Business processing method and device based on intention recognition, electronic equipment and medium

Abstract

The invention provides a business processing method and device based on intention recognition, electronic equipment and a medium. The business processing method comprises the steps of obtaining a query text input by a user on a business interaction interface, business data in a current business scene and business behavior data of the user; carrying out fusion processing on the query text, the service data in the current service scene and the service behavior data of the user; inputting the fused data into a pre-trained intention recognition model to obtain an intention recognition result; and calling a corresponding service interface and a service interface corresponding to the service interface based on the intention recognition result. According to the method, the desired function item can be directly found through the query text of the service interaction interface, so that the service processing efficiency is improved; and by fusing the business data and the business behavior data of the user, the information limitation of the input text can be made up, the intention recognition model is helped to perform intention recognition in combination with a specific business scene, and the intention recognition accuracy is greatly improved.

Description

Technical Field

[0001] The present invention relates to the field of computer technology, and in particular to a business processing method, device, electronic device and medium based on intent recognition. Background Art

[0002] In some large business systems, such as HR systems, the hierarchy is very deep and contains many menus and functions, such as performance homepage, employee performance, organizational performance, performance templates, indicator library, OKR management, goal management, goal setting, etc. The menus also contain many functional items, such as: "New performance appraisal activities, new model templates, new indicators, promote activity processes, add personnel, maintain appraisal information, set items" and common Web application functions. If users want to find a certain function item and process business through it, they must continuously click on many menus and function items, which is cumbersome and inefficient.

[0003] At present, some use AI dialogue assistants to help find relevant information, but they are mainly limited to text communication and provide some simple information and prompts. They are unable to accurately identify the user's intentions and accurately find the user's needs in combination with specific business data. For example, if the user only enters the "query" text, the AI ​​dialogue assistant cannot identify what is specifically being queried, resulting in inaccurate results. Summary of the invention

[0004] In view of this, an object of the present invention is to provide a business processing method, device, electronic device and medium based on intent recognition, so as to improve the efficiency of business processing and the accuracy of intent recognition in business processing.

[0005] In a first aspect, a service processing method based on intent recognition is provided, the method comprising:

[0006] Obtain the query text entered by the user on the business interaction interface, the business data of the current business scenario, and the user's business behavior data;

[0007] Integrate the query text, business data in the current business scenario, and user business behavior data;

[0008] Input the fused data into the pre-trained intent recognition model to obtain the intent recognition result;

[0009] Based on the intent recognition results, the corresponding business interface and the business interface corresponding to the business interface are called.

[0010] Optionally, fusing the query text, the business data in the current business scenario, and the business behavior data of the user includes:

[0011] Convert the query text, business data in the current business scenario, and user business behavior data into vector representations respectively;

[0012] The three vector representations are concatenated to obtain a concatenated feature vector;

[0013] The concatenated feature vector is normalized to obtain the target feature vector.

[0014] Optionally, converting the query text, the business data in the current business scenario, and the business behavior data of the user into vector representations respectively includes:

[0015] The three types of data are preprocessed separately, including:

[0016] Delete special characters in the query text and create a mapping table between abbreviations and full names;

[0017] And perform format verification on the business data in the current business scenario to unify the format;

[0018] and perform anomaly detection on users’ business behavior data to eliminate abnormal data;

[0019] The three types of preprocessed data are respectively subjected to feature extraction and encoding to obtain their respective vector representations, including:

[0020] Perform word segmentation, stem extraction, part-of-speech tagging and vector mapping on the preprocessed query text to obtain a first vector representation;

[0021] and encoding the process nodes and business operation types of the preprocessed business data to obtain node codes and operation type codes;

[0022] The node code and the operation type code are combined to obtain a second vector representation;

[0023] and encoding the user operation type, operation object, and operation time in the pre-processed user business behavior data to obtain an operation code, an operation object code, and an operation time code;

[0024] The operation code, the operation object code and the operation time code are combined to obtain a third vector representation.

