Interaction method and device of chat interface based on big data corpus, and medium

Abstract

The invention relates to the field of artificial intelligence and customer service systems, and discloses a chat interface interaction method and device based on a big data corpus and a medium, and the method comprises the following steps: integrating a plurality of chat tools, and realizing tool adsorption and message management through a unified interface; constructing a big data corpus, and carrying out cleaning, word segmentation, semantic vectorization and classification on corpora; performing semantic analysis on the message input by the user to generate a semantic feature vector; comprehensive matching cost is calculated based on semantic matching in combination with dynamic business data; factors such as semantic correlation, business consistency, response time and historical satisfaction are comprehensively considered, and the optimal verbal skill is recommended; and presenting a recommendation result to the user, and quickly sending the message through the integrated chat tool. According to the method, unified management of multi-platform messages can be realized, efficient and accurate dynamic verbal skill recommendation is provided, intelligent customer service requirements in complex scenes are met, and the customer service working efficiency and the user satisfaction are effectively improved.

Description

Technical Field

[0001] The present invention relates to the technical field of artificial intelligence and customer service systems, and specifically to an interactive method, device and medium for a chat interface based on a big data corpus. Background Art

[0002] With the rapid development of e-commerce, social networking and enterprise collaboration scenarios, chat tools have gradually become the core means of customer service communication. However, in the existing technology, customer service personnel often need to use multiple chat tools (such as WeChat, DingTalk, WeChat for Business, QQ, etc.) to communicate with customers at the same time. The switching and operation between different platforms are complicated, which not only increases the operational burden of customer service, but also easily leads to message omissions or response delays, seriously affecting work efficiency and customer experience.

[0003] In addition, in order to improve the response speed and quality, customer service usually uses preset speech templates. However, traditional template management systems mostly store and call speech content in a static manner, making it difficult to dynamically generate recommendation results based on real-time business needs. When faced with complex scenarios (such as after-sales service, order inquiries, or logistics issues), this fixed speech cannot meet the customer's needs for personalization and timeliness. At the same time, the semantic matching capabilities of the existing technology are limited, relying only on simple keyword retrieval or rough semantic analysis, and cannot fully understand the true intention of the user's input, resulting in the speech recommendation results not being consistent with the actual situation of customer needs.

[0004] Furthermore, when recommending words, the customer service system usually lacks effective linkage with business dynamic data (such as order status, logistics information, etc.), and cannot generate the best response content based on the customer's current actual problem. In addition, the recommendation results usually ignore the optimization of multiple factors such as response time and historical satisfaction, resulting in unreasonable priority of recommended content, affecting the efficiency and accuracy of the overall interaction.

[0005] Therefore, the existing technology has significant deficiencies in the integration of multiple chat tools, intelligent speech recommendation, dynamic business data combination and multi-objective optimization, and it is difficult to meet the needs of high efficiency and high accuracy in modern customer service scenarios. To this end, there is an urgent need for a system and method that can integrate multiple chat tools, combine corpus intelligent matching technology and dynamic business data analysis, and provide real-time, accurate and optimized speech recommendation to solve the above technical problems. Summary of the invention

[0006] In view of the shortcomings of the prior art, the present invention provides an interactive method, device and medium for a chat interface based on a big data corpus, which solves the limitations of integration and management of multiple chat tools, recommendation of static speech templates and the inability to combine dynamic business data for accurate speech matching.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solutions: an interactive method of a chat interface based on a big data corpus, comprising the following steps: Integrate multiple chat tools and realize adsorption, message receiving and sending of multiple chat tools through a unified management interface; Build a big data corpus, preprocess the corpus, and classify it according to semantic features; Obtain the user's input message, use the pre-trained language model to perform semantic analysis on the user's input message, and generate a semantic feature vector of the user's input; Calculate the cost value of semantic matching based on the semantic feature vector input by the user and the semantic vector of the speech template stored in the corpus; Perform multi-objective optimization based on semantic matching and recommend the best words; The recommended optimal words of speech are presented to users, and messages can be quickly sent through the integrated chat tool interface.

[0008] Preferably, the step of integrating multiple chat tools includes: For chat tools that support open interfaces, the official open API is called to receive, send, and adsorb messages. For chat tools that do not support open interfaces, the reading and sending of messages are accomplished by simulating user operations, including simulating keystrokes, capturing window events, and locating screen coordinates. Dynamically adsorb and arrange each chat tool window to form a unified message management interface.

[0009] Preferably, the step of constructing a big data corpus includes: Collect historical chat records from multiple chat tools, industry FAQ documents, and templates uploaded by users; Perform word segmentation on the corpus, remove invalid data, and clean and standardize the content; Use the pre-trained language model to perform semantic vectorization on the corpus, generate a high-dimensional semantic feature vector, and cluster and classify the speech templates in the corpus based on the semantic features.

[0010] Preferably, the step of generating a semantic feature vector input by the user comprises: Use the pre-trained language model to convert user input messages into semantic vectors, and perform high-dimensional vector representation of the overall semantics of the message; Using named entity recognition technology to extract specific keywords or phrases in user input, the keywords include product names, time information, price information, and other entities; Based on the context information input by the user, the semantic representation of the current input is dynamically adjusted so that it can contain contextual information.

[0011] Preferably, the step of calculating the cost value of comprehensive matching based on the semantic feature vector input by the user and the semantic vector of the speech template stored in the corpus, combined with dynamic business data, includes: Calculate the similarity between the semantic vector of the user input message and the semantic vector of each speech template in the corpus; By measuring the similarity between semantic vectors, the semantic matching cost between user input and each speech template is determined; Combine dynamic business data to score the logical consistency of the technical template to form a business consistency score; Based on the semantic matching cost and the business consistency score, the comprehensive matching cost is calculated, where the semantic matching cost is calculated based on the cosine similarity between vectors or the distance between vectors.

