Reservation process optimization method and system based on user feedback

Abstract

This application provides a method and system for optimizing the appointment process based on user feedback. First, it acquires a set of user feedback data for the target service scenario, including text feedback units and corresponding service node identifiers. Then, it performs sentiment analysis and intent recognition processing on the user feedback data to generate an optimization requirement feature set for each text feedback unit's associated service node identifier. Based on this optimization requirement feature set, it constructs a service optimization strategy set containing service node optimization strategies and their triggering conditions. According to the matching results between the real-time operating status data of the current service system and the strategy triggering conditions, it filters and executes the target service node optimization strategies to adjust the resource configuration rules or interaction logic of the corresponding service nodes in the appointment service process. The adjusted service node status is then updated to the service system, achieving real-time optimization of the appointment service process and improving user experience and service efficiency.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and more specifically, to a method and system for optimizing the appointment process based on user feedback. Background Technology

[0002] In today's service industry, appointment services have become a crucial means of enhancing customer experience and service efficiency. However, with the diversification of service scenarios and the continuous evolution of user needs, optimizing appointment service processes has become an ongoing challenge. Traditional methods for optimizing appointment service processes often rely on manual experience or periodic surveys. These methods are not only time-consuming and labor-intensive but also struggle to capture real-time user feedback and changes in demand, resulting in delayed optimization measures and limited effectiveness. Furthermore, given the diversity and complexity of user feedback across different service scenarios, accurately understanding user feedback and optimizing appointment service processes accordingly has become an urgent problem to be solved. Summary of the Invention

[0003] In view of this, the purpose of this application is to provide a method and system for optimizing the appointment process based on user feedback.

[0004] According to a first aspect of this application, a method for optimizing an appointment process based on user feedback is provided, the method comprising: Obtain a set of user feedback data for the target service scenario. The set of user feedback data includes multiple text feedback units submitted by users regarding the appointment service process and their corresponding service node identifiers. The user feedback data set is subjected to sentiment analysis and intent recognition processing to generate an optimized requirement feature set of associated service node identifiers for each text feedback unit; A service optimization strategy set is constructed based on the set of optimization requirement features of the associated service node identifier of each text feedback unit. The service optimization strategy set includes at least one service node optimization strategy and its corresponding strategy triggering conditions. Based on the matching results between the real-time operating status data of the current service system and the triggering conditions of the strategy, the optimization strategy for the target service node is selected from the set of service optimization strategies; The target service node optimization strategy is executed to adjust the resource configuration rules or interaction logic of the corresponding service node in the reservation service process, and the adjusted service node status is updated to the service system.

[0005] According to a second aspect of this application, an online consultation service system is provided, the online consultation service system including a machine-readable storage medium and a processor, the machine-readable storage medium storing machine-executable instructions, and the online consultation service system implementing the aforementioned appointment process optimization method based on user feedback when the processor executes the machine-executable instructions.

[0006] According to a third aspect of this application, a computer-readable storage medium is provided, wherein computer-executable instructions are stored therein, and when the computer-executable instructions are executed, the aforementioned user feedback-based appointment process optimization method is implemented.

[0007] Based on any of the above aspects, the technical effect of this application is as follows: By comprehensively acquiring user feedback data in target service scenarios and performing sentiment analysis and intent recognition processing, we can accurately capture users' optimization needs for the appointment service process. Based on this set of optimization needs characteristics, the constructed service optimization strategy set can flexibly address optimization needs in different service scenarios, ensuring the targeting and effectiveness of optimization measures. By matching the current service system's operational status data with strategy triggering conditions in real time, we can dynamically filter and execute optimization strategies for target service nodes, achieving real-time optimization of the appointment service process. After executing the target service node optimization strategy, adjustments to service node resource configuration rules or interaction logic can significantly improve user experience and service efficiency. Simultaneously, the adjusted service node status is updated to the service system, ensuring continuous optimization and iteration of the service process. This significantly improves the optimization efficiency and user satisfaction of the appointment service process, providing the service industry with a more intelligent and efficient appointment service solution. Attached Figure Description

[0008] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0009] Figure 1 A flowchart illustrating the appointment process optimization method based on user feedback provided in an embodiment of this application is shown. Figure 2 A schematic diagram of the component structure of the online consultation service system provided in the embodiments of this application is shown. Detailed Implementation

[0010] The embodiments of this application are described below with reference to the accompanying drawings. It should be understood that the embodiments described below with reference to the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of this application, and do not constitute a limitation on the technical solutions of the embodiments of this application.

[0011] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the terms “comprising” and “including” as used in embodiments of this application mean that the corresponding feature can be implemented as the presented feature, information, data, step, operation, element, and / or component, but do not exclude implementation as other features, information, data, step, operation, element, component, and / or combinations thereof supported by the art. It should be understood that when an element is said to be “connected” or “coupled” to another element, the element may be directly connected or coupled to the other element, or it may mean that the element and the other element are connected through an intermediate element. Furthermore, “connected” or “coupled” as used herein may include wireless connection or wireless coupling, and the term “and / or” as used herein indicates at least one of the items defined by the term; for example, “A and / or B” may be implemented as “A,” or as “B,” or as “A and B.”

[0012] To make the objectives, technical solutions, and advantages of this application clearer, the embodiments of this application will be described in further detail below with reference to the accompanying drawings. The technical solutions of the embodiments of this application and the technical effects produced by the technical solutions of this application will be explained below through the description of several exemplary embodiments. It should be noted that the following embodiments can be referenced, borrowed from, or combined with each other, and the same terms, similar features, and similar implementation steps in different embodiments will not be described again.

[0013] Figure 1 This document illustrates a flowchart of a user-feedback-based appointment process optimization method and system provided in an embodiment of this application. It should be understood that in other embodiments, the order of some steps in the user-feedback-based appointment process optimization method of this embodiment can be shared according to actual needs, or some steps can be omitted or maintained. The detailed steps of this user-feedback-based appointment process optimization method include: Step S110: Obtain a set of user feedback data for the target service scenario. The set of user feedback data includes multiple text feedback units submitted by users regarding the appointment service process and their corresponding service node identifiers.

[0014] Taking psychology education as an example, the appointment service process may involve key steps such as course booking, psychological testing booking, and expert consultation booking. To obtain user feedback data, various established channels can be used for collection. For instance, a dedicated feedback section can be set up on the official website of psychology education institutions, where users can fill in detailed text feedback after completing the booking process. Mobile applications will also have feedback entry points for convenient user submission at any time. In addition, user feedback will also be collected in the post-course evaluation process and online customer service chat logs.

[0015] For example, a user submitted the following text feedback: "When booking a psychological test, the options on the test selection page were not clearly categorized, making it difficult to quickly find the test I wanted. Then, when filling in my personal information, some of the required information was not directly related to the psychological test, which felt very cumbersome. Moreover, after submitting the booking, I did not receive a confirmation notification for a long time, which made me very anxious, wondering if the booking was successful." For each text feedback unit, a corresponding service node identifier can be associated with it. The service node identifier for "unclear option categorization on the assessment project selection page" is "Psychological Assessment Project Selection"; the service node identifier for "cumbersome personal information filling" is "Appointment Information Filling"; and the service node identifier for "no confirmation notification received after submitting an appointment for a long time" is "Appointment Confirmation Notification". By continuously collecting such text feedback units from various channels and accurately associating them with service node identifiers, a complete set of user feedback data can be formed.

[0016] It is worth noting that when acquiring user feedback data sets for target service scenarios, relevant laws and regulations must be strictly followed. Before users submit text feedback units, the purpose, use, and method of data collection, as well as the security measures for data storage, must be clearly explained to them. Legal authorization from users should be ensured by having them actively check an agreement to agree to the authorization. For example, on a psychology education appointment service platform, when a user enters the feedback page, a detailed data authorization pop-up window will appear. Users can only submit feedback information after carefully reading and agreeing to the relevant terms. Furthermore, the system will record the user's authorization information in detail, including the authorization time and content, for subsequent querying and auditing.