[0025] Optionally, the intent recognition model includes an LSTM input layer, a Transformer encoding and an output layer; the pre-training process of the intent recognition model includes:

[0026] Obtain historical query texts, historical business data, and historical user business behavior data, and label them with intent recognition labels. The three types of data and the corresponding intent recognition labels are used as training data sets.

[0027] The sequence information of the three types of data in the training data set is modeled through the LSTM input layer to obtain the vector representations corresponding to the three types of data;

[0028] After fusion preprocessing of each vector representation, the feature vector is obtained;

[0029] The feature vector is iteratively trained through the multi-head self-attention mechanism of the Transformer encoding layer, and the predicted intent recognition result is output through the output layer;

[0030] The model corresponding to the minimum loss value between the predicted intent recognition result and the intent recognition label is determined as the final intent recognition model.

[0031] Optionally, calling a corresponding business interface based on the intention recognition result and a business interface corresponding to the business interface include:

[0032] Find the corresponding business interface and the business interface corresponding to the business interface based on the intent recognition result;

[0033] Generate business cards for the searched business interfaces;

[0034] Send the business card to the business interaction interface for display;

[0035] In response to a click operation on a business card, the business interface corresponding to the interface is sent in the form of a pop-up window to the front-end page where the business interaction interface is located.

[0036] Optionally, the method further comprises:

[0037] If the business interface corresponding to the business interface is not found, a new business interface is generated based on the intent recognition results and the UI components in the front-end component library.

[0038] Optionally, generating a new business interface based on the intent recognition result and the UI components in the front-end component library includes:

[0039] Determine the basic entry items of the business interface based on the intent recognition results;

[0040] Search the UI components of each basic entry item in the front-end component library based on the type of the basic entry item;

[0041] After splicing the UI components, a new business interface is obtained.

[0042] In a second aspect, a service processing device based on intent recognition is provided, the device comprising:

[0043] An acquisition unit, used to acquire the query text entered by the user on the business interaction interface, the business data of the current business scenario, and the business behavior data of the user;

[0044] A fusion unit is used to fuse the query text, the business data in the current business scenario, and the business behavior data of the user;

[0045] The recognition unit is used to input the fused data into the pre-trained intent recognition model to obtain the intent recognition result;

[0046] The calling unit is used to call the corresponding business interface and the business interface corresponding to the business interface based on the intention recognition result.

[0047] In a third aspect, an electronic device is provided, comprising a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other via the communication bus;

[0048] Memory, used to store computer programs;

[0049] The processor is used to implement any method step described in the first aspect when executing a program stored in the memory.

[0050] According to a fourth aspect, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, and when the computer program is executed by a processor, any method step described in the first aspect is implemented.

[0051] The embodiments of the present invention provide a business processing method, device, electronic device and medium based on intent recognition, which obtains the query text input by the user on the business interaction interface, the business data in the current business scenario and the business behavior data of the user; fuses the query text, the business data in the current business scenario and the business behavior data of the user; inputs the fused data into the pre-trained intent recognition model to obtain the intent recognition result; and calls the corresponding business interface and the business interface corresponding to the business interface based on the intent recognition result. The present invention can directly find the desired function item through the query text of the business interaction interface, which greatly saves time and improves the efficiency of business processing; and by fusing the business data and the business behavior data of the user, it can make up for the information limitations of the input text, help the intent recognition model to perform intent recognition in combination with specific business scenarios, and greatly improve the accuracy of intent recognition.

[0052] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, preferred embodiments are given below and described in detail with reference to the accompanying drawings. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without creative work.

[0054] Figure 1 A flowchart of a service processing method based on intent recognition provided by an embodiment of the present invention is shown;

[0055] Figure 2 A schematic diagram of the structure of a service processing device based on intent recognition provided by an embodiment of the present invention is shown;

[0056] Figure 3 A schematic structural diagram of an electronic device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0057] In order to make the purpose, technical solutions and advantages of the embodiments of the present invention clearer, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. The components of the embodiments of the present invention generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of the present invention provided in the drawings is not intended to limit the scope of the claimed invention, but merely represents selected embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without making creative work belong to the scope of protection of the present invention.

