A Dynamic Analysis Method for Service Quality Incorporating Hypergraph Knowledge

Abstract

The present invention belongs to the technical field of natural language processing, and particularly relates to a method for dynamically analyzing service quality integrating hypergraph knowledge. This method uses the service quality dimension point mutual information algorithm to map the review topics to the five dimensions of the SERVQUAL model; constructs the customer reviews into a hypergraph, and then inputs the hypergraph into a hypergraph attention network for processing to obtain the customer service quality perception value contained in the customer reviews; uses the Wilson interval method to calculate the expected score of the merchant, and then corrects the expected score of the merchant through the Bayesian average method; adds the expected score of the merchant and the average value of the customer historical scores and normalizes them as the customer service quality expectation value to obtain the satisfaction of the customer in the five different dimensions of the SERVQUAL model respectively. The present invention proposes a customer service quality evaluation scheme that does not rely on questionnaire surveys, provides a feedback mechanism for merchants, and further enhances the competitiveness and influence of merchants.

Description

Technical Field

[0001] The present invention belongs to the technical field of natural language processing, and particularly relates to a method for dynamically analyzing service quality by integrating hypergraph knowledge. Background Art

[0002] Service quality is different from physical products that can be directly contacted. Its production and consumption are integrated. Therefore, it more depends on the interaction process between the customer experience and the service provider, and has the characteristic of intangibility. Since different people have different demands and expectations for services, the evaluation criteria of service quality also have a certain degree of subjectivity and flexibility. These characteristics make it difficult to evaluate service quality because multiple factors such as service providers, consumers, and service processes must be considered simultaneously. To better evaluate service quality, the SERVQUAL model based on the gap analysis method was proposed and gradually became the mainstream model in the field of service quality analysis. This model regards service quality as a multi-dimensional concept, specifically determining five dimensions: tangibility, reliability, responsiveness, assurance, and empathy. And service quality is defined as the gap between the customer's expectation of service and the actual service level experienced, that is, it is considered that the customer service quality perception value is equal to the difference between the customer service quality perception value and the customer service quality expectation value. Although the SERVQUAL model is widely used, there are still some deficiencies. First, the implementation of the SERVQUAL model requires designing a scale according to the specific field and expanding it into a questionnaire to be distributed to customers for investigation. However, in fact, the proportion of customers willing to fill out the questionnaire is relatively small. Subsequently, a series of complex operations such as data collection and result analysis are required, which requires a large amount of human and material costs. Second, the SERVQUAL model can only reflect the service quality level at the current moment and cannot track and predict the change trend of service quality, which limits its application in the continuous improvement of service quality. Summary of the Invention

[0003] The purpose of the present invention is to propose a method for dynamically analyzing service quality by integrating hypergraph knowledge. This method first obtains the comment topics and their keywords from large-scale online reviews through a comment topic extraction method based on a pre-trained model and maps them to the dimensions of the SERVQUAL model. Secondly, each comment is constructed into a hypergraph by using three hyperedge construction methods: text structure, internal semantics, and external knowledge, and a hypergraph attention network is used to process it to extract the service quality level actually perceived by the customer. Subsequently, the customer service quality expectation value is extracted, and the difference between it and the actual perception is used as the customer service quality perception value reflected by each comment. Then, combined with the comment topic corresponding to this evaluation, the satisfaction of the customer in the five different dimensions of the SERVQUAL model is obtained.

[0004] The present invention is realized by the following technical solutions.

[0005] A method for dynamically analyzing service quality integrating hypergraph knowledge, the steps are as follows:

[0006] Step 1: Collect online network platform data and construct a dataset; the online network platform data includes at least customer reviews;

[0007] Step 2: Extract the review themes contained in the customer reviews and determine the keywords and their weights for each review theme;

[0008] Step 3: Adjust the extracted review themes and keywords to optimize the review themes;

[0009] Step 4: Use the service quality dimension point mutual information algorithm to map the review themes to the five dimensions of the SERVQUAL model;

[0010] Step 5: Perform confidence learning on the dataset to find mislabeled samples and correct them manually;

[0011] Step 6: Construct the customer reviews into a hypergraph containing three types of hyperedges: text structure, intrinsic semantics, and external knowledge, and then input the hypergraph into a hypergraph attention network for processing, so as to obtain the customer service quality perception value contained in the customer reviews;

[0012] Step 7: Use the Wilson interval method to calculate the expected score of the merchant, and then correct the expected score of the merchant through the Bayesian average method; add the expected score of the merchant and the average value of the customer historical scores and normalize them as the customer service quality expectation value;

[0013] Step 8: Combine the three factors of review themes, customer service quality expectation value, and customer service quality perception value to obtain the satisfaction of customers in the five different dimensions of the SERVQUAL model.