[0017] During the data collection process, it is necessary to ensure that the collected data is only used for the purpose of optimizing the appointment process and not for other unauthorized uses without the user's consent. At the same time, the collected data should be relevant and necessary, avoiding excessive collection of the user's privacy-sensitive data. If it is indeed necessary to collect some privacy-sensitive data, such as descriptions of the user's psychological condition, strict privacy protection and anti-disclosure technical measures must be taken. For example, encrypt such privacy-sensitive data using advanced encryption algorithms, such as the combination of symmetric encryption and asymmetric encryption, to ensure the security of the data during transmission and storage. Implement strict access control over the data, allowing only authorized personnel to access and process the data, and record detailed logs of the access behavior.

[0018] Step S120: Perform sentiment analysis and intent recognition processing on the user feedback data set to generate an optimized requirement feature set of the associated service node identifiers for each text feedback unit.

[0019] This step is to deeply mine the user feedback data to understand the user's emotional attitude and potential needs. The following details each sub-step of this step.

[0020] Step S121: Segment the text feedback unit to obtain multiple semantic word units, and extract the词性标注结果及上下文依存关系 (pos-tagging results and context dependency relationships) for each semantic word unit.

[0021] For the above-mentioned user feedback text, a mature natural language processing word segmentation algorithm can be used for processing. The word segmentation algorithm will segment the text into individual semantic word units according to the semantic and grammatical rules between words. For example, "When making an appointment for a psychological assessment" will be segmented into "When", "make an appointment", "psychological assessment", "When"; "Feel that the option classification is not clear" will be segmented into "Feel", "option classification", "not clear", etc.

[0022] After word segmentation, pos-tagging will be performed on each semantic word unit. "When" is a preposition, "make an appointment" is a verb, "psychological assessment" is a noun, "When" is a particle, "Feel" is a verb, "option classification" is a noun, "not clear" is an adjective.

[0023] At the same time, the context dependency relationships between semantic word units will be analyzed. There is an object-predicate relationship between "make an appointment" and "psychological assessment", indicating that the object of the action "make an appointment" is "psychological assessment"; there is a subject-predicate relationship between "Feel" and "option classification is not clear", reflecting the connection between the user's subjective feeling and the description of the objective situation. Through the analysis of such context dependency relationships, the semantic structure of the text can be grasped more accurately.

[0024] Step S122: Call the pre-trained sentiment polarity prediction model to predict the sentiment polarity of the context statement containing the semantic word unit, and generate the sentiment polarity prediction result of the text feedback unit. The sentiment polarity prediction result includes the distribution ratio of positive sentiment intensity value and negative sentiment intensity value.

[0025] Step S1221: Input the semantic word units into the embedding layer of the sentiment polarity prediction model in the original order of the context statement to generate a word embedding vector corresponding to each semantic word unit.

[0026] In a psychology education context, the semantic word units, after word segmentation, are sequentially input into the embedding layer of the sentiment polarity prediction model according to their order in the original text. The role of the embedding layer is to transform each semantic word unit from a discrete text form into a continuous vector representation, i.e., a word embedding vector.

[0027] For example, for the semantic word unit "appointment," the embedding layer maps it to a vector of a specific dimension, say a 300-dimensional vector V1, based on a pre-trained word vector table. This vector V1 contains information about the word "appointment" in the semantic space, and the value of each dimension is obtained through training on a large amount of text data, reflecting the semantic relationship between this word and other words. Similarly, for the semantic word unit "psychological assessment," another 300-dimensional vector V2 is generated. This process continues, generating a corresponding word embedding vector for each semantic word unit. These vectors together constitute the initial vector representation of the context statement.

[0028] Step S1222: The word embedding vector is semantically encoded using the bidirectional long short-term memory network layer of the sentiment polarity prediction model, and the context encoding vector of each semantic word unit is output.

[0029] Bidirectional Long Short-Term Memory (Bi-LSTM) layers can capture the contextual information of semantic word units within a sentence. It consists of two LSTMs: a forward LSTM that processes word embedding vectors from left to right, and a backward LSTM that processes word embedding vectors from right to left.

[0030] Taking a forward LSTM as an example, when the first word embedding vector V1 (corresponding to "reservation") is input, the LSTM calculates the hidden state at the current time step based on the current input and the hidden state at the previous time step. Assuming the initial hidden state is H0, the hidden state H1 at the first time step can be calculated through the LSTM's gating mechanism, which includes an input gate, a forget gate, and an output gate, controlling the inflow, retention, and output of information.

[0031] The input gate determines how much information from the current input vector is added to the cell state, the forget gate determines how much information from the previous cell state is forgotten, and the output gate determines how much information from the current cell state is output to the hidden state. Through this gating mechanism, LSTM can effectively handle dependencies in long sequence data.

[0032] The inverse LSTM processes word embedding vectors in reverse order, similarly calculating the hidden state at each time step. Finally, the hidden states of the forward and inverse LSTMs at the same time step are concatenated to obtain the context encoding vector for each semantic word unit. For example, for the semantic word unit "appointment," its context encoding vector C1 is the concatenation result of the hidden states of the forward and inverse LSTMs at the corresponding time steps. In this way, the context encoding vector of each semantic word unit contains semantic information from its preceding and following context.

[0033] Step S1223: Input the context encoding vector into the attention mechanism layer of the sentiment polarity prediction model, calculate the attention weight of each semantic word unit in the context statement, and perform weighted fusion of the context encoding vector according to the attention weight to generate the statement representation vector of the context statement.

[0034] The role of the attention mechanism layer is to assign different attention weights to the context encoding vector of each semantic word unit to highlight important semantic word units. First, the similarity between each context encoding vector and a learnable query vector Q is calculated. Assuming the context encoding vectors are C1, C2, ..., Cn (where n is the number of semantic word units), the query vector Q is a vector with the same dimension as the context encoding vectors.

[0035] Similarity can be calculated using a dot product approach, specifically by calculating the dot product of Ci and Q to obtain the similarity score Si. Then, these similarity scores are processed using a softmax function to transform them into a probability distribution, i.e., attention weights Wi. The softmax function maps the similarity scores to the interval [0, 1] and ensures that the sum of all attention weights is 1.

[0036] After obtaining the attention weights, the context encoding vectors can be weighted and fused based on these weights. Assuming the context encoding vector Ci has dimension m, then the sentence representation vector S = "W1*C1; W2*C2; ...; Wn*Cn", where ";" represents the concatenation operation. Through this weighted fusion method, important semantic units contribute more to the sentence representation vector, thus better representing the semantics of the entire context sentence.

[0037] Step S1224: Input the sentence representation vector into the fully connected classification layer of the sentiment polarity prediction model, and output the positive sentiment score and negative sentiment score of the text feedback unit.

[0038] A fully connected classification layer is a multilayer perceptron consisting of multiple fully connected layers. The statement representation vector S is input into the first layer of the fully connected classification layer, which performs a linear transformation on the input vector, specifically calculating Z1 = S * W1 + b1 using a weight matrix W1 and a bias vector b1. Then, an activation function, such as the ReLU function, is applied to Z1 to introduce non-linearity.

[0039] Next, the vector Z1 processed by the activation function is input into the next fully connected layer, and the above linear transformation and activation function processing are repeated until the last layer. The output dimension of the last layer is 2, corresponding to the positive sentiment score and the negative sentiment score respectively. Assume that the output vector of the last layer is O=[P, N], where P represents the positive sentiment score and N represents the negative sentiment score.

[0040] Step S1225: Normalize the positive sentiment score and the negative sentiment score to generate the distribution ratio of the positive sentiment intensity value and the negative sentiment intensity value, which is used as the sentiment polarity prediction result.

[0041] To make positive and negative sentiment scores comparable, they need to be normalized. A common normalization method is to convert them into probability distributions. The softmax function can be used for normalization, i.e., positive sentiment intensity P' = exp(P) / (exp(P) + exp(N)), and negative sentiment intensity N' = exp(N) / (exp(P) + exp(N)).

[0042] Thus, P' and N' represent the distribution ratio of positive and negative sentiment intensity values, and their sum is 1. This distribution ratio is the sentiment polarity prediction result of this text feedback unit, which can intuitively reflect the emotional tendency expressed by the user in the text feedback.

