Response processing method, device and equipment based on fusion information and storage medium thereof
By breaking down and fusing user inquiries, a unique fused intent is generated, and response category tags are predicted and response agents are assigned. This solves the problem of accurate user allocation in financial business and improves the business efficiency and service quality of response agents.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-11-07
- Publication Date
- 2026-04-03
AI Technical Summary
In scenarios involving complex financial transactions and a large number of users, existing technologies are unable to accurately assign target answering agents to users, resulting in blind and inefficient agent allocation.
By acquiring user inquiry statements, performing segmentation and intent recognition, generating unique fusion intent, predicting response classification labels, and assigning response agents according to their levels, precise agent allocation can be achieved using response processing methods and devices based on fusion information.
This improved the efficiency of contact with answering agents, ensuring that users' inquiries are accurately identified and targeted answering agents are assigned, avoiding blind allocation and enhancing service quality in financial business scenarios.
Smart Images
Figure CN121787429A_ABST
Abstract
Description
Technical Field
[0001] This application relates to the field of artificial intelligence technology and is applied in business consultation scenarios. Specifically, it relates to a scenario of intelligently responding to business consultations of target users and allocating seats to agents, and involves a response processing method, device, equipment and storage medium based on fused information. Background Technology
[0002] In the field of financial business consultation, customer service agents have to handle numerous lists every day, each requiring individual contact, which is time-consuming and labor-intensive. Currently, to reduce the workload of customer service agents, most financial institutions have adopted a number of customer service robots to replace them.
[0003] However, the current use of customer service robots to replace human agents has generated a series of unresolved issues. For example, customer service robots tend to be more mechanical in their responses, lacking the flexibility of human agents in addressing user needs. Furthermore, the user lists assigned to agents often rely on a conventional, workload-based allocation method, failing to assign users to agents who are compatible with their needs or understand their true business intentions. Therefore, in complex financial business scenarios with large user volumes, there remains a technical challenge in accurately assigning targeted agents to users. Summary of the Invention
[0004] The purpose of this application is to propose a response processing method, apparatus, device and storage medium based on fused information, so as to solve the technical problem that the existing technology still cannot accurately allocate target response seats to users in business scenarios with complex financial business and a large number of users.
[0005] Firstly, embodiments of this application provide a response processing method based on fused information, which adopts the following technical solution: Response processing methods based on fused information include: Obtain user inquiry statements; The user's inquiry statement is split to obtain split data; Identify at least one consultation intent contained in each of the split data sets; The consultation intents contained in all the split data are merged to generate a unique merged intent corresponding to the user's consultation statement; Based on the unique fusion intent, the unique response category label that the user's inquiry statement hits is predicted; Identify the response level corresponding to the unique response classification label; Based on the response level, a response agent is assigned to the user's inquiry statement.
[0006] Secondly, embodiments of this application also provide a response processing device based on fused information, which adopts the following technical solution: A response processing device based on fused information includes: The query statement acquisition module is used to acquire user query statements; The statement splitting processing module is used to split the user's inquiry statement to obtain split data; The consultation intent recognition module is used to identify at least one consultation intent contained in each of the split data. The consultation intent fusion module is used to fuse the consultation intents contained in all the split data to generate a unique fused intent corresponding to the user's consultation statement. The response classification prediction module is used to predict the unique response classification label that the user's inquiry statement matches based on the unique fusion intent. The response level identification module is used to identify the response level corresponding to the unique response classification label; The response agent allocation module is used to allocate response agents according to the response level for the user's inquiry statement.
[0007] Thirdly, embodiments of this application also provide a computer device that adopts the technical solution described below: A computer device includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the response processing method based on fused information described above.
[0008] Fourthly, embodiments of this application also provide a computer-readable storage medium, which adopts the technical solutions described below: A computer-readable storage medium storing computer-readable instructions, which, when executed by a processor, implement the steps of the response processing method based on fused information as described above.
