Attention analysis method, device, equipment, medium and program product

Abstract

The invention relates to an appeal analysis method and device, computer equipment, a computer readable storage medium and a computer program product. The method comprises the following steps: acquiring user feedback data, and performing data desensitization on the user feedback data to obtain desensitized data; performing text splitting on the desensitized data to obtain split text data; performing data classification on the split text data to obtain scene classification data and sentiment classification data; performing appeal extraction on the split text data to obtain key appeal points, and performing vectorization processing on the key appeal points to obtain appeal vectors; performing appeal clustering on the appeal vectors according to a preset clustering model and the scene classification data to obtain an appeal clustering result; performing appeal summarization on the appeal clustering result to obtain an appeal summarization result; and performing appeal analysis on the sentiment classification data and the appeal clustering result according to an appeal summarization result to obtain an appeal analysis result. By adopting the method, the appeal processing efficiency and accuracy can be improved.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a claims analysis method, apparatus, computer device, computer-readable storage medium, and computer program product. Background Technology

[0002] With the development of artificial intelligence technology, users' demand for personalized and high-quality travel experiences is growing. Users' real needs, emotional tendencies, and improvement suggestions are key to understanding user needs and optimizing service experiences, providing precise guidance for service innovation and process optimization. However, traditional methods of manual analysis and small-scale model mining of user needs suffer from technical problems such as low processing efficiency, insufficient analytical dimensions, and limited semantic understanding capabilities when dealing with rapidly increasing unstructured text data. Summary of the Invention

[0003] Therefore, it is necessary to provide a method, apparatus, computer equipment, computer-readable storage medium, and computer program product that can improve the efficiency and accuracy of handling requests, addressing the aforementioned technical problems.

[0004] Firstly, this application provides a method for claim analysis, including:

[0005] Obtain user feedback data and perform data anonymization on the user feedback data to obtain anonymized data;

[0006] The de-identified data is then split into text data to obtain split text data.

[0007] The split text data is classified to obtain scene classification data and sentiment classification data;

[0008] The split text data is processed to extract the key appeal points, and these key appeal points are then vectorized to obtain appeal vectors.

[0009] The request vector is clustered according to a preset clustering model and scenario classification data to obtain the request clustering result;

[0010] The clustering results of the demands are used to summarize the demands, resulting in a demand summary.

[0011] Based on the summary of the demands, the emotional classification data and the clustering results of the demands are analyzed to obtain the demand analysis results.

[0012] In one embodiment, the step of classifying the split text data to obtain scene classification data and sentiment classification data includes:

[0013] The split text data is classified into service scenarios according to a preset service scenario classification table and a preset service scenario intelligent agent to obtain the scenario classification data;

[0014] The split text data is classified according to the preset sentiment classification agent to obtain the sentiment classification data.

[0015] In one embodiment, the sentiment classification data includes: sentiment tendency and emotion rating data; the step of performing sentiment classification on the split text data according to a preset sentiment classification agent to obtain the sentiment classification data includes:

[0016] The sentiment classification agent performs sentiment tendency classification on the split text data to obtain the sentiment tendency.

[0017] The emotion classification agent performs emotion classification on the split text data to obtain the emotion classification data.

[0018] In one embodiment, the step of extracting key appeal points from the split text data includes:

[0019] The structured text is obtained by preprocessing the split text data using a pre-defined demand mining agent.

[0020] Based on the stated requirements, the intelligent agent performs semantic understanding on the structured text to obtain deep semantic data;

[0021] The key appeal points are obtained by refining the deep semantic data based on the appeal point description model of the appeal mining agent.

[0022] In one embodiment, the step of clustering the request vector according to a preset clustering model and scene classification data to obtain the request clustering result includes:

[0023] The core distance of the claims is calculated by performing a core distance calculation on the claim vectors based on the clustering model;

[0024] Construct a minimum spanning tree based on the core distance of the stated requirements;

[0025] The minimum spanning tree is pruned according to the clustering model to obtain a hierarchical cluster structure.

[0026] Stability analysis was performed on the hierarchical cluster structure to obtain the clustering results of the claims.

[0027] In one embodiment, the step of summarizing the demands based on the clustering results to obtain the demand summary results includes:

[0028] Based on the preset demand summarizing agent, semantic analysis is performed on the text content with the same clustering label in the demand clustering results to obtain demand summarizing points;

[0029] The request summary points are annotated and generated according to the preset request summary agent to obtain request summary annotations; the request summary result includes the request summary points and the request summary annotations.

[0030] Secondly, this application also provides a claim analysis device, comprising:

[0031] The acquisition module is used to acquire user feedback data and perform data anonymization on the user feedback data to obtain anonymized data;

[0032] The text splitting module is used to split the de-identified data into split text data.

