Data processing method and device based on intention recognition, computer equipment and storage medium

Abstract

The invention relates to a data processing method and device based on intention recognition, computer equipment, a storage medium and a computer program product. The method comprises the steps of determining a target dialogue parameter and a scene based on a dialogue round corresponding to user question data; and processing the user question data and the scene through an answer reasoning model to obtain initial answer information, and obtaining dynamic scene information through the target dialogue parameter. And arranging the initial answer information through the dynamic scene information to obtain target answer information, and outputting the target answer information. By adopting the method, the output speed and flexibility of answer information can be improved, environment information can be collected in real time, learning and optimization can be performed through an intelligent algorithm, user requirements can be understood more deeply, efficient and accurate intelligent scheduling is realized, initial answer information is adjusted through dynamic scene information, and the user experience is improved. It can be ensured that the output target answer information can fit the instant intention of the user.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a data processing method, apparatus, computer equipment, storage medium and computer program product based on intent recognition. Background Art

[0002] With the continuous development of communication technology and the continuous growth of user needs, higher requirements are placed on the management of operator networks. The scale of operator networks is expanding, and the types of services and the number of users are growing. Operator networks can be managed through scheduling systems. The intelligent system in related technologies processes user question data in tasks based on preset rules and static algorithms, resulting in low accuracy of the processing results. Summary of the invention

[0003] Based on this, it is necessary to provide a data processing method, device, computer equipment, computer-readable storage medium and computer program product based on intent recognition, which can accurately identify user intentions, deeply understand user needs, and output answer information matching user intentions.

[0004] In a first aspect, the present application provides a data processing method based on intent recognition. The method comprises:

[0005] Determine the target conversation parameters and scenarios based on the conversation turns corresponding to the user's question data;

[0006] Processing the user question data and the scenario through an answer reasoning model to obtain initial answer information, and obtaining dynamic scenario information through the target dialogue parameters;

[0007] The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and the target answer information is output.

[0008] In one embodiment, determining the target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data includes:

[0009] If the user question data meets the compliance conditions, the target dialogue parameters and scenarios are determined based on the conversation turn corresponding to the user question data, and the user question data includes one or more of text data, voice data and image data.

[0010] In one embodiment, determining target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data includes:

[0011] If the conversation turn corresponding to the user question data is not the initial conversation turn, the target dialogue parameters and scenes corresponding to the user question data are extracted locally.

[0012] In one embodiment, determining target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data includes:

[0013] If the conversation turn corresponding to the user question data is an initial conversation turn, the target dialogue parameters are obtained from the information library, and the user question data is processed based on the intention recognition model to obtain the target intention, and the scenario corresponding to the target intention is determined.

[0014] In one embodiment, the processing of the user question data based on the intention recognition model to obtain the target intention includes:

[0015] Encode each word in the user's question data through a cyclic encoding module in the intention recognition model to obtain the intention vector corresponding to each word, wherein the cyclic encoding module includes a word-by-word encoding unit, a sentence encoding unit and a semantic slot encoding unit;

[0016] Performing weight screening on the intention vectors corresponding to the respective words to obtain a target intention vector;

[0017] Inputting the target intent vector and the pre-configured slot vector into a slot filling network for iterative updating to obtain the slot filling vectors corresponding to each of the words; inputting the target intent vector and the pre-configured slot vector into an intent classification network for iterative updating to obtain the intent classification vector;

[0018] Through the conditional random field model, the word slot filling vector and the intent classification vector are subjected to intent recognition and intent classification to obtain the target intent.

[0019] In one embodiment, obtaining dynamic scene information through the target dialogue parameter includes:

[0020] If the target dialogue parameter satisfies the preset complete condition, instantiating the target dialogue parameter to obtain the target capability;

[0021] Execute the target capability, obtain the execution result, and obtain dynamic scene information based on the execution result.

[0022] In one embodiment, the initial answer information is arranged by using the dynamic scene information to obtain the target answer information, including:

[0023] Arranging the initial answer information according to the dynamic scene information to obtain arranged initial answer information, and if the arranged initial answer information meets the compliance condition, determining the arranged initial answer information as the target answer information; or,

[0024] The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and if the target answer information meets the compliance condition, the target answer information is output.