[0014] Further preferably, the online network platform data further includes merchant information, customer information, and customer scores.

[0015] Further preferably, use a review theme modeling method based on the pre-trained model RoBERTa to extract the review themes contained in the customer reviews.

[0016] Further preferably, use the c-TF-IDF algorithm to determine the keywords and their weights for each review theme;

[0017] Further preferably, adjust the extracted review themes and keywords to optimize the review themes by setting stop words, changing the minimum word frequency, and adjusting the N-gram parameters.

[0018] Further preferably, in step four, seed words on the five dimensions of the SERVQUAL model are constructed, and the keyword in each comment topic and the seed words on the five dimensions of the SERVQUAL model are both constructed into a word set; the service quality dimension point mutual information algorithm (SQ-PMI) is used to calculate the correlation value between sets, and the comment topic is dynamically mapped to the five dimensions of the SERVQUAL model.

[0019] Further preferably, in step five, all customer comment data in the dataset are divided into five equal parts, and then the FastText model is used to perform five-fold cross-validation on the data. During the cross-validation process, the joint distribution matrix of the estimated noise label and the true label is obtained. Finally, the samples with a confidence threshold lower than the joint distribution matrix are screened out, which are the label error samples, and the rest are the label correct samples. Then the label error samples are manually corrected.

[0020] Further preferably, in step six, each customer comment is constructed into a hypergraph. First, the customer comment is tokenized, and words are used as the nodes of the graph. Then a fixed-size sliding window is used to obtain the global word co-occurrence as the sequential context, which is used as the text structure hyperedge of the hypergraph. The words belonging to the same comment topic in the nodes are connected by the same hyperedge as the internal semantic hyperedge. External knowledge is obtained from the ConceptNet knowledge graph and an external knowledge hyperedge is constructed; in the hypergraph processing link, the constructed hypergraph is fed into the hypergraph attention network, and the hypergraph attention network will output the customer service quality perception value contained in the customer comment.

[0021] On the basis of integrating the five dimensions of the SERVQUAL model and the calculation method of the customer service quality perception value, the present invention constructs a hypergraph by using the text sequential structure, internal semantic association of customer comment text and the triple knowledge of the knowledge graph, and then embeds the hypergraph into a graph neural network for processing, and has the following advantages:

[0022] 1. Combining the traditional SERVQUAL model with deep learning technology to perform sentiment analysis, comment topic modeling, etc. on online reviews, so as to dynamically and continuously analyze the customer satisfaction, avoiding the cumbersome process of designing, distributing and recycling questionnaires in the traditional analysis method.

[0023] 2. Designing a dynamic mapping method between a comment topic model and the five dimensions of the SERVQUAL model, so that each evaluation can be mapped to one dimension of the SERVQUAL model. And a method for extracting the customer service quality expectation value that is difficult to quantify and measure is proposed. The degree of fit between the deep learning model and the SERVQUAL model is improved, enabling them to better form an organic whole to jointly analyze the evaluation.

[0024] 3. Hyperedges are constructed for the evaluation hypergraph by respectively adopting the sequential structure of the evaluation text, the internal latent semantic information, and the external knowledge provided by the knowledge graph. The three different types of hyperedges are complementary, capable of providing multi-source heterogeneous features for the evaluation hypergraph and enriching the semantic information contained in the hypergraph.