[0033] The data classification module is used to classify the split text data to obtain scene classification data and sentiment classification data;

[0034] The appeal extraction module is used to extract appeals from the split text data, obtain key appeal points, and vectorize the key appeal points to obtain appeal vectors.

[0035] The appeal clustering module is used to cluster the appeal vector according to a preset clustering model and scenario classification data to obtain the appeal clustering result;

[0036] The demand summarization module is used to summarize the demands clustering results to obtain the demand summary results.

[0037] The appeal analysis module is used to perform appeal analysis on the sentiment classification data and the appeal clustering results based on the appeal summary results, and obtain appeal analysis results.

[0038] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0039] Obtain user feedback data and perform data anonymization on the user feedback data to obtain anonymized data;

[0040] The de-identified data is then split into text data to obtain split text data.

[0041] The split text data is classified to obtain scene classification data and sentiment classification data;

[0042] The split text data is processed to extract the key appeal points, and these key appeal points are then vectorized to obtain appeal vectors.

[0043] The request vector is clustered according to a preset clustering model and scenario classification data to obtain the request clustering result;

[0044] The clustering results of the demands are used to summarize the demands, resulting in a demand summary.

[0045] Based on the summary of the demands, the emotional classification data and the clustering results of the demands are analyzed to obtain the demand analysis results.

[0046] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, performs the following steps:

[0047] Obtain user feedback data and perform data anonymization on the user feedback data to obtain anonymized data;

[0048] The de-identified data is then split into text data to obtain split text data.

[0049] The split text data is classified to obtain scene classification data and sentiment classification data;

[0050] The split text data is processed to extract the key appeal points, and these key appeal points are then vectorized to obtain appeal vectors.

[0051] The request vector is clustered according to a preset clustering model and scenario classification data to obtain the request clustering result;

[0052] The clustering results of the demands are used to summarize the demands, resulting in a demand summary.

[0053] Based on the summary of the demands, the emotional classification data and the clustering results of the demands are analyzed to obtain the demand analysis results.

[0054] Fifthly, this application also provides a computer program product, including a computer program that, when executed by a processor, performs the following steps:

[0055] Obtain user feedback data and perform data anonymization on the user feedback data to obtain anonymized data;

[0056] The de-identified data is then split into text data to obtain split text data.

[0057] The split text data is classified to obtain scene classification data and sentiment classification data;

[0058] The split text data is processed to extract the key appeal points, and these key appeal points are then vectorized to obtain appeal vectors.

[0059] The request vector is clustered according to a preset clustering model and scenario classification data to obtain the request clustering result;

[0060] The clustering results of the demands are used to summarize the demands, resulting in a demand summary.

[0061] Based on the summary of the demands, the emotional classification data and the clustering results of the demands are analyzed to obtain the demand analysis results.

[0062] The aforementioned methods, apparatus, computer equipment, computer-readable storage media, and computer program products for analyzing user feedback data acquire user feedback data and anonymize it to obtain anonymized data; the anonymized data is then split into text data to obtain split text data; the split text data is then classified to obtain scenario classification data and sentiment classification data; the demands are extracted from the split text data to obtain key demands, and these key demands are then vectorized to obtain demand vectors; the demand vectors are then clustered according to a preset clustering model and the scenario classification data to obtain demand clustering results; the demand clustering results are then summarized to obtain demand summary results; and finally, the sentiment classification data and the demand clustering results are analyzed based on the demand summary results to obtain demand analysis results. Therefore, by splitting the anonymized data into text, and then classifying the split text data into scenario-based and sentiment-based data, and extracting key user requests from the split text data, we can achieve more refined and context-based analysis of user requests by utilizing a large language model for text splitting, data classification, and request extraction. Furthermore, by using a pre-defined clustering model—specifically, using a small model and scenario-based data—to cluster the request vectors, we can obtain request clustering results. Finally, we can summarize the requests based on these clustering results, and then analyze the sentiment-based data and clustering results to obtain request analysis results. This approach better leverages the synergy between the large language model and the small-scale clustering model, accurately identifying core user requests, thereby optimizing resource allocation and improving user experience. Attached Figure Description

[0063] To more clearly illustrate the technical solutions in the embodiments of this application or related technologies, the drawings used in the description of the embodiments of this application or related technologies will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0064] Figure 1 This is a diagram illustrating the application environment of the claims analysis method in one embodiment;

[0065] Figure 2 This is a flowchart illustrating the claims analysis method in one embodiment;

[0066] Figure 3 This is a schematic diagram of the appeal analysis and ranking process in one embodiment of the appeal analysis method;

[0067] Figure 4 This is a request analysis ranking diagram of the request analysis method in another embodiment;

[0068] Figure 5 This is a schematic diagram of the complete process of the claims analysis method in another embodiment;