[0009] Compared with the prior art, the embodiments of this application have the following main advantages: The response processing method based on fused information described in this application involves: acquiring user inquiry statements; splitting the user inquiry statements to obtain split data; identifying at least one inquiry intent contained in each of the split data; fusing the inquiry intents contained in each of the split data to generate a unique fused intent corresponding to the user inquiry statement; predicting a unique response category tag corresponding to the user inquiry statement based on the unique fused intent; identifying the response level corresponding to the unique response category tag; and assigning response agents to the user inquiry statement according to the response level. By analyzing the user inquiry statements to identify the user's inquiry intent, it is possible to subsequently assign response agents to users in a targeted manner based on the inquiry intent, avoiding blind assignment of response agents and improving the efficiency of agents' business contact. Applying this response processing method to the financial business field, it can accurately identify the user's true inquiry intent in business scenarios with complex financial business and a large number of users, thereby accurately assigning target response agents to users. Attached Figure Description
[0010] To more clearly illustrate the solutions in this application, the accompanying drawings used in the description of the embodiments of this application will be briefly introduced below. Obviously, the accompanying drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0011] Figure 1 This is an exemplary system architecture diagram to which this application can be applied; Figure 2 This is a flowchart of an embodiment of the response processing method based on fused information according to this application; Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 203 shown; Figure 4 This is a flowchart of a specific embodiment of the intent label recognition model training in the response processing method based on fused information described in this application; Figure 5 This is a flowchart of a specific embodiment of the intent label recognition model verification and optimization in the response processing method based on fused information described in this application; Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 204 shown; Figure 7 This is a schematic diagram of a structure of an embodiment of the response processing apparatus based on fused information according to this application; Figure 8This is a schematic diagram of the structure of one embodiment of the computer device according to this application. Detailed Implementation
[0012] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein in the specification of the application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application; the terms "comprising" and "having," and any variations thereof, in the specification, claims, and foregoing drawings of this application, are intended to cover non-exclusive inclusion. The terms "first," "second," etc., in the specification, claims, or foregoing drawings of this application are used to distinguish different objects, not to describe a particular order.
[0013] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0014] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings.
[0015] like Figure 1 As shown, system architecture 100 may include terminal device 101, network 102, and server 103. Terminal device 101 may be a laptop 1011, tablet 1012, or mobile phone 1013. Network 102 is used as a medium to provide a communication link between terminal device 101 and server 103. Network 102 may include various connection types, such as wired, wireless communication links, or fiber optic cables.
[0016] Users can use terminal device 101 to interact with server 103 via network 102 to receive or send messages, etc. Various communication client applications can be installed on terminal device 101, such as web browser applications, shopping applications, search applications, instant messaging tools, email clients, social media platform software, etc.
[0017] Terminal device 101 can be various electronic devices with a display screen and support web browsing. In addition to laptops 1011, tablets 1012, or mobile phones 1013, terminal device 101 can also be an e-book reader, an MP3 player (Moving Picture Experts Group Audio Layer III), an MP4 player (Moving Picture Experts Group Audio Layer IV), a laptop computer, and a desktop computer, etc.
[0018] Server 103 can be a server that provides various services, such as a backend server that provides support for the pages displayed on terminal device 101.
[0019] It should be noted that the response processing method based on fused information provided in this application embodiment is generally executed by the server, and correspondingly, the response processing device based on fused information is generally set in the server.
[0020] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.
[0021] Continue to refer to Figure 2 The diagram shows a flowchart of an embodiment of the response processing method based on fused information according to this application. The response processing method based on fused information includes the following steps: Step 201: Obtain the user's inquiry statement.
[0022] In this embodiment, the user inquiry statement includes a business inquiry statement sent by the user.
[0023] Specifically, in the financial business field, due to the diversification of financial services, the services that users inquire about are also different. Here, obtaining the user's inquiry statement is a preliminary step before clarifying the user's inquiry intention. It can be understood as using a unified message receiving server to obtain the inquiry statements sent by different users.
[0024] Step 202: The user's inquiry statement is split to obtain split data.
[0025] In this embodiment, the step of splitting the user's query statement to obtain split data specifically includes splitting the user's query statement according to a preset word segmentation strategy, such as combining a preset word segmentation dictionary, according to a preset word segmentation length, or according to preset part-of-speech analysis rules.
[0026] By breaking down the user's inquiry statement, split data is obtained, which can then be transformed into more granular fields, allowing for a clearer understanding of the user's business inquiry intent at a more refined level.
[0027] Step 203: Identify at least one consultation intent contained in each of the split data.
[0028] In this embodiment, each piece of data may correspond to only one consultation intent, such as a proprietary term in a specific business scenario; or a piece of data may correspond to multiple consultation intents, such as a common field used in multiple business scenarios. Therefore, by identifying at least one consultation intent contained in each piece of data, comprehensive identification of the consultation intent contained in all the piece of data is ensured, thereby assisting subsequent processing steps in a more comprehensive understanding of the user's business consultation intent.
[0029] Step 204: Perform fusion processing on the consultation intent contained in all the split data to generate a unique fused intent corresponding to the user's consultation statement.
[0030] In this embodiment, after identifying the consultation intent contained in each of the split data, a fusion processing method is adopted to merge the consultation intent contained in each of the split data, and finally obtain the unique fused intent corresponding to the user's consultation statement.