[0025] In a second aspect, the present application also provides a data processing device based on intent recognition. The device comprises:

[0026] A first determination module, used to determine target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data;

[0027] A first processing module is used to process the user question data and the scenario through an answer reasoning model to obtain initial answer information, and to obtain dynamic scenario information through the target dialogue parameter;

[0028] The arrangement module is used to arrange the initial answer information through the dynamic scene information to obtain the target answer information and output the target answer information.

[0029] In a third aspect, the present application further provides a computer device. The computer device includes a memory and a processor, the memory stores a computer program, and the processor implements the following steps when executing the computer program:

[0030] Determine the target conversation parameters and scenarios based on the conversation turns corresponding to the user's question data;

[0031] Processing the user question data and the scenario through an answer reasoning model to obtain initial answer information, and obtaining dynamic scenario information through the target dialogue parameters;

[0032] The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and the target answer information is output.

[0033] In a fourth aspect, the present application further provides a computer-readable storage medium. The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0034] Determine the target conversation parameters and scenarios based on the conversation turns corresponding to the user's question data;

[0035] Processing the user question data and the scenario through an answer reasoning model to obtain initial answer information, and obtaining dynamic scenario information through the target dialogue parameters;

[0036] The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and the target answer information is output.

[0037] In a fifth aspect, the present application further provides a computer program product. The computer program product includes a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0038] Determine the target conversation parameters and scenarios based on the conversation turns corresponding to the user's question data;

[0039] Processing the user question data and the scenario through an answer reasoning model to obtain initial answer information, and obtaining dynamic scenario information through the target dialogue parameters;

[0040] The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and the target answer information is output.

[0041] The above-mentioned data processing method, device, computer equipment, storage medium and computer program product based on intention recognition, the method includes: determining the target dialogue parameters and scenes based on the conversation turns corresponding to the user question data; processing the user question data and scenes through the answer reasoning model to obtain initial answer information, and obtaining dynamic scene information through the target dialogue parameters; arranging the initial answer information through the dynamic scene information to obtain the target answer information, and outputting the target answer information. By adopting this method, the output speed and flexibility of the answer information can be improved, the environmental information can be collected in real time, and the intelligent algorithm can be used for learning and optimization, so that the user needs can be understood more deeply, and efficient and accurate intelligent scheduling can be achieved. The initial answer information can be adjusted through the dynamic scene information, and the content can be dynamically adjusted according to the user's needs and scenes, and it can be ensured that the output target answer information can fit the user's immediate intention. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a data processing method based on intent recognition in one embodiment;

[0043] Figure 2 A schematic diagram of a flow chart of a step of obtaining a target intention vector in one embodiment;

[0044] Figure 3 A schematic diagram of a flow chart of steps for obtaining a dynamic scene in one embodiment;

[0045] Figure 4 A schematic diagram of a flow chart of a step of outputting target answer information in one embodiment;

[0046] Figure 5 Schematic diagram of the structure of a GRU model in one embodiment;

[0047] Figure 6 A schematic diagram of a joint slot filling and intent classification model used to obtain target intent in one embodiment;

[0048] Figure 7 The execution process of obtaining the target intention in an embodiment;

[0049] Figure 8 is a flowchart of a data processing method based on intent recognition in another embodiment;

[0050] Figure 9a A schematic diagram of an interactive page for inputting user question data to obtain target answer information in one embodiment;

[0051] Figure 9b A schematic diagram of an interactive page for inputting user question data to obtain target answer information in one embodiment;

[0052] Fig.10 is a structural block diagram of a data processing device based on intent recognition in one embodiment;

[0053] Fig.11 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0054] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0055] The present application embodiment provides a data processing method based on intent recognition, such as Figure 1As shown, the execution subject of the data processing method based on intent recognition can be a terminal device or a server or other processing device, wherein the terminal device can be a user equipment (User Equipment, UE), a mobile device, a user terminal, a terminal, a cellular phone, a cordless phone, a personal digital assistant (Personal Digital Assistant, PDA), a handheld device, a computing device, a vehicle-mounted device, a wearable device, etc. In some possible implementations, the data processing method based on intent recognition can be implemented by a processor calling a computer-readable instruction stored in a memory. It can be understood that the method can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server. In this embodiment, the data processing method based on intent recognition is applied to a terminal device as an example for explanation. The data processing method based on intent recognition includes the following steps:

[0056] Step 102: Determine target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data.