[0069] Figure 6 This is a structural block diagram of the claims analysis device in one embodiment;

[0070] Figure 7 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0071] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0072] The claims analysis method provided in this application embodiment can be applied to, for example, Figure 1In the application environment shown, terminal 102 communicates with server 104 via a network. A data storage system can store the data that server 104 needs to process. The data storage system can be integrated onto server 104 or placed on a cloud or other network server. Server 104 acquires user feedback data and performs data anonymization on it to obtain anonymized data; it then performs text splitting on the anonymized data to obtain split text data; it performs data classification on the split text data to obtain scene classification data and sentiment classification data; it extracts the demands from the split text data to obtain key demands, and then vectorizes these key demands to obtain demand vectors; it performs demand clustering on the demand vectors based on a preset clustering model and scene classification data to obtain demand clustering results; it summarizes the demands based on the demand clustering results to obtain demand summary results; and it performs demand analysis on the sentiment classification data and demand clustering results based on the demand summary results to obtain demand analysis results. The terminal 102 can be, but is not limited to, various personal computers, laptops, smartphones, tablets, IoT devices, and portable wearable devices. IoT devices can include smart speakers, smart TVs, smart air conditioners, smart in-vehicle systems, and projection devices. Portable wearable devices can include smartwatches, smart bracelets, and head-mounted displays. Head-mounted displays can be virtual reality (VR) devices, augmented reality (AR) devices, and smart glasses. The server 104 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0073] In one exemplary embodiment, such as Figure 2 As shown, a method for demand analysis is provided, which can be applied to... Figure 1 Taking server 104 as an example, the explanation includes the following steps 202 to 214. Wherein:

[0074] Step 202: Obtain user feedback data and perform data anonymization on the user feedback data to obtain anonymized data.

[0075] Here, "user" refers to the airline's passengers, and user feedback data can include customer service work orders, complaint channel information, social media information, cabin crew log data, and questionnaire survey data.

[0076] In some embodiments, user feedback data can be obtained from free text data generated at multiple touchpoints, including customer service tickets, complaint channels, social media, cabin crew logs, and questionnaires. This user feedback data is collected in real time through automated means and can cover the entire process of passenger feedback from ticket purchase, check-in, boarding to flight and after-sales service, laying a solid foundation for in-depth analysis and service optimization.

[0077] In some embodiments, sensitive information involved in user feedback data is strictly de-identified to obtain de-identified data. De-identified data will ensure the privacy and security of passengers, while not affecting the accuracy and availability of subsequent analysis. Sensitive information may include, but is not limited to, mobile phone numbers, email addresses, bank card numbers, ID card numbers, passport numbers, ticket numbers, addresses, and sensitive keywords.

[0078] It should be further noted that the desensitization process will employ a combination of automated tools and manual review to ensure the accuracy and consistency of the desensitization. All desensitization operations will be logged for subsequent auditing and tracking.

[0079] This embodiment can protect passenger privacy while ensuring the integrity and availability of data, providing reliable support for subsequent data analysis and business decisions.

[0080] Specifically, after the desensitization process, examples of desensitized data are shown in Table 1 below:

[0081] Table 1 Examples of Desensitization

[0082]

[0083] Step 204: Perform text splitting on the de-identified data to obtain split text data.

[0084] Text splitting refers to accurately splitting the original text of anonymized data into multiple independent text fragments to obtain split text data, and then presenting the split text data in a structured JSON format.

[0085] In some embodiments, a text-splitting agent pre-trained on a large language model performs text splitting on de-identified data, and the large model is required to output the split text data in JSON format. JSON format has good structure and scalability, and can clearly express the semantic boundaries and content of each split text data fragment.

[0086] It should be noted that in the process of building the text segmentation agent, it is first necessary to conduct in-depth research on the structural characteristics and semantic features of the anonymized data. Then, the text segmentation agent is built, and refined prompt words are designed. Through multiple rounds of iteration and debugging, the text segmentation agent is guided to complete the text segmentation task, ensuring that the large model can accurately understand and execute it. In order to verify the effectiveness of the agent, multiple rounds of experiments and evaluations were conducted on the text segmentation agent. The experimental results show that the agent performs well in terms of both accuracy and completeness of text segmentation.

[0087] Table 2 Examples of Text Segmentation for "Voice of the Customer"

[0088]

[0089] Step 206: Classify the split text data to obtain scene classification data and sentiment classification data.

[0090] It should be noted that passenger feedback data may involve different service scenarios and different emotional states, such as ticket sales, check-in and seat selection, meals, lounges, and luggage. Classifying the split text data into service scenarios can enable more refined analysis of the demands of different service scenarios.

