A Customer VOC Intelligent Response Method and System Based on an E-commerce Platform

Abstract

The present invention provides a method and system for intelligent response to customer VOC based on an e-commerce platform, and relates to the technical field of customer service of an e-commerce platform. The method comprises the steps of: recording the language text input by the customer and the initial response text fed back to the customer by the intelligent customer service in response to the language text; receiving the subsequent input text from the customer after receiving the initial response text; judging whether there is an understanding barrier based on the subsequent input text, and if so, generating an understanding barrier signal, and marking the initial response text as a text to be adjusted; adjusting the text to be adjusted based on the understanding barrier signal, generating an adjusted response text; and sending the adjusted response text to the corresponding customer. The method of the present invention overcomes the problem that elderly customers have difficulty understanding the content of standardized intelligent customer service responses, can effectively identify the understanding barriers of elderly customers, and automatically adjust the response method of the intelligent customer service to improve communication efficiency and user satisfaction.

Description

Technical Field

[0001] The present invention relates to the technical field of customer service of e-commerce platforms, and in particular to a method and system for intelligently responding to customer voice-of-content (VOC) responses based on e-commerce platforms. Background Art

[0002] To improve user experience and operational efficiency, e-commerce platforms have generally built intelligent customer service systems designed to receive and quickly respond to customer inquiries, suggestions, or complaints. These systems typically leverage advanced natural language processing (NLP) technology to parse customer input, identify their true intent, extract key information, and generate and send appropriate responses based on pre-defined knowledge bases or generative models. The effectiveness of this process is highly dependent on the system's accurate understanding of customer language and its ability to provide appropriate, timely, and clear responses.

[0003] However, when intelligent customer service targets elderly customers, existing systems face significant challenges. Due to factors such as age and cognitive habits, elderly customers are often unfamiliar with and confused by the professional terms and internet buzzwords widely used on e-commerce platforms. For example, terms like "SKU," "coupon," "flash sale," "group buying," and "live streaming" may be completely incomprehensible to elderly customers who are unfamiliar with e-commerce environments. When intelligent customer service systems include these terms in responses, elderly customers struggle to grasp the core message conveyed by the response, resulting in a disruption in information flow.

[0004] Due to difficulties in understanding the responses generated by intelligent customer service, elderly customers may exhibit specific behavioral patterns during communication. They may repeat the same questions as before or try to rephrase the original questions using language that is more familiar to them and different from the standard expression.

[0005] Existing intelligent systems often lack the ability to discern the user's understanding of these subsequent inputs. The system may mistake repeated or rephrased questions for new inquiries, generating responses that are similar to the previous ones but still contain difficult terminology. Alternatively, due to misunderstanding the rephrased questions, the system may generate responses that are not fully relevant to the customer's actual needs. This failure to recognize and adapt to user difficulties in understanding can easily lead to conversations becoming entangled in a "daffodil" situation, deviating from the goal of problem-solving and creating a cycle of ineffective communication.

[0006] Existing intelligent customer service systems typically respond using standardized language and terminology tailored to average users, failing to dynamically adapt to the cognitive characteristics and comprehension levels of elderly customers. The systems lack a mechanism to detect in real time or near real time when elderly customers fail to understand previous responses and adapt subsequent communication strategies and content accordingly. This rigid response model leaves elderly customers unable to obtain clear and understandable solutions even after repeated attempts to communicate with intelligent customer service representatives, effectively preventing their issues from being resolved.

[0007] Prolonged, ineffective communication not only wastes elderly customers' time and energy, but also leaves them feeling confused, frustrated, and even helpless. This can lead to serious doubts about the effectiveness, intelligence, and reliability of intelligent customer service systems, significantly reducing their trust in them. This decline in trust can cause elderly customers to abandon the convenient intelligent customer service channel and instead seek assistance from human agents, increasing the workload of the human customer service team. More importantly, failure to resolve issues promptly and effectively will directly impact elderly customers' shopping experience on e-commerce platforms, potentially leading to abandonment of purchases or even churn, which is detrimental to the long-term development of e-commerce platforms and the expansion of their user base. Summary of the Invention

[0008] The purpose of the present invention is to provide a customer VOC intelligent response method and system based on an e-commerce platform, which overcomes the problem that elderly customers have difficulty understanding the content of standardized intelligent customer service responses. It can effectively identify the comprehension barriers of elderly customers and automatically adjust the response method of intelligent customer service to improve communication efficiency and user satisfaction.

[0009] In a first aspect, the present invention provides a method for intelligently responding to customer VOCs based on an e-commerce platform, comprising the following steps:

[0010] S1. Record the language text entered by the customer and the initial response text of the intelligent customer service feedback to the customer for the language text;

[0011] S2 receives the customer's subsequent input text after receiving the initial response text;

[0012] S3. Determine whether there is a comprehension barrier based on the subsequent input text. If so, generate a comprehension barrier signal and mark the initial response text as a text to be adjusted. If not, input the subsequent input text as a new language text to the intelligent customer service and return to step S1.

[0013] S4. Adjust the text to be adjusted according to the comprehension barrier signal to generate an adjusted response text;

[0014] S5. Send the adjusted response text to the corresponding customer.

[0015] The customer VOC intelligent response method based on the e-commerce platform provided by the present invention analyzes the subsequent input of elderly customers after receiving the response of intelligent customer service, identifies their possible signs of comprehension difficulties, and adjusts the response content expression of intelligent customer service accordingly.

[0016] Furthermore, between step S2 and step S3, the following steps are also included:

[0017] S6. Determine the customer's current consultation intention based on the subsequent input text, specifically including the steps of:

[0018] S61. Utilize a preset intent recognition model to analyze the subsequent input text to determine the client's current consultation intent. This intent recognition model is based on historical elderly client conversation data and is fine-tuned using the BERT model. The model is capable of recognizing commonly used consultation intents among elderly clients and outputting a confidence score for the intent.

[0019] S62. Compare the current consultation intention with the historical consultation intention to determine whether the two are consistent; the historical consultation intention is obtained by analyzing the language text by the intention recognition model, and the historical consultation intention has a corresponding intention confidence;

[0020] S63. If the current consultation intention is inconsistent with the historical consultation intention, and the intention confidence of the current consultation intention is greater than the preset threshold and greater than the intention confidence of the historical consultation intention, it is determined that intention drift has occurred, and the subsequent input text is input as a new language text into the intelligent customer service, and then return to execute step S1, otherwise execute step S3.

[0021] Furthermore, the specific steps in step S61 include:

[0022] Constructing a product knowledge graph containing information about multiple products; the product information includes multiple product attributes of various products;

[0023] Extract all keywords from the subsequent input text by parsing the subsequent input text, and map each keyword to a corresponding product attribute vector based on the product knowledge graph;

[0024] Inputting all the product attribute vectors into the intention recognition model respectively to obtain multiple intention categories and corresponding intention confidences;

[0025] The intention category with the highest confidence is selected as the customer's current consultation intention.

[0026] Furthermore, the specific steps in step S3 include:

[0027] S31 extracts question words, negative words, and repeated words from the subsequent input text and constructs them into a text feature vector;

[0028] S32. Input the text feature vector into a preset comprehension barrier detection model to obtain a comprehension barrier probability; the comprehension barrier detection model is based on historical elderly customer conversation data and is trained using a support vector machine (SVM) algorithm to identify common expressions of elderly customers, such as questions, denials, and repetitions;

[0029] S33. Determine whether the probability of the comprehension barrier is greater than a preset probability threshold. If so, determine that there is an comprehension barrier, generate an comprehension barrier signal, and mark the initial response text as text to be adjusted; otherwise, determine that there is no comprehension barrier, input the subsequent input text as a new language text into the intelligent customer service, and return to execute step S1.