[0058] Considering that the current AI dialogue assistants are mainly limited to text communication and provide some simple information and prompts, they are unable to accurately identify the user's intentions and accurately find the user's needs in combination with specific business data; for example, if the user only enters the "query" text, the AI ​​dialogue assistant cannot identify what is specifically being queried, resulting in inaccurate results.

[0059] Based on this, an embodiment of the present invention provides a service processing method and device based on intent recognition, which is described below through an embodiment.

[0060] The embodiment of the present invention provides a service processing method based on intent recognition, such as Figure 1 As shown, the method comprises the following steps:

[0061] Step S101: Acquire the query text input by the user on the business interaction interface, the business data of the current business scenario and the business behavior data of the user.

[0062] In the embodiment of the present invention, the business interaction interface refers to an interface set in the business system to assist business processing during business processing, such as a text dialog box, in which a text input box is set, and the intent recognition result can be obtained by inputting the text, instruction, etc. to be queried through the text input box. For example, in the performance system, the user enters the text "create performance activity" to directly obtain the entrance to performance creation, and clicks the entrance to enter the performance creation interface.

[0063] The business data in the current business scenario can be extracted from the business process management system, in which a mapping relationship between business domains and business domain tags, a mapping relationship between business process nodes and business process node tags under each business domain, and a mapping relationship between business operation types and business operation type tags in each business process node are set; when the user operates in the business module, each data will have a corresponding tag, and the system will detect the data tag, for example, determine the user's business scenario based on the business domain tag, determine the business process node under the scenario based on the business node tag, and determine the business operation type in the node based on the business operation type tag, etc.

[0064] The user's business behavior data can be obtained from the system's user operation log, including the user's click operation records in the system, click operation objects, stay time on different interfaces, frequency of use of different functions, historical operation habits, etc.

[0065] In a feasible implementation, the user's business behavior data can also be quantified, for example, the proportion of the number of operations in different business modules to the total number of operations is calculated to measure the user's attention to different business modules. Behavior quantification can help the intent recognition model recognize the user's intent more quickly.

[0066] Step S102: The query text, the business data in the current business scenario, and the business behavior data of the user are integrated and processed.

[0067] By fusing the three types of data, we can make up for the information limitations in the existing text dialog box, add auxiliary information to help the intent recognition model better identify user intent, and improve the accuracy of intent recognition.

[0068] The specific fusion method will be described in detail in the following embodiments and will not be repeated here.

[0069] Step S103: input the fused data into the pre-trained intent recognition model to obtain the intent recognition result.

[0070] The intent recognition model not only recognizes based on the input text, but also recognizes the intent based on the user's specific business scenario, business node, and business operation type, and further integrates some of the user's business behavior data as auxiliary information to help with recognition, greatly improving the accuracy of intent recognition. For example, if the user only enters the "query" text in the dialog box, which is too vague and unclear about what to query, the intent recognition model can determine the business scenario to be handled based on the business data in the current business scenario. For example, in the performance system, it is more inclined to query performance data, and then further combined with user behavior data, such as click operation objects or historical operation habits, to infer the specific performance indicators that the user may want to query.

[0071] Step S104: Call the corresponding business interface and the business interface corresponding to the business interface based on the intention recognition result.

[0072] After obtaining the intent recognition result, you can call the corresponding business interface in the business library according to the intent recognition result. For example, if the intent recognition result is "create business performance", you can call the performance creation interface and the business interface linked to the interface; if the intent recognition result is "query historical performance", you can call the performance database interface and the historical performance record interface linked to the performance database interface.