[0057] The user question data may be question data input by a user terminal, and the user question data includes multiple representation forms, such as text, voice, and interactive operation with a graphical interface. The conversation round corresponding to the user question data may be the conversation round in which the question data input by the user is currently located, such as a new conversation, that is, an initial conversation round, or an existing conversation, that is, the user question data is input based on an existing conversation, and the existing conversation represents a non-new conversation. The target conversation parameter may include user attributes and parameters, and the user attributes may include identity feature information of the user device that inputs the user question data, and the parameter may be a location parameter that represents the current abnormality, such as a link location parameter where a fault occurs, and the like. The scenario may be a service scenario corresponding to the session, such as a dial-up Internet service scenario, a broadband service scenario, and the like.

[0058] Specifically, the user device can input user question data to the terminal device, and the terminal device can determine the current conversation turn corresponding to the user question data based on the acquired user question data. After determining the conversation turn corresponding to the current user question data, the terminal device can obtain the target object parameters and scenes corresponding to the user question data based on the conversation turn.

[0059] In one example, the terminal device may respond to an input operation by a user and obtain text data corresponding to the input operation or voice data corresponding to the input operation, or image data corresponding to the input operation, etc.; in another example, the terminal device may display a questioning interface, and the terminal device may respond to a triggering operation of the questioning interface by a user and obtain user question data corresponding to the triggering operation; for example, a first question option and a second question option may be displayed on the questioning interface, and the terminal device may respond to a triggering operation of the first question option by the user and determine that the user question data is the question data corresponding to the first question option.

[0060] Step 104, the user question data and the scene are processed by the answer reasoning model to obtain initial answer information, and dynamic scene information is obtained through the target dialogue parameters.

[0061] Among them, the answer reasoning model can be a machine learning model that can contain multiple parameters. The answer reasoning model can be trained based on sample question data, the scenarios corresponding to the sample question data, and the sample answer information corresponding to the sample question data; the initial answer information can be determined based on the output result of the answer reasoning model; the dynamic scene information can be dynamic scene content, which can characterize the user's personalized needs and the scenario in which he is in, and is used to adjust the initial answer information output by the answer reasoning model. The dynamic scene information includes a multi-dimensional complex information set, which characterizes the state and behavior of the user asking the question, as well as the environmental state of the user asking the question and the change data of the environmental state, etc., where the environmental state can be the scene in which the user asking the question is in.

[0062] Specifically, the terminal device can input the user's question data and the scene corresponding to the user's question data into the answer reasoning model, and infer or predict the user's question data and the scene through the various parameters contained in the answer reasoning model to obtain the initial answer information for answering the user's question data. And the terminal device can process based on the target dialogue parameter to obtain the dynamic scene information corresponding to the target dialogue parameter.

[0063] Step 106, the initial answer information is arranged by using the dynamic scene information to obtain the target answer information, and the target answer information is output.

[0064] Specifically, the terminal device can adjust the initial answer information through dynamic scene information. For example, the content of the initial answer information can be dynamically adjusted based on the user's needs and scenes represented by the dynamic scene information to obtain the target answer information; the dependency and interaction method between the initial answer information and the scene can also be defined to obtain the target answer information; the target answer information can also be adjusted based on the context of the conversation corresponding to the user's question data to obtain the adjusted target answer information, and the adjusted target answer information can be output, for example, the target answer information can be displayed on the terminal device. The process of arranging the initial answer information to obtain the target answer information can make the output target answer information match the user's real-time intention.