[0030] Furthermore, the specific steps in step S31 include:

[0031] Constructing a stop word list; the stop word list includes common modal particles, auxiliary words and meaningless words;

[0032] Cleaning the subsequent input text using the stop word list to remove stop words in the subsequent input text to obtain a cleaned subsequent input text;

[0033] Using a word segmentation algorithm to perform word segmentation processing on the subsequent input text after the cleaning to obtain multiple text segmentations;

[0034] For each of the text segmentation words, determine whether it belongs to a preset question word library, a negative word library or a repeated word library; if it belongs to the question word library, mark the text segmentation word as a question word; if it belongs to the negative word library, mark the text segmentation word as a negative word; if it belongs to the repeated word library, mark the text segmentation word as a repeated word;

[0035] Counting the number of question words, negative words, and repeated words in the subsequent input text, which are recorded as the number of question words, the number of negative words, and the number of repeated words, respectively;

[0036] The text feature vector is constructed according to the number of question words, the number of negative words, and the number of repeated words.

[0037] Furthermore, the specific steps in step S4 include:

[0038] S41. Determining an adjustment method for the text to be adjusted based on the comprehension barrier signal; the adjustment method includes terminology simplification and sentence simplification. The terminology simplification is used when the comprehension barrier signal indicates that the customer has difficulty understanding professional terminology, and the sentence simplification is used when the comprehension barrier signal indicates that the customer has difficulty understanding long sentences.

[0039] If it is determined that the adjustment direction is only term simplification, a preset elderly-friendly vocabulary library is consulted and the semantic similarity between each standard term in the text to be adjusted and the vocabulary in the elderly-friendly vocabulary library is calculated. Vocabulary with a semantic similarity higher than a preset value is selected as a replacement vocabulary, and the priority of the replacement vocabulary is adjusted according to the weight value, with the replacement vocabulary with the highest weight value being preferentially used to generate the adjusted response text; each vocabulary in the elderly-friendly vocabulary library is assigned a weight value, which represents the frequency of use of the vocabulary among the elderly customer group;

[0040] S43. If the adjustment direction is determined to be sentence simplification only, the long sentence in the text to be adjusted is split into multiple short sentences and the word order is adjusted according to the preset simplification rule set to generate the adjusted response text;

[0041] S44. If the adjustment direction is determined to be terminology simplification and sentence simplification, the standard terms in the text to be adjusted are replaced with simplified expressions by referring to the elderly customer-friendly vocabulary library, and the sentence and structure of the text to be adjusted after the replacement of the terms are adjusted by applying the simplification rule set to generate an adjusted response text.

[0042] In a second aspect, the present invention provides a customer VOC intelligent response system based on an e-commerce platform, comprising:

[0043] A recording module is used to record the language text input by the customer and the initial response text fed back to the customer by the intelligent customer service based on the language text;

[0044] A receiving module, configured to receive subsequent input text from the client after receiving the initial response text;

[0045] a first judgment module, configured to determine whether there is an understanding barrier based on the subsequent input text; if so, generate an understanding barrier signal and mark the initial response text as a text to be adjusted; if not, input the subsequent input text as a new language text into the intelligent customer service and return to run the recording module;

[0046] a generating module, configured to adjust the text to be adjusted according to the comprehension impairment signal and generate an adjusted response text;

[0047] The sending module is used to send the adjusted response text to the corresponding customer.

[0048] Furthermore, the system further includes a second judgment module, which is configured to judge the customer's current consultation intention based on the subsequent input text, and specifically performs the following steps:

[0049] S61. Utilize a preset intent recognition model to analyze the subsequent input text to determine the client's current consultation intent. This intent recognition model is based on historical elderly client conversation data and is fine-tuned using the BERT model. The model is capable of recognizing commonly used consultation intents among elderly clients and outputting a confidence score for the intent.

[0050] S62. Compare the current consultation intention with the historical consultation intention to determine whether the two are consistent; the historical consultation intention is obtained by analyzing the language text by the intention recognition model, and the historical consultation intention has a corresponding intention confidence;

[0051] S63. If the current consultation intention is inconsistent with the historical consultation intention, and the intention confidence of the current consultation intention is greater than the preset threshold and greater than the intention confidence of the historical consultation intention, it is determined that intention drift has occurred, and the subsequent input text is input as a new language text into the intelligent customer service, and then the recording module is returned to run, otherwise the first judgment module is run.

[0052] Furthermore, the second judgment module is executed when it is used to obtain the customer's current consultation intention by analyzing the subsequent input text using a preset intention recognition model:

[0053] Constructing a product knowledge graph containing information about multiple products; the product information includes multiple product attributes of various products;

[0054] Extract all keywords from the subsequent input text by parsing the subsequent input text, and map each keyword to a corresponding product attribute vector based on the product knowledge graph;

[0055] Inputting all the product attribute vectors into the intention recognition model respectively to obtain multiple intention categories and corresponding intention confidences;

[0056] The intention category with the highest confidence is selected as the customer's current consultation intention.

[0057] Furthermore, the first judgment module is configured to determine whether there is an understanding barrier based on the subsequent input text. If so, it generates an understanding barrier signal and marks the initial response text as a text to be adjusted. If not, it inputs the subsequent input text as a new language text into the intelligent customer service and returns to the execution of the recording module:

[0058] S31 extracts question words, negative words, and repeated words from the subsequent input text and constructs them into a text feature vector;

[0059] S32. Input the text feature vector into a preset comprehension barrier detection model to obtain a comprehension barrier probability; the comprehension barrier detection model is based on historical elderly customer conversation data and is trained using a support vector machine (SVM) algorithm to identify common expressions of elderly customers, such as questions, denials, and repetitions;

[0060] S33. Determine whether the probability of the comprehension barrier is greater than a preset probability threshold. If so, determine that there is an comprehension barrier, generate an comprehension barrier signal, and mark the initial response text as text to be adjusted. Otherwise, determine that there is no comprehension barrier, input the subsequent input text as a new language text into the intelligent customer service, and return to run the recording module.

[0061] As can be seen from the above, the customer VOC intelligent response method based on the e-commerce platform provided by the present invention can proactively and intelligently identify whether elderly customers have comprehension difficulties by analyzing their subsequent conversational behavior after receiving a response from intelligent customer service. By dynamically adjusting the response content based on an elderly-friendly vocabulary library and a set of response simplification rules, standardized complex expressions are converted into simplified expressions that are easier for elderly customers to understand. This effectively avoids ineffective repetitive communication and conversation deviations caused by language barriers, improves the accuracy and efficiency of intelligent customer service information transmission, significantly improves the experience of elderly customers using intelligent customer service, enhances their trust in intelligent customer service, and indirectly reduces the pressure on manual customer service.

[0062] Other features and advantages of the present invention will be described in the following description, and in part will become apparent from the description, or will be understood by practicing the embodiments of the present invention. The purposes and other advantages of the present invention can be realized and obtained by the structures particularly pointed out in the written description and the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 A flowchart of a method for intelligently responding to customer VOCs based on an e-commerce platform provided in an embodiment of the present invention.

[0064] Figure 2 A schematic diagram of the structure of a customer VOC intelligent response system based on an e-commerce platform provided in an embodiment of the present invention.

[0065] Description of labels:

[0066] 100, recording module; 200, receiving module; 300, first judgment module; 400, generating module; 500, sending module; 600, second judgment module. DETAILED DESCRIPTION

[0067] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings herein can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work are within the scope of protection of the present invention.

[0068] It should be noted that similar reference numerals and letters represent similar items in the following drawings. Therefore, once an item is defined in one drawing, it does not need to be further defined or explained in subsequent drawings. At the same time, in the description of the present invention, the terms "first", "second", etc. are used only to distinguish the description and should not be understood as indicating or implying relative importance.