[0091] In some embodiments, data classification is performed on the split text data to obtain scene classification data and sentiment classification data, including: classifying the split text data into service scenarios according to a preset service scenario classification table and a preset service scenario agent to obtain scene classification data; and classifying the split text data into sentiments according to a preset sentiment classification agent to obtain sentiment classification data.

[0092] The service scenario classification table can be a list of air passenger service scenarios. By designing and debugging prompt words, the model service scenario agent is guided to understand the semantics of the passenger's text and classify it into a certain category of service scenarios in the aviation field. In this way, the semantic understanding capabilities of the large model are used to build an agent specifically for identifying different service scenarios in the aviation field.

[0093] Specifically, based on business practices and user needs analysis, 16 core service scenarios were identified. See Table 3 for a detailed classification of these service scenarios.

[0094] Table 3 List of Air Passenger Service Scenarios

[0095]

[0096] In some embodiments, after batch testing and verification to confirm that the service scenario agent can stably and accurately classify scenarios, the agent is encapsulated into an interface and integrated into the customer experience management platform to achieve automated real-time processing of passenger feedback text data, with an average processing time of 0.3 seconds per passenger feedback data.

[0097] In some embodiments, the sentiment classification data includes: sentiment tendency and emotion rating data; performing sentiment classification on the split text data according to a preset sentiment classification agent to obtain sentiment classification data includes: performing sentiment tendency classification on the split text data according to the sentiment classification agent to obtain sentiment tendency; and performing emotion rating on the split text data according to the sentiment classification agent to obtain emotion rating data.

[0098] Among them, sentiment can be divided into four dimensions: positive sentiment, neutral sentiment, negative sentiment, and mixed sentiment. It should be further explained that positive sentiment feedback text usually contains positive evaluations of airline services, facilities, or experiences, and has obvious positive characteristics. Neutral sentiment feedback is expressed as objective statements or constructive opinions, lacking obvious sentiment bias. Negative sentiment text contains clear dissatisfaction and criticism, often using negative sentiment words such as "delay," "unfriendly," and "terrible." Mixed sentiment feedback presents the complexity of sentiment bias, containing both positive and negative evaluations in the same text, or showing contradictory sentiment biases in different service dimensions.

[0099] In some embodiments, the sentiment classification agent categorizes the segmented text data into sentiment tendencies. Simultaneously, it grades the user emotions reflected in the segmented text data, obtaining sentiment grade data, which quantifies user emotions into floating-point values ​​between 0 and 10. It should be further noted that a smaller sentiment grade value indicates a worse travel experience and stronger negative emotions; conversely, a larger sentiment grade value indicates a better travel experience and weaker negative emotions.

[0100] In some embodiments, a sentiment classification agent specifically designed for the aviation field is constructed based on large model technology. This agent achieves efficient understanding of aviation terminology and context by combining a pre-trained large language model with domain adaptation technology. As shown in Table 4, this agent can accurately identify the sentiment tendencies of various passenger feedback contents and provide numerical values ​​for sentiment rating, providing accurate data insights for airline service optimization.

[0101] Table 4 Examples of Sentiment Classification

[0102]

[0103] Step 208: Extract the demands from the split text data to obtain the key demands, and then vectorize the key demands to obtain the demand vector.

[0104] Among them, demand extraction refers to efficiently extracting specific demands from split text data.

[0105] In some embodiments, extracting demands from split text data to obtain key demands includes: performing structured preprocessing on the split text data using a pre-defined demand mining agent to obtain structured text; performing semantic understanding on the structured text using the demand mining agent to obtain deep semantic data; and extracting demands from the deep semantic data using the demand point description model of the demand mining agent to obtain key demands.

[0106] In some embodiments, a request mining agent, pre-trained on a large model and specifically designed for mining passenger requests in the aviation field, first performs structured preprocessing on the input split text data, including word segmentation, part-of-speech tagging, and syntactic analysis, to obtain structured text and ensure its structured representation. Subsequently, the large model, through its multi-layered neural network architecture, performs deep semantic understanding on the structured text to obtain deep semantic data. Its core is a Transformer-based model, and the core formula is the Transformer encoder's self-attention mechanism, as shown in the following formula:

[0107]

[0108] in, , , These represent the query, key, and value matrices, respectively. It is the dimension of the key vector. Through the multi-head attention mechanism, the model can capture multiple semantic relationships in the text, and can comprehensively understand the meaning of the sentence from multiple perspectives, capturing rich semantic relationships.

[0109] In some embodiments, after obtaining deep semantic data of the text through the Transformer encoder, the appeal point description model of the appeal mining agent is used to refine the appeal from the deep semantic data. The model further identifies key entities and potential appeal points in the text based on the deep semantic data, and then generates a concise appeal point description based on the Transformer decoder, thereby obtaining the key appeal points. The core formula of this step is as follows:

[0110]

[0111] in, It is the input text. It is a description of the generated appeal points. It is the generated first One word, It is the previously generated word sequence.