[0073] The present invention can directly find the desired function items through the query text of the business interaction interface, and by integrating business data and user business behavior data, it can make up for the information limitations of the input text, help the intent recognition model to perform intent recognition in combination with specific business scenarios, and greatly improve the accuracy of intent recognition.

[0074] Based on the above embodiment, the query text, the business data in the current business scenario and the business behavior data of the user are integrated and processed, including the following steps:

[0075] Step S1021: Convert the query text, the business data in the current business scenario, and the business behavior data of the user into vector representations respectively.

[0076] In this step, different types of data may adopt different vector conversion methods. The vector conversion will be described in detail in the following embodiments and will not be repeated here.

[0077] Step S1022: concatenate the three vector representations to obtain a concatenated feature vector.

[0078] For example, the query text vector is X text , the business data vector is X process , the user business behavior data vector is X behavior , then the fused feature vector is X=(Xtext , X process , X behavior ).

[0079] Step S1023: normalize the concatenated feature vector to obtain a target feature vector.

[0080] The fused concatenated feature vector X is normalized so that the value of each element is in the range of [0, 1]. In one example, the normalization formula used is: Where x is an element in the concatenated feature vector X, min(x) is the minimum element value in the concatenated feature vector, and max(x) is the maximum element value in the concatenated feature vector.

[0081] Based on the above embodiment, converting the query text, the business data in the current business scenario, and the business behavior data of the user into vector representations respectively includes the following steps:

[0082] Step A: Preprocess the three types of data separately, including:

[0083] Delete special characters in the query text and create a mapping table between abbreviations and full names.

[0084] Among them, special characters include garbled characters, spaces, etc.

[0085] For the query text, a feasible implementation method for deleting special characters is as follows: Step 1: Count the occurrence frequency of special characters in the text, set as f specoal Step 2: When f special When the value is >0.1 (0.1 is the threshold value, which can be adjusted according to the actual situation), special characters are removed. For example, "Create performance activity" is processed as "Create performance activity".

[0086] In addition, by establishing a mapping table of common abbreviations and full names, for example, "perf" is mapped to "performance". When encountering abbreviated query text, it can be normalized according to the mapping table.

[0087] And perform format verification on the business data in the current business scenario to unify the format.

[0088] In one feasible implementation, data entry errors can be checked using a data verification algorithm. For example, for process time data, a time format verification algorithm is used, and if the data does not conform to the YYYY-MM-DD HH:MM:SS format, corrections are made.

[0089] In another feasible implementation, the number of types of business process time formats submitted by different departments may also be counted and set to K. When K≥3 (indicating that the format is confusing), the data format needs to be unified.

[0090] And perform anomaly detection on users’ business behavior data to eliminate abnormal data.

[0091] In a feasible implementation, a threshold for abnormal operations can be set based on the user's average operation frequency and operation time distribution, and data greater than the abnormal threshold can be removed. For example, if the user's click operation frequency in a certain period of time exceeds 3 times the average frequency and the operation duration is less than 1 second, it is judged as an abnormal data point and deleted.

[0092] Step B: Extract and encode the three types of preprocessed data to obtain their respective vector representations, including:

[0093] The preprocessed query text is segmented, stemmed, POS tagged and vector mapped to obtain the first vector representation.

[0094] In an embodiment of the present invention, the process of converting the query text into the first vector representation is described in detail. First, a word segmentation algorithm based on rules or machine learning (such as a maximum matching method or a neural network word segmentation model) is used to split the query text into individual words. For example, "create performance activities" is split into "create", "performance" and "activity". Then, the Porter algorithm or the Snowball algorithm is used to extract stems and convert the words into their basic forms, such as "activities" into "activity". Then, a part-of-speech tagging tool (such as the Stanford NLP part-of-speech tagger) is used to determine the part of speech of each word. Finally, a pre-trained word vector model (such as Word2Vec, BERT, etc.) is used to map each word to a high-dimensional vector. Assuming the word vector dimension is d, and a text contains n words, it is represented as an n×d matrix.