[0069] Reference Attachment Figure 1 The present invention provides a customer VOC intelligent response method based on an e-commerce platform, comprising the following steps:

[0070] S1. Record the language text entered by the customer and the initial response text fed back to the customer by the intelligent customer service based on the language text;

[0071] S2. Receive subsequent input text from the customer after receiving the initial response text;

[0072] S3. Based on the subsequent input text, determine whether there is a comprehension barrier. If so, generate a comprehension barrier signal and mark the initial response text as a text to be adjusted. If not, input the subsequent input text as the new language text to the intelligent customer service and return to step S1.

[0073] S4. Adjusting the text to be adjusted based on the comprehension barrier signal and generating an adjusted response text;

[0074] S5. Send the adjusted response text to the corresponding customer.

[0075] Step S1 captures the interaction between the customer and the intelligent customer service, including the customer's original question and the intelligent customer service's response. These records provide basic data for subsequent analysis.

[0076] Step S2 is responsible for obtaining further input from the customer after seeing the intelligent customer service response. This subsequent input is the key source of information for the system to determine whether the customer understands the previous response.

[0077] Step S3 is the core step in understanding barrier detection. It analyzes the customer's understanding status based on subsequent input. If subsequent input indicates that the customer did not understand the previous response, such as repeating a question or expressing confusion, the system generates a comprehension barrier signal and identifies the initial response text that caused the customer confusion, preparing for modification. If subsequent input indicates that the customer understands the previous response or raises a new question, the system treats this subsequent input as a new starting point for the conversation and passes it to the intelligent customer service representative for processing. The conversation then returns to step S1 and begins a new round of interaction.

[0078] Step S4, initiated upon receiving the comprehension barrier signal, modifies the content or format of the identified text to be adjusted, generating an adjusted response that is more easily understood by the customer. Step S5 sends the adjusted, clearer response to the customer, aiming to resolve the previous communication barrier and enable the customer to understand and continue the conversation.

[0079] Specifically, this method incorporates a step to assess the customer's understanding level and, based on this assessment, triggers adjustments to the response text, creating a closed feedback loop. First, the intelligent customer service system receives the customer's spoken text and generates an initial response text. These interactions are recorded. Next, the system receives subsequent input from the customer after receiving the initial response. Based on this subsequent input, the system analyzes whether the customer has difficulty understanding the initial response. If a comprehension barrier is detected, the system generates a comprehension barrier signal and marks the initial response text as a text to be adjusted. Based on this comprehension barrier signal, the system modifies the text to be adjusted and generates an adjusted response text. Finally, the adjusted response text is sent to the customer. If no comprehension barrier is detected, the subsequent input text is input as the new spoken text to the intelligent customer service system, and the system returns to the step of recording the initial interaction, where the intelligent customer service system continues the conversation normally. This mechanism enables the intelligent customer service system to dynamically adapt to the customer's comprehension level. In particular, when a customer fails to understand the standard response, the system can promptly adjust the communication strategy to provide more understandable information. This effectively addresses the problem of ineffective communication caused by response text comprehension barriers and improves communication efficiency.

[0080] In some specific embodiments, a customer enters the text "How do I buy this phone?" and the intelligent customer service responds with an initial response: "Please click the 'Buy Now' button on the product page, select the SKU and quantity, and then submit the order." The system records this interaction. After receiving the initial response, the customer enters a subsequent text: "What is the SKU? Where is the button?" The system receives this subsequent text. Based on analysis of the subsequent text, the system detects that the customer has questions about "SKU" and "button," determines that there is a comprehension barrier, generates a comprehension barrier signal, and marks the initial response as requiring adjustment. Based on the comprehension barrier signal, the system adjusts the required text. For example, "Select the SKU and quantity" can be adjusted to "Select the product model and purchase quantity," and "Click the 'Buy Now' button on the product page" can be adjusted to "Find the 'Buy Now' button on the product page and tap it." The adjusted response is generated: "Please find the 'Buy Now' button on the product page, tap it, then select the product model and purchase quantity, and then submit the order." The system then sends the adjusted response to the customer.

[0081] In some embodiments, the method further includes between step S2 and step S3:

[0082] S6. Determine the customer's current consultation intention based on the subsequent input text, specifically including the following steps:

[0083] S61. Utilize a pre-set intent recognition model to analyze subsequent input text to determine the customer's current consultation intent. This intent recognition model is based on historical conversation data with elderly customers and is fine-tuned using the BERT model. It can identify common consultation intents used by elderly customers and output a confidence score for the intent.

[0084] S62. Compare the current consultation intention with the historical consultation intention to determine whether the two are consistent; the historical consultation intention is obtained by analyzing the language text through the intention recognition model, and the historical consultation intention has a corresponding intention confidence;

[0085] S63. If the current consultation intention is inconsistent with the historical consultation intention, and the intention confidence of the current consultation intention is greater than the preset threshold and greater than the intention confidence of the historical consultation intention, it is determined that intention drift has occurred, and the subsequent input text is input as a new language text into the intelligent customer service, and then return to execute step S1, otherwise execute step S3.

[0086] The intent recognition model is trained to identify common inquiry intentions from elderly customers, such as inquiring about product prices, understanding product attributes, and inquiring about order status. By learning from a large amount of historical conversation data with elderly customers, the model is able to understand their unique expressions. The intent confidence output by the intent recognition model indicates the model's degree of certainty in the recognition result. The current inquiry intent is compared with historical inquiry intents through string matching or intent ID comparison. The logic for determining intent drift combines intent consistency and the confidence of the current intent. When the intent is inconsistent and the confidence of the current intent is sufficiently high (above a preset threshold) and higher than the confidence of the historical intent, the system deems that the customer's communication goal has clearly changed.

[0087] Specifically, when the intelligent customer service receives the customer's initial language text, it first identifies and records his historical consulting intentions. The intelligent customer service generates an initial response text and sends it to the customer. Subsequently, the system receives the customer's subsequent input text. Before determining whether there are any comprehension barriers to the subsequent input text, the system uses the intent recognition model to analyze the subsequent input text to obtain the customer's current consulting intention and its confidence. The system compares the current consulting intention with the previously recorded historical consulting intention. If the comparison result shows that the intentions are inconsistent, and the confidence of the current intention meets the preset conditions (greater than the preset threshold and greater than the historical intention confidence), the system determines that the customer's consulting intention has drifted. At this point, the system no longer attempts to determine whether the subsequent input text is an understanding barrier to the initial response text, but instead regards the subsequent input text as the starting language text for a new round of consultation initiated by the customer, and inputs it into the beginning of the intelligent customer service processing flow, that is, returns to the step of recording customer input and the initial response of the intelligent customer service. Thus, the intelligent customer service will generate a corresponding response based on the customer's new consulting intention. If the intention Figure 1 If the intent is inconsistent, or if the confidence level of the current intent is insufficient to determine a clear intent drift, the system proceeds to determine whether there is a comprehension barrier to subsequent input text. By adding intent drift detection before determining comprehension barriers, the system can promptly identify shifts in customer communication goals, avoid ineffective communication based on the old intent, and improve the accuracy of intelligent customer service responses.

[0088] In some specific embodiments, a customer enters the initial language text "How do I use this rice cooker?" The intelligent customer service identifies the historical intent as "inquiring usage instructions" with a confidence level of 0.9 and sends an initial response text containing usage instructions. The customer then enters the subsequent input text "How much is the price?" The system receives this subsequent input text. The intent recognition model analyzes "How much is the price?" and identifies the current intent as "inquiring price" with a confidence level of 0.95. The system compares the current intent "inquiring price" with the historical intent "inquiring usage instructions" and finds that the two are inconsistent. The current intent confidence level of 0.95 is greater than the preset threshold of 0.7 and greater than the historical intent confidence level of 0.9. The system determines that intent drift has occurred. Therefore, the system inputs "How much is the price?" as the new language text to the intelligent customer service and returns to the recording step. The intelligent customer service will then generate a new response for the "inquiring price" intent.