[0025] 4. Focal loss is introduced to improve the robustness of the model on the imbalanced sample dataset. On the sample dataset with extreme positive and negative ratios, dynamic weight adjustment can be achieved, increasing the weight of the loss value of the minority class samples to improve the network performance. Brief Description of the Drawings

[0026] Figure 1 is a flowchart of a method for dynamically analyzing service quality by integrating hypergraph knowledge provided by the present invention;

[0027] Figure 2 is a schematic diagram of the framework structure of the method for dynamically analyzing service quality by integrating hypergraph knowledge;

[0028] Figure 3 is a scatter plot of the distribution of customer service quality perception values of comments;

[0029] Figure 4 is a histogram of the number of comments in each interval of customer service quality perception values;

[0030] Figure 5 is a graph of the change in the service quality level of merchants over time series. Detailed Embodiment

[0031] The present invention will be further described in detail below with reference to the drawings and embodiments.

[0032] Refer to Figure 1 , a method for dynamically analyzing service quality by integrating hypergraph knowledge, the steps are as follows:

[0033] Step 1: Collect online network platform data such as merchant information, customer information, customer comments, and customer ratings from the online network platform, and construct a dataset after cleaning and preprocessing the online network platform data;

[0034] Step 2: Use the comment topic modeling method based on the pre-trained model RoBERTa to extract the comment topics contained in the customer comments, and use the c-TF-IDF algorithm to determine the keywords and their weights of each comment topic;

[0035] Step 3: Adjust the extracted comment topics and keywords in three ways: setting stop words, changing the minimum word frequency, and adjusting the N-gram parameters to optimize the comment topics;

[0036] Step 4: Use the service quality dimension point mutual information algorithm to map the comment topics to the five dimensions of the SERVQUAL model;

[0037] Step 5: Conduct confidence learning on the dataset to find mislabeled samples and correct them manually;

[0038] Step 6: Construct the customer comments into a hypergraph containing three types of hyperedges: text structure, intrinsic semantics, and external knowledge, and then input the hypergraph into the hypergraph attention network for processing to obtain the customer service quality perception value contained in the customer comments;

[0039] Step 7: Use the Wilson interval method to calculate the expected score of the merchant, and then correct the expected score of the merchant through the Bayesian average method; Add the expected score of the merchant and the average value of the customer's historical scores and normalize them as the customer service quality expectation value;

[0040] Step 8: Combine the three factors of comment topics, customer service quality expectation value, and customer service quality perception value to obtain the satisfaction of customers in the five different dimensions of the SERVQUAL model.

[0041] Figure 2 Fig. shows the basic framework of the present invention, which has three branches: SERVQUAL dimension mapping branch, customer service quality perception value prediction branch, and customer service quality expectation value prediction branch.

[0042] For the dataset composed of online network platform data, in the SERVQUAL dimension mapping branch, first use the pre-trained RoBERTa model to perform embedding operations on the words in the text. The embedded word vectors will be used for comment topic extraction and constructing the nodes of the hypergraph. Subsequently, the comment topics contained in the dataset are obtained by reducing the dimensionality and clustering of the word vectors, and then the keywords and weights of each comment topic are determined through the c-TF-IDF algorithm. This algorithm can be represented by formula (1):

[0043] (1);

[0044] Where, represents the operation result of the c-TF-IDF algorithm of keyword t under clustering c, t represents the keyword in the comment topic clustering, c represents the clustering, represents the word frequency of keyword t under clustering c, represents the word frequency of keyword t in all corpora, is the average number of words in clustering c. To ensure that the value is only positive, add 1 to the division in the logarithm.

[0045] After obtaining the preliminary comment themes and their keywords, the extracted comment themes and keywords are adjusted in three ways: setting stop words, changing the minimum word frequency, and adjusting the N-gram parameters, so as to optimize the representation of comment themes. After obtaining the optimized comment themes and keywords, there may be a problem that the focus of the comment themes is inconsistent with the dimensions of the SERVQUAL evaluation model. Based on this, the present invention constructs seed words for the five dimensions of the SERVQUAL model, and constructs a set of words for each keyword in the comment theme and the seed words for the five dimensions of the SERVQUAL model. The present invention uses the service quality dimension point mutual information algorithm (SQ-PMI) to calculate the correlation value between sets, which can dynamically map the comment theme to the five dimensions of the SERVQUAL model. This algorithm can be expressed by formula (2):

[0046] (2);