[0012] Preferably, the step of performing multi-objective optimization based on semantic matching includes: Generate business consistency scores for speech templates based on semantic matching cost values ​​entered by users and combined with dynamic business data; Further comprehensively consider the response time cost of the speech template and optimize it according to the priority of response time; Combine the historical usage data of the speech template and adjust the historical satisfaction score weight according to the user feedback score; The optimal speech template is recommended based on a comprehensive evaluation of semantic matching cost, business consistency score, response time cost, and historical satisfaction score.

[0013] Preferably, the step of presenting the recommended optimal speech to the user and completing the quick sending of the message through the integrated chat tool interface includes: Sort the recommended words according to the priority of comprehensive cost and display them in the user interface in the form of a list; Support users to edit the recommended words again, including modifying the text content, adjusting the graphic format, and adding hyperlinks; provide a one-click sending function, which can send the selected words directly to the corresponding chat tool window by calling the adapter interface of multiple chat tools or simulating operations; For speech that contains pictures or other rich media content, it supports the display and sending of mixed text and pictures.

[0014] Preferably, the recommended optimal speech supports multiple types of presentation methods, including: Text form; Mixed text and image format; Hyperlink embed form.

[0015] The present invention also provides an interactive device for a chat interface based on a big data corpus, comprising: Multi-tool integration module, used to absorb multiple chat tool windows to achieve unified reception and sending of messages; The corpus management module is used to collect and process corpora from multiple sources, vectorize their semantics, and store them in categories; Semantic analysis module, used to perform semantic vectorization processing on messages input by users; The speech recommendation module is used to recommend the best speech based on the semantic matching and multi-objective optimization between the user input message and the speech templates in the corpus; Output module for displaying recommended words and sending messages through the integrated chat tool.

[0016] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the above method is implemented.

[0017] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the method described above is implemented.

[0018] The present invention provides an interactive method, device and medium for a chat interface based on a big data corpus, which has the following beneficial effects: 1. The present invention realizes the adsorption and message management of multiple chat tools such as WeChat, DingTalk, Enterprise WeChat, QQ, etc. through a unified management interface. It not only supports the centralized reception and sending of messages, but also solves the problem of frequent switching between multiple platforms through the multi-opening function and tool adaptation module. The integrated interface simplifies the operation process of customer service and greatly improves work efficiency.

[0019] 2. The present invention combines corpus semantic matching with dynamic business data to achieve dynamic matching recommendation of speech templates. By introducing dynamic business data (such as order status, logistics information, etc.), the accuracy of recommendations is improved, ensuring that the recommendation results are highly consistent with the actual needs of users, overcoming the problem that traditional static templates cannot meet real-time business needs.

[0020] 3. The present invention adopts a multi-objective optimization model, comprehensively considers factors such as semantic matching, business consistency, response time, historical satisfaction, etc., and constructs a flexible and adjustable comprehensive cost model. The model can dynamically adjust the weight parameters according to the scene requirements, thereby realizing the intelligent recommendation of the optimal speech, meeting the complex needs of different business scenarios.

[0021] 4. The present invention supports mixed text and image layout, dynamic data binding and user secondary editing when recommending words, which improves the flexibility and operability of the recommendation results. At the same time, through the design of highlighting key fields and priority prompts, it helps users quickly identify and select appropriate words, optimizing the user interaction experience.

[0022] 5. The present invention realizes seamless message sending for different chat tools by calling open interfaces or simulating user operations. For tools that do not support open interfaces, automated operation technology is used to complete the rapid sending of messages, ensuring the compatibility of multiple platforms. At the same time, the built-in message verification and retry mechanism improves the reliability of sending.

[0023] 6. The present invention significantly reduces the duplication of customer service and optimizes the operation process by efficiently integrating multiple tools, intelligent speech recommendations and fast message sending. At the same time, it improves user experience, user satisfaction and business processing efficiency through intelligent recommendations and high-precision replies. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] Figure 1 One of the schematic flow charts of the method of the present invention; Figure 2 This is the second schematic diagram of the method flow of the present invention; Figure 3 It is a schematic diagram of the computer device structure of the present invention.

[0025] Among them, 100, multi-tool integration module; 200, corpus management module; 300, semantic analysis module; 400, speech recommendation module; 500, output module; 40, computer equipment; 41, processor; 42, memory; 43, storage medium. DETAILED DESCRIPTION

[0026] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0027] Please see attached Figure 1 The present invention provides an interactive method for a chat interface based on a big data corpus, which aims to achieve the integration of multiple chat tools and intelligent speech recommendation through corpus construction and semantic analysis technology, and improve the work efficiency of application scenarios such as e-commerce customer service. The following will combine the steps of the method proposed by the present invention to gradually explain its implementation principle and specific technical content in detail.

[0028] like Figure 1As shown, the interactive method of the chat interface based on the big data corpus may include the following steps: S1. Integrate multiple chat tools and realize adsorption, message receiving and sending of multiple chat tools through a unified management interface; S2. Build a big data corpus, preprocess the corpus, and classify it according to semantic features; S3, obtaining the user's input message, using the pre-trained language model to perform semantic analysis on the user's input message, and generating a semantic feature vector of the user's input; S4. Calculate the cost value of comprehensive matching based on the semantic feature vector input by the user and the semantic vector of the speech template stored in the corpus, combined with dynamic business data; S5. Perform multi-objective optimization based on semantic matching and recommend the best words; S6. Present the recommended optimal words to the user, and quickly send the message through the integrated chat tool interface.

[0029] For step S1, in this embodiment, based on the actual use requirements of multiple mainstream chat tools (such as WeChat, WeChat for Enterprises, DingTalk, QQ, Feishu, etc.), a chat tool integration implementation method is proposed, which realizes window adsorption, message reception and sending of multiple chat tools through a unified management interface. The following describes in detail the technical implementation of the present invention in step S1 from the aspects of tool adsorption, message reception and sending, interface adaptation, and multi-opening support.

[0030] In this embodiment, the first step of integrating multiple chat tools is to implement the adsorption operation of each chat tool window so that these tools can be centrally operated through a unified management interface.