[0112] In some embodiments, if the appeal mining agent is unable to extract a clear appeal from the text, the appeal mining agent will return "Unable to extract passenger appeal" to ensure the reliability of key appeal points.

[0113] In some embodiments, it should be noted that, in order to perform systematic clustering analysis on key appeals, these textual key appeals first need to be converted into machine-processable numerical representations. A pre-trained large-scale language model from the BERT series is used as a feature extractor to map each key appeal into a high-dimensional semantic space. Specifically, each appeal is transformed into a 1024-dimensional dense vector (retaining 3 decimal places), resulting in an appeal vector. This vector representation can effectively capture the semantic information and contextual relationships of the text, laying the foundation for subsequent clustering analysis. For the vectorization task of Chinese short sentences, the BERT series models are a better choice because they can capture the bidirectional semantic relationships of Chinese short sentences well and have high computational efficiency.

[0114] It's worth further explaining that during vectorization, we input the key appeal points into a pre-trained large language model and obtain the output of its last hidden layer as the vector representation of the text. This representation method not only considers the surface features of words but also captures the dependencies and contextual information between words through a self-attention mechanism. The self-attention mechanism can dynamically allocate weights, enabling the model to focus on the most important semantic information in the text, thus more accurately reflecting the semantic connotation of the appeal points. For example, for the appeal point "passengers need airlines to improve the quality and service of business class meals," the model can identify keywords such as "business class" and "meal quality" and highlight these semantic features in the vector representation.

[0115] Specifically, Table 5 shows examples of the claims and their corresponding 1024-dimensional vector representations. Each vector is an array of floating-point values ​​ranging from -1 to 1, representing the specific location of the claim in the semantic space. For example, the vector corresponding to the claim "passengers need airlines to improve the quality and service of business class meals" may have high values ​​in dimensions related to "business class" and "meal quality," while the vector corresponding to "passengers expect flexible meal cancellation services" may show significant features in dimensions related to "meal cancellation" and "flexibility." This vectorized representation can not only capture the core semantics of the claims but also reflect the similarities and differences between the claims and other claims.

[0116] Table 5. Examples of Vectorization of Claim Point Embedding

[0117]

[0118] In this embodiment, through this vectorized representation, we can further utilize small-model clustering algorithms to group the requests, thereby identifying clusters of requests with similar semantics. Therefore, by converting textual requests into vector representations and combining them with subsequent clustering algorithms for analysis, we can systematically identify and classify passenger requests, providing a scientific basis for subsequent decision-making.

[0119] Step 210: Cluster the request vectors according to the preset clustering model and scene classification data to obtain the request clustering results.

[0120] The preset clustering model can be the HDBSCAN (Hierarchical Density-Based Spatial Clustering of Applications with Noise) clustering model. It should be noted that HDBSCAN is a density-based hierarchical clustering algorithm that combines the density clustering idea of ​​DBSCAN with the hierarchical structure of AHC (Agglomerative Hierarchical Clustering), and introduces the concept of stability, thus exhibiting greater flexibility and robustness when handling clusters with different densities and shapes.

[0121] It is worth noting that a significant advantage of HDBSCAN is its ability to automatically identify and label outliers (i.e., data points that do not belong to any cluster). This feature is particularly important in airline passenger complaint analysis scenarios, as it can effectively filter out anomalous complaints (outliers) from individual passengers, thus focusing more on the complaints that passengers are generally concerned about. Unlike many traditional clustering algorithms, HDBSCAN does not require pre-specifying the number of clusters; it can automatically determine the appropriate number of clusters based on the density distribution of the data. This feature makes it particularly suitable for scenarios where the number of natural clusters in the data cannot be determined in advance, which perfectly meets the needs of airline passenger complaint analysis.

[0122] In some embodiments, the appeal vectors are clustered according to a preset clustering model and scenario classification data to obtain appeal clustering results, including: calculating the core distance of the appeal vectors according to the clustering model to obtain the core distance of the appeals; constructing a minimum spanning tree according to the core distance of the appeals; pruning the minimum spanning tree according to the clustering model to obtain a hierarchical cluster structure; and performing stability analysis on the hierarchical cluster structure to obtain the appeal clustering results.

[0123] Specifically, HDBSCAN identifies cluster structures in data by constructing a hierarchical density tree. The algorithm first calculates the core distance for each data point, then constructs a minimum spanning tree based on these core distances. Next, HDBSCAN transforms the minimum spanning tree into a hierarchical cluster structure through pruning operations. Finally, it determines the optimal cluster partition through stability analysis. This process can not only automatically identify the number of clusters, but also effectively handle noisy data, ensuring the accuracy and reliability of clustering results.