[0047] Where T represents the set of comment theme keywords, D represents the set of dimension seed words, represents the service quality dimension point mutual information value of the comment theme T on the dimension D, d represents a dimension seed word, represents the probability that the comment theme keyword t appears, represents the probability that the dimension seed word d appears, represents the probability that the comment theme keyword t and the dimension seed word d appear simultaneously. A positive service quality dimension point mutual information value indicates that the two sets are positively correlated, and the larger the value, the stronger the correlation. If it is 0, it means that the two are almost statistically independent and have no correlation. If it is negative, it means that the two are negatively correlated, that is, mutually exclusive. For each comment theme, the SQ-PMI is used to calculate the correlation value between the comment theme and each of the five dimensions of the SERVQUAL model, and the comment theme is mapped to the dimension with the largest correlation value.

[0048] The customer service quality perception value prediction branch constructs hyperedges for the hypergraph according to the text structure, internal semantics, and external knowledge. Then, low-confidence samples are found through confidence learning and corrected to prevent the spread of wrong labels. Subsequently, the hypergraph is fed into the hypergraph attention network, and the hypergraph attention network outputs the customer service quality perception value. The reason for using confidence learning is that when customers give scores online, the scores given do not match the emotions contained in the comments. For example, consumers give negative comments but give full marks, or give positive comments but only give low scores. This will reduce the credibility of the labels of customer comment data and affect subsequent model training and the extraction of customer service expectation values. To alleviate this situation, the present invention divides all customer comment data in the dataset into five equal parts, and then uses the FastText model to perform five-fold cross-validation on the data, and obtains the joint distribution matrix of the estimated noise labels and the true labels during the cross-validation process. This process can be expressed by formula (3) and formula (4):

[0049] (3);

[0050] (4);

[0051] Wherein represents the sample number, m represents the total number of label categories, represents one of the categories. represents the sample 's predicted value, represents the category 's number of labels. is an m-by-m count matrix, the values of which are statistically obtained from the samples, and the columns represent the number of labels on the category and the rows represent the number of samples whose model predicted value is the category . represents the total number of samples with pseudo-label . is the calibration count matrix, and formula (4) shows the process of regularizing it. After regularization, the joint distribution matrix of the estimated noisy labels and the true labels is obtained . Finally, samples below the confidence threshold are screened out according to the joint distribution matrix, which are the mislabeled samples, and the rest are correctly labeled samples. Then the mislabeled samples are corrected manually.

[0052] The customer service quality expectation prediction branch first combines the Wilson interval method with Bayesian averaging to obtain the expected score of the merchant. Specifically, the present invention uses the Wilson interval method to calculate the expected score of the merchant at a 95% confidence level. Since some merchants have fewer reviews, using only the Wilson interval algorithm will cause unpopular merchants to be restricted by the lower limit of the interval and result in a lower score. Therefore, the present invention then uses the Bayesian averaging method to provide a compensation value for unpopular merchants so that their scores will not be unreliable due to a small number of evaluations. This process can be represented by formula (5) and formula (6):

[0053] (5);

[0054] (6);

[0055] Wherein, represents the Wilson score, is the upper limit of the score and plays a scaling role. Denote the favorable rate, which is obtained by dividing the average value by the total score value. Here, n is the total number of evaluations, and K is the statistic of the normal distribution at a fixed confidence level. h represents the number of merchant evaluations, and C represents the correction value for the merchant, which is equal to the average number of votes for all items. The smaller the C value, the more the result of Bayesian correction will tend to the original data, the influence on niche merchants will be weakened, and the overall score will be closer to the Wilson score. The larger the C value, the greater the correction value for niche merchants, making the score differences among all merchants smaller. H represents the total average score of all merchants. Denote the final expected score of the merchant.

[0056] After obtaining the expected score of the merchant, sum it with the customer's historical average score and then normalize it to obtain the expected value of the customer service quality extracted. Finally, calculate the difference between the perceived value of the customer service quality and the expected value of the customer service quality in the five dimensions of the SERVQUAL model to obtain the customer's satisfaction in the five dimensions.

