 is specifically used for: Obtain the current set of all categories and use the category set as the recognition framework; Obtain the confidence score for each category within the identification framework; Based on the evidence theory synthesis rules, the confidence levels of various categories are fused to generate a fused fuzzy confidence distribution.

[0084] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0085] In one embodiment, the computing module 500 is specifically used for: Construct an optimization model for rule weights and feature weights; Obtain the rule weights and feature weights, and optimize the rule weights and feature weights based on the rule weight and feature weight priority model to generate target rule weights and target feature weights; Based on the target rule weights and target feature weights, the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain are calculated.

[0086] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0087] In one embodiment, the computing module 500 is further configured to: Based on the optimization model of the rule weights and feature weights, and the target rule weights, the matching degree of the historical search records in the rule fuzzy domain is obtained; Obtain the set of confidence rules activated for all historical search records in the confidence rule base model; Based on the matching degree and the target rule weight, the activation degree of each confidence rule in the historical search records and confidence rule set is calculated.

[0088] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0089] In one embodiment, the data sending module is specifically used for: A corresponding confidence level is generated based on the activation level of each confidence rule; The activated confidence rules are discounted using a discount operator to generate the target confidence level; The identification decision is made using the principle of maximizing confidence and the target confidence, and the predicted intent classification is output. The predicted intent is categorized and sent to the agent terminal.

[0090] For specific implementation details, please refer to the method embodiment; they will not be repeated here.

[0091] This invention provides an artificial intelligence-based auxiliary response device. By fusing and processing the confidence distributions of different categories within the identification framework, it constructs the input-output relationship between the feature space and the category space. It adopts the principle of maximizing confidence for identification decisions, which can filter invalid or low-quality lists, identify the user's true intent, and improve the contact efficiency of the agent list.

[0092] Another embodiment of the present invention provides a computer device, which may be a server, and its internal structure diagram may be as follows. Figure 4 As shown, the computer device includes a processor, memory, network interface, and database connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile and / or volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and database. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage media. The network interface is used to communicate with external clients via a network connection. When the computer program is executed by the processor, it implements the functions or steps of a server-side artificial intelligence-based assisted response method.

[0093] In one embodiment, a computer device is provided, which may be a client, and its internal structure diagram may be as follows: Figure 5As shown, the computer device includes a processor, memory, network interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The network interface is used to communicate with an external server via a network connection. When the computer program is executed by the processor, it implements client-side functions or steps of an artificial intelligence-based assisted response method.

[0094] In one embodiment, a computer device is provided, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to perform the following steps: The fuzzy set segmentation method and membership function of the confidence rule base are set in advance to generate the confidence rule base model; Obtain the list of breakpoints, and retrieve the user's historical search records based on the breakpoint list; The historical search records are input into the confidence rule base model, and a fuzzy confidence distribution corresponding to the historical search records is generated based on the fuzzy set segmentation method and membership function. Fusion identification of fuzzy confidence distributions of different categories within the framework; Construct a rule weight and feature weight optimization model, and based on the rule weight and feature weight optimization model, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain; The confidence level is calculated based on the matching degree and activation degree, and the predicted intent classification is output according to the principle of maximizing the confidence level. The predicted intent classification is then sent to the agent terminal.

[0095] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, the computer program performing the following steps when executed by a processor: The fuzzy set segmentation method and membership function of the confidence rule base are set in advance to generate the confidence rule base model; Obtain the list of breakpoints, and retrieve the user's historical search records based on the breakpoint list; The historical search records are input into the confidence rule base model, and a fuzzy confidence distribution corresponding to the historical search records is generated based on the fuzzy set segmentation method and membership function. Fusion identification of fuzzy confidence distributions of different categories within the framework; Construct a rule weight and feature weight optimization model, and based on the rule weight and feature weight optimization model, calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain; The confidence level is calculated based on the matching degree and activation degree, and the predicted intent classification is output according to the principle of maximizing the confidence level. The predicted intent classification is then sent to the agent terminal.

[0096] It should be noted that the functions or steps that can be implemented by the computer-readable storage medium or computer device described above can be referred to the relevant descriptions on the server side and client side in the foregoing method embodiments. To avoid repetition, they will not be described one by one here.

[0097] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory may include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in a variety of forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), dual data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), RAMbus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

[0098] The embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs.

[0099] Through the description of the above embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus a general-purpose hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the parts that contribute to the related technology, can be embodied in the form of a software product. This computer software product can exist in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0100] It should be noted that if any software tools or components not belonging to our company appear in the embodiments of this application, they are merely for illustrative purposes and do not represent actual use.

[0101] Among other things, conditional language such as “can,” “may,” “may,” or “may,” unless otherwise specifically stated or otherwise understood as in the context in which they are used, is generally intended to convey that a particular implementation may include (but not others) certain features, elements, and / or operations. Therefore, such conditional language is also generally intended to imply that features, elements, and / or operations are necessary for one or more implementations in any way, or that one or more implementations must include logic for determining, with or without input or prompting, whether such features, elements, and / or operations are included or will be performed in any particular implementation.