[0031] Specifically, as an implementation method, for chat tools that support open API or SDK (such as DingTalk, Feishu, and Enterprise WeChat), the present invention preferentially calls the official API interface to manage the chat window. The window handle information of the tool is obtained through the API interface, and the system call (such as WinAPI of the Windows platform) is used to realize the adsorption and unified arrangement of the window.

[0032] In another possible implementation, for chat tools that do not provide an official API (such as WeChat, QQ), this embodiment implements the window adsorption function by capturing screen events and simulating operations. For example: Use system window monitoring technology (such as MacOS's Accessibility API or Windows's window event capture tool) to detect the chat tool's active window information in real time; By detecting features such as window title and window handle, the window of the target chat tool is located, and window management tools (such as AutoIt or PyWin32) are used to adsorb, arrange and adjust the window.

[0033] It should be noted that in this embodiment, a unique window identifier can be assigned to different chat tool windows for subsequent message management and tool calling. This design can ensure that when multiple tools are running at the same time, there will be no confusion in operation due to window conflicts.

[0034] In this embodiment, messages of multiple chat tools are received and sent through a unified management interface, ensuring unified processing and rapid response of messages.

[0035] Specifically, for a chat tool that supports an open interface, the present invention directly calls the interface to complete message sending and receiving. For example: Call the message monitoring interface provided by DingTalk or WeChat for Enterprise to capture the message content from the chat window in real time; Call the message sending interface to send the preset or manually entered content to the target chat object.

[0036] As an option, for chat tools that do not support open interfaces, this embodiment uses a method of simulating user operations to complete the reception and transmission of messages. The specific technical implementation includes: Use screen capture technology to read text content in chat windows in real time; Use OCR (Optical Character Recognition) technology to extract text from the captured message area and generate message data based on the extraction results; By simulating key input (such as using tools such as PyAutoGUI and Sikuli), the generated message or preset words are sent to the corresponding chat window.

[0037] It should be noted that in some embodiments, the reception and transmission of messages are synchronized with the window adsorption logic. For example, after the window adsorption is completed, the system will immediately initialize the message monitoring and sending module to ensure that the user can obtain the message in real time and reply.

[0038] In this embodiment, the unified management interface interacts with multiple chat tools through the adaptation module, ensuring that the operation interfaces of different chat tools can be uniformly packaged to facilitate subsequent function expansion and maintenance.

[0039] In a possible implementation, the interface adaptation modules are divided into two categories according to the characteristics of the tool: For open interface tools (such as Feishu and DingTalk), directly adapt their API interfaces and call functions such as message processing and window management through encapsulated modules; For tools with closed interfaces (such as WeChat and QQ), the adaptation module simulates user operations through scripting tools (such as AutoHotKey and PyWin32) to implement message sending and window management.

[0040] For example, for the WeChat tool, the adaptation module will simulate the following operation process: Detect the focus position of the chat window and switch it to the target chat window; Use key simulation technology to input the message content into the chat input box character by character; Simulate sending shortcut keys (such as the Enter key) to complete message sending.

[0041] It is understandable that the design of the adaptation module can be further expanded to support the compatibility requirements of new chat tools or different platforms (such as Windows, MacOS) in the future.

[0042] In this embodiment, in order to meet the needs of running multiple accounts at the same time, a multi-opening support function is provided, which allows users to manage multiple accounts in a unified interface by creating multiple independent chat tool instances.

[0043] Specifically, as an option, this embodiment implements the multi-opening function through sandbox technology (such as Sandboxie or Docker). Each chat tool instance runs in an independent sandbox environment to ensure that the data and operating environment between different instances are completely isolated. The creation logic of the sandbox environment includes: Allocate an independent temporary running directory for each instance to store account-related cache and configuration; Start multiple process instances of the chat tool, each instance is bound to a unique window identifier.

[0044] In another implementation, this embodiment uses virtual machine technology (such as VMware, VirtualBox) to create an independent virtual environment for each chat tool account to ensure that different accounts do not interfere with each other.

[0045] It should be noted that, in the process of implementing the multi-opening function, the adsorption and arrangement of windows can also be dynamically adjusted in this embodiment. For example: Arrange multiple instance windows of the same tool in a unified group according to user operation habits; The windows of different tools are displayed in partitions to ensure that users can quickly switch the target window.

[0046] As an operation example, the operation process of this embodiment can be described as follows: First, the user starts the unified management interface of the present invention, and the interface loads the chat tool list configured by the user through the adaptation module.

[0047] The system automatically detects the running status of each chat tool, and selects API call or simulation operation method according to the tool type to complete window adsorption.

[0048] After the adsorption is completed, the system initializes the message monitoring module to receive the message content from various chat tools in real time.

[0049] Users can directly select the target chat tool in the unified interface, enter or select preset content, and then complete the message sending operation through the adaptation module.

[0050] If the user needs to log in to multiple accounts at the same time, the system will start the multi-opening module, create an independent running instance for each account, and adsorb the instance windows to a unified interface.

[0051] This embodiment integrates the adsorption, message reception and sending of multiple chat tools through a unified management interface, thereby solving the problem of low efficiency caused by tool switching in the prior art.

[0052] For step S2, in this embodiment, in order to build a big data corpus, a series of data collection, cleaning, vectorization processing and classification steps are performed to form a structured corpus that meets the needs of intelligent speech recommendation. This step mainly includes the source of the corpus, the preprocessing method and the specific implementation of semantic classification, which will be described in detail one by one below.

[0053] In this embodiment, the sources of the big data corpus are diversified to ensure that the corpus covers a wide range and meets the needs of different chat scenarios.

[0054] As a choice, the main sources of the corpus include the following categories: Extract corpus from historical chat records of multiple chat tools (such as WeChat, DingTalk, QQ, etc.); Collect industry standard FAQ documents, such as common questions and standard answers on e-commerce platforms; The template scripts uploaded by users are customized to suit specific scenarios or industry requirements.

