E-commerce customer service response system and method based on multi-model routing technology

Abstract

The invention discloses an e-commerce customer service response system and method based on a multi-model routing technology, and relates to the technical field of e-commerce, and the method comprises the steps: collecting a user input text and user behavior data, carrying out the preliminary intention matching of a user input text request through an e-commerce field regular expression rule library, and obtaining a regular matching result; inputting the regular matching result and the user request text into a BERT multi-mode classification model to obtain an accurate intention recognition result; calculating a routing decision score according to the regular matching result and the accurate intention recognition result by adopting a dynamic weight formula to obtain a target service module; and querying the product knowledge graph through the target service module, and generating personalized response content in combination with the user behavior data. According to the method, a routing decision score is calculated by adopting a dynamic weight formula according to a regular matching result and an accurate intention recognition result, a product knowledge graph is inquired through a target service module, and personalized response content is generated in combination with user behavior data.

Description

Technical Field

[0001] The present invention relates to the field of e-commerce technology, and in particular to an e-commerce customer service response system and method based on multi-model routing technology. Background Art

[0002] In the e-commerce sector, intelligent customer service systems have become a key technology for improving user service efficiency. Existing conventional approaches typically combine rule-based preliminary intent recognition with precise classification using a single deep learning model. For example, after matching high-frequency keywords using regular expressions, they leverage pre-trained language models (such as BERT) for fine-grained intent judgment and route inquiries to the appropriate business module based on static rules. This approach effectively handles user inquiries in standard scenarios.

[0003] However, existing conventional methods mainly lack dynamic adaptability in the collaborative mechanism of rules and models. Fixed weight fusion methods are difficult to cope with the changing user expressions and complex contexts in e-commerce scenarios. Static routing strategies may ignore the impact of visual information on decision-making. Personalized response generation is separated from routing decisions. User behavior data is only used in the final response stage and is not involved in routing optimization. As a result, the specific needs of high-value users cannot be identified in early decision-making. Summary of the Invention

[0004] In view of the above existing problems, the present invention is proposed.

[0005] Therefore, the present invention provides an e-commerce customer service response method based on multi-model routing technology to solve the problems of insufficient adaptability of rules and models in the existing technology and the separation of personalized responses and routing decisions.

[0006] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0007] In a first aspect, the present invention provides an e-commerce customer service response method based on multi-model routing technology, which includes: Collect user input text and user behavior data, perform preliminary intent matching on user input text requests using the regular expression rule library in the e-commerce field, and obtain regular matching results; Input the regular matching results and user request text into the BERT multimodal classification model to obtain accurate intent recognition results; The regular matching results and precise intent recognition results are used to calculate the routing decision score using a dynamic weight formula to obtain the target business module; Generate personalized response content by querying the product knowledge graph through the target business module and combining it with user behavior data; Adapt the personalized response content to cross-platform formats to form customer service responses.

[0008] As a preferred solution of the e-commerce customer service response method based on multi-model routing technology described in the present invention, wherein: collecting user input text and user behavior data includes the following steps: Receive the original text request input by the user, perform basic cleaning, and generate standardized text; The behavior log database of the e-commerce platform is called through the user ID to obtain user behavior data.

[0009] As a preferred solution of the e-commerce customer service response method based on multi-model routing technology described in the present invention, wherein: preliminary intent matching is performed on the user input text request through the e-commerce field regular expression rule library to obtain a regular matching result, including the following steps: Compare the user input text request with the regular template of the e-commerce regular expression rule library to match the keyword combination; According to the matched keyword combination, the confidence of the current regular match is calculated to obtain the regular match result.

[0010] As a preferred solution of the e-commerce customer service response method based on multi-model routing technology described in the present invention, the regular matching result and the user request text are input into the BERT multimodal classification model to obtain an accurate intent recognition result, including the following steps: Combine the regular expression matching results and the user request text to generate a multimodal input text sequence; Perform BERT standard preprocessing on the multimodal input text sequence to generate text vectors; Load the BERT base model through PyTorch, add a multimodal fusion layer, and output the improved BERT base model architecture; Use e-commerce domain corpus to pre-train the improved BERT basic model architecture for domain adaptation, output the BERT initial model, encode the regular matching results and user behavior data into feature vectors, and output a multimodal training dataset; The multimodal training dataset is input into the BERT initial model, fine-tuned using the joint loss function, and the BERT multimodal classification model is output; The text vector is input into the BERT multimodal classification model to obtain accurate intent recognition results.