[0095] And the process nodes and business operation types of the preprocessed business data are encoded to obtain node codes and operation type codes.

[0096] In a feasible implementation, different business process nodes are numbered in sequence, and the total number of nodes is N. node , numbered from 1 to N node . The business operation types in the process (such as approval, submission, modification, etc.) are represented by specific numbers, for example, approval code is 1, submission code is 2, etc. Let the total number of operation types be N op , the encoding range is from 1 to N op .

[0097] The node code and the operation type code are combined to obtain a second vector representation.

[0098] Combine node code, operation type code and other information into a vector representation of business data. Assuming the vector dimension of business data is m, it is represented as a 1×m vector;

[0099] And the user operation type, operation object, and operation time in the pre-processed user's business behavior data are encoded to obtain an operation code, an operation object code, and an operation time code.

[0100] In the embodiment of the present invention, the user operation type is such as click, input, selection, etc.; the operation object is such as performance indicators, department selection box, etc.; and the operation time is such as the stay time in a certain module, etc. In one example, for example, the click operation is coded as 1 and the input operation is coded as 2. Suppose the total number of operation codes is N act , the encoding range is from 1 to N act In another example, the operation time is discretized into minutes, for example, 0-5 minutes is coded as 1, 6-10 minutes is coded as 2, etc. Let the total number of time codes be N time , the encoding range is from 1 to N time .

[0101] The operation code, the operation object code and the operation time code are combined to obtain a third vector representation.

[0102] Assuming that the dimension of the user's business behavior data vector is p, it is represented as a 1×p vector.

[0103] The present invention obtains a vector representation of each data by performing corresponding preprocessing and feature encoding on different data, which is convenient for providing diverse and rich input data for the subsequent intent recognition model so that the intent recognition model can perform accurate intent recognition.

[0104] Based on the above embodiment, the intent recognition model includes an LSTM input layer, a Transformer encoding layer, and an output layer; the pre-training process of the intent recognition model includes:

[0105] Step S1031: Obtain historical query text, historical business data and historical user business behavior data, and mark them with intent recognition labels, and use the three types of data and corresponding intent recognition labels as training data sets.

[0106] In this step, user business behavior data can be obtained in the user operation log, which can be understood to include business behavior data of multiple users. The user's historical business data and historical query text are obtained in the business process management system. Each user's historical query text, historical business data and historical user business behavior data can be taken as a group of data, and the training data set includes multiple groups of data.

[0107] The intent recognition labels can be manually labeled by professionals, or the rule engine can be used to automatically label the data with intent recognition labels based on predefined rules.

[0108] Step S1032: Model the sequence information of the three types of data in the training data set through the LSTM input layer to obtain vector representations corresponding to the three types of data.

[0109] In this step, the sequence information in the three input data, such as time series information, node sequence, etc., is encoded through the LSTM feature extraction layer to obtain the corresponding vector representation.

[0110] Step S1033: performing fusion preprocessing on the vector representations to obtain a feature vector.

[0111] The fusion method can refer to the method in the above embodiment, which will not be described again here.

[0112] Step S1034: Iteratively train the feature vector through the multi-head self-attention mechanism of the Transformer encoding layer, and output the predicted intent recognition result through the output layer.

[0113] In the embodiment of the present invention, the Transformer encoding layer sets a three-head attention mechanism to focus on the keywords in the query text, the key links in the business process data, and the key operations in the user behavior. The multi-head attention mechanism can focus on the importance of data from different sources.

[0114] Step S1035: Determine the model corresponding to the minimum loss value between the predicted intent recognition result and the intent recognition label as the final intent recognition model.

[0115] In this step, the cross entropy loss function can be used as the loss function for model training. The model is trained using the training data set, and the gradient of the loss function to the model parameters is calculated using the back propagation algorithm. In each iteration, the model parameters are continuously adjusted according to the gradient descent algorithm. Assuming the learning rate is α and the model parameter is θ, the model parameter update formula is:

[0116]

[0117] Among them, θ is the updated model parameter; θ t is the model parameter in the previous iteration; α is the learning rate; is the gradient of the loss function with respect to the model parameters.