[0055] Specifically, the collection of historical chat records is achieved by calling the message export function of each chat tool or by screen capture technology. In a possible implementation, the system obtains a standardized question-and-answer data set from FAQ documents or e-commerce platforms through a distributed crawling tool or data interface.

[0056] It should be noted that in the management of user-defined speech templates, this embodiment supports users to upload templates through a unified interface, and these templates will be automatically added to the corpus and merged with the existing corpus.

[0057] In this embodiment, in order to ensure the high quality and consistency of the corpus data, the collected original corpus is preprocessed, including data cleaning, word segmentation and standardization operations.

[0058] During the data cleaning process, as an option, the system performs the following operations on the original corpus: Remove meaningless data such as blank messages, invalid characters (such as extra punctuation marks), and repeated sentences; For multilingual scenarios, unify the language format of corpus (such as conversion between simplified and traditional Chinese, and conversion between uppercase and lowercase English); Filter special symbols (such as "#" and "@") or noise data (such as error codes).

[0059] Specifically, the word segmentation process uses Chinese word segmentation tools (such as Jieba word segmentation or Spacy) to segment the text into words. In a possible implementation, for specific terms in the field of e-commerce customer service, this embodiment introduces a field-defined dictionary, such as "order number", "express logistics", "invoice" and other high-frequency words, to improve the accuracy of word segmentation.

[0060] It should be noted that in the data standardization process, this embodiment converts the text content into a unified format. For example, time-related sentences (such as "tomorrow" and "the day after tomorrow") are standardized to specific dates; and digital expressions (such as "ten yuan" and "￥10") are uniformly converted to digital formats.

[0061] In this embodiment, the preprocessed corpus is converted into a high-dimensional semantic feature representation through semantic vectorization technology so as to be used for semantic matching and recommendation in subsequent steps.

[0062] Specifically, semantic vectorization uses a pre-trained language model (such as BERT or GPT) to embed each corpus into a high-dimensional vector. i The embedding process can be expressed as: in, For corpus t i is the high-dimensional semantic vector of , and d is the dimension of the embedding space.

[0063] As an option, in order to improve the efficiency and accuracy of semantic vectorization, this embodiment combines the following technologies in the corpus embedding process: Use word embedding to capture the basic semantic relationship between words; Contextual modeling (such as the Transformer mechanism in BERT) is used to consider the contextual information of words in the corpus, thereby generating a more accurate sentence-level semantic representation.

[0064] It should be noted that when processing a long text corpus, this embodiment splits the text into multiple sentences through a sentence segmentation operation, and performs embedding processing on each sentence separately. The final semantic vector is fused by weighted averaging of sentence vectors.

[0065] In this embodiment, in order to facilitate subsequent semantic matching and recommendation, an unsupervised clustering algorithm is used to semantically classify the corpus in the corpus.

[0066] In a possible implementation, the classification process includes the following steps: Through the semantic vector of the corpus, the cosine similarity is used to measure the semantic similarity between different corpora; Based on the semantic similarity matrix, an unsupervised clustering algorithm (such as K-means or DBSCAN) is used to automatically classify the corpus.

[0067] Specifically, for the K-means algorithm, the goal of clustering is to cluster n pieces of data {V t1 ,V t2 ,…,V tn} into k categories, so that the semantic similarity of the corpus within the same category is maximized, while the semantic similarity of the corpus between different categories is minimized. Its objective function is: Among them, μ i is the center of the ith cluster, C i is the corpus set of the i-th cluster.

[0068] As an option, to improve the interpretability of the classification results, this embodiment generates keyword labels for each category. These labels are based on the category center point μ i The most important feature words are extracted, such as "after-sales service" and "logistics inquiry".

[0069] It should be noted that after the classification is completed, the system will dynamically maintain the classification results and allow the category boundaries to be automatically adjusted when new corpus is added.

[0070] As an operation example, the operation process of this embodiment includes the following contents: The system extracts 1 million pieces of data from WeChat and DingTalk chat history; These corpora were cleaned and 850,000 records were retained after removing invalid data; Use the BERT model to perform semantic vectorization on each corpus and map it to a 768-dimensional embedding space; The K-means algorithm was used to divide the corpus into 10 categories and generate keyword tags; The classified corpus is stored in a distributed database for use by subsequent semantic matching modules.

[0071] This embodiment constructs a big data corpus that meets the requirements of the present invention through preprocessing, vectorization and semantic classification, ensuring that the quality and organizational structure of the corpus data can support semantic matching and recommendation operations in subsequent steps.

[0072] For step S3, in this embodiment, for the message input by the user, by combining the pre-trained language model with the semantic feature extraction technology, the input text is converted into a high-dimensional semantic vector, providing standardized input data for the subsequent semantic matching step. The following describes the technical content in detail from aspects such as message reception, semantic vector generation, named entity extraction and context fusion.

[0073] In this embodiment, the message input by the user is received in real time through the unified management interface described in step S1. It should be noted that the message input by the user can come from multiple chat tools, including WeChat, DingTalk, WeChat for Enterprise, etc.

[0074] Specifically, when a user inputs and sends a message in the chat tool window, the system captures the input content through the message monitoring module and passes it to the semantic analysis module as a string.

[0075] As an option, if the user's message contains multiple paragraphs or sentences, the system will segment the message according to the sentence segmentation rules in natural language processing (such as punctuation splitting rules) and perform subsequent processing on each paragraph separately.

[0076] In this embodiment, a pre-trained language model (such as BERT, GPT) is used to semantically vectorize the message input by the user to generate a high-dimensional semantic feature vector. It should be noted that the main purpose of semantic vectorization is to map the input message to a high-dimensional embedding space for subsequent semantic matching.