[0102] The contents already described herein in this specification and accompanying drawings include examples of methods and apparatuses capable of providing AI-based assisted response. It is certainly not possible to describe every conceivable combination of elements and / or methods for the purpose of describing the various features of this disclosure, but it will be appreciated that many other combinations and substitutions of the disclosed features are possible. Therefore, it will be apparent that various modifications can be made to this disclosure without departing from the scope or spirit of this disclosure. Furthermore, or in alternatives, other embodiments of this disclosure may become apparent from consideration of this specification and accompanying drawings and from practice of this disclosure as presented herein. It is intended that the examples presented in this specification and accompanying drawings be considered illustrative rather than restrictive in all respects. Although specific terminology is used herein, it is used in a general and descriptive sense and is not intended for limiting purposes.

Claims

1. An artificial intelligence-based assisted answering method, characterized by The method comprises: The fuzzy set segmentation method and membership function of the confidence rule base are set in advance to generate a confidence rule base model; Obtain a breakpoint list, and obtain historical search records of a user based on the breakpoint list; Input the historical search records into the confidence rule base model, and generate fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and the membership function; Fuse the fuzzy confidence distribution of different categories in the recognition framework; Build a rule weight and feature weight optimization model, and calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain based on the rule weight and feature weight optimization model; Calculate the confidence degree based on the matching degree and the activation degree, output the predicted intent classification according to the maximum confidence principle, and send the predicted intent classification to a terminal of an agent. 2.The AI-based assistant response method of claim 1, wherein, The fuzzy set segmentation method and membership function of the confidence rule base are set in advance to generate a confidence rule base model, comprising: The fuzzy set segmentation method of the confidence rule base is set in advance according to the statistical distribution characteristics of the data set; A membership function based on Cauchy distribution is constructed in advance; Generate a confidence rule base model according to the fuzzy set segmentation method and the membership function. 3.The AI-based assistant response method of claim 1, wherein, The historical search records are input into the confidence rule base model, and the fuzzy confidence distribution corresponding to the historical search records is generated based on the fuzzy set segmentation method and the membership function, comprising: Obtain a training data set for training the confidence rule base; Build a confidence rule base model based on the training data set; Input the historical search records into the trained confidence rule base model, and generate the fuzzy confidence distribution corresponding to the historical search records based on the fuzzy set segmentation method and the membership function of the confidence rule base model. 4.The AI-based assistant response method of claim 1, wherein, Fuse the fuzzy confidence distribution of different categories in the recognition framework, comprising: Obtain all category sets at present, and take the category set as a recognition framework; Obtain the confidence degree corresponding to each category under the recognition framework; Fuse the confidence degrees of all categories based on the evidence theory synthesis rule to generate a fused fuzzy confidence distribution. 5.The AI-based assistant response method of claim 1, wherein, The rule weight and feature weight optimization model is built, and the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain are calculated based on the rule weight and feature weight optimization model, comprising: Build a rule weight and feature weight optimization model; Obtain rule weights and feature weights, optimize the rule weights and feature weights based on the rule weight and feature weight optimization model, generate target rule weights and target feature weights, and calculate the matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain based on the target rule weights and the target feature weights. The matching degree and activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain are calculated based on the target rule weights and the target feature weights, comprising: 6.The AI-based assistant response method of claim 5, wherein, Obtain the matching degree of the historical search records in the rule fuzzy domain based on the rule weight and feature weight optimization model and the target rule weights; Obtain a confidence rule set activated by all historical search records in the confidence rule base model; ​ Based on the matching degree and the target rule weight, an activation degree of the historical search record and each confidence rule in a confidence rule set is calculated. 7.The AI-based assistant response method of claim 6, wherein, The confidence degree is calculated based on the matching degree and the activation degree, and a predicted intent classification is output according to a maximum confidence principle, and the predicted intent classification is sent to a service terminal, including: The confidence degree corresponding to each confidence rule is generated based on the activation degree of each confidence rule; The activated confidence rule is discounted by using a discount operator to generate a target confidence degree; The predicted intent classification is output by using the maximum confidence principle and the target confidence degree for identification decision; The predicted intent classification is sent to the service terminal.

8. An artificial intelligence-based auxiliary answering apparatus, characterized by, The device includes: The setting module sets the fuzzy set segmentation method and the membership function of the confidence rule library in advance to generate a confidence rule library model; The data acquisition module acquires a breakpoint list and acquires the historical search record of a user based on the breakpoint list; The data distribution generation module inputs the historical search record into the confidence rule library model to generate a fuzzy confidence distribution corresponding to the historical search record based on the fuzzy set segmentation method and the membership function; The fusion module fuses the fuzzy confidence distributions of different categories in the recognition framework; The calculation module constructs a rule weight and feature weight optimization model, and calculates the matching degree and the activation degree of the fuzzy confidence distribution in the corresponding rule fuzzy domain based on the rule weight and feature weight optimization model; The data sending module calculates the confidence degree based on the matching degree and the activation degree, and outputs the predicted intent classification according to the maximum confidence principle, and sends the predicted intent classification to the service terminal.

9. A computer device, comprising: The computer device includes at least one processor; and The memory is in communication connection with the at least one processor; wherein The memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to execute the steps of the artificial intelligence-based auxiliary answering method in any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer readable storage medium stores computer executable instructions, and the computer executable instructions are executed by one or more processors to enable the one or more processors to execute the steps of the artificial intelligence-based auxiliary answering method in any one of claims 1-7.

Images