[0118] The model parameters with the smallest loss function value are determined as the parameters of the final intent recognition model.

[0119] The learning rate is a hyperparameter that determines the step size of parameter updates. A learning rate that is too large will result in too fast parameter updates, which may exceed the optimal solution; a learning rate that is too small will result in slow convergence and too long training time. In a feasible implementation, a dynamic adjustment strategy is adopted, such as using a larger learning rate for rapid convergence in the early stage of training, and using a smaller learning rate for fine-tuning in the later stage.

[0120] Based on the above embodiment, calling the corresponding business interface and the business interface corresponding to the business interface based on the intention recognition result includes the following steps:

[0121] Step S1041: Search for the corresponding business interface and the business interface corresponding to the business interface based on the intention recognition result.

[0122] For example, if the intent recognition result is to create performance, the server will search for the business interface for creating performance in the performance system.

[0123] Step S1042: Generate a business card for the searched business interface.

[0124] In this step, the business card is a user interface design pattern. Business cards usually contain a brief description, an icon or a picture, and an action button to guide users to further understand or use the function. Because intent recognition is only to infer the user's intention, it is impossible to be 100% correct, but it is only close to the user's intention with the greatest probability. Therefore, it is pre-displayed to the user in the form of a business card, and the user confirms whether it is what he needs. If so, he can enter the business interface by clicking on the business card. If not, he can continue to enter other query text to continue intent recognition. Improve the user experience and enable users to quickly find and access the functions they are interested in.

[0125] Step S1043: Send the business card to the business interaction interface for display.

[0126] In this step, the business card can be displayed on the business interaction interface in the form of a pop-up window, or it can be directly displayed in the main area of ​​the business interaction interface.

[0127] Step S1044: In response to a click operation on the business card, the business interface corresponding to the interface is sent in the form of a pop-up window to the front-end page where the business interaction interface is located.

[0128] If the user confirms that the content on the business card is what he needs, he can click the button of the business card to enter the business details interface.

[0129] Based on the above embodiment, the method further includes:

[0130] Step S105: If the business interface corresponding to the business interface is not found, a new business interface is generated based on the intent recognition result and the UI components in the front-end component library.

[0131] If there is no existing business interface, a new business interface can be generated based on the UI components in the front-end component library to meet user needs in a timely manner, make business interactions more humane and intelligent, and enhance user experience.

[0132] Based on the above embodiment, generating a new business interface based on the intention recognition result and the UI component in the front-end component library includes:

[0133] Step S1051: Determine basic input items of the business interface based on the intention recognition result.

[0134] For example, to create employee performance, but the performance management system only has interfaces and screens for creating organizational performance, and does not have business interfaces and screens for creating employee performance, then you can refer to the template for creating organizational performance to determine some basic entry items required for performance creation, such as performance name, indicator description, indicator classification, evaluator, etc.

[0135] Step S1052: searching for UI components of each basic entry item in the front-end component library based on the type of the basic entry item.

[0136] In this step, the UI components include a number input box, a text input box, and a drop-down selection box. For example, the number input box is used to configure some number entry items; the text input box is used to configure some names, indicator descriptions, measurement indicators, etc., and the drop-down selection item is used to configure indicator classification, evaluator, status, etc.

[0137] Step S1053: Connect the UI components to obtain a new business interface.

[0138] After the user fills in the information in the business interface and submits it, the filled data will be sent back to the database and presented in the front-end interface for review.

[0139] In another embodiment, historical business behavior data may be used in conjunction with generating a new business interface. For example, when configuring the options in a performance indicator selection drop-down box, drop-down item recommendations may be made in conjunction with the performance indicator types that the user has frequently used.