[0031] Specifically, a multi-head attention mechanism can be introduced to calculate the contribution of each split data point to the intent recognition of the user's consultation statement. This can also be understood as the intent recognition weight of each split data point to the user's consultation statement. Then, based on this contribution, the consultation intent contained in all the split data points is integrated. Here, it is necessary to combine the consultation intent probability corresponding to each split data point. Finally, the consultation intent with the highest probability corresponding to the user's consultation statement is selected through a dot product or accumulation mode as the unique fused intent.
[0032] Step 205: Based on the unique fusion intent, predict the unique response category label that the user's inquiry statement matches.
[0033] In this embodiment, the unique response category label can be understood as a pre-organized business category label. Specifically, it can be in the form of a form, which pre-organizes the unique fusion intent corresponding to different business category names. Subsequently, after identifying the unique fusion intent corresponding to the user's inquiry statement, the business category label hit by the user's inquiry statement is identified according to the unique fusion intent, that is, the unique response category label is identified.
[0034] Step 206: Identify the response level corresponding to the unique response classification label.
[0035] In this embodiment, the response level refers to the pre-set answer level for different business category names. For example, if the answer level corresponding to the current response category label is low, it is possible to assign an intelligent robot to answer the user's inquiry. If the answer level corresponding to the current response category label is high and the user's inquiry is tricky and complicated, it is possible to assign a human agent to the user.
[0036] Step 207: Assign a response agent to the user's inquiry statement according to the response level.
[0037] In this embodiment, the provided response processing method based on fused information analyzes the user's inquiry statement to identify the user's inquiry intent, which facilitates the subsequent targeted allocation of response agents to the user based on the inquiry intent, avoiding blind allocation of response agents and improving the efficiency of agents' business contact.
[0038] In this embodiment, the following steps are taken: First, a user's inquiry statement is acquired. This statement is then broken down to obtain split data. At least one inquiry intent is identified in each of the split data segments. These intents are then fused to generate a unique fused intent corresponding to the user's inquiry statement. Based on this unique fused intent, a unique response category tag is predicted for the user's inquiry statement. The response level corresponding to the unique response category tag is then identified. Finally, response agents are assigned to the user's inquiry statement based on the response level. By analyzing the user's inquiry statement to identify their intent, the system can effectively assign response agents to users based on their intent, avoiding blind assignment and improving the efficiency of agents' business interactions. Applying this response processing method to the financial business sector allows for accurate identification of users' true inquiry intent in complex financial scenarios with a large user base, enabling precise assignment of target response agents.
[0039] In this embodiment, the step of splitting the user inquiry statement to obtain split data specifically includes: splitting the user inquiry statement into words according to the part-of-speech segmentation method to obtain the split fields contained in the user inquiry statement, wherein the split fields include the split text and / or split words.
[0040] Specifically, the part-of-speech segmentation method refers to segmenting the words in the user's query statement according to different parts of speech such as nouns, pronouns, verbs, and modal words, thereby obtaining all the segmented fields contained in the user's query statement.
[0041] In this embodiment, after performing the step of splitting the user's inquiry statement to obtain split data, the method further includes: labeling the parts of speech corresponding to the split text and / or the split words respectively; cleaning the split text and / or the split words according to the part-of-speech labeling results, cleaning out text and / or words that are meaningless to intent recognition, and updating the split data.
[0042] Specifically, after segmenting the user's inquiry statement using part-of-speech tagging, the parts of speech of each segmented character and / or word are labeled. Since different parts of speech contribute differently to the business inquiry intent—for example, nouns often correspond to business names, and verbs often correspond to business actions—these types of words are more important for identifying the business inquiry intent. However, modal particles, prepositions, or conjunctions in the user's inquiry statement often do not involve the business inquiry intent; they are merely words used for coherence in the statement. Therefore, based on the part-of-speech tagging results, the segmented characters and / or words are cleaned to remove those that are meaningless for intent identification, and the segmented data is updated. This allows for processing fewer fields of data when using the segmented data for subsequent business inquiry intent prediction or identification.
[0043] Continue to refer to Figure 3 , Figure 3 yes Figure 2 A flowchart of a specific embodiment of step 203 shown includes: Step 301: Input all the split data into the pre-trained intent label recognition model in sequence; Step 302: Obtain the set of intent labels output by the pre-trained intent label recognition model for each split data, wherein the set of intent labels also records the weights corresponding to different intent labels. Specifically, the pre-trained intent tag recognition model may output a different number of intent tags for each split data. Therefore, the different number of intent tags corresponding to each split data are then aggregated to obtain a set of intent tags corresponding to each split data.
[0044] Step 303: For each element in the intent tag set, summarize and organize the intents to obtain the consultation intent.
[0045] In this embodiment, the step of summarizing and organizing the intents for each element in the intent tag set to obtain the consultation intent refers to summarizing all the intent tags that have been hit, and finally retaining only the most important and unique consultation intent for intent identification.