[0011] As a preferred solution of the e-commerce customer service response method based on multi-model routing technology described in the present invention, the regular matching results and the precise intent recognition results are used to calculate the routing decision score using a dynamic weight formula to obtain the target business module, including the following steps: Extract the intent type and confidence level from the regular matching results, as well as the BERT intent score and context coherence score from the precise intent recognition results, to form structured routing input data. Analyze user behavior data to obtain session status, dynamically calculate weight coefficient based on session status, and obtain dynamic weight; The confidence score and BERT intent score are combined according to dynamic weights to calculate the routing decision score. The accuracy-response delay balance point of different intent categories is extracted from user behavior data, a consultation intent threshold is set, and the routing decision score is compared with the consultation intent threshold to obtain the target business module.

[0012] As a preferred solution of the e-commerce customer service response method based on multi-model routing technology described in the present invention, wherein: querying the product knowledge graph through the target business module and combining user behavior data to generate personalized response content, including the following steps: Use the knowledge graph query interface according to the target business module to obtain initial product attribute data; Extract recent behavior data from user behavior data and output user portrait feature vectors; Input product attribute data and user portrait feature vectors into the strategy engine and generate response strategies through rule templates; According to the response logic, adapted content is called from the multimedia resource library to form original response materials, which are then personalized and sorted to generate personalized response content.

[0013] As a preferred solution of the e-commerce customer service response method based on multi-model routing technology described in the present invention, wherein: the personalized response content is adapted to a cross-platform format to form a customer service response, including the following steps: Parse the text in the personalized response content, match the UI adaptation template according to the terminal type, and obtain rendering rules; The personalized response content is converted according to the rendering rules to obtain an optimized content package, which is embedded with a session tracking identifier to form a customer service response.

[0014] In a second aspect, the present invention provides an e-commerce customer service response system based on multi-model routing technology, including a regular expression matching result module that collects user input text and user behavior data, performs preliminary intent matching on user input text requests through a regular expression rule library in the e-commerce field, and obtains a regular expression matching result; The precise intent recognition result module inputs the regular matching results and user request text into the BERT multimodal classification model to obtain precise intent recognition results; The target business module uses the regular matching results and the precise intent recognition results to calculate the routing decision score using a dynamic weight formula to obtain the target business module; The personalized response content module queries the product knowledge graph through the target business module and combines it with user behavior data to generate personalized response content; The customer service response module adapts the personalized response content to the cross-platform format to form a customer service response.

[0015] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program is executed by the processor, any step of the e-commerce customer service response method based on multi-model routing technology as described in the first aspect of the present invention is implemented.

[0016] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program is executed by a processor, it implements any step of the e-commerce customer service response method based on multi-model routing technology as described in the first aspect of the present invention.

[0017] The beneficial effects of the present invention are as follows: the routing decision score is calculated by using a dynamic weight formula through regular matching results and precise intent recognition results, and the product knowledge graph is queried through the target business module and personalized response content is generated in combination with user behavior data. The routing decision score is calculated by the dynamic weight formula, which realizes the dynamic fusion of rule matching and deep learning results, and the weight distribution is adaptively adjusted according to the real-time session state, so that the routing decision retains the high efficiency of rule matching and the semantic understanding depth of model recognition, achieving the effect of accurately matching business scenario requirements, and generating personalized response content through the collaborative processing of knowledge graph and user behavior data, which realizes the deep correlation analysis of product attributes and user portraits, and dynamically combines multimodal response materials based on the strategy engine, so that the generated customer service answers ensure both professional accuracy and meet the user's personalized preferences, which significantly improves the service quality and user experience of e-commerce customer service. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0019] Figure 1 This is a flowchart of an e-commerce customer service response method based on multi-model routing technology.