[0124] In some embodiments, HDBSCAN is applied in airline passenger demand analysis to enable us to cluster similar demands in the same service scenario into the same category, obtain demand clustering results, and thus more clearly identify the main concerns and needs of passengers.

[0125] Step 212: Summarize the demands based on the clustering results to obtain the demand summary results.

[0126] The "demand summary" refers to generating a concise summary with corresponding explanations by semantically analyzing and aggregating multiple similar demands from passengers.

[0127] In some embodiments, a special agent for request summarization is created using a big oracle model. This agent performs deep content summarization on texts with the same clustering labels in the request clustering results. Specifically, the request summarization agent generates a concise summary with corresponding explanations by performing semantic analysis and aggregation on multiple similar requests from passengers, in order to better understand the core needs of passengers.

[0128] In some embodiments, as shown in Table 6, the request summarizing agent summarizes the content of 9 similar request points, concisely summarizing the passenger's request as "the passenger needs to successfully book various special meals within 24 hours before departure", and adds an explanation.

[0129] Table 6 Examples of Summary of Request Points

[0130]

[0131] Step 214: Based on the summary of demands, perform demand analysis on the emotional classification data and the demand clustering results to obtain the demand analysis results.

[0132] In some embodiments, in-depth statistical analysis is performed on the emotional classification data and the emotional clustering results based on the summary of the demands, and the distribution of passengers' emotional tendencies at different times is continuously monitored. Combined with the demands analysis results, the focus is on the feedback of passengers with strong negative emotions (i.e. passengers with very small emotional rating values ​​in the emotional classification data) to obtain demands analysis results. The feedback of these demands analysis results often reveals more serious or urgent service problems, which may require early warning and priority handling to prevent the situation from escalating.

[0133] In the above-mentioned appeal analysis method, user feedback data is obtained and anonymized to obtain anonymized data; the anonymized data is then split into text data; the split text data is classified into scenario classification data and sentiment classification data; the appeals are extracted from the split text data to obtain key appeal points, which are then vectorized to obtain appeal vectors; the appeal vectors are clustered according to a preset clustering model and scenario classification data to obtain appeal clustering results; the appeal clustering results are summarized to obtain appeal summary results; and appeal analysis is performed on the sentiment classification data and appeal clustering results based on the appeal summary results to obtain appeal analysis results. Therefore, by splitting the anonymized data into text, and then classifying the split text data into scenario-based and sentiment-based data, and extracting key user requests from the split text data, we can achieve more refined and context-based analysis of user requests by utilizing a large language model for text splitting, data classification, and request extraction. Furthermore, by using a pre-defined clustering model—specifically, using a small model and scenario-based data—to cluster the request vectors, we can obtain request clustering results. Finally, we can summarize the requests based on these clustering results, and then analyze the sentiment-based data and clustering results to obtain request analysis results. This approach better leverages the synergy between the large language model and the small-scale clustering model, accurately identifying core user requests, thereby optimizing resource allocation and improving user experience.

[0134] In one exemplary embodiment, such as Figure 3 As shown, this demand analysis method also includes steps 302 to 304. Wherein:

[0135] Step 302: Sort the requests from largest to smallest based on the number of requests received according to the request analysis results to obtain the request ranking results.

[0136] In some embodiments, such as Figure 4 As shown, further statistical analysis of the appeal analysis results is performed, and the appeal points are arranged from largest to smallest in terms of the number of responses, thus presenting the appeal analysis results more clearly and accurately.

[0137] Step 304: Construct a demand analysis ranking chart based on the demand ranking results, and display the demand analysis ranking chart.

[0138] In some embodiments, a demand analysis ranking chart is constructed based on the demand ranking results, and the demand analysis ranking chart is displayed. Specifically, for example... Figure 4 As shown, taking the meal scenario in August as an example, after removing text from which no request could be extracted and outliers, the data is sorted in descending order of request percentage. The analysis results show that passengers' primary request regarding meals is: "Passengers must successfully book various special meals within 24 hours before departure," including vegetarian meals, halal meals, children's meals, and infant meals. This finding indicates that airlines have room for improvement in managing the time limits for special meal bookings.

[0139] In some embodiments, all steps of this solution are integrated into an automated process, visually displaying passengers' core needs. Business departments can clearly identify passenger needs that should be prioritized for resource allocation in the near future, thereby achieving more precise service optimization and more efficient resource allocation. For example, in the aforementioned meal scenario, the focus of service optimization should be on optimizing the time limit for special meal reservations, and airlines need to provide more flexible services to meet different dietary needs and religious customs.