[0046] Continue to refer to Figure 4 In some specific implementations, before step 301, a step of training an intent label recognition model is also included. Figure 4 This is a flowchart of a specific embodiment of the intent label recognition model training in the response processing method based on fused information described in this application, including: Step 401: Obtain a training dataset for intent label recognition training, wherein each training sample in the training dataset is in the format of "sentence + intent label"; Specifically, each training sample in the training dataset is in the format of "sentence + intent label", which ensures that the intent label recognition model is trained in a supervised mode.
[0047] Step 402: Input the training dataset into the intent label recognition model to be trained, wherein the intent label recognition model to be trained includes a language recognition model based on pre-trained BERT. Step 403: Using the pre-trained BERT-based language recognition model, calculate the word vectors for each sentence and each intent label in all training samples; In this embodiment, word vector recognition and calculation of the BERT language recognition model are used to calculate the word vector of each sentence and each intent label in all training samples. Specifically, it can be understood that each character or each word corresponds to an encoding vector value. Then, based on the encoding vector value corresponding to each character or each word, the encoding vector sequence corresponding to each sentence is calculated, as well as the encoding vector value or encoding vector sequence corresponding to each intent label is calculated.
[0048] Step 404: Calculate the weight of each word for each intent label based on the word vectors of each sentence and each intent label in all training samples; Specifically, the entire training dataset can be treated as a whole, and the TF-IDF algorithm can be used to calculate the weight of each word for different intent labels. For example, if all samples in a certain intent label contain a specific word, while other samples do not, it can be understood that the samples in which the specific word appears must be labeled with that intent label. Therefore, the weight of the intent label for that specific word is 100%, meaning that the correlation between the two is high. On the other hand, if a specific word appears in all samples, but only 10% of the samples are labeled with that specific intent label, it can be understood that the weight of the intent label for that specific word is only 10%, meaning that the correlation between the two is low.
[0049] Step 405: Based on the different intent tag weights corresponding to each word, accumulate the weights of the same intent tags corresponding to different words in the sentence to obtain at least one intent tag weight corresponding to each sentence. Specifically, by calculating the weights of different intent labels corresponding to different words in each training sample, and then summing the weights of the same intent labels corresponding to different words in a sentence, the weights and values of different intent labels corresponding to each sentence can be identified.
[0050] Step 406: Obtain the pre-trained intent label recognition model.
[0051] Continue to refer to Figure 5 In some specific implementations, after step 405, a step of verifying and optimizing the intent label recognition model is also included. Figure 5 This is a flowchart of a specific embodiment of the intent label recognition model verification and optimization in the response processing method based on fused information described in this application, including: Step 501: For each sentence, filter the intent tag corresponding to the maximum weight of at least one intent tag to obtain the unique target intent tag corresponding to each sentence. Specifically, in step 405, after obtaining the weights and values of different intent tags corresponding to each sentence, in step 501, the maximum weight and value of the intent tag corresponding to each sentence is selected, and this intent tag is used as the unique target intent tag corresponding to each sentence.
[0052] Step 502: Organize the unique target intent labels corresponding to each sentence to construct a verification dataset; Specifically, the "sentence + intent label" format of the training dataset is used, and the same "sentence + intent label" format is used when constructing the validation dataset. Here, the "intent label" in the validation dataset is the aforementioned unique target intent label.
[0053] Step 503: Compare the intent label consistency between the verification dataset and the training dataset, and calculate the training loss value based on the comparison result; Specifically, after obtaining the training data, the consistency of the intent labels of the same sentences in the verification dataset and the training dataset is compared, and the training loss value is calculated.
[0054] Step 504: Identify whether the training loss value meets the preset loss threshold; Step 505: If the training loss value does not meet the preset loss threshold, then the training parameters of the intent label recognition model are tuned and the model is retrained. Step 506: If the training loss value meets the preset loss threshold, then the sentences corresponding to the same intent label are aggregated and sorted, and a response classification label is set for each aggregated result according to the preset response classification form, wherein the response classification form contains the mapping relationship between the response classification label and all intent labels.
[0055] Specifically, sentences corresponding to the same intent tag are grouped and organized. Then, a response classification tag is set for each grouped result according to a preset response classification form. That is, a business name tag is set for each grouped sentence set. Here, by setting a business name tag for each grouped sentence set, a unique mapping relationship between unique intent tags and business name tags is established.
[0056] Continue to refer to Figure 6 , Figure 6 yes Figure 2 A flowchart of a specific embodiment of step 204 shown includes: Step 601: Identify the consultation intent contained in all the split data in the user consultation statement, and the intent tag weight corresponding to each consultation intent on different split data. Step 602: Accumulate the weights of all intent tags corresponding to the same consultation intent to obtain the unique accumulated weights corresponding to different consultation intents. Step 603: Sort the consultation intentions according to the unique cumulative weights corresponding to different consultation intentions, and based on the sorting results, obtain the consultation intention with the maximum unique cumulative weight as the unique fusion intention corresponding to the user's consultation statement.