[0043] Step S123: Construct an intent recognition feature vector based on the named entity recognition result of the semantic word unit and the part-of-speech tagging result, and call the intent recognition model to determine the intent category label of the text feedback unit according to the intent recognition feature vector.

[0044] Step S1231: Identify service operation verb units, service object noun units, and service condition adverbial units from the semantic word units.

[0045] In the aforementioned user feedback text, semantic word unit analysis identified service operation verb units, such as "appointment," "fill in," and "submit," which represent the actions performed by the user during the appointment service process. Service object noun units included "psychological assessment," "personal information," and "appointment," which are the objects of the service operations. Service condition adverbial units, such as "on the assessment item selection page" and "long time," limited the conditions or timeframes for the service operations.

[0046] Step S1232: Determine the basic intent category based on the matching result of the service operation verb unit in the preset intent category mapping table, and further classify the basic intent category based on the type label of the service object noun unit.

[0047] The pre-defined intent category mapping table associates different service operation verbs with basic intent categories. For example, the basic intent category corresponding to "appointment" might be "service appointment," and the basic intent category corresponding to "fill in" might be "information filling."

[0048] Then, the basic intent categories are further refined based on the type label of the service recipient noun unit. If the service recipient noun unit is "psychological assessment", then the basic intent category of "service appointment" can be refined to "psychological assessment appointment"; if the service recipient noun unit is "personal information", then the basic intent category of "information filling" can be refined to "personal information filling".

[0049] Step S1233: Extract the logical connective units and comparison condition description fragments from the service condition adverbial unit, and combine and encode the type of the logical connective unit and the parameter range of the comparison condition description fragment to generate a condition constraint encoding sequence.

[0050] The service condition adverbial unit may contain logical connective units such as "first," "next," and "moreover," which indicate the order or logical relationship of service operations. Comparative condition descriptive fragments such as "unclear," "cumbersome," and "long time" provide comparative descriptions of the service's state or time.

[0051] The types of logical connective units and the parameter ranges of comparison condition description fragments are combined and encoded. For example, the logical connective "first" is encoded as 1, "then" as 2, and "moreover" as 3; the comparison condition description fragments "unclear" as 101, "cumbersome" as 102, and "long time" as 103. These are then combined to form a condition constraint encoding sequence, such as [1, 101, 2, 102, 3, 103].

[0052] Step S1234: Perform category encoding and numerical normalization on the service operation verb unit, the service object noun unit, and the condition constraint encoding sequence, respectively, and then perform vectorization concatenation according to a preset dimension to generate the intent recognition feature vector.

[0053] The verb units for service operations are categorized and coded, for example, "appointment" is coded as 201, "fill in" as 202, and "submit" as 203. The noun units for service recipients are also categorized and coded, for example, "psychological assessment" as 301, "personal information" as 302, and "appointment" as 303.

[0054] The encoded service operation verb units, service object noun units, and condition constraint encoding sequences are numerically normalized, mapping their values ​​to the interval [0, 1]. Then, these normalized vectors are concatenated according to a preset dimensionality requirement. Assuming a preset dimension of 100, the normalized vectors of the service operation verb units, the service object noun units, and the condition constraint encoding sequences are concatenated sequentially to form a 100-dimensional intent recognition feature vector.

[0055] Step S1235: Input the intent recognition feature vector into the intent recognition model for nonlinear transformation to obtain a higher-order intent feature vector.

[0056] An intent recognition model is typically a deep neural network. The intent recognition feature vector is input into the first layer of the model. This layer performs a non-linear transformation on the input vector, using a non-linear activation function, such as the sigmoid or tanh function, to transform each element of the input vector. Assuming the input vector is X, the weight matrix of the first layer is W, and the bias vector is b, then the output vector Y1 of the first layer can be obtained by calculating Y1 = f(X*W+b), where f is the activation function.

[0057] Next, Y1 is input to the next layer, and the above nonlinear transformation process is repeated. After multiple transformations, a higher-order intent feature vector is obtained. The higher-order intent feature vector contains more complex semantic information and can better represent the potential intent of the text feedback unit.

[0058] Step S1236: Perform dimensionality compression processing on the high-order intent feature vector to generate an intent category probability distribution vector, which contains probability values ​​of multiple preset intent categories.

[0059] To reduce the dimensionality of high-order intent feature vectors while retaining important information, dimensionality compression is performed. Common methods include using techniques such as principal component analysis (PCA) or autoencoders.

[0060] Taking PCA as an example, firstly, the covariance matrix of the higher-order intent feature vectors is calculated. Then, the covariance matrix is ​​decomposed into eigenvalues ​​to obtain eigenvalues ​​and eigenvectors. The top k eigenvectors with larger eigenvalues ​​are selected, and the higher-order intent feature vectors are projected onto the subspace spanned by these eigenvectors to obtain the compressed vectors.

[0061] Next, the compressed vector is input into a fully connected layer, whose output dimension is equal to the number of preset intent categories. The softmax function is then used to convert the output of this layer into a probability distribution, i.e., an intent category probability distribution vector. Each element represents the probability value of the corresponding preset intent category.

[0062] Step S1237: Sort the multiple preset intent categories according to the magnitude of the probability values, and select the top N candidate intent categories to form a candidate intent set.

[0063] The preset intent categories are sorted in descending order based on the magnitude of each element in the intent category probability distribution vector. For example, suppose the preset intent categories are "improve appointment process," "optimize information filling," and "improve notification timeliness," with corresponding probability values ​​of 0.6, 0.3, and 0.1, respectively. After sorting by probability value from largest to smallest, the top N (assuming N=2) preset intent categories with the largest probability values ​​are selected, namely "improve appointment process" and "optimize information filling," to form the candidate intent set.

[0064] Step S1238: Calculate the category confidence score based on the co-occurrence frequency of each category in the candidate intent set in the historical feedback data and the resource configuration status of the current service node.

[0065] In historical feedback data, the co-occurrence frequency of each category in the candidate intent set with other intent categories is statistically analyzed. For example, if "improve appointment process" and "optimize information filling" frequently appear together in historical feedback data, then their co-occurrence frequency is relatively high.

[0066] Additionally, the resource configuration status of the current service node should be considered. If the current service node has sufficient resources to optimize the appointment process and information entry, then the confidence levels for these two intent categories will be relatively high.

[0067] Based on the combined co-occurrence frequency and resource allocation status, a category confidence score is calculated for each candidate intent category. This can be calculated using a weighted fusion method. Assuming the weight of co-occurrence frequency is 0.6 and the weight of resource allocation status is 0.4, then the category confidence score for "Improve Appointment Process" = co-occurrence frequency score * 0.6 + resource allocation status score * 0.4.

[0068] Step S1239: Use the candidate intent category with the highest category confidence score as the intent category label of the text feedback unit.

[0069] The category confidence scores of each candidate intent category in the candidate intent set are compared, and the candidate intent category with the highest score is selected as the intent category label of the text feedback unit. For example, if "improve the appointment process" has the highest category confidence score, then the intent category label of the text feedback unit is "improve the appointment process".

[0070] Step S124: Based on the preset time entity recognition rules and service timeliness keyword library, identify the service timeliness requirement description fragment from the context dependency relationship, and extract the priority parameters and time constraints corresponding to the service timeliness requirement description fragment. The service timeliness requirement description fragment includes explicit time constraint description and implicit timeliness requirement description fragment.

[0071] The preset time entity recognition rules will define some time-related grammatical structures and keywords, and the service timeliness keyword library will contain time-related words such as "immediately", "as soon as possible", and "long time".

[0072] In the aforementioned user feedback text, "no confirmation notification received for a long time" is an explicit time constraint description. This description fragment can be identified using time entity recognition rules and a service timeliness keyword database. The priority parameter for this description fragment can be determined based on the user's tone and business rules. If the user exhibits significant anxiety, the priority parameter can be set higher. The time constraint condition is "long time," which can be further quantified into a specific time range, such as exceeding 30 minutes, based on business experience or historical data.