[0077] Specifically, in a possible implementation, the process of generating a semantic vector of the user input message uuu includes the following steps: First, the system decomposes the message u into a word sequence {w1,w2,…,w n}, where n is the number of words in the message; Then, the pre-trained language model is used to train each word w i Perform word embedding and generate word vector v i ; Finally, through the context modeling capability of the language model (such as the Transformer structure in BERT), the context information of all words in the message is combined to generate the semantic vector V of the complete message. u : Vu =f context ({v1,v2,…,v n}) Among them, f context Represents the context encoding function of the language model.

[0078] In one possible implementation, a 768-dimensional embedding space is used to represent the semantic vector, i.e.

[0079] It should be noted that the semantic vector generation process can select different language models according to specific application scenarios. For example, in general scenarios, the BERT model can be selected; in scenarios targeting the e-commerce field, a pre-trained model that has been fine-tuned with e-commerce data (such as ERNIE, Roberta) can be selected.

[0080] In this embodiment, in order to further improve the accuracy of semantic representation, named entity recognition (NER) technology is combined to extract key entity information from the message input by the user and embed it into the semantic vector.

[0081] Specifically, the main goal of named entity extraction is to identify specific entities contained in the message, such as product name, time information, price information, etc. For example, for the input message "I want to check whether the goods I bought yesterday have been shipped", the system can extract the following entities: Time entity: "yesterday"; Action Entity: "query"; Target entity: "Product".

[0082] As an implementation method, the implementation process of NER includes: Perform word segmentation and part-of-speech tagging on messages input by users; Use a sequence labeling based deep learning model (such as BiLSTM-CRF) to identify named entities and output entity categories and scopes; The extracted entity information is encoded into a feature vector and fused with the overall semantic vector of the message.

[0083] In an exemplary operation, for the message “Is order number 123456 shipped?”, the system extracts “order number 123456” as a key entity through NER and appends it to the semantic vector V u In this paper, an enhanced semantic vector is formed.

[0084] It should be noted that the goal of named entity extraction is to retain the key information in the input message, thereby improving the accuracy of semantic matching, especially when it comes to scenarios such as goods and orders.

[0085] In this embodiment, when processing multiple rounds of dialogue or continuous input, the semantics of the user input message may depend on context information. Therefore, the system dynamically adjusts the semantic representation of the current input through context fusion technology.

[0086] In a possible implementation, the context fusion process includes the following: First, the system maintains the historical input sequence {u t-1 ,u t-2 ,…,u t-n}, where t is the current time step; then, the semantic vector of the historical input Perform weighted average or attention mechanism calculation to generate context representation V context ; Finally, the context representation is combined with the semantic vector V of the current input message u Fusion, generate the final semantic representation V u ′ : V u ′ =Concat(V u ,V context ) or V u ′ =αV u +(1-α)V context Among them, α is the fusion coefficient, which can be obtained through data learning.

[0087] As an option, for scenarios with less contextual information (such as single-round question and answer), the system can omit the context fusion step and directly use the current input semantic vector for subsequent processing.

[0088] In an example, assuming that the user inputs a message "When will the goods be shipped?", the system processes the message as follows: First, the system splits the message into the word sequence "when" and "shipping" through word segmentation; Use the pre-trained language model to encode the word sequence and generate a semantic vector V u ; Combine named entity recognition technology to extract "shipping" as the action entity; If the user has previously entered "I placed an order yesterday", the system will fuse its contextual information with the current input to generate an enhanced semantic vector V u ′ .

[0089] Finally, the system will V u ′As output, it is used by subsequent steps.

[0090] This embodiment can fully mine the semantic features of user input messages through semantic vector generation, named entity extraction and context fusion technology, and ensure that the generated semantic vector has high accuracy and context relevance.

[0091] For step S4, in this embodiment, the semantic matching technology is combined with the query of dynamic business data to calculate the matching cost of the semantic feature vector of the user input message and the semantic vector of the speech template in the corpus, ensuring that the generated recommendation results can not only reflect the semantic relevance, but also maintain consistency with the business logic. The following describes the specific implementation in detail from three aspects: the definition of matching cost, the combination of dynamic business data and the optimization of comprehensive cost.

[0092] In this embodiment, the semantic feature vector of the user input message is represented by V u Indicates that the semantic feature vector of each speech template in the corpus is V ti Indicates that the matching cost is a measure of V u and V ti The smaller the matching cost, the higher the semantic relevance between the input message and the template.

[0093] As an option, the matching cost can be defined by cosine similarity. Specifically, cosine similarity measures the similarity of the directions of two vectors, and the larger the value, the closer the two vectors are, so it can be converted into a matching cost: in, <V u ,V ti > indicates V u and V ti The dot product of ∥V u ∥ and ∥V ti ∥ represents the Euclidean norm of two vectors respectively.

[0094] For the convenience of expressing the cost, the matching cost is defined as: c(V u ,V ti )=1-CosineSim(V u ,V ti ) In another possible implementation, the matching cost can also be calculated by the Euclidean distance, which is defined as: It should be noted that cosine similarity and Euclidean distance have their own advantages in different scenarios and can be selected according to actual needs.

[0095] In this embodiment, in order to improve the relevance of the matching results to the actual business needs, the query and judgment logic of dynamic business data is combined. The introduction of dynamic business data mainly solves the problem that fixed templates cannot reflect the real-time business status.

[0096] Specifically, when the user input message contains business-related content (such as orders, logistics or after-sales service), the system will extract key information through named entity recognition (NER) technology and retrieve dynamic business status through query interfaces or databases. For example: For the input "When will the order be shipped?", the system can extract "order number" or "user account" as the query key field; Query the business system to obtain the current status information of the order (such as "shipped", "pending shipment" or "processing"); The returned dynamic business data will be combined with the semantic matching process to dynamically adjust the recommended matching results.

[0097] In a possible implementation, the logic of dynamic business data query is as follows: 1. Use the entity information (such as order number, logistics order number, etc.) input by the user to construct a query request; 2. Query the backend database or call the external interface to obtain dynamic status data; 3. Logically compare the returned status data with the corpus template to generate a business consistency score B(t i ).