[0057] As a generalization of graphs, hypergraphs can model complex network structures more flexibly and exhibit higher-order relationships that graphs cannot fully depict. In the present invention, each customer review is constructed into a hypergraph. First, the customer review is segmented, with words used as the nodes of the graph, and then a fixed-size sliding window is used to obtain the global word co-occurrence as sequential context, which is used as the text structure hyperedge of the hypergraph. To capture the higher-order correlation between words and potential review topics, words belonging to the same review topic in the nodes are connected by the same hyperedge as the internal semantic hyperedge. Since most customer reviews are short texts, to enrich the semantics therein, the present invention obtains external knowledge from the ConceptNet knowledge graph and constructs external knowledge hyperedges. The hypergraph obtained by using the above three hyperedge construction methods can be represented by G(V, E), where V = {v 1 , …, v a} represents the set of a nodes in the hypergraph, v 1 represents the first node, v a represents the a-th node, E = {e 1 , …, e b} represents the set of b hyperedges, e 1 represents the first hyperedge, e b represents the b-th hyperedge. For any hyperedge e, it can connect two or more nodes. The topological structure of the hypergraph G can be represented by an incidence matrix of a rows and b columns as follows,

[0058] (7);

[0059] where p is the node number and q is the hyperedge number, represents the p-th node, represents the p-th hyperedge.

[0060] In the hypergraph processing stage, the constructed hypergraph is fed into the hypergraph attention network, which outputs the perceived customer service quality value contained in this customer review. The hypergraph attention network will aggregate the vertices connected to the hyperedge to obtain the representation vector of the hyperedge, and then aggregate the hyperedge vectors to update the node vectors. Since in the process of node aggregation, the importance of each node to the hyperedge is not the same; in the process of hyperedge aggregation, the influence of each hyperedge on the node is not consistent either. Therefore, the present invention introduces an attention mechanism in the hypergraph attention network to dynamically adjust the weights in the aggregation process. Equation (8) represents the process of aggregating all the nodes connected to the hyperedge to obtain the hyperedge representation vector. Equation (9) represents the process of aggregating the hyperedge vectors connected to the node to update the node representation. In this process, the number of hyperedges aggregated by the node is always equal to the degree of this node.

[0061] (8);

[0062] (9);

[0063] Among them, represents the hyperedge in the -th layer of the hypergraph attention network. represents the hyperedge connected to the node, represents the representation vector of this node in the -th layer, as the attention coefficient represents the importance of the node to the hyperedge . W 1 and W 2 both represent weight matrices. represents the aggregation operation of the -th layer, represents the process of aggregating all the nodes connected to each hyperedge in the -th layer to obtain the hyperedge representation vector . represents the process of aggregating the hyperedges connected to each node in the -th layer to update the node representation vector, represents the representation vector of node p in the -th layer, represents the representation vector of the node in the -th layer. represents the set of edges connected to the node , represents the hyperedge The importance of nodes is important.

[0064] The perceived value of customer service quality often shows an uneven distribution. Generally speaking, positive perceptions are much more than negative perceptions. This imbalance will lead to the minority class samples being overwhelmed by the majority class, making the output of the hypergraph attention network all the majority class and resulting in prediction failure. To address the problem of sample imbalance faced in the hypergraph processing link, the present invention uses the focal loss (FL) with dynamic class weights to train the hypergraph attention network. The focal loss is an improvement of the cross-entropy loss. By setting a dynamic scaling factor, the weight of the loss value of the minority class samples is increased to avoid being overwhelmed by the majority class.

[0065] The data used in the embodiments of the present invention comes from the yelp website. The reviews on this website are different from the reviews related to physical products. In the former, customers are more inclined to evaluate the service quality level of service providers and other service entities, while the latter focuses on evaluating an actual existing product itself. Due to the intangibility of services and the diversity of evaluation criteria, it is more difficult to measure the level. The present invention stores the merchant information, customer information, customer reviews, and customer ratings provided by the yelp website in different tables of the database respectively. Among them, the merchant information table contains three fields: merchant number, merchant category, and the review number corresponding to the merchant. The present invention will collect all relevant data of the merchants engaged in the tree care industry on the yelp website and integrate them into a dataset, and then preprocess the text in the dataset. First, stop words are removed, mainly including some prepositions and conjunctions without actual meaning, then all letters are uniformly converted to lowercase, and finally, part-of-speech reduction is performed.