[0057] In this embodiment, the step of predicting the unique response category label of the user's inquiry statement based on the unique fusion intent specifically includes: identifying the sentence collection and organization result corresponding to the unique fusion intent; and determining the unique response category label of the user's inquiry statement based on the response category label previously set for the sentence collection and organization result according to the response category form.
[0058] Specifically, using the unique mapping relationship between the unique intent tag and the business name tag in step 506 above, after identifying the sentence set organization result corresponding to the unique fusion intent, the unique response category tag, i.e. the business category name, is determined according to the response category tag corresponding to the sentence set organization result.
[0059] In this embodiment, the following steps are taken: First, a user's inquiry statement is acquired. This statement is then broken down to obtain split data. At least one inquiry intent is identified in each of the split data segments. These intents are then fused to generate a unique fused intent corresponding to the user's inquiry statement. Based on this unique fused intent, a unique response category tag is predicted for the user's inquiry statement. The response level corresponding to the unique response category tag is then identified. Finally, response agents are assigned to the user's inquiry statement based on the response level. By analyzing the user's inquiry statement to identify their intent, the system can effectively assign response agents to users based on their intent, avoiding blind assignment and improving the efficiency of agents' business interactions. Applying this response processing method to the financial business sector allows for accurate identification of users' true inquiry intent in complex financial scenarios with a large user base, enabling precise assignment of target response agents.
[0060] The embodiments of this application can acquire and process relevant data based on artificial intelligence technology. Artificial intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to obtain optimal results.
[0061] Foundational technologies for artificial intelligence generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing, operating / interactive systems, and mechatronics. AI software technologies mainly encompass computer vision, robotics, biometrics, speech processing, natural language processing, and machine learning / deep learning.
[0062] In this embodiment, the following steps are taken: First, a user's inquiry statement is acquired. This statement is then broken down to obtain split data. At least one inquiry intent is identified in each of the split data segments. These intents are then fused to generate a unique fused intent corresponding to the user's inquiry statement. Based on this unique fused intent, a unique response category tag is predicted for the user's inquiry statement. The response level corresponding to the unique response category tag is then identified. Finally, response agents are assigned to the user's inquiry statement based on the response level. By analyzing the user's inquiry statement to identify their intent, the system can effectively assign response agents to users based on their intent, avoiding blind assignment and improving the efficiency of agents' business interactions. Applying this response processing method to the financial business sector allows for accurate identification of users' true inquiry intent in complex financial scenarios with a large user base, enabling precise assignment of target response agents.
[0063] Further reference Figure 7 As a response to the above Figure 2 To implement the method shown, this application provides an embodiment of a response processing device based on fused information, which is similar to... Figure 2 Corresponding to the method embodiments shown, this device can be specifically applied to various electronic devices.
[0064] like Figure 7 As shown, the response processing device 700 based on fused information described in this embodiment includes: a consultation statement acquisition module 701, a statement splitting and processing module 702, a consultation intent recognition module 703, a consultation intent fusion module 704, a response classification prediction module 705, a response level recognition module 706, and a response agent allocation module 707. Wherein: The consultation statement acquisition module 701 is used to acquire user consultation statements. The statement splitting processing module 702 is used to split the user inquiry statement to obtain split data; Consultation intent recognition module 703 is used to identify at least one consultation intent contained in each of the split data; The consultation intent fusion module 704 is used to fuse the consultation intents contained in all the split data to generate a unique fused intent corresponding to the user's consultation statement. The response classification prediction module 705 is used to predict the unique response classification label that the user's inquiry statement hits based on the unique fusion intent. The response level identification module 706 is used to identify the response level corresponding to the unique response classification label; The response agent allocation module 707 is used to allocate response agents according to the response level for the user's inquiry statement.
[0065] This application obtains user inquiry statements; breaks down the user inquiry statements to obtain split data; identifies at least one inquiry intent contained in each of the split data; fuses the inquiry intents contained in each of the split data to generate a unique fused intent corresponding to the user inquiry statement; predicts a unique response category tag matched by the user inquiry statement based on the unique fused intent; identifies the response level corresponding to the unique response category tag; and assigns response agents to the user inquiry statement according to the response level. By analyzing the user inquiry statements to identify the user's inquiry intent, it is possible to subsequently assign response agents to users in a targeted manner based on the inquiry intent, avoiding blind assignment of response agents and improving the efficiency of agents' business contact. Applying this response processing method to the financial business field, it can accurately identify the user's true inquiry intent in business scenarios with complex financial business and a large number of users, thereby accurately assigning target response agents to users.