[0020] Figure 2 Schematic diagram of an e-commerce customer service response system based on multi-model routing technology.

[0021] Figure 3 This is a flowchart of the regular expression matching results.

[0022] Figure 4 It is the flowchart of the target business module. DETAILED DESCRIPTION

[0023] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0024] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0025] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0026] Reference Figures 1 to 4 , is an embodiment of the present invention, which provides an e-commerce customer service response method based on multi-model routing technology, including the following steps:

[0027] S1. Collect user input text and user behavior data.

[0028] S1.1. Receive the original text request input by the user, perform basic cleaning, and generate standardized text.

[0029] Furthermore, we use regular expressions to match and remove invisible control characters, use the English stop word list of the NLTK library and the Chinese stop word list of Harbin Institute of Technology to filter bilingual stop words, and use Unicode Normalization Form C (NFC) to convert the text into a standardized encoding to ensure character encoding consistency in subsequent processing and generate standardized text.

[0030] S1.2. Call the behavior log database of the e-commerce platform through the user ID to obtain user behavior data.

[0031] Furthermore, an SSL encrypted channel is established to connect to the e-commerce platform database, and then parameterized SQL query statements are executed to prevent SQL injection. The query results are encapsulated in the JSON-LD standard format. The fields include core dimensions such as timestamp, behavior type, and associated product ID. The data acquisition process uses the OAuth2.0 protocol for authentication to ensure compliance with GDPR data privacy regulations and generate user behavior data.

[0032] S2. Perform preliminary intent matching on the user input text request through the regular expression rule library in the e-commerce field to obtain regular matching results.

[0033] S2.1. Compare the user input text request with the regular template of the e-commerce field regular expression rule library to match the keyword combination.

[0034] Furthermore, the user input text request is compared one by one with the regular templates in the regular expression rule library in the e-commerce field, and the match function of the re library is used to perform pattern matching to identify text fragments containing specific keyword combinations. The regular expression rule library in the e-commerce field contains product consultation templates and after-sales service templates. During the matching process, the position and matching length of the hit keywords are recorded at the same time. When the main keyword (such as "return") and the auxiliary keyword (such as "policy") are matched at the same time, it is judged as a strong match. The comparison result is output as a structured matching record. The fields contain the matched regular template ID, the hit keyword list and the matching position, and the matching keyword combination is obtained.

[0035] S2.2. Calculate the confidence of the current regular match based on the matched keyword combination to obtain the regular match result.

[0036] Furthermore, the number of hits for core keywords is counted (for example, "return" and "policy" each count as 1 point). A position weight is then calculated based on the keyword position (for example, a core keyword appearing at the beginning of a sentence receives a 0.2 increase in weight). This is then combined with a weighted calculation using the match length coefficient (for example, a coefficient of 1.2 is used when the matching text accounts for more than 50% of the entire sentence). This is used to calculate the confidence level of the current regular expression match and obtain the regular expression match result. Specifically, the expression is, ; in, is the confidence level, As the basis, is the position weight, is the matching length coefficient, is the normalization factor.

[0037] S3. Input the regular matching results and user request text into the BERT multimodal classification model to obtain accurate intent recognition results.

[0038] S3.1. Combine the regular expression matching results and the user request text to generate a multimodal input text sequence.

[0039] Furthermore, the intent type field in the regular matching result (for example, return policy consultation) is concatenated with the user request text to form an annotated enhanced text sequence in the format of [intent label] original request text (for example, [return policy consultation] What is the return policy?). The confidence value in the regular matching result is converted into a 32-bit floating-point feature vector, which is aligned in dimension with the BERT word vector of the user request text to generate a multimodal input text sequence.

[0040] S3.2. Perform BERT standard preprocessing on the multimodal input text sequence to generate text vectors.

[0041] Furthermore, the enhanced text sequence is converted into a subword ID sequence through BERTTokenizer, position encoding is added to special tags, and the regular matching confidence feature vector and user behavior statistical features are respectively subjected to Min-Max normalization to ensure that the numerical range is consistent with the BERT word vector distribution. A 768-dimensional text vector representation is generated, of which the first 512 dimensions are BERT word vectors, the middle 128 dimensions are normalized confidence features, and the last 128 dimensions are user behavior features to generate a text vector.