[0140] It should be noted that passenger needs change dynamically over time in different service scenarios. Therefore, optimizing air travel services is a continuous iterative process. By conducting automated demand analysis on the latest passenger feedback daily, a closed-loop link of "feedback → intelligent analysis → optimization → feedback" is constructed to achieve timely response to demands and precise improvement of services. Therefore, this solution makes this process automated, intelligent, efficient, and timely, and can continuously drive the improvement of service quality.

[0141] In some embodiments, for a clearer understanding of the claims process of this application, please refer to [link to relevant documentation]. Figure 5 To gain a more concrete understanding of the complete process of demand analysis.

[0142] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0143] Based on the same inventive concept, this application also provides a claims analysis apparatus for implementing the claims analysis method described above. The solution provided by this apparatus is similar to the solution described in the above method; therefore, the specific limitations in one or more claims analysis apparatus embodiments provided below can be found in the limitations of the claims analysis method described above, and will not be repeated here.

[0144] In one exemplary embodiment, such as Figure 6 As shown, a request analysis device is provided, including: an acquisition module 602, a text segmentation module 604, a data classification module 606, a request extraction module 608, a request clustering module 610, a request summarization module 612, and a request analysis module 614, wherein:

[0145] The acquisition module 602 is used to acquire user feedback data and perform data anonymization on the user feedback data to obtain anonymized data;

[0146] The text splitting module 604 is used to split the de-identified data into split text data.

[0147] The data classification module 606 is used to classify the split text data to obtain scene classification data and sentiment classification data.

[0148] The appeal extraction module 608 is used to extract appeals from the split text data, obtain key appeal points, and vectorize the key appeal points to obtain appeal vectors.

[0149] The appeal clustering module 610 is used to cluster appeal vectors according to a preset clustering model and scenario classification data to obtain appeal clustering results;

[0150] The demand summarization module 612 is used to summarize the demands from the demand clustering results to obtain the demand summary results; the demand analysis module 614 is used to analyze the sentiment classification data and the demand clustering results based on the demand summary results to obtain the demand analysis results.

[0151] In some embodiments, the data classification module 606 is further configured to classify the split text data according to a preset service scenario classification table and a preset service scenario agent to obtain scenario classification data; and to classify the split text data according to a preset sentiment classification agent to obtain sentiment classification data.

[0152] In some embodiments, the sentiment classification data includes: sentiment tendency and emotion rating data; the data classification module 606 is further configured to classify the split text data into sentiment tendencies according to the sentiment classification agent to obtain sentiment tendencies; and to classify the split text data into emotions according to the sentiment classification agent to obtain emotion rating data.

[0153] In some embodiments, the appeal extraction module 608 is further configured to perform structured preprocessing on the split text data through a preset appeal mining agent to obtain structured text; perform semantic understanding on the structured text according to the appeal mining agent to obtain deep semantic data; and extract appeals from the deep semantic data according to the appeal point description model of the appeal mining agent to obtain key appeal points.

[0154] In some embodiments, the appeal clustering module 610 is further configured to calculate the core distance of the appeal vectors according to the clustering model to obtain the core distance of the appeals; construct a minimum spanning tree according to the core distance of the appeals; perform pruning operations on the minimum spanning tree according to the clustering model to obtain a hierarchical cluster structure; and perform stability analysis on the hierarchical cluster structure to obtain the appeal clustering result.

[0155] In some embodiments, the appeal summarization module 612 is further configured to perform semantic analysis on the text content with the same clustering label in the appeal clustering results according to the preset appeal summarization intelligent agent to obtain appeal summarization points; and generate appeal summarization annotations by annotating the appeal summarization points according to the preset appeal summarization intelligent agent; the appeal summarization results include appeal summarization points and appeal summarization annotations.

[0156] In the aforementioned appeal analysis device, user feedback data is acquired and anonymized to obtain anonymized data; the anonymized data is then split into text data; the split text data is categorized to obtain scenario-based and sentiment-based data; appeals are extracted from the split text data to obtain key appeal points, which are then vectorized to obtain appeal vectors; appeal vectors are clustered based on a preset clustering model and scenario-based data to obtain appeal clustering results; appeal clustering results are summarized to obtain appeal summary results; and appeal analysis is performed on the sentiment-based data and appeal clustering results based on the appeal summary results to obtain appeal analysis results. Therefore, by splitting the anonymized data into text, and then classifying the split text data into scenario-based and sentiment-based data, and extracting key user requests from the split text data, we can achieve more refined and context-based analysis of user requests by utilizing a large language model for text splitting, data classification, and request extraction. Furthermore, by using a pre-defined clustering model—specifically, using a small model and scenario-based data—to cluster the request vectors, we can obtain request clustering results. Finally, we can summarize the requests based on these clustering results, and then analyze the sentiment-based data and clustering results to obtain request analysis results. This approach better leverages the synergy between the large language model and the small-scale clustering model, accurately identifying core user requests, thereby optimizing resource allocation and improving user experience.