[0073] Additionally, there may be implicit timeliness requirements in the descriptions. For example, when a user describes the unclear categorization of assessment options on the selection page, although they don't explicitly mention time, they implicitly express a desire to complete the assessment selection quickly and efficiently. This is an example of an implicit timeliness requirement. For these implicit timeliness requirement descriptions, priority parameters and time constraints also need to be extracted. Priority parameters can be determined based on the overall tone of the user feedback and the importance of this service node in the entire appointment process. For example, if the assessment selection is a crucial prerequisite step for subsequent processes, the priority parameter can be set higher. Time constraints can be determined based on industry standards or the processing time of similar past transactions. For instance, the expectation is that users can complete the assessment selection within a short timeframe, and this "shorter timeframe" can be set within a reasonable range based on business analysis.

[0074] Step S125: Perform association mapping processing on the sentiment polarity prediction result, the intent category label, the priority parameter and the time constraint condition with the corresponding service node identifier to generate an optimized requirement feature set of the associated service node identifier.

[0075] In the context of psychology education, the previously obtained emotional polarity prediction results, intent category labels, priority parameters, and time constraints are associated with their respective service node identifiers. For example, for the user feedback mentioned earlier, the "Psychological Assessment Item Selection" service node has a relatively high negative emotional intensity value in its emotional polarity prediction result, an intent category label of "Optimizing Assessment Item Selection" within the "Improving Appointment Process" subcategory, a relatively high priority parameter, and a time constraint that users are expected to complete the selection within a short period of time. This information is combined to form the optimization requirement feature set for this service node. The "Appointment Information Filling" and "Appointment Confirmation Notification" service nodes are also associated and mapped in the same way, generating their respective optimization requirement feature sets. In this way, each service node has a feature set containing its optimization-related information, providing a concrete basis for subsequent service optimization strategy formulation.

[0076] Step S130: Construct a service optimization strategy set based on the set of optimization requirement features of the associated service node identifier of each text feedback unit. The service optimization strategy set includes at least one service node optimization strategy and its corresponding strategy triggering conditions.

[0077] Step S131: Based on the sentiment polarity prediction results, the priority parameters, and the time constraints, select the target service node identifiers that need to be processed first, and extract the intent category tag set corresponding to the target service node identifiers.

[0078] In the psychology education appointment service, target service nodes are selected based on the emotional polarity prediction results, priority parameters, and time constraints of each service node. Service nodes with high negative emotional intensity values, high priority parameters, and urgent time constraints are identified as priority target service nodes. For example, the "appointment confirmation notification" service node, if it has a high negative emotional intensity value, user feedback indicating excessively long waiting times and anxiety, a high priority parameter, and a time constraint of resolving the notification delay issue as quickly as possible, will be selected as a target service node.

[0079] For the selected target service nodes, extract their corresponding intent category tag set. For example, the intent category tags for the "appointment confirmation notification" service node may include "improve notification timeliness," and these tags reflect the areas where this service node needs to be optimized.

[0080] Step S132: Perform service logic defect analysis on each intent category label in the intent recognition result to determine the service process defect type and the scope of the defect's impact that leads to negative emotions.

[0081] For each intent category tag, conduct an in-depth analysis of the service logic to identify service process defects that lead to negative emotions. Taking the intent category tag "improve notification timeliness" as an example, analyze the service logic of the "appointment confirmation notification" service node. Potential service process defects may include data synchronization issues between the notification system and the appointment system, resulting in delayed notification generation; or a single notification channel, failing to reach users in a timely manner.

[0082] Simultaneously, the scope of impact of these defects needs to be determined. If the notification delay issue only affects users during certain appointment slots, the impact is relatively small; if it affects all users with appointments, the impact is larger. By analyzing service logic defects based on intent category tags, the specific problems and their extent of impact can be clearly identified, providing precise direction for subsequent optimization strategy development.

[0083] Step S133: Match a predefined optimization operation template according to the service process defect type. The optimization operation template includes resource allocation adjustment parameters, service response logic adjustment rules, and coordination mechanism between service nodes.

[0084] The predefined optimization operation templates are a series of optimization schemes pre-designed based on common service process defect types. For the data synchronization problem between the notification system and the reservation system in the "appointment confirmation notification" service node analyzed earlier, a corresponding optimization operation template can be matched. This template may include resource allocation adjustment parameters, such as increasing the computing resources of the data synchronization server to improve data synchronization efficiency; service response logic adjustment rules, such as optimizing the notification generation algorithm to ensure that a notification is generated immediately after a successful reservation; and coordination mechanisms between service nodes, such as establishing a real-time data interaction channel between the reservation system and the notification system to ensure timely data updates and sharing.

[0085] For different types of service process defects, there are corresponding optimization operation templates available for matching. These templates provide detailed operation guidance for solving specific problems.

[0086] Step S134: Calculate the adjustment range of each parameter in the optimization operation template based on the defect impact range, and generate strategy triggering conditions in combination with the time constraint conditions.

[0087] Based on the previously determined scope of the defect's impact, calculate the adjustment range for each parameter in the optimization operation template. If the notification delay issue of the "Appointment Confirmation Notification" service node has a significant impact, then when increasing the computing resources of the data synchronization server, it is necessary to significantly adjust the resource allocation parameters to ensure that the notification needs of a large number of users can be met.

[0088] Combine time constraints to generate strategy trigger conditions. For example, if the time constraint is to resolve the notification delay issue as quickly as possible, the strategy trigger condition could be set to immediately trigger the optimization strategy and initiate operations such as resource allocation adjustments and service response logic adjustments if no notification is generated within a specified time after a successful reservation.

[0089] Step S135: Bind the optimized operation template with the adjusted parameters and the strategy triggering conditions to form the service node optimization strategy and add it to the service optimization strategy set.

[0090] The matched optimization operation template, the parameters adjusted according to the scope of the defect's impact, and the strategy triggering conditions generated in conjunction with time constraints are bound together. For the "appointment confirmation notification" service node, the optimization operation template, which includes resource allocation adjustments, service response logic adjustments, and coordination mechanisms between service nodes, is combined with the adjusted resource allocation parameters and strategy triggering conditions to form the optimization strategy for that service node.

[0091] Add the optimization strategy for each service node to the service optimization strategy set, which contains optimization strategies for each service node.

[0092] Step S140: Based on the matching result between the real-time operating status data of the current service system and the policy triggering conditions, select the target service node optimization strategy from the service optimization strategy set.

[0093] Step S141: Monitor the real-time load data, resource utilization indicators, and service request queue length of each service node in the service system.

[0094] In the psychology education appointment service system, specialized monitoring tools are used to monitor the real-time load data, resource utilization indicators, and service request queue length of each service node. For the "Psychological Assessment Item Selection" service node, real-time load data reflects the node's workload in processing user requests at any given moment, resource utilization indicators show the usage of computing and storage resources, and the service request queue length indicates how many user requests are waiting to be processed. Similarly, real-time monitoring is also performed on service nodes such as "Appointment Information Filling" and "Appointment Confirmation Notification" to obtain these key operational status data.

[0095] Step S142: Parse the required operating status parameter thresholds and logical judgment conditions according to the policy triggering conditions in the service node optimization strategy.

[0096] For each service node optimization strategy in the service optimization strategy set, analyze the required runtime status parameter thresholds and logical judgment conditions in its strategy triggering conditions. Taking the optimization strategy of the "appointment confirmation notification" service node as an example, the strategy triggering condition may stipulate that the optimization strategy is triggered when no notification is generated after a successful appointment and the resource utilization of the service node is lower than a certain threshold. Here, "a certain time" is the runtime status parameter threshold, and "no notification is generated after a certain time and the resource utilization is lower than a certain threshold" is the logical judgment condition.

[0097] Step S143: Compare the real-time load data, the resource utilization index, and the service request queue length with their respective corresponding operating status parameter thresholds to generate multiple condition-satisfied status identifiers.

[0098] The system compares the real-time load data, resource utilization metrics, and service request queue length of each service node with the parsed operational status parameter thresholds. For the "Psychological Assessment Item Selection" service node, if its real-time load data exceeds the preset load threshold, the corresponding condition is marked as "Satisfied"; if the resource utilization metric does not reach the preset utilization threshold, the corresponding condition is marked as "Not Satisfied." This process continues, generating a condition satisfaction status flag for each service node based on the comparison results of its operational status data with the thresholds.