[0089] In some embodiments, the specific steps in step S61 include:

[0090] Build a product knowledge graph containing multiple product information; product information includes various product attributes such as category, brand, price range, and target audience;

[0091] By parsing the subsequent input text, all keywords in the subsequent input text are extracted, and based on the product knowledge graph, each keyword is mapped to the corresponding product attribute vector;

[0092] All product attribute vectors are input into the intent recognition model to obtain multiple intent categories and corresponding intent confidences;

[0093] The intention category with the highest confidence is selected as the customer's current consultation intention.

[0094] The construction of a product knowledge graph can be achieved using graph database technology, where nodes represent products, attribute types, and attribute values, and edges represent the associations between products and attribute types, and between attribute types and attribute values. Keywords can be extracted by parsing subsequent input text, which can be achieved using word segmentation technology combined with part-of-speech tagging and named entity recognition technology. Mapping keywords to product attribute vectors can be vectorized based on the knowledge graph. For example, knowledge graph embedding technology can be used to map entities and relationships in the knowledge graph to a low-dimensional vector space, so that semantically related entities and relationships are close in distance in the vector space. As a result, the entities or attributes corresponding to the extracted keywords can find corresponding representations in the vector space. The product attribute vector is input into the intent recognition model, which can be a classification model that receives the vector input and outputs the intent category and its confidence. The intent category with the highest confidence is selected, that is, the intent with the largest probability value output by the model is selected as the final recognition result.

[0095] Specifically, to address the issue of inaccurate intent recognition caused by elderly customers' unfamiliarity with products on e-commerce platforms, this solution first builds a knowledge graph containing rich product attribute information. When receiving subsequent customer input, the system parses the text and extracts key information terms. These terms may not be standard product names or attribute descriptions. However, by comparing and mapping them with the product knowledge graph, the system can convert these non-standard terms into standardized product attribute representations, or product attribute vectors, that the system can understand. For example, if a customer asks for "a machine that can watch TV," the system extracts "watch TV" and "machine" as keywords. Using the knowledge graph, it associates "watch TV" with "video playback" under the "function" attribute and "machine" with "TV" under the "category" attribute. This association is converted into a vector representation. These product attribute vectors are then fed into an intent recognition model trained on conversational data from elderly customers. Based on these vectors, the model analyzes the customer's true intent, such as whether they are inquiring about price, features, or purchase method. The model outputs multiple possible intent categories and a confidence score for each category. The system selects the intent with the highest confidence as the customer's current inquiry intent. In this way, even if the language used by elderly customers is not precise enough or contains daily vocabulary they are familiar with, the system can use the assistance of the product knowledge graph to more accurately understand their consultation intentions about the product, thereby improving the accuracy of intent recognition.

[0096] In some specific implementations, when constructing a product knowledge graph, attributes such as product ID, product name, category (e.g., home appliances, apparel), brand (e.g., Haier, Midea), price range (e.g., 0-100 RMB, 100-500 RMB), target audience (e.g., seniors, youth), functions (e.g., voice control, video playback), and material may be included. For example, for a smart TV, the knowledge graph may record its category as "TV," brand as "a certain brand," functions as "video playback," "smart connectivity," and "voice control," and target audience as "home users." When an elderly customer enters "how much is a TV that can be controlled by speaking?", the system extracts the keywords "mouth control," "watch TV," "machine," and "price." Based on the knowledge graph, "mouth control" is mapped to "function: voice control," "watch TV" is mapped to "function: video playback," "machine" is mapped to "category: TV," and "price" is mapped to "intent type: price query." These mapping results are converted into a product attribute vector, such as a vector containing dimensional information such as category, function, and intent type. This vector is then input into the intent recognition model. After analyzing the vector, the model outputs the intent category and its confidence level. For example, "Query the price of a TV" has a confidence level of 0.95, while "Query the features of a TV" has a confidence level of 0.03. The system selects "Query the price of a TV," which has the highest confidence level, as the customer's current inquiry intent. This allows the system to accurately understand the customer's intent, even if the customer uses nonstandard or colloquial expressions.

[0097] In some embodiments, the specific steps in step S3 include:

[0098] S31. Extract question words, negative words, and repeated words from the subsequent input text and construct them into a text feature vector;

[0099] S32. Input the text feature vector into a preset comprehension barrier detection model to obtain the comprehension barrier probability; the comprehension barrier detection model is based on historical elderly customer conversation data and is trained using a support vector machine (SVM) algorithm to identify common expressions of elderly customers, such as questions, denials, and repetitions;

[0100] S33. Determine whether the probability of comprehension disorder is greater than a preset probability threshold. If so, determine that there is an comprehension disorder, generate an comprehension disorder signal, and mark the initial response text as text to be adjusted; otherwise, determine that there is no comprehension disorder, input the subsequent input text as a new language text to the intelligent customer service, and return to execute step S1.

[0101] This embodiment analyzes the customer's subsequent input text to identify specific language features that may express confusion or misunderstanding. First, in step S31, the system extracts question words, negation words, and repetition words from the subsequent input text. These vocabulary types are often used by customers to express questions, negate previous information, or emphasize parts they don't understand. These words are extracted and their occurrence is constructed into a text feature vector, providing a quantitative basis for subsequent judgment. This allows the capture of non-standard or unique expressions that elderly customers may use. Next, in step S32, the constructed text feature vector is input into a pre-set comprehension barrier detection model. This model is trained specifically based on historical conversation data from elderly customers and utilizes the support vector machine (SVM) algorithm. This training enables the model to identify common ways elderly customers express questions, negation, or repetition. After receiving the feature vector, the model outputs a comprehension barrier probability, which indicates the likelihood that the customer has a comprehension barrier. By using a model trained on elderly customer data, the system improves its ability to identify unique expressions of elderly customers. Finally, in step S33, the system compares the comprehension barrier probability obtained in step S32 with a pre-set probability threshold. If the calculated probability is greater than the threshold, the system determines that the customer has an understanding barrier and generates an understanding barrier signal. At the same time, the initial response text previously sent by the intelligent customer service is marked as text to be adjusted and prepared for modification. If the probability is not greater than the threshold, it is determined that the customer does not have an understanding barrier, and the system regards the customer's subsequent input text as a new valid input, inputs it as a new language text to the intelligent customer service, and returns to step S1 to start a new round of interaction. By making judgments based on probability thresholds, this technical solution provides a specific, operational and targeted understanding barrier detection mechanism, which improves the system's ability to identify customer confusion, thereby being able to trigger adjustments to the response text more promptly, improve communication effectiveness, and effectively improve the accuracy and robustness of understanding barrier identification.

[0102] In some specific embodiments, determining whether there is a comprehension obstacle can be achieved as follows: After receiving the subsequent input text from the customer, the system first performs text cleaning and word segmentation. For example, a stop word list is used to remove common modal particles and meaningless words. Then,词性 tagging and lexicon matching are performed on the word segmentation results. For example, a question word lexicon (such as "ma", "ne", "what"), a negation word lexicon (such as "not", "mei", "not be"), and a repeated word lexicon (such as the same words appearing consecutively) are maintained. The number of words in the subsequent input text belonging to these three lexicons is counted to obtain the number of question words, the number of negation words, and the number of repeated words respectively. These three quantities are constructed into a three-dimensional text feature vector, for example, [number of question words, number of negation words, number of repeated words]. This feature vector is input into an SVM model that has been pre-trained using elderly customer dialogue data. The model outputs a probability value between 0 and 1, indicating the likelihood of a comprehension obstacle. This probability value is compared with a preset probability threshold (such as 0.7). If the probability value is greater than 0.7, the system determines that the customer has a comprehension obstacle, generates a comprehension obstacle signal, and marks the response text previously sent by the intelligent customer service as pending adjustment. If the probability value is less than or equal to 0.7, it is determined that there is no comprehension obstacle, and the subsequent input text is processed as a new input.