[0042] S3.3. Load the BERT base model through PyTorch, add a multimodal fusion layer, and output the improved BERT base model architecture.

[0043] Furthermore, the from_pretrained method of the PyTorch framework is used to load the bert-base-uncased pre-trained model weights, and a multimodal fusion layer is inserted after the corresponding position of the [CLS] tag of the BERT base model. The fusion layer consists of two fully connected layers. The first layer concatenates the BERT hidden layer output (768 dimensions) with the regular matching confidence features (128 dimensions) and user behavior features (128 dimensions) into a 1024-dimensional vector. The second layer reduces the dimension to 768 dimensions through the ReLU activation function. The improved BERT base model architecture maintains the original 12-layer Transformer structure unchanged, and only adds a multimodal feature fusion path in the sequence output part. The final output dimension is consistent with the original BERT model.

[0044] S3.4. Use e-commerce domain corpus to perform domain adaptation pre-training on the improved BERT basic model architecture, output the BERT initial model, encode the regular matching results and user behavior data into feature vectors, and output the multimodal training dataset.

[0045] Furthermore, the improved BERT basic model architecture is pre-trained for domain adaptation using e-commerce domain corpus (including product description text, customer service conversation records, and user review data). The masked language modeling task and the next sentence prediction task are jointly optimized. During the training process, the Embedding layer parameters of the original BERT model are frozen, and only the parameters of the multimodal fusion layer and the last three layers of Transformer are updated. The domain-adapted BERT initial model is output, and the intent type field in the regular matching result is converted into a 32-dimensional sparse vector through One-Hot encoding. The numerical features in the user behavior data (such as the number of returns in the past 30 days) are normalized into a 32-dimensional dense vector through Z-score. The two types of feature vectors are spliced ​​to form a 64-dimensional multimodal feature vector, which together with the preprocessed e-commerce domain text corpus constitutes a multimodal training dataset.

[0046] S3.5. Input the multimodal training dataset into the BERT initial model, use the joint loss function for fine-tuning, and output the BERT multimodal classification model.

[0047] Furthermore, the multimodal training dataset is input into the BERT initial model for fine-tuning training, and a joint loss function consisting of the cross-entropy loss function and the KL divergence loss function is adopted. The cross-entropy loss function calculates the difference between the predicted intent and the true label, and the KL divergence loss function constrains the distribution consistency of the multimodal feature vector and the BERT text representation. The AdamW optimizer is used during training, the learning rate is set to 5e-5, the batch_size is 32, and 10 epochs are iterated on the training set. After fine-tuning is completed, the optimal checkpoint is retained and the BERT multimodal classification model is output.

[0048] S3.6. Input the text vector into the BERT multimodal classification model to obtain accurate intent recognition results.

[0049] Furthermore, the text vector is input into the BERT multimodal classification model to perform forward reasoning. The text vector first obtains contextual representation through the BERT model's 12-layer Transformer encoder, and then interacts with the multimodal feature vector in the fusion layer. The fused representation is passed through the classification head (fully connected layer + Softmax) to calculate the probability distribution of 28 types of intents. The intent with the highest probability value is selected as the main image output, and the second highest probability intent is retained as the auxiliary result to generate a structured and accurate intent recognition result.

[0050] S4. Calculate the routing decision score using a dynamic weight formula based on the regular matching results and the precise intent recognition results to obtain the target business module.

[0051] S4.1. Extract the intent type and confidence from the regular matching results, as well as the BERT intent score and context coherence score from the precise intent recognition results, to form structured routing input data.

[0052] Furthermore, the intent type field and normalized confidence value are extracted from the regular matching results, and the BERT intent score is obtained from the precise intent recognition results. The four types of data are encapsulated according to the preset field structure, and the context coherence score is obtained by analyzing the intent of the last three interactions in the conversation history. Figure 1 The consistency is obtained, and the output is structured routing input data.

[0053] S4.2. Analyze user behavior data to obtain session status, dynamically calculate weight coefficients based on the session status, and obtain dynamic weights.