[0066] In this embodiment, the response processing device 700 based on fused information further includes a part-of-speech tagging module and a data splitting and cleaning module. Wherein: The part-of-speech tagging module is used to tag the parts of speech corresponding to the split text and / or the split words respectively; The data cleaning module is used to clean the split text and / or the split words based on the part-of-speech tagging results, remove text and / or words that are meaningless for intent recognition, and update the split data.
[0067] In this embodiment, the response processing device 700 based on fused information further includes a training dataset acquisition module, a training dataset input module, a target word vector calculation module, an intent label weight calculation module, a comprehensive calculation module, and a model training completion module. Wherein: The training dataset acquisition module is used to acquire a training dataset for intent label recognition training, wherein each training sample in the training dataset is in the format of "sentence + intent label"; The training dataset input module is used to input the training dataset into the intent label recognition model to be trained, wherein the intent label recognition model to be trained includes a language recognition model based on pre-trained BERT. The target word vector calculation module is used to calculate the word vector of each sentence and each intent label in all training samples using the language recognition model based on pre-trained BERT. The intent label weight calculation module is used to calculate the weight of each word for different intent labels based on the word vectors of each sentence and each intent label in all training samples. The comprehensive calculation module is used to accumulate the weights of the same intent tags corresponding to different words in a sentence based on the different intent tag weights corresponding to each word, so as to obtain at least one intent tag weight corresponding to each sentence. The model training completion module is used to obtain a pre-trained intent label recognition model.
[0068] In this embodiment, the response processing device 700 based on fused information further includes an intent label filtering module, a verification dataset construction module, a training loss value calculation module, a loss recognition and judgment module, a first branch processing module, and a second branch processing module. Wherein: The intent tag filtering module is used to filter the intent tag corresponding to the maximum weight of at least one intent tag for each sentence, so as to obtain the unique target intent tag corresponding to each sentence. The validation dataset building module is used to organize the unique target intent labels corresponding to each sentence and build the validation dataset. The training loss calculation module is used to compare the consistency of intent labels between the validation dataset and the training dataset, and calculate the training loss value based on the comparison results. The loss identification and judgment module is used to identify whether the training loss value meets the preset loss threshold; The first branch processing module is used to perform training parameter tuning on the intent label recognition model and retrain the model if the training loss value does not meet the preset loss threshold. The second branch processing module is used to, if the training loss value meets the preset loss threshold, aggregate the sentences corresponding to the same intent label, and set a response classification label for each aggregated result according to the preset response classification form, wherein the response classification form contains the mapping relationship between the response classification label and all intent labels.
[0069] In this embodiment, the consultation intent fusion module 704 includes an intent-related identification unit, a weight accumulation processing unit, and a unique fusion intent acquisition unit. Wherein: The intent-related identification unit is used to identify the consultation intent contained in all the split data in the user consultation statement, and the intent tag weight corresponding to each consultation intent on different split data. The weight accumulation processing unit is used to accumulate the weights of all intent tags corresponding to the same consultation intent, and obtain the unique accumulated weights corresponding to different consultation intents respectively. The unique fusion intent acquisition unit is used to sort the consultation intents according to the unique cumulative weights corresponding to different consultation intents, and based on the sorting results, obtain the consultation intent with the maximum unique cumulative weight as the unique fusion intent corresponding to the user's consultation statement.
[0070] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by instructing related hardware through computer-readable instructions. These computer-readable instructions can be stored in a computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the methods described above. The aforementioned storage medium can be a non-volatile storage medium such as a magnetic disk, optical disk, or read-only memory (ROM), or random access memory (RAM).
[0071] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.
[0072] To address the aforementioned technical problems, embodiments of this application also provide a computer device. Please refer to [link / reference needed]. Figure 8 , Figure 8 This is a basic structural block diagram of the computer device in this embodiment.
[0073] The computer device 8 includes a memory 8a, a processor 8b, and a network interface 8c that are interconnected via a system bus. It should be noted that... Figure 8Only a computer device 8 with component memory 8a, processor 8b, and network interface 8c is shown. However, it should be understood that it is not required to implement all the components shown, and more or fewer components can be implemented instead. Those skilled in the art will understand that the computer device described herein is a device capable of automatically performing numerical calculations and / or information processing according to pre-set or stored instructions. Its hardware includes, but is not limited to, microprocessors, application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), digital signal processors (DSPs), embedded devices, etc.
[0074] The computer device can be a desktop computer, laptop, handheld computer, or cloud server, etc. The computer device can interact with the user via a keyboard, mouse, remote control, touchpad, or voice control.