[0103] The following is an example. Suppose an elderly customer consults product details on an e-commerce platform, and the intelligent customer service replies with a text containing terms such as "parameters" and "specifications". After receiving the reply, the customer inputs "I still don't understand. What does that parameter mean?".

[0104] The system receives the customer's subsequent input "I still don't understand. What does that parameter mean?", and extracts the question words "don't understand", "what does it mean", and the repeated word "still". These words are converted into a numerical vector. For example, the bag-of-words model or TF-IDF, or a more complex word embedding vector is used.

[0105] This vector is input into the pre-trained SVM comprehension obstacle detection model. This model is trained on a large amount of elderly customer dialogue data and has learned the patterns of elderly customers expressing comprehension obstacles such as "didn't understand", "not clear", "say it again". After processing the vector, the model outputs a comprehension obstacle probability, for example, 0.9.

[0106] The system compares the probability of 0.9 with a preset probability threshold (for example, 0.7). Because 0.9 is greater than 0.7, the system determines that the customer has difficulty understanding the message. At this point, the system generates a difficulty understanding signal and marks the response text previously sent by the intelligent customer service representative, including "parameters" and "specifications," as requiring adjustment. If the customer enters "OK, thank you," the model might output a low probability (for example, 0.1), which is below the threshold. The system then does not generate a difficulty understanding signal and does not adjust the response.

[0107] In this way, even if elderly customers' expressions are not standard or complete, the system can accurately identify their comprehension barriers by extracting key features and using probability-based model judgments, triggering subsequent response adjustments. Specific implementation methods may include: using natural language processing toolkits for part-of-speech tagging and keyword extraction; using pre-trained models such as Word2Vec or BERT to generate word or sentence vectors; and using libraries such as Scikit-learn to implement SVM classifiers and train them using labeled elderly customer conversation data.

[0108] In some embodiments, the specific steps in step S31 include:

[0109] Construct a stop word list; the stop word list includes common modal particles, auxiliary words and meaningless words;

[0110] Cleaning the subsequent input text using a stop word list to remove stop words in the subsequent input text to obtain a cleaned subsequent input text;

[0111] Use the word segmentation algorithm to perform word segmentation on the subsequent input text after cleaning to obtain multiple text segmentations;

[0112] For each text segmentation, determine whether it belongs to the preset question word library, negative word library or repeated word library; if it belongs to the question word library, mark the text segmentation as a question word; if it belongs to the negative word library, mark the text segmentation as a negative word; if it belongs to the repeated word library, mark the text segmentation as a repeated word;

[0113] Count the number of question words, negative words, and repeated words in the subsequent input text, and record them as the number of question words, the number of negative words, and the number of repeated words respectively;

[0114] Construct a text feature vector based on the number of question words, negative words, and repeated words.

[0115] A stop word list is constructed, which contains common modal particles, auxiliary words, and meaningless words to reduce interference information in the text. The input text is then cleaned to obtain a more refined text by removing stop words. A word segmentation algorithm is applied to break the cleaned text into independent lexical units, providing a basis for lexical-level analysis. Pre-set question word libraries, negative word libraries, and repeated word libraries are used to provide criteria for identifying specific words. Text segmentation words are judged one by one and marked according to whether they belong to these lexicons. The number of question words, negative words, and repeated words is counted to convert the text information into numerical features. A text feature vector is constructed and, based on the statistical number, provides input data for the subsequent comprehension disorder detection model. As a result, the accuracy of question word, negative word, and repeated word extraction is improved, and the robustness of comprehension disorder identification is enhanced.

[0116] Specifically, when receiving subsequent text input from a customer, the system first cleans the text using a pre-set stop word list to remove words that are not useful for determining comprehension difficulties, such as modal particles or auxiliary words. The cleaned text is then processed by a word segmentation algorithm, breaking it down into a series of independent words or phrases. The system then iterates through these text segmentations, comparing each one against a pre-set database of question words, negation words, and repeated words. If a segmentation word appears in the question word database, it is marked as a question word; if it appears in the negation word database, it is marked as a negation word; and if it appears in the repeated word database, it is marked as a repeated word. After all segmentations have been marked, the system counts the total number of question words, negation words, and repeated words marked in the subsequent input text. Finally, a numerical text feature vector is constructed based on the number of question words, negation words, and repeated words obtained. This text feature vector is then input into the comprehension difficulty detection model to determine whether the customer has comprehension difficulties. In this way, key features reflecting comprehension difficulties can be more accurately extracted from the unique language expressions of elderly customers, thereby improving the effectiveness of comprehension disorder detection.

[0117] In some specific embodiments, it is assumed that the subsequent input text of the customer is "Why don't I understand? Isn't there none?". First, a stop word list is constructed, for example, including "呢", "是不是", "啊", etc. The input text is cleaned using this stop word list, and the cleaned text "Why don't I understand? None?" is obtained. Then, a word segmentation algorithm is used to segment the cleaned text, and a text word segmentation list is obtained, for example, ["我", "怎么", "弄", "不", "明白", "没有"]. Then, a preset question word library (for example, including "怎么", "如何") and a negative word library (for example, including "不", "没有") are consulted. In the text word segmentation list, "怎么" belongs to the question word library and is marked as a question word; "不" belongs to the negative word library and is marked as a negative word; "没有" belongs to the negative word library and is marked as a negative word. The statistical result is: the number of question words is 1, the number of negative words is 2, and the number of repeated words is 0. Finally, a text feature vector is constructed based on these quantities, for example, [1, 2, 0]. This vector is input into the understanding obstacle detection model to calculate the understanding obstacle probability. By removing stop words and using a targeted word library, the intention of the elderly customer to express questions and negations can be more accurately captured, improving the accuracy of feature extraction.

[0118] In certain embodiments, the specific steps in step S4 include:

[0119] S41. Determine the adjustment method of the text to be adjusted according to the understanding obstacle signal; the adjustment methods include term simplification and sentence pattern simplification. Term simplification is used when the understanding obstacle signal indicates that the customer has an obstacle in understanding professional terms, and sentence pattern simplification is used when the understanding obstacle signal indicates that the customer has an obstacle in understanding long sentences;

[0120] S42. If it is determined that the adjustment direction is only term simplification, then consult the preset elderly customer-friendly vocabulary library, calculate the semantic similarity between each standard term in the text to be adjusted and the vocabulary in the elderly customer-friendly vocabulary library, and select the vocabulary with a semantic similarity higher than the preset value as the replacement vocabulary, and adjust the priority of the replacement vocabulary according to the weight value, and preferentially use the replacement vocabulary with the highest weight value to generate the adjusted response text; the vocabulary in the elderly customer-friendly vocabulary library is all set with weight values, and the weight value represents the usage frequency of the vocabulary in the elderly customer group;

[0121] S43. If it is determined that the adjustment direction is only sentence pattern simplification, then according to the preset simplification rule set, split the long sentence in the text to be adjusted into multiple short sentences and adjust the word order to generate the adjusted response text to make it conform to the reading habits of elderly customers, delete unnecessary modifying components, and ensure that the sentence meaning is clear;

[0122] S44. If the adjustment direction is determined to be terminology simplification and sentence simplification, the standard terms in the text to be adjusted will be replaced with simplified expressions by referring to the elderly customer-friendly vocabulary library, and the sentence and structure of the text to be adjusted after the replacement of terms will be adjusted by applying the simplification rule set to generate the adjusted response text.