[0157] Each module in the aforementioned claims analysis device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the memory of a computer device as software, so that the processor can call and execute the corresponding operations of each module.

[0158] In one exemplary embodiment, a computer device is provided, which may be a server, and its internal structure diagram may be as follows: Figure 7As shown, this computer device includes a processor, memory, input / output (I / O) interfaces, and a communication interface. The processor, memory, and I / O interfaces are connected via a system bus, and the communication interface is also connected to the system bus via the I / O interfaces. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operating system and computer programs stored in the non-volatile storage media. The database stores user feedback data. The I / O interfaces are used for exchanging information between the processor and external devices. The communication interface is used for communicating with external terminals via a network connection. When the computer program is executed by the processor, it implements a request analysis method.

[0159] Those skilled in the art will understand that Figure 7 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0160] In one exemplary embodiment, a computer device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the above-described claims analysis method.

[0161] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored, which, when executed by a processor, implements the steps of the above-described claims analysis method.

[0162] In one embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps of the above-described claims analysis method.

[0163] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of the relevant data must comply with relevant regulations.

[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, artificial intelligence (AI) processors, etc., and are not limited to these.

[0165] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this application.

[0166] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for analyzing demands, characterized in that, The method includes: Obtain user feedback data and perform data anonymization on the user feedback data to obtain anonymized data; The de-identified data is then split into text data to obtain split text data. The split text data is classified to obtain scene classification data and sentiment classification data; The split text data is processed to extract the key appeal points, and these key appeal points are then vectorized to obtain appeal vectors. The request vector is clustered according to the preset clustering model and scenario classification data to obtain the request clustering result; The clustering results of the demands are used to summarize the demands, resulting in a summary of the demands. Based on the summary of the demands, the emotional classification data and the clustering results of the demands are analyzed to obtain the demand analysis results.

2. The method according to claim 1, characterized in that, The process of classifying the split text data to obtain scene classification data and sentiment classification data includes: The split text data is classified into service scenarios according to a preset service scenario classification table and a preset service scenario intelligent agent to obtain the scenario classification data; The split text data is classified according to the preset sentiment classification agent to obtain the sentiment classification data.

3. The method according to claim 2, characterized in that, The sentiment classification data includes: sentiment tendency and emotion level data; the process of classifying the split text data according to a preset sentiment classification agent to obtain the sentiment classification data includes: The sentiment classification agent performs sentiment tendency classification on the split text data to obtain the sentiment tendency. The emotion classification agent performs emotion classification on the split text data to obtain the emotion classification data.

4. The method according to claim 1, characterized in that, The process of extracting the demands from the split text data yields key demands, including: The structured text is obtained by preprocessing the split text data using a pre-defined demand mining agent. Based on the stated requirements, the intelligent agent performs semantic understanding on the structured text to obtain deep semantic data; The key appeal points are obtained by refining the deep semantic data based on the appeal point description model of the appeal mining agent.

5. The method according to claim 1, characterized in that, The step of clustering the request vector according to a preset clustering model and scene classification data to obtain the request clustering result includes: The core distance of the claims is calculated by performing a core distance calculation on the claim vectors based on the clustering model; Construct a minimum spanning tree based on the core distance of the stated requirements; The minimum spanning tree is pruned according to the clustering model to obtain a hierarchical cluster structure. Stability analysis was performed on the hierarchical cluster structure to obtain the clustering results of the claims.

6. The method according to claim 1, characterized in that, The process of summarizing the demands based on the clustering results to obtain the demand summary results includes: Based on the preset demand summarizing agent, semantic analysis is performed on the text content with the same clustering label in the demand clustering results to obtain demand summarizing points; The request summary points are annotated and generated according to the preset request summary agent to obtain request summary annotations; the request summary result includes the request summary points and the request summary annotations.

7. A demand analysis device, characterized in that, The device includes: The acquisition module is used to acquire user feedback data and perform data anonymization on the user feedback data to obtain anonymized data; The text splitting module is used to split the de-identified data into split text data. The data classification module is used to classify the split text data to obtain scene classification data and sentiment classification data; The appeal extraction module is used to extract appeals from the split text data, obtain key appeal points, and vectorize the key appeal points to obtain appeal vectors. The appeal clustering module is used to cluster the appeal vector according to a preset clustering model and scenario classification data to obtain the appeal clustering result; The demand summarization module is used to summarize the demands clustering results to obtain the demand summary results. The appeal analysis module is used to perform appeal analysis on the sentiment classification data and the appeal clustering results based on the appeal summary results, and obtain appeal analysis results.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 6.

Images