[0075] The memory 8a includes at least one type of readable storage medium, including flash memory, hard disk, multimedia card, card-type memory (e.g., SD or DX memory), random access memory (RAM), static random access memory (SRAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), programmable read-only memory (PROM), magnetic memory, magnetic disk, optical disk, etc. In some embodiments, the memory 8a may be an internal storage unit of the computer device 8, such as the hard disk or memory of the computer device 8. In other embodiments, the memory 8a may also be an external storage device of the computer device 8, such as a plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, etc., equipped on the computer device 8. Of course, the memory 8a may also include both the internal storage unit and its external storage device of the computer device 8. In this embodiment, the memory 8a is typically used to store the operating system and various application software installed on the computer device 8, such as computer-readable instructions based on a response processing method for fused information. In addition, the memory 8a can also be used to temporarily store various types of data that have been output or will be output.
[0076] In some embodiments, the processor 8b may be a central processing unit (CPU), controller, microcontroller, microprocessor, or other data processing chip. The processor 8b is typically used to control the overall operation of the computer device 8. In this embodiment, the processor 8b is used to execute computer-readable instructions stored in the memory 8a or to process data, for example, to execute computer-readable instructions of the response processing method based on fused information.
[0077] The network interface 8c may include a wireless network interface or a wired network interface, which is typically used to establish communication connections between the computer device 8 and other electronic devices.
[0078] The computer device proposed in this embodiment belongs to the field of artificial intelligence technology and is applied in business consultation scenarios, specifically, in the scenario of intelligently allocating response seats to target users' business consultations. This application obtains user consultation statements; splits the user consultation statements to obtain split data; identifies at least one consultation intent contained in each of the split data; fuses the consultation intents contained in each of the split data to generate a unique fused intent corresponding to the user consultation statement; predicts a unique response category tag corresponding to the user consultation statement based on the unique fused intent; identifies the response level corresponding to the unique response category tag; and allocates response seats to the user consultation statement according to the response level. By analyzing the consultation intent of user consultation statements, the user's consultation intent is identified, facilitating targeted allocation of response seats to users based on the consultation intent, avoiding blind allocation of response seats, and improving the efficiency of effective business contact for agents. Applying this response processing method to the financial business field, it can accurately identify the user's true consultation intent in business scenarios with complex financial business and a large number of users, thereby accurately allocating target response seats to users.
[0079] This application also provides another embodiment, namely, providing a computer-readable storage medium storing computer-readable instructions that can be executed by a processor to cause the processor to perform the steps of the response processing method based on fused information as described above.
[0080] The computer-readable storage medium proposed in this embodiment belongs to the field of artificial intelligence technology and is applied in business consultation scenarios, specifically, in the scenario of intelligently allocating response agents to business consultations of target users. This application obtains user consultation statements; splits the user consultation statements to obtain split data; identifies at least one consultation intent contained in each of the split data; fuses the consultation intents contained in each of the split data to generate a unique fused intent corresponding to the user consultation statement; predicts a unique response category tag corresponding to the user consultation statement based on the unique fused intent; identifies the response level corresponding to the unique response category tag; and allocates response agents to the user consultation statement according to the response level. By analyzing the consultation intent of user consultation statements, the user's consultation intent is identified, facilitating targeted allocation of response agents to users based on the consultation intent, avoiding blind allocation of response agents, and improving the efficiency of effective business contact for agents. Applying this response processing method to the financial business field, it can accurately identify the user's true consultation intent in business scenarios with complex financial business and a large number of users, thereby accurately allocating target response agents to users.
[0081] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.
[0082] Obviously, the embodiments described above are only some embodiments of this application, not all embodiments. The accompanying drawings show preferred embodiments of this application, but do not limit the patent scope of this application. This application can be implemented in many different forms; rather, the purpose of providing these embodiments is to make the disclosure of this application more thorough and comprehensive. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing specific embodiments, or make equivalent substitutions for some of the technical features. Any equivalent structures made using the content of this application's specification and drawings, directly or indirectly applied to other related technical fields, are similarly within the scope of patent protection of this application. Software tools or components not belonging to this company that appear in the embodiments of this application are merely illustrative examples and do not represent actual use.
Claims
1. A response processing method based on fused information, characterized in that, Includes the following steps: Obtain user inquiry statements; The user's inquiry statement is split to obtain split data; Identify at least one consultation intent contained in each of the split data sets; The consultation intents contained in all the split data are merged to generate a unique merged intent corresponding to the user's consultation statement; Based on the unique fusion intent, the unique response category label that the user's inquiry statement hits is predicted; Identify the response level corresponding to the unique response classification label; Based on the response level, a response agent is assigned to the user's inquiry statement.