[0140] Based on the same inventive concept, a service processing device based on intent recognition is provided, such as Figure 2 As shown, the device comprises:

[0141] The acquisition form 201 is used to acquire the query text input by the user on the business interaction interface, the business data of the current business scenario and the business behavior data of the user.

[0142] The fusion unit 202 is used to fuse the query text, the business data in the current business scenario and the business behavior data of the user.

[0143] The recognition unit 203 is used to input the fused data into the pre-trained intention recognition model to obtain the intention recognition result.

[0144] The calling unit 204 is used to call the corresponding business interface and the business interface corresponding to the business interface based on the intention recognition result.

[0145] Based on the same technical concept, an embodiment of the present invention further provides an electronic device, such as Figure 3 As shown, it includes a processor 301 , a communication interface 302 , a memory 303 and a communication bus 304 , wherein the processor 301 , the communication interface 302 , and the memory 303 communicate with each other via the communication bus 304 .

[0146] Memory 303, used for storing computer programs;

[0147] The processor 301 is used to implement the steps of the business processing method based on intent recognition when executing the program stored in the memory 303.

[0148] The communication bus mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. The communication bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, only one thick line is used in the figure, but it does not mean that there is only one bus or one type of bus.

[0149] The communication interface is used for communication between the above electronic device and other devices.

[0150] The memory may include a random access memory (RAM) or a non-volatile memory (NVM), such as at least one disk memory. Optionally, the memory may also be at least one storage device located away from the aforementioned processor.

[0151] The above-mentioned processor can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application specific integrated circuit (ASIC), a field programmable gate array (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components.

[0152] The computer program product for performing business processing based on intent recognition provided in an embodiment of the present invention includes a computer-readable storage medium storing program code. The instructions included in the program code can be used to execute the methods described in the previous method embodiments. The specific implementation can be found in the method embodiments, which will not be repeated here.

[0153] The device for business processing based on intent recognition provided in the embodiment of the present invention can be specific hardware on the device or software or firmware installed on the device. The implementation principle and technical effects of the device provided in the embodiment of the present invention are the same as those of the aforementioned method embodiment. For the sake of brief description, for matters not mentioned in the device embodiment, reference can be made to the corresponding contents in the aforementioned method embodiment. Technical personnel in the relevant field can clearly understand that for the convenience and conciseness of description, the specific working processes of the systems, devices and units described above can all refer to the corresponding processes in the aforementioned method embodiment, and will not be repeated here.

[0154] In the embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of devices or units can be electrical, mechanical or other forms.

[0155] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0156] In addition, each functional unit in the embodiment provided by the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0157] If the functions are implemented in the form of software functional units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product. The computer software product is stored in a storage medium, including several instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the methods described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc., which can store program codes.

[0158] It should be noted that similar numbers and letters represent similar items in the following figures. Therefore, once an item is defined in one figure, it does not need to be further defined and explained in subsequent figures. In addition, the terms "first", "second", "third", etc. are only used to distinguish the description and are not to be understood as indicating or implying relative importance.

[0159] Finally, it should be noted that the above-described embodiments are only specific implementations of the present invention, which are used to illustrate the technical solutions of the present invention, rather than to limit them. The protection scope of the present invention is not limited thereto. Although the present invention is described in detail with reference to the above-described embodiments, ordinary technicians in the field should understand that any technician familiar with the technical field can still modify the technical solutions recorded in the above-described embodiments within the technical scope disclosed by the present invention, or can easily think of changes, or make equivalent replacements for some of the technical features therein; and these modifications, changes or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention. They should all be covered within the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A business processing method based on intent recognition, characterized in that: The method comprises: Obtain the query text entered by the user on the business interaction interface, the business data of the current business scenario, and the user's business behavior data; The query text, the business data in the current business scenario and the business behavior data of the user are integrated and processed; Input the fused data into the pre-trained intent recognition model to obtain the intent recognition result; Based on the intention recognition result, the corresponding business interface and the business interface corresponding to the business interface are called.