[0099] Step S144: Perform logical operations on the condition satisfaction status identifier based on the logical judgment condition to determine the candidate service node optimization strategy list that meets the strategy triggering condition.

[0100] Based on the logical judgment conditions in the policy triggering conditions, logical operations are performed on the generated condition satisfaction status indicators. For example, for the optimization policy of the "appointment confirmation notification" service node, the logical judgment condition is "no notification generated for a certain period of time and resource utilization is lower than a certain threshold." If the condition "no notification generated for a certain period of time" is satisfied with the status indicator "satisfied," and the condition "resource utilization is lower than a certain threshold" is also satisfied with the status indicator "satisfied," then the optimization policy of this service node satisfies the policy triggering conditions, and it is added to the candidate service node optimization policy list. By performing such logical operations on the optimization policies of all service nodes, the candidate service node optimization policy list that satisfies the policy triggering conditions is determined.

[0101] Step S145: Based on the expected benefits and implementation costs of each candidate service node optimization strategy in the candidate service node optimization strategy list, a trade-off analysis is performed, and the candidate service node optimization strategy with the highest benefit-cost ratio is selected as the target service node optimization strategy.

[0102] For each candidate service node optimization strategy in the candidate service node optimization strategy list, evaluate its expected optimization benefits and implementation costs. Expected optimization benefits can be assessed from aspects such as improved user satisfaction, reduced user complaints, and increased business volume. Implementation costs include increased hardware resource costs, software development costs, and personnel training costs.

[0103] The optimization strategies for each candidate service node are ranked by calculating their benefit-cost ratio (the ratio of expected optimization benefit to implementation cost). The optimization strategy with the highest benefit-cost ratio is selected as the target service node optimization strategy. For example, for the "appointment information filling" service node, one candidate optimization strategy is expected to significantly improve the efficiency of users filling in information, thereby increasing the appointment success rate, but with a relatively low implementation cost; while another candidate optimization strategy can also improve efficiency to some extent, but with a higher implementation cost. After calculating and comparing the benefit-cost ratio, the strategy with the higher benefit-cost ratio is selected as the target service node optimization strategy.

[0104] Step S146: If there are multiple conflicting strategies for the same service node in the candidate service node optimization strategy list, then select the candidate service node optimization strategy with the least conflict impact as the target service node optimization strategy according to the priority ranking of the expected benefit value of strategy optimization and the current load status of the service node.

[0105] In the list of candidate service node optimization strategies, there may be multiple conflicting strategies for the same service node. For example, for the "Psychological Assessment Item Selection" service node, one candidate optimization strategy is to increase the number of assessment items displayed to provide users with more choices; another candidate optimization strategy is to reduce the number of assessment items displayed to improve selection efficiency. These two strategies conflict to some extent.

[0106] The strategy with the least conflicting impact is selected based on the priority ranking of the expected benefits of strategy optimization and the current load status of the service node. If the current service node is under high load and users prefer to quickly select evaluation items, then the strategy of reducing the number of evaluation items displayed may be more suitable, as it has less impact on system load and can meet user needs to some extent. In this way, it is ensured that the selected optimization strategy for the target service node achieves the optimization effect while minimizing the negative impact on the service system.

[0107] Step S150: Execute the target service node optimization strategy to adjust the resource configuration rules or interaction logic of the corresponding service node in the reservation service process, and update the adjusted service node status to the service system.

[0108] Step S151: Parse the resource allocation adjustment parameters in the target service node optimization strategy and generate a resource reallocation instruction to adjust the computing resource quota or human resource scheduling rules of the service node.

[0109] For the selected target service node optimization strategy, the resource allocation adjustment parameters are parsed. Taking the target service node optimization strategy for the "appointment confirmation notification" service node as an example, if the strategy requires increasing the computing resources of the data synchronization server, then the specific resource allocation adjustment parameters are parsed, such as the number of computing cores and memory size that need to be increased.

[0110] Based on these parameters, resource reallocation instructions are generated and sent to the resource management module of the service system. The resource management module adjusts the computing resource quotas of the service nodes according to the instructions, such as allocating more computing cores and memory to the data synchronization server. If adjustments to human resource scheduling rules are involved, such as increasing the number of personnel responsible for sending notifications or adjusting personnel's work schedules, corresponding adjustments will also be made according to the requirements in the strategy.

[0111] Step S152: Modify the business process processing order or exception handling mechanism of the service node according to the service response logic adjustment rules in the target service node optimization strategy, and update the state transition conditions of the service node.

[0112] Based on the service response logic adjustment rules in the target service node optimization strategy, the business process processing order of the service nodes is modified. For example, for the "Psychological Assessment Item Selection" service node, if the strategy requires optimizing the display logic of assessment items, the business process processing order will be modified to prioritize the display of frequently used assessment items, or to sort and display the assessment items according to their popularity.

[0113] Simultaneously, the exception handling mechanism for service nodes has been updated. If network anomalies or data errors occur during the selection of evaluation items, the new exception handling mechanism will be implemented, such as providing more user-friendly error messages to guide users to reselect. Furthermore, the state transition conditions for service nodes have been updated to ensure that the service process flows normally according to the optimized logic.

[0114] Step S153: Based on the inter-service node coordination mechanism in the target service node optimization strategy, establish an information synchronization channel and task handover protocol between service nodes to ensure data consistency between optimized service nodes.

[0115] Based on the service node coordination mechanism in the target service node optimization strategy, an information synchronization channel and task handover agreement are established between service nodes. Taking the "Reservation Information Filling" and "Reservation Confirmation Notification" service nodes as examples, an information synchronization channel is established to ensure that when a user completes and submits the information at the "Reservation Information Filling" service node, the "Reservation Confirmation Notification" service node can promptly obtain the latest reservation information.

[0116] Develop a task handover agreement that clearly defines the data transmission format, time requirements, and error handling methods between the two service nodes. For example, stipulate that after a user submits information, the "Reservation Information Filling" service node must accurately transmit the information to the "Reservation Confirmation Notification" service node within a certain time. If a data transmission error occurs, there should be a corresponding retry mechanism or error reporting mechanism. By establishing an information synchronization channel and a task handover agreement, ensure data consistency between the optimized service nodes and avoid service problems caused by data inconsistency.

[0117] Step S154: Simulate a real-world multi-node collaborative load and sudden traffic fluctuation model in a preset gray-scale release environment, perform stress tests on the adjusted service nodes, and collect performance index data and abnormal event records during the test process.

[0118] In a pre-defined canary release environment, simulate multi-node collaborative load and sudden traffic fluctuation models under real-world scenarios. For example, simulate a scenario where a large number of users simultaneously make reservations within a certain time period, or simulate the load on the service system during peak business hours.

[0119] The adjusted service nodes were deployed to a canary release environment for stress testing. During the testing process, performance metrics such as service node response time, throughput, and resource utilization were collected. Simultaneously, abnormal events such as service node crashes, data loss, and inaccurate error messages were recorded. Through stress testing and data collection, the performance and stability of the adjusted service nodes in real-world scenarios were evaluated.

[0120] Step S155: Verify whether the optimization effect has reached the expected threshold based on the performance index data. If it has, synchronize the adjusted service node status to the production environment; otherwise, roll back to the pre-optimization state and regenerate the service optimization strategy.

[0121] Based on the collected performance metrics data, verify whether the optimization effect of the adjusted service nodes has reached the expected threshold. For example, for the "appointment confirmation notification" service node, the expected average response time of the notification should be shortened to a certain extent after optimization. If the average response time obtained in the test reaches this expected threshold, it indicates that the optimization effect is good. The adjusted service node status will then be synchronized to the production environment so that more users can enjoy the optimized service.

[0122] If the optimization effect does not reach the expected threshold, it indicates that the current optimization strategy may have a problem, and it is necessary to roll back to the state before optimization. The rollback operation restores the resource configuration rules, business process processing order, state transition conditions, etc. of the service nodes to the state before optimization. Then, user feedback data is re-analyzed, and a new service optimization strategy is generated based on the new situation, and optimization attempts are made again.