[0123] Based on the comprehension difficulty signals it receives, the system determines how to adjust the initial response text. Comprehension difficulty signals can indicate that the customer has difficulty understanding specialized terminology, long sentences, or both. Based on the signal type, the system selects an appropriate adjustment strategy: term simplification, sentence simplification, or a combination of both. When term simplification is needed, the system accesses a pre-built senior-friendly vocabulary library. This vocabulary library stores terms that are more easily understood by seniors and their frequency weights among this demographic. The system calculates the semantic similarity between the specialized terminology in the text to be adjusted and the vocabulary in the vocabulary library, selecting terms that meet a certain similarity threshold as potential replacements. The system then ranks these replacements based on their weights, prioritizing the highest-weighted terms for replacement. When sentence simplification is needed, the system applies a set of pre-set simplification rules. These rules guide the system in identifying long sentences in the text to be adjusted, breaking them down into multiple, shorter sentences, and adjusting the sentence order to accommodate seniors' reading habits. The rules also guide the system in removing non-essential modifiers to ensure the core meaning of the sentence is clear. When both terminology and sentence simplification are required, the system first performs term replacement and then applies sentence simplification rules to the replaced text.

[0124] Specifically, when the intelligent customer service system receives a signal indicating a customer has difficulty understanding the text, it triggers an adjustment to the initial response text previously sent. First, based on the specific type of difficulty signal, the system determines whether the customer is struggling with specialized terminology, long sentences, or both. For example, if the difficulty signal stems from repeated inquiries about specific terms, the system may determine that terminology simplification is necessary; if the signal stems from confusion about entire sentences or paragraphs, the system may determine that sentence structure simplification is necessary. After determining the adjustment method, if terminology simplification is required, the system consults a user-friendly vocabulary library designed specifically for older customers. This vocabulary library contains alternative terms that correspond to commonly used professional terms in the e-commerce field and are more familiar to older customers. Each alternative term is associated with a weighted value that reflects its popularity among older customers. The system calculates the semantic similarity between the professional terms in the text to be adjusted and the alternative terms in the vocabulary library, selecting the terms with the highest similarity as replacement candidates. To ensure the effectiveness of the replacement, the system prioritizes the alternative terms with the highest weights, transforming difficult terms into easily understood expressions. This effectively lowers the threshold for understanding the terminology. If sentence simplification is necessary, the system applies a set of pre-set simplification rules. These rules define how to identify and handle complex sentences. For example, they break down long sentences containing multiple clauses into separate, simpler ones, adjust the order of sentence components to better suit the reading habits of older customers, and remove modifiers or parentheses that do not affect the core meaning. By applying these rules, the structure and presentation of the text to be adjusted are optimized, becoming more concise and clear, and reducing the difficulty of understanding caused by sentence complexity. If the comprehension barrier signal indicates that the customer has difficulty understanding both terminology and sentence structure, the system first performs term replacement, replacing specialized terms in the text with user-friendly terms. Based on this, sentence simplification rules are then applied to adjust the structure and presentation of the replaced text. This step-by-step approach ensures that the final response text is free of difficult terminology and complex sentence structure, maximizing comprehension efficiency for older customers. Through this adjustment process, the system generates a response text that better suits the cognitive characteristics of older customers, effectively resolving communication barriers caused by language complexity and improving the interaction between intelligent customer service and older customers.

[0125] In some specific implementations, suppose the initial response sent by the intelligent customer service to an elderly customer is, "The SKU of the product you purchased has been received and is expected to ship tomorrow. Please check your coupon information." The system detects that the customer has difficulty understanding the terms "SKU" and "coupon" and determines that terminology and sentence structure simplification are necessary. The system first performs terminology simplification. By consulting a database of friendly terms for elderly customers, it finds that the friendly terms for "SKU" include "product number" with a weight of 0.9 and "product number" with a weight of 0.7. The friendly terms for "coupon" include "discount coupon" with a weight of 0.95 and "discount coupon" with a weight of 0.8. The system calculates semantic similarity and confirms that "product number" and "discount coupon" have high similarity to the original term. Based on the weights, the system prioritizes "product number" to replace "SKU" and "discount coupon" to replace "coupon." The resulting text becomes, "The product number you purchased has been received and is expected to ship tomorrow. Please check your discount coupon information." The system then applies sentence simplification rules. Recognizing this as a long sentence containing multiple pieces of information, the system breaks it down into shorter sentences and adjusts the word order. For example, a rule can split the original sentence into: "The item number you purchased has arrived in the warehouse. It's expected to ship tomorrow. Please check your discount coupon information." At the same time, it removes the phrase "Please check it" that may be difficult for elderly customers to understand and adjusts it to a more direct expression. The resulting adjusted response text is "The item number you purchased has arrived in the warehouse. It's expected to ship tomorrow. Please check your discount coupon information." This adjusted text uses vocabulary more familiar to elderly customers and simpler sentence structures, significantly improving their comprehension.

[0126] In some specific implementations, suppose an elderly customer is shopping on an e-commerce platform and sees a "discount on purchases above a certain amount" promotion on a product detail page, but does not understand its meaning. The customer then asks the intelligent customer service representative, "What does this discount on purchases above a certain amount mean?"

[0127] The AI customer service representative initially responded: "Hello, a discount means that when your order amount reaches a certain amount, the system will automatically deduct the corresponding discount amount. For example, 10 off 100 means that if your total order amount reaches 100 yuan, you can enjoy a discount of 10 yuan."

[0128] When elderly customers receive this response, they may be confused by terms like "order amount" and "discount amount," or find the explanation unintuitive. They may then type in: "What is order amount? How is the discount calculated?"

[0129] After the system receives subsequent input and the comprehension barrier detection module analyzes it, it generates a comprehension barrier signal, indicating that the customer has comprehension barriers to terms such as "order amount" and "discount amount", and indicates that the initial response text is the text to be adjusted.

[0130] Step S4 is executed:

[0131] 1. Based on the signals of understanding barriers, the adjustment direction is determined to be terminology simplification.

[0132] 2. A search of the elderly-friendly vocabulary database revealed that simplified expressions for "order amount" include "total purchase price" and "the total amount you will pay." The one with the highest weight is "total purchase price." Simplified expressions for "discount amount" include "money saved" and "money reduced." The system prioritizes the simplified expression with the highest weight.

[0133] 3. Apply a response simplification rule set, which may include rules such as "replace terminology" and "simplify sentence structure." The system replaces "order amount" in the original response with "total purchase price" and "discount amount" with "amount saved."

[0134] 4. Generate the adjusted response text: "Hello, the discount you're asking about means that once the total purchase amount reaches a certain amount, the system will automatically deduct the savings. For example, 'Save 10 Yuan on Spend 100 Yuan' means that if the total purchase amount reaches 100 Yuan, you can save 10 Yuan."

[0135] The intelligent customer service representative sent the adjusted response to the customer. The elderly customer noticed the new response used more accessible terms and successfully grasped the concept of discounts.

[0136] The specific implementation of the elderly customer-friendly vocabulary library can be a database or file that stores standard terms and their corresponding simplified expressions and weight values. For example:

[0137] "Order amount": [{"expression":"total price of the purchase","weight":0.8},{"expression":"the total amount you need to pay","weight":0.6}],

[0138] "Discount amount": [{"expression":"money saved","weight":0.9},{"expression":"money reduced","weight":0.7}],

[0139] The response simplification rule set can be a set of rule-based or model-based text transformation methods. For example, a rule-based method might define rules such as "remove content within brackets" or "convert a definition sentence beginning with 'is' to an explanation sentence beginning with 'Simply put,'" while a model-based method might train a sequence-to-sequence model that takes complex text as input and generates simplified text as output.