2. The method according to claim 1, characterized in that The fusion processing of the query text, the business data in the current business scenario and the business behavior data of the user includes: The query text, the business data in the current business scenario and the business behavior data of the user are respectively converted into vector representations; The three vector representations are concatenated to obtain a concatenated feature vector; The concatenated feature vector is normalized to obtain a target feature vector.

3. The method according to claim 2, characterized in that The converting the query text, the business data in the current business scenario and the business behavior data of the user into vector representations respectively includes: The three types of data are preprocessed separately, including: Delete the special characters in the query text and establish a mapping table between abbreviations and full names; the special characters are at least one of the following: garbled characters, spaces; and performing format verification on the business data in the current business scenario to unify the format; and performing anomaly detection on the business behavior data of the user to eliminate abnormal data; The three types of preprocessed data are respectively subjected to feature extraction and encoding to obtain their respective vector representations, including: Perform word segmentation, stem extraction, part-of-speech tagging and vector mapping on the preprocessed query text to obtain a first vector representation; and encoding the process nodes and business operation types of the preprocessed business data to obtain node codes and operation type codes; Combining the node code and the operation type code to obtain a second vector representation; and encoding the user operation type, operation object, and operation time in the pre-processed user business behavior data to obtain an operation code, an operation object code, and an operation time code; The operation code, operation object code and operation time code are combined to obtain a third vector representation.

4. The method according to claim 1, characterized in that: The intent recognition model includes an LSTM input layer, a Transformer encoding layer, and an output layer. The pre-training process of the intent recognition model includes: Obtain historical query texts, historical business data, and historical user business behavior data, and label them with intent recognition labels. The three types of data and the corresponding intent recognition labels are used as training data sets. Modeling the sequence information of the three types of data in the training data set through the LSTM input layer to obtain vector representations corresponding to the three types of data; Each vector representation is fused and preprocessed to obtain a feature vector; Iteratively train the feature vector through the multi-head self-attention mechanism of the Transformer encoding layer, and output the predicted intent recognition result through the output layer; The model corresponding to the minimum loss value between the predicted intent recognition result and the intent recognition label is determined as the final intent recognition model.

5. The method according to claim 1, characterized in that The calling of the corresponding business interface based on the intention recognition result and the business interface corresponding to the business interface include: Find the corresponding business interface and the business interface corresponding to the business interface based on the intention recognition result; Generate business cards for the searched business interfaces; Sending the business card to the business interaction interface for display; In response to a click operation on the business card, the business interface corresponding to the interface is sent in a pop-up window to the front-end page where the business interaction interface is located.

6. The method according to claim 1, characterized in that The method further comprises: If the business interface corresponding to the business interface is not found, a new business interface is generated based on the intent recognition results and the UI components in the front-end component library.

7. The method according to claim 6, characterized in that The generating of a new business interface based on the intention recognition result and the UI components in the front-end component library includes: Determine the basic entry items of the business interface based on the intent recognition results; Searching for UI components of each basic entry item in the front-end component library based on the type of the basic entry item; After splicing the UI components, a new business interface is obtained.

8. A service processing device based on intent recognition, characterized in that: The device comprises: An acquisition unit, used to acquire the query text entered by the user on the business interaction interface, the business data of the current business scenario, and the business behavior data of the user; A fusion unit, used for fusing the query text, the business data in the current business scenario and the business behavior data of the user; The recognition unit is used to input the fused data into the pre-trained intent recognition model to obtain the intent recognition result; A calling unit is used to call a corresponding business interface and a business interface corresponding to the business interface based on the intention recognition result.

9. An electronic device, characterized in that: It includes a processor, a communication interface, a memory and a communication bus, wherein the processor, the communication interface and the memory communicate with each other through the communication bus; Memory, used to store computer programs; A processor, for implementing the method steps described in any one of claims 1 to 7 when executing a program stored in a memory.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method steps described in any one of claims 1 to 7 are implemented.

Images