[0054] Furthermore, the number of user questions in the current session, the average time interval between adjacent requests, and the operation type distribution are counted. The session state indicators are input into the dynamic weight to obtain -0.1 time interval average + 0.05 operation type entropy. The operation type entropy reflects the degree of dispersion of the operation distribution, and the dynamic weight coefficient (for example, 0.45) is obtained. The weight value range is limited to the interval [0.2, 0.8] to ensure routing stability, and the normalized dynamic weight is output.

[0055] S4.3. Confidence and BERT intent scores are fused according to dynamic weights to calculate the routing decision score.

[0056] Furthermore, the confidence of the regular expression matching result and the BERT intent score are weighted and fused using a dynamic weight coefficient. The input parameters are range-checked to ensure that the confidence and BERT intent scores are in the range of [0, 1] and the dynamic weight coefficient is in the range of [0.2, 0.8]. The normalized routing decision score is then output. Specifically, the expression is, ; in, is the routing decision score, is the dynamic weight coefficient, is the regular matching confidence, Scoring BERT intent.

[0057] S4.4. Extract the accuracy-response delay balance point of different intent categories from the user behavior data, set the consultation intent threshold, compare the routing decision score with the consultation intent threshold, and obtain the target business module.

[0058] Furthermore, based on user behavior data, the relationship between accuracy and response delay of each intent category is analyzed, and an accuracy-delay curve is drawn. The inflection point is selected as the balance point. The threshold of consulting intent is set according to the balance point, and the routing decision score is compared with the threshold of the corresponding intent category. When the routing decision score ≥ the threshold, the target business module is obtained.

[0059] S5. Query the product knowledge graph through the target business module and combine it with user behavior data to generate personalized response content.

[0060] S5.1. Use the knowledge graph query interface according to the target business module to obtain initial product attribute data.

[0061] Furthermore, a SPARQL query statement is constructed based on the target business module, and the RDF endpoint interface of the e-commerce knowledge graph is accessed through the HTTP protocol. The query statement contains the product ID and the required attribute fields (such as material composition and washing method). The knowledge graph returns the initial product attribute data in JSON-LD format. The data fields strictly follow the Schema.org vocabulary specification. During the query process, the product ID is verified for validity, and the returned data is checked for integrity to obtain the initial product attribute data.

[0062] S5.2. Extract recent behavior data from user behavior data and output user portrait feature vector.

[0063] Furthermore, the behavioral records of the last 30 days (including browsed product categories, purchase records, and customer service consultation types) are extracted from the user behavior database. The behavioral keyword weights (for example, pure cotton has a weight of 0.12, and return has a weight of 0.08) are calculated using the TF-IDF algorithm. Combined with the time decay factor (daily decay coefficient of 0.95), weighted behavioral features are generated. Numerical features (for example, 3 returns in the past 7 days) are Z-score standardized, and categorical features (for example, preferred material type) are One-Hot encoded to output the user portrait feature vector.

[0064] S5.3. Input product attribute data and user portrait feature vectors into the strategy engine and generate response strategies through rule templates.

[0065] Furthermore, the product attribute data and user portrait feature vectors are input into the policy engine based on the Drools rule engine to match the predefined business rule templates. During the execution of the rule engine, high-priority rules are matched first, and the conflicting rules are handled using the most recently activated principle to generate a response strategy.

[0066] S5.4. Based on the response logic, adapted content is called from the multimedia resource library to form original response materials, and the original response materials are personalized sorted and cut to generate personalized response content.

[0067] Furthermore, based on the basic reply content and additional service information fields in the response strategy, matching graphic, text, and video materials are retrieved from the multimedia resource library. Multiple conditions are combined through Elasticsearch's Boolean query to accurately match resources, and the original response material set is output. The materials are sorted based on the preference weights in the user portrait feature vector, and the top N items (for example, N=3 on mobile devices) are intercepted to generate personalized response content.

[0068] S6. Adapt the personalized response content to a cross-platform format to form a customer service response.

[0069] S6.1. Parse the text in the personalized response content, match the UI adaptation template according to the terminal type, and obtain rendering rules.