2. The response processing method based on fused information according to claim 1, characterized in that, The step of splitting the user's query statement to obtain split data specifically includes: The user's query statement is split into words and phrases according to the part-of-speech segmentation method to obtain the split fields contained in the user's query statement, wherein the split fields include split text and / or split words; After performing the step of splitting the user's query statement to obtain split data, the method further includes: Label the parts of speech corresponding to the split text and / or the split words respectively; Based on the part-of-speech tagging results, the segmented text and / or the segmented words are cleaned to remove text and / or words that are meaningless for intent recognition, and the segmentation data is updated.
3. The response processing method based on fused information according to claim 1, characterized in that, The step of identifying at least one consultation intent contained in each of the split data segments specifically includes: All the split data are sequentially input into the pre-trained intent label recognition model; Obtain the set of intent labels output by the pre-trained intent label recognition model for each split data, wherein the set of intent labels also records the weights corresponding to different intent labels; For each element in the intent tag set, the intents are summarized and organized to obtain the consultation intent.
4. The response processing method based on fused information according to claim 3, characterized in that, Before performing the step of sequentially inputting all the split data into the pre-trained intent label recognition model, the method further includes: Obtain a training dataset for intent label recognition training, wherein each training sample in the training dataset is in the format of "sentence + intent label"; The training dataset is input into the intent label recognition model to be trained, wherein the intent label recognition model to be trained includes a language recognition model based on pre-trained BERT. Using the pre-trained BERT-based language recognition model, word vectors for each sentence and each intent label in all training samples are calculated; Based on the word vectors of each sentence and each intent label in all training samples, calculate the weight of each word for different intent labels; Based on the different intent label weights corresponding to each word, the weights of the same intent labels corresponding to different words in the sentence are accumulated to obtain at least one intent label weight corresponding to each sentence. Obtain a pre-trained intent label recognition model.
5. The response processing method based on fused information according to claim 4, characterized in that, After performing the step of accumulating the weights of the same intent tags corresponding to different words in a sentence based on the different intent tag weights corresponding to each word, and obtaining at least one intent tag weight corresponding to each sentence, the method further includes: For each sentence, the intent tag corresponding to the maximum weight of at least one intent tag is selected to obtain the unique target intent tag corresponding to each sentence. The unique target intent labels corresponding to each sentence are compiled to construct a validation dataset; The consistency of intent labels is compared between the validation dataset and the training dataset, and the training loss value is calculated based on the comparison results. Identify whether the training loss value meets a preset loss threshold; If the training loss value does not meet the preset loss threshold, the training parameters of the intent label recognition model are tuned and the model is retrained. If the training loss value meets the preset loss threshold, the sentences corresponding to the same intent label are aggregated and organized, and a response classification label is set for each aggregated result according to the preset response classification form, wherein the response classification form contains the mapping relationship between the response classification label and all intent labels.
6. The response processing method based on fused information according to claim 1, characterized in that, The step of fusing the consultation intents contained in all the split data to generate a unique fused intent corresponding to the user's consultation statement specifically includes: Identify the consultation intent contained in each of the segmented data in the user's consultation statement, and the intent tag weight corresponding to each consultation intent on different segmented data. The weights of all intent tags corresponding to the same consultation intent are summed to obtain the unique summed weights corresponding to different consultation intents. The user's consultation statement is selected as the unique fused intent based on the unique cumulative weight corresponding to each consultation intent.
7. The response processing method based on fused information according to claim 5, characterized in that, The step of predicting the unique response category label matched by the user's inquiry statement based on the unique fusion intent specifically includes: Identify and organize the sentence set corresponding to the unique fusion intent; Based on the response classification tags previously set for the aggregated results of the sentences according to the response classification form, the unique response classification tag that the user's inquiry statement matches is determined.
8. A response processing device based on fused information, characterized in that, include: The query statement acquisition module is used to acquire user query statements; The statement splitting processing module is used to split the user's inquiry statement to obtain split data; The consultation intent recognition module is used to identify at least one consultation intent contained in each of the split data. The consultation intent fusion module is used to fuse the consultation intents contained in all the split data to generate a unique fused intent corresponding to the user's consultation statement. The response classification prediction module is used to predict the unique response classification label that the user's inquiry statement matches based on the unique fusion intent. The response level identification module is used to identify the response level corresponding to the unique response classification label; The response agent allocation module is used to allocate response agents according to the response level for the user's inquiry statement.
9. A computer device, characterized in that, It includes a memory and a processor, wherein the memory stores computer-readable instructions, and the processor executes the computer-readable instructions to implement the steps of the response processing method based on fused information as described in any one of claims 1 to 7.
10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-readable instructions, which, when executed by a processor, implement the steps of the response processing method based on fused information as described in any one of claims 1 to 7.