[0123] Step S156: If the number of rollbacks exceeds the preset threshold, a manual intervention process is triggered and an optimization strategy failure alarm is generated. At the same time, the automated optimization function of the target service node is frozen.

[0124] If the number of rollbacks exceeds a preset threshold during optimization attempts, it indicates that the current automated optimization strategy may not be effective in resolving the issue. In this case, a manual intervention process is triggered, notifying relevant technical and business personnel to conduct an in-depth analysis of the service node problem.

[0125] Simultaneously, an optimization strategy failure alert is generated, reminding team members to monitor the optimization status of the service node. To prevent further issues, the automated optimization function of the target service node is frozen to prevent continued ineffective optimization operations. The human team will then develop a new optimization plan based on the specific circumstances, employing more in-depth analysis methods and specialized knowledge to resolve the problems existing in the service node.

[0126] Step S210: Continuously monitor the trend of user feedback data changes after implementing the target service node optimization strategy, and extract newly generated text feedback units for sentiment analysis.

[0127] After implementing the target service node optimization strategy, we continuously monitor user feedback data. We collect newly generated user feedback data through various feedback channels mentioned earlier, such as the official website feedback section and mobile application feedback entry. We then perform sentiment analysis on these newly generated text feedback units, employing the same process as before, including word segmentation, part-of-speech tagging, contextual dependency analysis, and sentiment polarity prediction, to understand users' emotional attitudes towards the service after optimization.

[0128] Step S220: Compare the change in negative sentiment intensity value before and after optimization. If the change does not reach the preset optimization effect threshold, trigger the iterative update mechanism of the service optimization strategy set.

[0129] The sentiment polarity prediction results of the newly generated text feedback units after optimization are compared with the results before optimization, with a focus on the magnitude of change in negative sentiment intensity values. If the magnitude of change in negative sentiment intensity values ​​does not reach the preset optimization effect threshold, it indicates that the current optimization strategy has not achieved the desired effect, and the iterative update mechanism of the service optimization strategy set needs to be triggered.

[0130] The iterative update mechanism re-examines previous optimization processes and analyzes potential problems. For example, the parameters for adjusting the optimization strategy might not be accurate enough, or the defect analysis of the service process might not be comprehensive enough. Based on these analysis results, the set of service optimization strategies is updated, optimization strategies are reselected or adjusted, and optimization attempts are made again.

[0131] Step S230: Expand the database of the optimized operation template based on the newly identified service process defect types and the newly added intent category tags in the user feedback data.

[0132] During the continuous monitoring of user feedback data, new types of service process defects and new intent category tags may be identified. For example, as the business develops and user needs change, new service process issues may arise, such as users reporting that they cannot select consecutive courses within a specific time period when booking a course. This would be a newly identified type of service process defect. The new intent category tag might be "Support consecutive course bookings".

[0133] For these newly identified service process defect types and newly added intent category tags, they should be incorporated into the optimization operation template database. First, a detailed analysis of the new service process defects should be conducted to determine possible optimization directions. For example, for the problem of being unable to select consecutive courses, possible optimization directions include adjusting the course selection algorithm and updating the storage structure of course data. Then, based on these optimization directions, new optimization operation templates should be designed. The new optimization operation templates will also include resource allocation adjustment parameters, service response logic adjustment rules, and coordination mechanisms between service nodes. For example, to support consecutive course reservations, it may be necessary to increase server computing resources to handle more complex course selection logic; this is a resource allocation adjustment parameter; modifying the interaction logic of the course selection page to make it easier for users to select consecutive courses is a service response logic adjustment rule; establishing a more efficient information synchronization mechanism between course management service nodes and reservation service nodes to ensure the consistency of course data is a coordination mechanism between service nodes. These new optimization operation templates should be added to the optimization operation template database for use in subsequent optimization processes.

[0134] Step S240: Perform statistical analysis on the implementation effect of historical service optimization strategies, and establish a strategy effect prediction model to optimize the benefit-cost ratio calculation logic for subsequent strategy selection.

[0135] For example, step S241: collect implementation record data of historical service optimization strategies, the implementation record data including strategy parameters, performance index comparison data before and after implementation, and user feedback change indicators.

[0136] In the psychology education appointment service system, collecting historical service optimization strategy implementation records is fundamental to building a predictive model for strategy effectiveness. Relevant data is collected through various means, including system logs, performance monitoring tools, and user feedback channels. For each historical service optimization strategy, specific strategy parameters are recorded, such as the specific values ​​of resource allocation adjustments and the rules governing service response logic adjustments. Simultaneously, performance metrics comparison data before and after strategy implementation are collected, such as changes in service node response time, throughput, and error rate. Additionally, user feedback metrics are collected, including changes in user satisfaction ratings, complaint rates, and negative sentiment intensity values ​​before and after strategy implementation. For example, for an optimization strategy targeting the "appointment information filling" service node, the strategy parameters such as the number of information fields adjusted and format requirements are recorded, along with data on the average response time of the service node before and after implementation, and the proportion of user complaints about cumbersome information filling.

[0137] Step S242: Perform one-hot encoding on the categorical strategy parameters in the implementation record data, and standardize the continuous parameters to generate strategy feature vectors, effect feature vectors, and cost feature vectors.

[0138] After collecting historical service optimization strategy implementation records, different types of parameters need to be preprocessed. For categorical strategy parameters, such as service node types (course booking, psychological test booking, etc.) and optimization operation types (resource allocation adjustment, service response logic adjustment, etc.), one-hot encoding is used. One-hot encoding converts each categorical parameter into a binary vector, where only one bit is 1 and the rest are 0, to represent the category to which the parameter belongs. For example, if there are three service node types: "course booking," "psychological test booking," and "expert consultation booking," and a strategy targets the "course booking" service node, then its one-hot encoded vector might be [1, 0, 0].

[0139] For continuous parameters, such as specific values ​​for resource allocation and performance indicators, standardization is performed. The purpose of standardization is to convert continuous parameters with different ranges into values ​​with the same scale for subsequent analysis and calculation. A common standardization method is to subtract the mean from the parameter value and then divide by its standard deviation. This process ensures that the mean of each continuous parameter is 0 and the standard deviation is 1.

[0140] After the above processing, a strategy feature vector, an effect feature vector, and a cost feature vector are generated. The strategy feature vector contains various parameter information of the strategy, such as categorical strategy parameters after one-hot encoding and standardized continuous strategy parameters; the effect feature vector contains various effect indicators after the strategy is implemented, such as changes in performance indicators and changes in user feedback indicators; the cost feature vector contains various cost information required to implement the strategy, such as hardware resource costs and software development costs.

[0141] Step S243: Construct a multiple regression model to analyze the correlation between the strategy feature vector and the effect feature vector, and generate an effect prediction coefficient matrix.

[0142] A multiple regression model is constructed to analyze the relationship between the strategy feature vector and the effect feature vector. The multiple regression model can be expressed as the effect feature vector equal to the product of the strategy feature vector and the effect prediction coefficient matrix, plus an error term. In constructing the model, the strategy feature vector of historical service optimization strategies is used as input, and the effect feature vector is used as output. The effect prediction coefficient matrix is ​​estimated by minimizing the sum of squares of the error term.

[0143] Specifically, historical data is used for model training. A portion of the historical data is used as the training set, and the other portion as the test set. During training, the values ​​of the performance prediction coefficient matrix are continuously adjusted to minimize the error between the model's predicted performance feature vectors and the actual performance feature vectors. After multiple iterations of training, an optimal performance prediction coefficient matrix is ​​obtained. This matrix reflects the degree of influence of each policy feature on the performance feature. For example, if the coefficient for a certain policy feature is large, it indicates that the policy feature has a large influence on the performance feature; if the coefficient is small, it indicates a small influence.

[0144] Step S244: Train the classification model to predict the benefit-cost ratio of the strategy implementation based on the strategy feature vector and the cost feature vector.

[0145] Train a classification model to predict the benefit-cost ratio rank of policy implementation. First, calculate the benefit-cost ratio for each policy based on historical data and categorize the benefit-cost ratio into different ranks, such as high, medium, and low. Then, use the policy feature vector and cost feature vector as input, and the benefit-cost ratio rank as output, to train the model using a classification algorithm.