[0070] Furthermore, the text fields and multimedia resource identifiers in the personalized response content are parsed, the terminal type is determined through User-Agent detection, and the UI adaptation template that matches the current terminal type and content type is queried from the UI template library. The UI adaptation template query conditions include the terminal resolution range, content presentation form, and interactive control requirements, and structured rendering rules are output.

[0071] S6.2. Convert the personalized response content according to the rendering rules to obtain an optimized content package, embed the session tracking identifier, and form a customer service response.

[0072] Furthermore, the text and multimedia resources in the personalized response content are formatted according to rendering rules, the text content is reformatted according to the specified font size and layout position, the multimedia resources are converted into an encoding format that adapts to the terminal resolution, and the interactive controls add platform-specific event bindings. After the conversion is completed, an optimized content package containing text, media, and metadata is generated, and then a session tracking identifier is embedded in the content package header to form a customer service response that complies with the HTTP protocol.

[0073] This embodiment also provides an e-commerce customer service response system based on multi-model routing technology, including: The regular expression matching result module collects user input text and user behavior data, performs preliminary intent matching on user input text requests using the regular expression rule library in the e-commerce field, and obtains regular expression matching results; The precise intent recognition result module inputs the regular matching results and user request text into the BERT multimodal classification model to obtain precise intent recognition results; The target business module uses the regular matching results and the precise intent recognition results to calculate the routing decision score using a dynamic weight formula to obtain the target business module; The personalized response content module queries the product knowledge graph through the target business module and combines it with user behavior data to generate personalized response content; The customer service response module adapts the personalized response content to the cross-platform format to form a customer service response.

[0074] This embodiment also provides a computer device, which is suitable for the e-commerce customer service response method based on multi-model routing technology, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute computer-executable instructions to implement the e-commerce customer service response method based on multi-model routing technology proposed in the above embodiment.

[0075] The computer device may be a terminal, comprising a processor, memory, a communication interface, a display, and an input device connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores an operating system and computer programs. The internal memory provides an environment for the operating system and computer programs stored in the non-volatile storage media. The communication interface of the computer device is used to communicate with external terminals via wired or wireless communication. Wireless communication may be achieved via Wi-Fi, a carrier network, NFC (near-field communication), or other technologies. The display of the computer device may be a liquid crystal display or an electronic ink display. The input device may be a touchscreen overlay on the display, buttons, a trackball, or a touchpad on the computer device housing, or an external keyboard, touchpad, or mouse.

[0076] This embodiment also provides a storage medium having a computer program stored thereon. When the program is executed by a processor, the program implements the e-commerce customer service response method based on multi-model routing technology as proposed in the above embodiment. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0077] In summary, the present invention calculates the routing decision score by using a dynamic weight formula through regular matching results and precise intent recognition results, queries the product knowledge graph through the target business module, and generates personalized response content in combination with user behavior data. The routing decision score is calculated by the dynamic weight formula, which realizes the dynamic fusion of rule matching and deep learning results, and adaptively adjusts the weight distribution according to the real-time session state, so that the routing decision retains the high efficiency of rule matching and the semantic understanding depth of model recognition, achieving the effect of accurately matching the needs of business scenarios, and generates personalized response content through the collaborative processing of knowledge graph and user behavior data, which realizes the deep correlation analysis of product attributes and user portraits, and dynamically combines multimodal response materials based on the strategy engine, so that the generated customer service answers ensure both professional accuracy and meet the user's personalized preferences, which significantly improves the service quality and user experience of e-commerce customer service.

[0078] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An e-commerce customer service response method based on multi-model routing technology, characterized by: include, Collect user input text and user behavior data, perform preliminary intent matching on user input text requests using the regular expression rule library in the e-commerce field, and obtain regular matching results; Input the regular matching results and user request text into the BERT multimodal classification model to obtain accurate intent recognition results; The regular matching results and precise intent recognition results are used to calculate the routing decision score using a dynamic weight formula to obtain the target business module; Generate personalized response content by querying the product knowledge graph through the target business module and combining it with user behavior data; Adapt the personalized response content to cross-platform formats to form customer service responses.