[0098] It should be noted that the calculation of the business consistency score is based on the degree of match between the template content and the dynamic business status. For example: If the template contains "Order shipped" and the dynamic status is "Shipped", a high score is given; If the template content does not match the status (for example, the template is "order shipped", but the actual status is "pending shipment"), the score will be lowered or even excluded.

[0099] In this embodiment, business consistency scoring is introduced on the basis of semantic matching cost to construct a new comprehensive matching cost model. The definition of comprehensive cost is: in: Semantic matching cost, which measures the semantic relevance between the input and the template; B(t i ): Business consistency score, which measures the logical consistency between the template and the dynamic business status; α, β: weight parameters used to balance the priority of semantic matching and business logic.

[0100] As an option, the business consistency score B(t i ) can be calculated based on predefined rules or machine learning models. For example: If the template content directly matches the dynamic data (e.g. the "Order Shipped" template is consistent with the "Shipped" status), then B(t i ) takes a higher value; If there is a logical conflict between the template content and the dynamic data, then B(t i ) takes a lower value or even 0.

[0101] In one possible implementation, the system dynamically adjusts the weight parameters α and β according to the business scenario. For example, in an order query scenario, the weight of the business consistency score is increased to prioritize the recommendation of templates that meet the order status.

[0102] In one example, assuming that the user inputs a message: "When will the order be shipped?", the processing is as follows: 1. Semantic analysis: The system converts the input message "When will the order be shipped?" into a semantic vector V u .

[0103] 2. Dynamic business data query: Extract the "order number" in the message as the key entity and obtain the order status through database query.

[0104] The query result returned is: "Order Status: Shipped".

[0105] 3. Semantic matching: According to the semantic vector V u With corpus template Calculating semantic matching cost

[0106] 4. Business consistency score: Based on the dynamic state of the query, the template "Your order has been shipped" is assigned a high business consistency score B(t i ).

[0107] 5. Comprehensive cost optimization: Combine the semantic matching cost and business consistency score to calculate the comprehensive cost C total .

[0108] After sorting, the recommended template is "Your order has been shipped, the logistics order number is 123456, please check it later."

[0109] This embodiment combines semantic matching with dynamic business data query to achieve the accuracy of template recommendation results and consistency of business logic. Especially in dynamic state-driven scenarios (such as order delivery and logistics query), it can generate responses that are more in line with user needs based on real-time data, significantly improving the system's intelligence level and user satisfaction.

[0110] For step S5, in this embodiment, based on the semantic matching cost value between the user input and the speech template in the corpus, a multi-objective optimization strategy is introduced to comprehensively calculate and recommend the optimal speech template by combining the score of dynamic business data, historical satisfaction and response time. The specific implementation method will be described in detail below.

[0111] In this embodiment, in order to achieve multi-objective optimization, the semantic matching cost, the logical consistency score of dynamic business data and other factors (such as response time and historical satisfaction) are comprehensively considered to define a comprehensive cost function C total Specifically, the comprehensive cost function is in the form of: in: Semantic matching cost, which measures the relevance between user input and template semantics; B(t i ): Business consistency score, which measures the consistency between template scripts and dynamic business data; T(t i ): Response time cost, which is used to indicate the estimated processing time of the speech template; S(t i ): Historical satisfaction score, indicating the customer satisfaction of the script template in historical use; α, β, γ, δ: weight coefficients, which are adjusted dynamically according to the specific scenario requirements.

[0112] It should be noted that the design of the comprehensive cost function can be adjusted according to the specific application scenario. For example, in a scenario with high real-time requirements, the weight of γ can be increased to give priority to templates with shorter response time.

[0113] In this embodiment, the semantic matching cost It is the core part of the comprehensive cost function and directly affects the accuracy of the matching results.

[0114] As an option, in order to improve the recommendation accuracy, the system normalizes the semantic matching cost. The normalization formula is: in, Represents the set of matching costs between user input and all templates.

[0115] It should be noted that the normalized semantic matching cost value It is normalized to the interval [0,1] to facilitate uniform weighting with other objectives.

[0116] In this embodiment, the service consistency score B(t i ) is used to measure whether the template content is consistent with the dynamic business data logic. Its calculation logic is based on the degree of match between the template and the real-time data.

[0117] Specifically, in one possible implementation, the value of the business consistency score depends on the following rules: If the template content completely matches the dynamic business data, for example, the template "Order Shipped" is consistent with the queried order status "Shipped", then the value B(t i )=1; If the template content partially matches the dynamic business data, for example, the template "order in progress" matches the status "to be shipped", then the value B(t i )∈(0,1); If the template content does not match the dynamic business data, then B(t i )=0.

[0118] It should be noted that the business consistency score is generated by combining logical judgment and semantic analysis, which can effectively improve the business relevance of the recommended template.

[0119] In this embodiment, the response time cost T(t i ) indicates the estimated delay time when sending or processing a speech template. For example, a template with additional query operations (such as database retrieval or API call) has a higher response time cost.

[0120] Alternatively, the response time cost is calculated as: Among them, T(t i ) is normalized to the interval [0,1], and smaller values ​​indicate shorter response times.

[0121] Historical satisfaction score S(t i ) indicates the customer satisfaction of the template in historical use. The specific score is obtained through user feedback data or click-through rate statistics. For example: Templates with high satisfaction (such as a historical favorable rating of more than 90%) are assigned a value of S(t i )=0.9; Templates with lower satisfaction (such as higher complaint rate) are assigned S(t i )<0.5.

[0122] It should be noted that the satisfaction score is used as a penalty factor in the comprehensive cost function through 1-S(t i ), that is, the lower the satisfaction level of the template, the higher the comprehensive cost value.

[0123] In this embodiment, the comprehensive cost value C of each template is calculated. total , select k templates with the smallest comprehensive cost as the recommendation results.

[0124] In a possible implementation, the recommendation process includes the following steps: According to the comprehensive cost formula C total Calculate the cost values ​​of all templates; Sort by cost value from low to high; The top k ranked templates are selected as recommendation candidates and returned to the user.