[0140] The direction of adjustment can be determined based on the specific content of the comprehension barrier signal. For example, if the signal contains "terminology not understood," term simplification is triggered; if it contains "too long," sentence simplification is triggered. If both are present, both simplification methods are triggered.

[0141] Reference Attachment Figure 2 The present invention provides a customer VOC intelligent response system based on an e-commerce platform, comprising:

[0142] The recording module 100 is used to record the language text input by the customer and the initial response text fed back to the customer by the intelligent customer service based on the language text;

[0143] Receiving module 200, for receiving subsequent input text from the client after receiving the initial response text;

[0144] The first judgment module 300 is configured to determine whether there is a comprehension barrier based on the subsequent input text. If so, it generates a comprehension barrier signal and marks the initial response text as a text to be adjusted. If not, it inputs the subsequent input text as a new language text to the intelligent customer service and returns it to the operation record module.

[0145] A generating module 400 is configured to adjust the text to be adjusted according to the comprehension impairment signal and generate an adjusted response text;

[0146] The sending module 500 is configured to send the adjusted response text to the corresponding customer.

[0147] In some embodiments, a second judgment module 600 is further included. The second judgment module 600 is used to judge the customer's current consultation intention based on the subsequent input text, and specifically performs the following steps:

[0148] S61. Utilize a pre-set intent recognition model to analyze subsequent input text to determine the customer's current consultation intent. This intent recognition model is based on historical conversation data with elderly customers and is fine-tuned using the BERT model. It can identify common consultation intents used by elderly customers and output a confidence score for the intent.

[0149] S62. Compare the current consultation intention with the historical consultation intention to determine whether the two are consistent; the historical consultation intention is obtained by analyzing the language text through the intention recognition model, and the historical consultation intention has a corresponding intention confidence;

[0150] S63. If the current consultation intention is inconsistent with the historical consultation intention, and the intention confidence of the current consultation intention is greater than the preset threshold and greater than the intention confidence of the historical consultation intention, it is determined that intention drift has occurred, and the subsequent input text is input as a new language text into the intelligent customer service, and then the operation record module is returned, otherwise the first judgment module is run.

[0151] In some embodiments, the second determination module 600 is executed when being used to obtain the client's current consultation intention by analyzing subsequent input text using a preset intention recognition model:

[0152] Build a product knowledge graph containing multiple product information; product information includes various product attributes such as category, brand, price range, and target audience;

[0153] By parsing the subsequent input text, all keywords in the subsequent input text are extracted, and based on the product knowledge graph, each keyword is mapped to the corresponding product attribute vector;

[0154] All product attribute vectors are input into the intent recognition model to obtain multiple intent categories and corresponding intent confidences;

[0155] The intention category with the highest confidence is selected as the customer's current consultation intention.

[0156] In some embodiments, the first determination module 300 is configured to determine whether there is an understanding barrier based on the subsequent input text. If so, it generates an understanding barrier signal and marks the initial response text as text to be adjusted. If not, it inputs the subsequent input text as a new language text into the intelligent customer service and returns to the operation recording module.

[0157] S31. Extract question words, negative words, and repeated words from the subsequent input text and construct them into a text feature vector;

[0158] S32. Input the text feature vector into a preset comprehension barrier detection model to obtain the comprehension barrier probability; the comprehension barrier detection model is based on historical elderly customer conversation data and is trained using a support vector machine (SVM) algorithm to identify common expressions of elderly customers, such as questions, denials, and repetitions;

[0159] S33. Determine whether the probability of comprehension disorder is greater than a preset probability threshold. If so, it is determined that there is an comprehension disorder, an comprehension disorder signal is generated, and the initial response text is marked as text to be adjusted. Otherwise, it is determined that there is no comprehension disorder, and the subsequent input text is input as a new language text into the intelligent customer service, and the operation record module is returned.

[0160] In some embodiments, the first determination module 300 is executed when extracting question words, negative words, and repeated words from a subsequent input text and constructing them into a text feature vector:

[0161] Construct a stop word list; the stop word list includes common modal particles, auxiliary words and meaningless words;

[0162] Cleaning the subsequent input text using a stop word list to remove stop words in the subsequent input text to obtain a cleaned subsequent input text;

[0163] Use the word segmentation algorithm to perform word segmentation on the subsequent input text after cleaning to obtain multiple text segmentations;

[0164] For each text segmentation, determine whether it belongs to the preset question word library, negative word library or repeated word library; if it belongs to the question word library, mark the text segmentation as a question word; if it belongs to the negative word library, mark the text segmentation as a negative word; if it belongs to the repeated word library, mark the text segmentation as a repeated word;

[0165] Count the number of question words, negative words, and repeated words in the subsequent input text, and record them as the number of question words, the number of negative words, and the number of repeated words respectively;

[0166] Construct a text feature vector based on the number of question words, negative words, and repeated words.

[0167] In some embodiments, when the generation module 400 is used to adjust the text to be adjusted based on the comprehension impairment signal and generate the adjusted response text, the following steps are performed:

[0168] S41. Determine an adjustment method for the text to be adjusted based on the comprehension barrier signal; the adjustment methods include terminology simplification and sentence simplification. Term simplification is used when the comprehension barrier signal indicates that the customer has difficulty understanding professional terminology, and sentence simplification is used when the comprehension barrier signal indicates that the customer has difficulty understanding long sentences.

[0169] If it is determined that the adjustment direction is only term simplification, then consult a preset elderly-friendly vocabulary library and calculate the semantic similarity between each standard term in the text to be adjusted and the vocabulary in the elderly-friendly vocabulary library. Select the vocabulary with a semantic similarity higher than the preset value as the replacement vocabulary, and adjust the priority of the replacement vocabulary according to the weight value, giving priority to the replacement vocabulary with the highest weight value, and generate the adjusted response text; each vocabulary in the elderly-friendly vocabulary library is assigned a weight value, which represents the frequency of use of the vocabulary among the elderly customer group;

[0170] If the adjustment direction is determined to be sentence simplification only, then, based on a preset set of simplification rules, the long sentence in the text to be adjusted is split into multiple shorter sentences and the word order is adjusted to generate an adjusted response text that conforms to the reading habits of elderly customers, removing unnecessary modifiers and ensuring clarity of sentence meaning.

[0171] S44. If the adjustment direction is determined to be terminology simplification and sentence simplification, the standard terms in the text to be adjusted will be replaced with simplified expressions by referring to the elderly customer-friendly vocabulary library, and the sentence and structure of the text to be adjusted after the replacement of terms will be adjusted by applying the simplification rule set to generate the adjusted response text.

[0172] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any actual relationship or order between these entities or operations.

[0173] The foregoing description is merely an embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Those skilled in the art will readily appreciate that the present invention is susceptible to various modifications and variations. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A customer VOC intelligent response method based on an e-commerce platform, characterized in that: The following steps are involved: S1. Record the language text entered by the customer and the initial response text of the intelligent customer service feedback to the customer for the language text; S2 receives the subsequent input text after the customer receives the initial response text; S3. Determine whether there is a comprehension barrier based on the subsequent input text. If so, generate a comprehension barrier signal and mark the initial response text as a text to be adjusted. If not, input the subsequent input text as a new language text to the intelligent customer service and return to step S1. S4. Adjust the text to be adjusted according to the comprehension barrier signal to generate an adjusted response text; S5. Send the adjusted response text to the corresponding customer; The specific steps in step S3 include: S31 extracts question words, negative words, and repeated words from the subsequent input text and constructs them into a text feature vector; S32. Input the text feature vector into a preset comprehension barrier detection model to obtain a comprehension barrier probability; the comprehension barrier detection model is based on historical elderly customer conversation data and is trained using a support vector machine (SVM) algorithm to identify common expressions of elderly customers, such as questions, denials, and repetitions; S33. Determine whether the probability of the comprehension barrier is greater than a preset probability threshold. If so, determine that there is an comprehension barrier, generate an comprehension barrier signal, and mark the initial response text as text to be adjusted; otherwise, determine that there is no comprehension barrier, input the subsequent input text as a new language text into the intelligent customer service, and return to execute step S1.