[0066] The effects of the present invention are measured by 4 indicators, namely accuracy, weighted precision, recall, and F1 value. They use the number of samples in each category as the weight. Compared with the macro method of directly averaging the indicators of each category, it can ensure that more important categories have a greater influence when evaluating the model effect.

[0067] The present invention automatically extracts 9 review topics, as well as the keywords and weights under each review topic, from the dataset through a pre-trained RoBERTa model. The keywords and their weights of the first three review topics are shown in Table 1.

[0068] Table 1 Review topic keywords and their weights of the first three extracted review topics

[0069]

[0070] Subsequently, the service quality dimension point mutual information algorithm is used to map each review topic to the five dimensions of the SERVQUAL model. The service quality dimension point mutual information values (SQ-PMI values) of each review topic on the five dimensions of the SERVQUAL model are shown in Table 2. The dimension with the largest value is the SERVQUAL dimension to which this review topic belongs.

[0071] Table 2 SQ-PMI values of each review topic on the five dimensions of the SERVQUAL model

[0072]

[0073] Then, the hypergraph attention network is used to obtain the service quality perception value of each review. The hypergraph attention network is trained on a part of the dataset and is used to analyze new samples after training to ensure that the proposed method in this paper can incrementally extract customer perception. The accuracy of the hypergraph attention network on the test set reaches 85.2%, and the weighted precision, recall, and F1 values are 83.3%, 81.5%, and 82.6% respectively. Then, the customer's expected value of customer service quality is calculated, and their difference is used as the evaluation of the corresponding customer service quality perception value. The distribution of the customer service quality perception values of all reviews in the dataset is as Figure 3 shown. The number of customer reviews in each customer service quality perception value interval is as Figure 4 shown.

[0074] To verify the effectiveness of the method proposed in this study in standardized document generation, ablation experiments were designed, and 6 ablation schemes were developed. In Scheme 1, a review is regarded as a node, the entire dataset is constructed as a graph, and then the node classification task of the graph is performed. The focal loss is used to avoid the classification result from shifting to multiple classes. On the basis of Scheme 1, Scheme 2 introduces an attention mechanism in the process of graph node aggregation. In Scheme 3, each evaluation is constructed as a graph, and the hyperedges of the graph are constructed according to the text structure, and then the graph convolutional network is used to process the hypergraph. On the basis of Scheme 3, Scheme 4 fuses the internal semantics and external knowledge of the review text, and constructs the internal potential review topics and external knowledge triples as hyperedges respectively. On the basis of Scheme 5, Scheme 6 removes the focal loss and uses the cross-entropy loss function to observe whether the model prediction results will be imbalanced. Scheme 7 is the proposed scheme of the present invention. Confidence learning is performed on the dataset before model training, and some samples with possible label errors are corrected to prevent the wrong labels from spreading in the model and interfering with the prediction effect of the model. The results of the ablation experiments are shown in Table 3. It can be seen from this that the indicators of the method of the present invention are all better than the other 6 ablation schemes, which reflects the effectiveness of the proposed scheme of the present invention and the rationality of the module combination.

[0075] Table 3 Ablation Experiment

[0076]

[0077] Since the present invention can quickly obtain the satisfaction of customers on the five dimensions of the SERVQUAL evaluation model, the comment sequence can be dynamically analyzed in the time dimension. The change of satisfaction on the Tree Services dataset over time is as Figure 5 shown. It can be seen from this that the service level of the empathy evaluation dimension of merchants in this dataset is always lower than the expected level, which reflects that merchants should strengthen personalized care and services for customers, improve the understanding and attention to customer needs, and give priority to the interests of customers at the same time, so as to establish a deeper emotional connection and thus improve customer satisfaction in terms of empathy. There have also been situations where customers were dissatisfied with the responsiveness dimension, which also requires key attention from merchants and continuous improvement.

[0078] This embodiment provides a computer-readable storage medium, on which computer instructions are stored. When the instructions are executed by a processor, the dynamic service quality analysis method integrating hypergraph knowledge is implemented.