[0125] For example, when the user enters "when will the order be shipped" and the order status is "shipped", the system calculates the comprehensive cost of the template as follows: Template 1: “Your order has been shipped”; C total =0.2×0.1+0.5×1+0.2×0.3+0.1×0.1=0.36 Template 2: “Your order will be shipped soon”; C total =0.2×0.3+0.5×0.7+0.2×0.4+0.1×0.2=0.45 Final recommended template 1.

[0126] In one example, the user inputs a message: "When will the order be shipped?" The operation process is as follows: The system calculates the semantic matching cost c(V u ,V ti ); The order status is "shipped", and the business consistency score B(t i ); Combine the template's response time and historical satisfaction to calculate the comprehensive cost value C of each template. total ; Sort by comprehensive cost value and recommend the template with the lowest comprehensive cost: "Your order has been shipped, the logistics order number is 123456".

[0127] This embodiment uses a multi-objective optimization strategy to comprehensively consider factors such as semantic relevance, business logic consistency, response time, and historical satisfaction to generate an optimal template that meets user needs.

[0128] For step S6, in this embodiment, after semantic matching and multi-objective optimization, the system presents the generated optimal speech recommendation to the user, and completes the rapid sending of messages through the integrated chat tool interface. This step focuses on the display of recommended speech, user interaction support, and the implementation process of final message sending.

[0129] In this embodiment, the recommended optimal speech words are presented to the user through a unified interface, and the display method supports list format and mixed text and image format.

[0130] As an option, the recommended optimal words are sorted by comprehensive cost and displayed in the interface. Specifically, the content of each recommended word displayed on the interface includes: The complete text content of the template; Related dynamic business data (such as order status, logistics information); Comprehensive cost scoring details (such as semantic matching cost, business consistency score).

[0131] In a possible implementation, the system provides the user with visual prompts for the recommended words, for example: Highlight key fields of business data (such as logistics order number, estimated delivery time); Use colors or icons to indicate the priority of recommended words (e.g. green for high priority).

[0132] It should be noted that the recommended optimal words are not limited to text display, but also support combination with multimedia content, such as product pictures, hyperlinks, etc., to meet the needs of different application scenarios.

[0133] In this embodiment, the user can re-edit the recommended words or send them directly.

[0134] Specifically, the system supports the following user interaction functions: Editing of the content of the recommended speech: Users can modify the content of the recommended speech according to actual needs, including adding or deleting text, adjusting the format, etc.; Quick template selection: users can select a recommended script as the final content to be sent by clicking; Adjustment of mixed text and image layout: For recommendation scripts containing images or hyperlinks, users can preview and edit their layout.

[0135] As a possible implementation method, after the user completes editing, the system will update the modified content to the sending preview area in real time, and enter the message sending stage after the user confirms.

[0136] It should be noted that the design of user interaction aims to improve the flexibility and adaptability of recommendation language, while ensuring that the content sent meets actual needs.

[0137] In this embodiment, the sending of the recommended words is completed through the integrated chat tool interface. The specific implementation includes message interface calling and automated operation support.

[0138] As an option, for chat tools that support open interfaces (such as DingTalk and WeChat for Enterprise), the system can quickly send messages by calling their message sending API. The specific parameters of the API call include: The identity of the target contact or group chat; Confirmed content to be sent (including text, images, hyperlinks, etc.); Receipt information of the sending status.

[0139] In another possible implementation, for chat tools that do not support open interfaces (such as WeChat and QQ), the system uses a method of simulating user operations to complete message sending. For example: Use window event capture technology to locate the target chat window; By simulating keystroke operations, the content is input into the input box of the chat window character by character; Simulate sending shortcut keys (such as the Enter key) to complete message sending.

[0140] It should be noted that this embodiment protects user privacy and data security during the sending process, for example, by storing and transmitting message content in a local encrypted manner to ensure that data is not accessed without authorization.

[0141] In this embodiment, in order to improve the efficiency and reliability of message sending, an optimization strategy for the sending process is designed.

[0142] As an option, the system will automatically verify the integrity and legality of the content before sending the message. For example: Check whether the content contains key fields (such as order number, logistics order number); Verify the validity of hyperlinks to avoid sending invalid or erroneous links.

[0143] In a possible implementation, the system also supports a retry mechanism for sending failures. For example: If the sending fails due to network reasons, the system will record the failure message and automatically retry after the network is restored; If the attempt fails after multiple attempts, the system will remind the user and provide the option to resend manually.

[0144] It should be noted that the optimized design of the sending process not only improves the stability of the system, but also reduces the error rate of users in operation.

[0145] In one running example, after the user enters the message "When will the order be shipped?", the system recommends the optimal phrase "Your order has been shipped, the logistics order number is 123456, please check it later", and completes the message sending through the following steps: The system highlights the recommended words in the recommendation interface and attaches the logistics order number; The user clicks on the content of the speech to preview it and chooses to send it directly; The system calls WeChat's simulation operation module and inputs the recommended words into the target chat window; The simulated key operation completes the message sending and returns a status prompt indicating successful sending.

[0146] This embodiment realizes efficient management and flexible adjustment of sent content through the presentation of recommended words and user interaction design. The implementation method of message sending not only supports the seamless connection of multiple chat tools, but also provides high reliability and security. It should be noted that the display and sending process of recommended words can be customized according to specific application scenarios to further enhance the practicality of the system and user experience.

[0147] In general, the present invention realizes intelligent and real-time customer service interaction functions through the integration of multiple chat tools, corpus construction and semantic vectorization, semantic matching combined with dynamic data, optimal speech recommendation based on multi-objective optimization, and efficient presentation and sending of recommendation results. The present invention takes the pre-trained language model as the core technology, combines dynamic business data and historical feedback to optimize the recommendation results, and supports the seamless integration of multiple chat tools through interface adaptation and automated operation, which significantly improves the efficiency and accuracy of the customer service system, and has a wide range of application scenarios and powerful expansion capabilities.