2. The e-commerce customer service response method based on multi-model routing technology according to claim 1, characterized in that: Collect user input text and user behavior data, The following steps are included: Receive the original text request input by the user, perform basic cleaning, and generate standardized text; The behavior log database of the e-commerce platform is called through the user ID to obtain user behavior data.

3. The e-commerce customer service response method based on multi-model routing technology according to claim 2, characterized in that: The regular expression rule library in the e-commerce field is used to perform preliminary intent matching on the user input text request to obtain regular matching results. The following steps are included: Compare the user input text request with the regular template of the e-commerce regular expression rule library to match the keyword combination; According to the matched keyword combination, the confidence of the current regular match is calculated to obtain the regular match result.

4. The e-commerce customer service response method based on multi-model routing technology according to claim 3, characterized in that: Input the regular matching results and user request text into the BERT multimodal classification model to obtain accurate intent recognition results, including the following steps: Combine the regular expression matching results and the user request text to generate a multimodal input text sequence; Perform BERT standard preprocessing on the multimodal input text sequence to generate text vectors; Load the BERT base model through PyTorch, add a multimodal fusion layer, and output the improved BERT base model architecture; Use e-commerce domain corpus to pre-train the improved BERT basic model architecture for domain adaptation, output the BERT initial model, encode the regular matching results and user behavior data into feature vectors, and output a multimodal training dataset; The multimodal training dataset is input into the BERT initial model, fine-tuned using the joint loss function, and the BERT multimodal classification model is output; The text vector is input into the BERT multimodal classification model to obtain accurate intent recognition results.

5. The e-commerce customer service response method based on multi-model routing technology according to claim 4, characterized in that: The regular matching results and the precise intent recognition results are used to calculate the routing decision score using a dynamic weight formula to obtain the target business module. The following steps are included: Extract the intent type and confidence level from the regular matching results, as well as the BERT intent score and context coherence score from the precise intent recognition results, to form structured routing input data. Analyze user behavior data to obtain session status, dynamically calculate weight coefficient based on session status, and obtain dynamic weight; The confidence score and BERT intent score are combined according to dynamic weights to calculate the routing decision score. The accuracy-response delay balance point of different intent categories is extracted from user behavior data, a consultation intent threshold is set, and the routing decision score is compared with the consultation intent threshold to obtain the target business module.

6. The e-commerce customer service response method based on multi-model routing technology according to claim 5, characterized in that: Generate personalized response content by querying product knowledge graphs through target business modules and combining them with user behavior data. The following steps are included: Use the knowledge graph query interface according to the target business module to obtain initial product attribute data; Extract recent behavior data from user behavior data and output user portrait feature vectors; Input product attribute data and user portrait feature vectors into the strategy engine and generate response strategies through rule templates; According to the response logic, adapted content is called from the multimedia resource library to form original response materials, which are then personalized and sorted to generate personalized response content.

7. The e-commerce customer service response method based on multi-model routing technology according to claim 6, characterized in that: Adapt the personalized response content to the cross-platform format to form a customer service response, including the following steps: Parse the text in the personalized response content, match the UI adaptation template according to the terminal type, and obtain rendering rules; The personalized response content is converted according to the rendering rules to obtain an optimized content package, which is embedded with a session tracking identifier to form a customer service response.

8. An e-commerce customer service response system based on multi-model routing technology, based on the e-commerce customer service response method based on multi-model routing technology according to any one of claims 1 to 7, characterized in that: include, The regular expression matching result module collects user input text and user behavior data, performs preliminary intent matching on user input text requests using the regular expression rule library in the e-commerce field, and obtains regular expression matching results; The precise intent recognition result module inputs the regular matching results and user request text into the BERT multimodal classification model to obtain precise intent recognition results; The target business module uses the regular matching results and the precise intent recognition results to calculate the routing decision score using a dynamic weight formula to obtain the target business module; The personalized response content module queries the product knowledge graph through the target business module and combines it with user behavior data to generate personalized response content; The customer service response module adapts the personalized response content to the cross-platform format to form a customer service response.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the e-commerce customer service response method based on multi-model routing technology described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the e-commerce customer service response method based on multi-model routing technology described in any one of claims 1 to 7 are implemented.

Images