[0146] Commonly used classification algorithms include decision trees, support vector machines, and neural networks. Taking the decision tree algorithm as an example, a decision tree constructs a tree-structured classification model based on the different values ​​of the policy feature vector and cost feature vector. During training, the decision tree continuously partitions the data, ensuring that samples within each subset have similar or identical benefit-cost ratio levels. By learning from historical data, the decision tree model can find the relationship between policy features, cost features, and benefit-cost ratio levels.

[0147] After training, the classification model can predict its benefit-cost ratio level based on the new policy feature vector and cost feature vector, providing a reference for subsequent policy selection.

[0148] Step S245: Integrate the effect prediction coefficient matrix with the benefit-cost ratio level to generate a strategy effect prediction value to replace the manually set benefit-cost ratio calculation rule.

[0149] The effect prediction coefficient matrix and the benefit-cost ratio rank are fused to generate a strategy effect prediction value. A weighted fusion method can be used to generate a comprehensive strategy effect prediction value for each strategy based on the degree of influence of strategy characteristics reflected in the effect prediction coefficient matrix and the benefit-cost relationship of the strategy represented by the benefit-cost ratio rank.

[0150] For example, for a new service optimization strategy, the potential effects are first predicted using an effect prediction coefficient matrix based on the strategy's feature vector, resulting in a preliminary effect prediction value. Then, the preliminary effect prediction value is adjusted based on the benefit-cost ratio ranking of the strategy predicted by the classification model. If the benefit-cost ratio ranking is high, it indicates that the strategy may achieve higher benefits at a given cost, so the preliminary effect prediction value can be appropriately increased; if the benefit-cost ratio ranking is low, it indicates that the strategy's benefits are relatively low, so the preliminary effect prediction value can be appropriately decreased.

[0151] The strategy effect predictions generated in this way can more accurately reflect the actual effect and cost-benefit relationship of the strategy, replacing manually set cost-benefit ratio calculation rules, making subsequent strategy selection more scientific and reasonable.

[0152] Step S250: When an architecture upgrade or business rule change of the service system is detected, the policy triggering conditions and optimization operation templates in the service optimization policy set are automatically adjusted for compatibility adaptation.

[0153] During the operation of the psychology education appointment service system, the system's architecture upgrades and business rule changes are monitored in real time. When an architecture upgrade is detected, such as replacing server hardware, upgrading the operating system or database management system, or when business rules change, such as adjusting course fees or altering certain steps in the appointment process, the strategy triggering conditions and optimization operation templates in the service optimization strategy set need to be adjusted for compatibility.

[0154] For policy triggering conditions, it's necessary to check whether they still apply to the new system architecture and business rules. For example, after a system architecture upgrade, the meaning or value range of certain operational status parameters may have changed, so the parts of the policy triggering conditions involving these parameters need to be adjusted accordingly. If changes to business rules lead to changes in the functionality or processing logic of certain service nodes, then the logical judgment conditions related to these service nodes in the policy triggering conditions also need to be modified.

[0155] Optimized operation templates also require compatibility adjustments. After a system architecture upgrade, the resource allocation parameters in the optimized operation templates may need to be reset based on the new hardware configuration and performance. For example, if the new server hardware has higher performance, the amount of computing resources allocated can be appropriately reduced. After business rules change, the service response logic adjustment rules and the coordination mechanism between service nodes may need to be modified. For instance, after the course fee standard is adjusted, the processing order of the reservation service node's business process may need to be changed to ensure the accuracy of fee calculation; the corresponding service response logic adjustment rules in the optimized operation templates also need to be updated.

[0156] By automatically adapting and adjusting the policy triggering conditions and optimization operation templates for compatibility, we ensure that the set of service optimization policies can still work normally after the service system architecture is upgraded or the business rules are changed, and continue to provide effective support for the optimization of the service system.

[0157] Figure 2 An online consultation service system 100 provided in an embodiment of this application is shown, including a processor 1001 and a memory 1003, and program code stored in the memory 1003. The processor 1001 executes the program code to implement the steps of the appointment process optimization method based on user feedback.

[0158] Figure 2The online consultation service system 100 shown includes a processor 1001 and a memory 1003. The processor 1001 and the memory 1003 are connected, for example, via a bus 1002. Optionally, the online consultation service system 100 may further include a transceiver 1004, which can be used for data interaction between the online consultation service system and other online consultation service systems, such as sending and / or receiving data. It should be noted that in actual scheduling, the transceiver 1004 is not limited to one, and the structure of this online consultation service system 100 does not constitute a limitation on the embodiments of this application.

[0159] Processor 1001 may be a CPU (Central Processing Unit), a general-purpose processor, a DSP (Digital Signal Processor), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute the various exemplary logic blocks, modules, and circuits described in conjunction with the disclosure of this application. Processor 1001 may also be a combination that implements computing functions, such as including one or more microprocessor combinations, a combination of a DSP and a microprocessor, etc.

[0160] Bus 1002 may include a pathway for transmitting information between the aforementioned components. Bus 1002 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. Bus 1002 may be divided into address bus, data bus, control bus, etc.

[0161] The memory 1003 may be ROM (Read Only Memory) or other types of static storage devices capable of storing static information and instructions, RAM (Random Access Memory) or other types of dynamic storage devices capable of storing information and instructions, or EEPROM (Electrically Erasable Programmable Read Only Memory), CD-ROM (Compact Disc Read Only Memory) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, digital universal optical discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium capable of having or storing program code and capable of being read by a computer, without limitation herein.

[0162] The memory 1003 is used to store program code for executing the embodiments of this application, and its execution is controlled by the processor 1001. The processor 1001 is used to execute the program code stored in the memory 1003 to implement the steps shown in the foregoing method embodiments.

[0163] This application provides a computer-readable storage medium storing program code, which, when executed by a processor, can implement the steps and corresponding content of the aforementioned method embodiments.

[0164] It should be understood that although arrows indicate various operation steps in the flowcharts of the embodiments of this application, the order in which these steps are implemented is not limited to the order indicated by the arrows. Unless explicitly stated herein, in some implementation scenarios of the embodiments of this application, the implementation steps in each flowchart may be executed in other orders based on requirements. Furthermore, some or all steps in each flowchart may include multiple sub-steps or multiple stages depending on the actual implementation scenario. Some or all of these sub-steps or stages may be executed at the same time, and each sub-step or stage may also be executed at different times. In scenarios where execution times differ, the execution order of these sub-steps or stages can be flexibly configured based on requirements, and the embodiments of this application do not limit this.

[0165] The above description is only an optional implementation method for some implementation scenarios of this application. It should be noted that for those skilled in the art, other similar implementation methods based on the technical concept of this application, without departing from the technical concept of this application, also fall within the protection scope of the embodiments of this application.

Claims

1. A method for optimizing the appointment process based on user feedback, characterized in that, The method includes: Obtain a set of user feedback data for the target service scenario. The set of user feedback data includes multiple text feedback units submitted by users regarding the appointment service process and their corresponding service node identifiers. The user feedback data set is subjected to sentiment analysis and intent recognition processing to generate an optimized requirement feature set of associated service node identifiers for each text feedback unit; A service optimization strategy set is constructed based on the set of optimization requirement features of the associated service node identifier of each text feedback unit. The service optimization strategy set includes at least one service node optimization strategy and its corresponding strategy triggering conditions. Based on the matching results between the real-time operating status data of the current service system and the triggering conditions of the strategy, the optimization strategy for the target service node is selected from the set of service optimization strategies; The target service node optimization strategy is executed to adjust the resource configuration rules or interaction logic of the corresponding service node in the reservation service process, and the adjusted service node status is updated to the service system.