[0079] The above-described invention only expresses the implementation manners of the embodiments of the present invention, and thus cannot be construed as a limitation on the scope of the invention patent, nor is it a limitation on the structure of the embodiments of the present invention in any form. It should be noted that for those of ordinary skill in the art, without departing from the concept of the embodiments of the present invention, several changes and improvements can still be made, and these all belong to the protection scope of the embodiments of the present invention.

Claims

1. A method for dynamic analysis of service quality integrating hypergraph knowledge, characterized in that: Here are the steps: Step 1: Collect online network platform data and construct a data set; the online network platform data at least includes customer reviews; Step 2: Extract the review topics contained in the customer reviews and determine the keywords and their weights for each review topic; Step 3: Adjust the extracted review topics and keywords to optimize the review topics; Step 4: Construct seed words on the five dimensions of the SERVQUAL model, and use the service quality dimension point mutual information algorithm to dynamically map the review topics to the five dimensions of the SERVQUAL model: ； in T Represents a collection of comment topic keywords. D represents the set of dimension seed words, Indicates the comment topic T In Dimension D The mutual information value of the service quality dimension point on d Represents a dimension seed word, Indicates the comment topic keywords t The probability of occurrence, Represents dimension seed word d The probability of occurrence, Indicates the comment topic keywords t and dimension seed words d Probability of simultaneous occurrence; Step 5: Perform confidence learning on the dataset to find samples with incorrect labels and manually correct them; Step 6: Construct customer reviews into a hypergraph containing three types of hyperedges: text structure, intrinsic semantics, and external knowledge. Then input the hypergraph into the hypergraph attention network for processing, thereby obtaining the customer service quality perception value contained in the customer reviews. Step 7: Use the Wilson interval method to calculate the expected score of the merchant, and then use the Bayesian average method to correct the expected score of the merchant; add the expected score of the merchant and the average of the customer's historical score and normalize them as the customer's service quality expectation; Step 8: Combining the three factors of review topic, customer service quality expectations, and customer service quality perception, we can derive customer satisfaction on the five different dimensions of the SERVQUAL model.

2. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: The online network platform data also includes merchant information, customer information and customer ratings.

3. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: A review topic modeling method based on the pre-trained model RoBERTa is used to extract review topics contained in customer reviews.

4. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: The c-TF-IDF algorithm is used to determine the keywords and their weights for each review topic.

5. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: By setting stop words, changing the minimum word frequency, and adjusting N-gram parameters, the extracted comment topics and keywords are adjusted to optimize the comment topics.

6. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: In step five, all customer review data in the dataset are divided into five equal parts, and then the FastText model is used to perform a five-fold cross-validation on the data. During the cross-validation process, the joint distribution matrix of the estimated noise labels and the true labels is obtained. Finally, the samples below the confidence threshold are screened out against the joint distribution matrix, which are the incorrectly labeled samples, and the rest are correctly labeled samples. The incorrectly labeled samples are then manually corrected.

7. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: The Wilson interval method is used to find the expected score of the merchant, which is expressed as follows: ； in, represents the Wilson score, is the upper limit of the rating. Indicates the rate of favorable comments, n is the total number of reviews, K It is a statistic of the normal distribution at a fixed confidence level.

8. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 7 is characterized in that: The expected rating of the merchant is corrected by the Bayesian average method, which is expressed as follows: ； in, h Indicates the number of merchant reviews. C Indicates the correction value for the merchant. H Represents the total average score of all merchants. Indicates the final expected merchant rating.

9. The method for dynamic analysis of service quality integrating hypergraph knowledge according to claim 1 is characterized in that: In step six, each customer review is constructed as a hypergraph. First, the customer reviews are segmented and the words are used as nodes of the graph. Then, a fixed-size sliding window is used to obtain global word co-occurrence as sequential context, which is used as the text structure hyperedge of the hypergraph. The words belonging to the same review topic in the node are connected with the same hyperedge as the intrinsic semantic hyperedge. External knowledge is obtained from the ConceptNet knowledge graph and external knowledge hyperedges are constructed. In the hypergraph processing stage, the constructed hypergraph is sent to the hypergraph attention network, which will output the customer service quality perception value contained in the customer reviews.

Images