[0148] The interactive device of the chat interface based on the big data corpus described below and the interactive method of the chat interface based on the big data corpus described above can refer to each other.

[0149] Please see attached Figure 2 The present invention also provides an interactive device for a chat interface based on a big data corpus, comprising: The multi-tool integration module 100 is used to absorb multiple chat tool windows to achieve unified reception and sending of messages; The corpus management module 200 is used to collect and process corpora from multiple sources, vectorize their semantics and store them in categories; The semantic analysis module 300 is used to perform semantic vectorization processing on the message input by the user; The speech recommendation module 400 is used to recommend the optimal speech based on the semantic matching and multi-objective optimization between the user input message and the speech template in the corpus; The output module 500 is used to display the recommended words and send messages through the integrated chat tool.

[0150] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0151] Please see attached Figure 3 The present invention further provides a computer device 40, comprising: a processor 41 and a memory 42, wherein the memory 42 stores a computer program executable by the processor, and when the computer program is executed by the processor, the above method is executed.

[0152] The present invention further provides a storage medium 43 on which a computer program is stored. When the computer program is run by the processor 41, the above method is executed.

[0153] Among them, the storage medium 43 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 red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0154] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An interactive method for a chat interface based on a big data corpus, characterized in that: The following steps are involved: Integrate multiple chat tools and realize adsorption, message receiving and sending of multiple chat tools through a unified management interface; Build a big data corpus, preprocess the corpus, and classify it according to semantic features; Obtain the user's input message, use the pre-trained language model to perform semantic analysis on the user's input message, and generate a semantic feature vector of the user's input; Based on the semantic feature vector input by the user and the semantic vector of the speech template stored in the corpus, combined with dynamic business data, the cost value of comprehensive matching is calculated; Perform multi-objective optimization based on semantic matching and recommend the best words; The recommended optimal words of speech are presented to users, and messages can be quickly sent through the integrated chat tool interface.

2. The interactive method of the chat interface based on the big data corpus according to claim 1 is characterized in that: The steps of integrating multiple chat tools include: For chat tools that support open interfaces, the official open API is called to receive, send and adsorb messages. For chat tools that do not support open interfaces, the reading and sending of messages are completed by simulating user operations, including simulating keystrokes, capturing window events, and locating screen coordinates; Dynamically adsorb and arrange each chat tool window to form a unified message management interface.

3. The interactive method of the chat interface based on the big data corpus according to claim 1, characterized in that: The steps of constructing the big data corpus include: Collect historical chat records from multiple chat tools, industry FAQ documents, and templates uploaded by users; Perform word segmentation on the corpus, remove invalid data, and clean and standardize the content; Use the pre-trained language model to perform semantic vectorization on the corpus, generate a high-dimensional semantic feature vector, and cluster and classify the speech templates in the corpus based on the semantic features.

4. The interactive method of the chat interface based on the big data corpus according to claim 1 is characterized in that: The step of generating a semantic feature vector input by the user comprises: Use the pre-trained language model to convert user input messages into semantic vectors, and perform high-dimensional vector representation of the overall semantics of the message; Using named entity recognition technology to extract specific keywords or phrases in user input, the keywords include product names, time information, price information, and other entities; Based on the context information input by the user, the semantic representation of the current input is dynamically adjusted so that it can contain contextual information.

5. The interactive method of the chat interface based on the big data corpus according to claim 1 is characterized in that: The step of calculating the cost value of comprehensive matching based on the semantic feature vector input by the user and the semantic vector of the speech template stored in the corpus, combined with dynamic business data, includes: Calculate the similarity between the semantic vector of the user input message and the semantic vector of each speech template in the corpus; By measuring the similarity between semantic vectors, the semantic matching cost between user input and each speech template is determined; Combine dynamic business data to score the logical consistency of the technical template to form a business consistency score; Based on the semantic matching cost and the business consistency score, the comprehensive matching cost is calculated, where the semantic matching cost is calculated based on the cosine similarity between vectors or the distance between vectors.

6. The interactive method of chat interface based on big data corpus according to claim 1, characterized in that: The steps of performing multi-objective optimization based on semantic matching include: Generate business consistency scores for speech templates based on semantic matching cost values ​​entered by users and combined with dynamic business data; Further comprehensively consider the response time cost of the speech template and optimize it according to the priority of response time; Combine the historical usage data of the speech template and adjust the historical satisfaction score weight according to the user feedback score; The optimal speech template is recommended based on a comprehensive evaluation of semantic matching cost, business consistency score, response time cost and historical satisfaction score.

7. The interactive method of chat interface based on big data corpus according to claim 1, characterized in that: The step of presenting the recommended optimal speech to the user and completing the quick sending of the message through the integrated chat tool interface includes: Sort the recommended words according to the priority of comprehensive cost and display them in the user interface in the form of a list; Support users to edit the recommendation words, including modifying the text content, adjusting the graphic format, and adding hyperlinks; Provide a one-click sending function, which can directly send the selected words to the corresponding chat tool window by calling the adaptation interface of multiple chat tools or simulating operations; For speech that contains pictures or other rich media content, it supports the display and sending of mixed text and pictures.

8. The interactive method of the chat interface based on the big data corpus according to claim 7 is characterized in that: The recommended optimal speech supports multiple types of presentation methods, including: Text form; Mixed text and image format; Hyperlink embed form.

9. An interactive device for a chat interface based on a big data corpus, applied to the method according to any one of claims 1 to 8, characterized in that: include: Multi-tool integration module, used to absorb multiple chat tool windows to achieve unified reception and sending of messages; The corpus management module is used to collect and process corpora from multiple sources, vectorize their semantics, and store them in categories; Semantic analysis module, used to perform semantic vectorization processing on messages input by users; The speech recommendation module is used to recommend the best speech based on the semantic matching and multi-objective optimization between the user input message and the speech templates in the corpus; Output module for displaying recommended words and sending messages through the integrated chat tool.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 8 is implemented.

Images