2. The appointment process optimization method based on user feedback according to claim 1, characterized in that, The process of performing sentiment analysis and intent recognition on the user feedback data set to generate an optimized requirement feature set of associated service node identifiers for each text feedback unit includes: The text feedback unit is segmented to obtain multiple semantic word units, and the part-of-speech tagging result and contextual dependency relationship of each semantic word unit are extracted. The pre-trained sentiment polarity prediction model is invoked to predict the sentiment polarity of the context statement containing the semantic word unit, and the sentiment polarity prediction result of the text feedback unit is generated. The sentiment polarity prediction result includes the distribution ratio of positive sentiment intensity value and negative sentiment intensity value. Based on the named entity recognition results of the semantic word units and the part-of-speech tagging results, an intent recognition feature vector is constructed, and the intent recognition model is called to determine the intent category label of the text feedback unit according to the intent recognition feature vector; Based on the preset time entity recognition rules and service timeliness keyword library, the service timeliness requirement description fragment is identified from the context dependency relationship, and the priority parameters and time constraints corresponding to the service timeliness requirement description fragment are extracted. The service timeliness requirement description fragment includes explicit time constraint description and implicit timeliness requirement description fragment. The emotion polarity prediction result, the intent category label, the priority parameter, and the time constraint are associated and mapped with the corresponding service node identifier to generate an optimized requirement feature set of the associated service node identifier.

3. The appointment process optimization method based on user feedback according to claim 2, characterized in that, The invocation of the pre-trained sentiment polarity prediction model to predict the sentiment polarity of the context statement containing the semantic word unit, generating the sentiment polarity prediction result of the text feedback unit, includes: The semantic word units are input into the embedding layer of the sentiment polarity prediction model in the original order of the context statement to generate a word embedding vector corresponding to each semantic word unit; The word embedding vector is semantically encoded using the bidirectional long short-term memory network layer of the sentiment polarity prediction model, and the context encoding vector of each semantic word unit is output. The context encoding vector is input into the attention mechanism layer of the sentiment polarity prediction model, the attention weight of each semantic word unit in the context sentence is calculated, and the context encoding vector is weighted and fused according to the attention weight to generate the sentence representation vector of the context sentence; The sentence representation vector is input into the fully connected classification layer of the sentiment polarity prediction model, and the positive sentiment score and negative sentiment score of the text feedback unit are output. The positive and negative sentiment scores are normalized to generate the distribution ratio of the positive and negative sentiment intensity values, which is used as the sentiment polarity prediction result.

4. The appointment process optimization method based on user feedback according to claim 2, characterized in that, The construction of the intent recognition feature vector based on the named entity recognition result of the semantic word unit and the part-of-speech tagging result includes: Identify service operation verb units, service object noun units, and service condition adverbial units from the semantic word units; The basic intent category is determined based on the matching result of the service operation verb unit in the preset intent category mapping table, and the basic intent category is further classified based on the type label of the service object noun unit. Extract the logical connective units and comparison condition description fragments from the service condition adverbial units, and combine and encode the type of the logical connective units and the parameter range of the comparison condition description fragments to generate a condition constraint encoding sequence; The service operation verb unit, the service object noun unit, and the condition constraint encoding sequence are respectively subjected to category encoding and numerical normalization processing, and then vectorized and concatenated according to a preset dimension to generate the intent recognition feature vector.

5. The appointment process optimization method based on user feedback according to claim 2, characterized in that, The intent recognition model determines the intent category label of the text feedback unit based on the intent recognition feature vector, including: The intent recognition feature vector is input into the intent recognition model and subjected to nonlinear transformation to obtain a higher-order intent feature vector. The higher-order intent feature vector is subjected to dimensionality compression to generate an intent category probability distribution vector, which contains probability values ​​of multiple preset intent categories. The multiple preset intent categories are sorted according to the magnitude of the probability values, and the top N candidate intent categories are selected to form a candidate intent set; The category confidence score is calculated based on the co-occurrence frequency of each category in the candidate intent set in historical feedback data and the resource configuration status of the current service node; The candidate intent category with the highest confidence score is used as the intent category label for the text feedback unit.

6. The appointment process optimization method based on user feedback according to claim 2, characterized in that, The service optimization strategy set is constructed based on the set of optimization requirement features associated with the service node identifiers of each text feedback unit, including: Based on the sentiment polarity prediction results, the priority parameters, and the time constraints, the target service node identifiers that need to be prioritized are selected, and the intent category tag set corresponding to the target service node identifiers is extracted. Service logic defect analysis is performed on each intent category label in the intent recognition results to determine the service process defect type and the scope of the defect's impact that leads to negative emotions; Based on the service process defect type, a predefined optimization operation template is matched. The optimization operation template includes resource allocation adjustment parameters, service response logic adjustment rules, and coordination mechanisms between service nodes. Based on the scope of the defect's impact, the adjustment range of each parameter in the optimization operation template is calculated, and the strategy triggering condition is generated in combination with the time constraint condition. The optimized operation template is bound with the adjusted parameters and the strategy triggering conditions to form the service node optimization strategy and added to the service optimization strategy set.

7. The appointment process optimization method based on user feedback according to claim 1, characterized in that, The step of selecting target service node optimization strategies from the service optimization strategy set based on the matching result between the real-time operating status data of the current service system and the strategy triggering conditions includes: Monitor the real-time load data, resource utilization indicators, and service request queue length of each service node in the service system; Based on the strategy triggering conditions in the service node optimization strategy, the required running status parameter thresholds and logical judgment conditions are analyzed. The real-time load data, the resource utilization index, and the service request queue length are compared with their respective corresponding operating status parameter thresholds to generate multiple condition-satisfied status identifiers. Based on the logical judgment conditions, logical operations are performed on the condition satisfaction status indicators to determine a list of candidate service node optimization strategies that meet the policy triggering conditions; Based on the expected benefits and implementation costs of each candidate service node optimization strategy in the candidate service node optimization strategy list, a trade-off analysis is performed, and the candidate service node optimization strategy with the highest benefit-cost ratio is selected as the target service node optimization strategy. If there are multiple conflicting strategies for the same service node in the candidate service node optimization strategy list, the candidate service node optimization strategy with the least conflict impact is selected as the target service node optimization strategy based on the priority ranking of the expected benefit value of the strategy optimization and the current load status of the service node.

8. The appointment process optimization method based on user feedback according to claim 1, characterized in that, The step of executing the target service node optimization strategy to adjust the resource configuration rules or interaction logic of the corresponding service node in the reservation service process includes: The resource allocation adjustment parameters in the target service node optimization strategy are analyzed to generate resource reallocation instructions to adjust the computing resource quota or human resource scheduling rules of the service node. Modify the business process processing order or exception handling mechanism of the service node according to the service response logic adjustment rules in the target service node optimization strategy, and update the state transition conditions of the service node. Based on the inter-service node coordination mechanism in the target service node optimization strategy, an information synchronization channel and task handover protocol are established between service nodes to ensure data consistency among optimized service nodes. In a pre-defined canary release environment, simulate a real-world scenario of multi-node collaborative load and sudden traffic fluctuations, conduct stress tests on the adjusted service nodes, and collect performance metrics data and abnormal event records during the test process. Verify whether the optimization effect has reached the expected threshold based on the performance index data. If it has, synchronize the adjusted service node status to the production environment; otherwise, roll back to the state before optimization and regenerate the service optimization strategy. If the number of rollbacks exceeds a preset threshold, a manual intervention process is triggered and an optimization strategy failure alarm is generated, while the automated optimization function of the target service node is frozen.

9. The appointment process optimization method based on user feedback according to claim 6, characterized in that, After updating the adjusted service node status to the service system, the method further includes: Continuously monitor the changing trends of user feedback data after implementing the target service node optimization strategy, and extract newly generated text feedback units for sentiment analysis. Compare the changes in negative sentiment intensity before and after optimization. If the changes do not reach the preset optimization effect threshold, the iterative update mechanism of the service optimization strategy set is triggered. The database of the optimized operation templates is expanded based on the newly identified service process defect types and the newly added intent category tags in the user feedback data; Statistical analysis was conducted on the implementation effects of historical service optimization strategies, and a strategy effect prediction model was established to optimize the cost-benefit ratio calculation logic for subsequent strategy selection. When an architectural upgrade or business rule change is detected in the service system, the policy triggering conditions and optimization operation templates in the service optimization policy set are automatically adjusted for compatibility.

10. An online consultation service system, characterized in that, The device includes a processor and a computer-readable storage medium storing machine-executable instructions that, when executed by the processor, implement the user feedback-based appointment process optimization method as described in any one of claims 1-9.

Images