2. The customer VOC intelligent response method based on the e-commerce platform according to claim 1 is characterized in that: The following steps are included between step S2 and step S3: S6. Determine the customer's current consultation intention based on the subsequent input text, specifically including the steps of: S61. Utilize a preset intent recognition model to analyze the subsequent input text to determine the client's current consultation intent. This intent recognition model is based on historical elderly client conversation data and is fine-tuned using the BERT model. The model is capable of recognizing commonly used consultation intents among elderly clients and outputting a confidence score for the intent. S62. Compare the current consultation intention with the historical consultation intention to determine whether the two are consistent; the historical consultation intention is obtained by analyzing the language text by the intention recognition model, and the historical consultation intention has a corresponding intention confidence; S63. If the current consultation intention is inconsistent with the historical consultation intention, and the intention confidence of the current consultation intention is greater than the preset threshold and greater than the intention confidence of the historical consultation intention, it is determined that intention drift has occurred, and the subsequent input text is input as a new language text into the intelligent customer service, and then return to execute step S1, otherwise execute step S3.

3. The customer VOC intelligent response method based on the e-commerce platform according to claim 2 is characterized in that: The specific steps in step S61 include: Constructing a product knowledge graph containing information about multiple products; the product information includes multiple product attributes of various products; Extract all keywords from the subsequent input text by parsing the subsequent input text, and map each keyword into a corresponding product attribute vector based on the product knowledge graph; Inputting all the product attribute vectors into the intention recognition model respectively to obtain multiple intention categories and corresponding intention confidences; The intention category with the highest confidence is selected as the customer's current consultation intention.

4. The customer VOC intelligent response method based on the e-commerce platform according to claim 1 is characterized in that: The specific steps in step S31 include: Constructing a stop word list; the stop word list includes common modal particles, auxiliary words and meaningless words; Cleaning the subsequent input text using the stop word list to remove stop words in the subsequent input text to obtain a cleaned subsequent input text; Using a word segmentation algorithm to perform word segmentation processing on the subsequent input text after the cleaning to obtain multiple text segmentations; For each of the text segmentation words, determine whether it belongs to a preset question word library, a negative word library or a repeated word library; if it belongs to the question word library, mark the text segmentation word as a question word; if it belongs to the negative word library, mark the text segmentation word as a negative word; if it belongs to the repeated word library, mark the text segmentation word as a repeated word; Counting the number of question words, negative words, and repeated words in the subsequent input text, which are recorded as the number of question words, the number of negative words, and the number of repeated words, respectively; The text feature vector is constructed according to the number of question words, the number of negative words, and the number of repeated words.

5. The customer VOC intelligent response method based on the e-commerce platform according to claim 1 is characterized in that: The specific steps in step S4 include: S41. Determining an adjustment method for the text to be adjusted based on the comprehension barrier signal; the adjustment method includes terminology simplification and sentence simplification. The terminology simplification is used when the comprehension barrier signal indicates that the customer has difficulty understanding professional terminology, and the sentence simplification is used when the comprehension barrier signal indicates that the customer has difficulty understanding long sentences. If it is determined that the adjustment direction is only term simplification, a preset elderly-friendly vocabulary library is consulted and the semantic similarity between each standard term in the text to be adjusted and the vocabulary in the elderly-friendly vocabulary library is calculated. Vocabulary with a semantic similarity higher than a preset value is selected as a replacement vocabulary, and the priority of the replacement vocabulary is adjusted according to the weight value, with the replacement vocabulary with the highest weight value being preferentially used to generate the adjusted response text; each vocabulary in the elderly-friendly vocabulary library is assigned a weight value, which represents the frequency of use of the vocabulary among the elderly customer group; S43. If the adjustment direction is determined to be sentence simplification only, the long sentence in the text to be adjusted is split into multiple short sentences and the word order is adjusted according to the preset simplification rule set to generate the adjusted response text; S44. If the adjustment direction is determined to be terminology simplification and sentence simplification, the standard terms in the text to be adjusted are replaced with simplified expressions by referring to the elderly customer-friendly vocabulary library, and the sentence and structure of the text to be adjusted after the replacement of the terms are adjusted by applying the simplification rule set to generate an adjusted response text.

6. A customer VOC intelligent response system based on an e-commerce platform, characterized in that: include: A recording module is used to record the language text input by the customer and the initial response text fed back to the customer by the intelligent customer service based on the language text; A receiving module, configured to receive subsequent input text from the client after receiving the initial response text; a first judgment module, configured to determine whether there is an understanding barrier based on the subsequent input text; if so, generate an understanding barrier signal and mark the initial response text as a text to be adjusted; if not, input the subsequent input text as a new language text into the intelligent customer service and return to run the recording module; a generating module, configured to adjust the text to be adjusted according to the comprehension impairment signal and generate an adjusted response text; A sending module, configured to send the adjusted response text to the corresponding customer; The first judgment module is configured to determine whether there is an understanding barrier based on the subsequent input text. If so, it generates an understanding barrier signal and marks the initial response text as a text to be adjusted. If not, it inputs the subsequent input text as a new language text into the intelligent customer service and returns to the execution time of the recording module: S31 extracts question words, negative words, and repeated words from the subsequent input text and constructs them into a text feature vector; S32. Input the text feature vector into a preset comprehension barrier detection model to obtain a comprehension barrier probability; the comprehension barrier detection model is based on historical elderly customer conversation data and is trained using a support vector machine (SVM) algorithm to identify common expressions of elderly customers, such as questions, denials, and repetitions; S33. Determine whether the probability of the comprehension barrier is greater than a preset probability threshold. If so, determine that there is an comprehension barrier, generate an comprehension barrier signal, and mark the initial response text as text to be adjusted. Otherwise, determine that there is no comprehension barrier, input the subsequent input text as a new language text into the intelligent customer service, and return to run the recording module.

7. The customer VOC intelligent response system based on the e-commerce platform according to claim 6 is characterized in that: The second judging module is further included, and the second judging module is used to judge the current consultation intention of the customer based on the subsequent input text, and specifically performs the following steps: S61. Using a preset intention recognition model, by analyzing the subsequent input text, obtain the customer's current consultation intention; The intent recognition model is based on historical elderly customer conversation data and is fine-tuned using the BERT model. It can identify the common consultation intents of elderly customers and output the intent confidence level. S62. Compare the current consultation intention with the historical consultation intention to determine whether the two are consistent; the historical consultation intention is obtained by analyzing the language text by the intention recognition model, and the historical consultation intention has a corresponding intention confidence; S63. If the current consultation intention is inconsistent with the historical consultation intention, and the intention confidence of the current consultation intention is greater than the preset threshold and greater than the intention confidence of the historical consultation intention, it is determined that intention drift has occurred, and the subsequent input text is input as a new language text into the intelligent customer service, and then the recording module is returned to run, otherwise the first judgment module is run.

8. The customer VOC intelligent response system based on the e-commerce platform according to claim 7 is characterized in that: The second judgment module is executed when it is used to obtain the customer's current consultation intention by analyzing the subsequent input text using a preset intention recognition model: Constructing a product knowledge graph containing information about multiple products; the product information includes multiple product attributes of various products; Extract all keywords from the subsequent input text by parsing the subsequent input text, and map each keyword into a corresponding product attribute vector based on the product knowledge graph; Inputting all the product attribute vectors into the intention recognition model respectively to obtain multiple intention categories and corresponding intention confidences; The intention category with the highest confidence is selected as the customer's current consultation intention.

Images