[0065] In the above-mentioned data processing method based on intention recognition, the target dialogue parameters and scenes are determined based on the conversation turns corresponding to the user question data. The user question data and scenes are processed by the answer reasoning model to obtain the initial answer information, and the dynamic scene information is obtained by the target dialogue parameters. The initial answer information is arranged by the dynamic scene information to obtain the target answer information, and the target answer information is output. By adopting this method, the output speed and flexibility of the answer information can be improved, the environmental information can be collected in real time, and learning and optimization can be performed through intelligent algorithms. The user needs can be understood more deeply, and efficient and accurate intelligent scheduling can be achieved. The initial answer information can be adjusted by dynamic scene information, and the content can be dynamically adjusted according to the user's needs and scenes, which can ensure that the output target answer information can fit the user's immediate intentions.

[0066] In one embodiment, the specific processing of the step of "determining target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data" includes:

[0067] If the user question data meets the compliance conditions, the target conversation parameters and scenarios are determined based on the conversation turn corresponding to the user question data.

[0068] The user question data includes one or more of text data, voice data and image data, and the user question data includes multiple representation forms, such as text, voice and interactive operation with a graphical interface. The compliance condition can be a condition for determining whether the user question data is compliant, such as determining whether there are no prohibited words in the user question data and the format complies with the specification.

[0069] In one example, the terminal device may respond to an input operation by a user and obtain text data corresponding to the input operation or voice data corresponding to the input operation, or image data corresponding to the input operation, etc.; in another example, the terminal device may display a questioning interface, and the terminal device may respond to a triggering operation of the questioning interface by a user and obtain user question data corresponding to the triggering operation; for example, a first question option and a second question option may be displayed on the questioning interface, and the terminal device may respond to a triggering operation of the first question option by the user and determine that the user question data is the question data corresponding to the first question option.

[0070] Specifically, after obtaining the user question data, the terminal device can determine whether the user question data meets the preset compliance conditions. If it is determined that the user question data is compliant, the terminal device can determine the current conversation turn of the user question data, and based on the current conversation turn of the user question data, determine the target dialogue parameters corresponding to the user question data and the scenario in which the user question data is located.

[0071] In one example, the current session round of the user question data may be an initial session round, that is, the user question data is a new session initiated by the user to the terminal device; the current session round of the user question data may also be a non-initial session round, that is, the user question data is sent by the user to the terminal device through an existing session between the user and the terminal device.

[0072] In another example, the process of the terminal device determining whether the user question data is compliant may be that the terminal device may determine whether the format of the user question data complies with the specification, and may also determine whether the user question data contains prohibited words, which may be words that are unsafe, irregular, or have bad meanings; if the terminal device determines that the format of the user question data complies with the specification, and the user question data does not contain prohibited words, then the terminal device may determine that the user question data is compliant, that is, it meets the preset compliance conditions.

[0073] In this embodiment, by processing the user question data after determining its compliance, the processing efficiency of the user question data can be guaranteed and resource waste can be avoided. By determining the corresponding target dialogue parameters and scenarios through the determined conversation turns, the flexibility of parameter determination can be improved.

[0074] In one embodiment, the specific processing of the step of "determining target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data" includes:

[0075] If the conversation turn corresponding to the user question data is not the initial conversation turn, the target conversation parameters and scenes corresponding to the user question data are extracted locally.

[0076] The initial session round indicates that the user question data is sent to the terminal device through a new session established between the user and the terminal device, and there is no session between the user and the terminal device, or the existing session has ended, etc. The local may be a local terminal device, for example, a database storing terminal device data, and the database may be set inside the terminal device or outside the terminal device and connected to the terminal device.

[0077] Specifically, the terminal device can judge the conversation round of the user question data. If it is determined that the current conversation round of the user question data is not the initial conversation round, that is, it is determined that the user question data is input to the terminal device through an existing conversation between the user and the terminal device, the terminal device can locally extract the target conversation parameters and scenes corresponding to the user question data. In one example, the terminal device locally stores one or more target conversation parameters and scenes of existing conversations. If the terminal device determines that the conversation round corresponding to the user question data is not the initial conversation round, that is, the user question data is input through an existing conversation (for example, it can be a first existing conversation), the terminal device can query the first existing conversation in the locally stored data and extract the target conversation parameters and scenes corresponding to the first existing conversation.

[0078] In this embodiment, by determining the conversation turn of the user's question data, when the conversation turn of the user's question data is an existing conversation, the target conversation parameters and scenario of the existing conversation can be directly obtained locally, thereby ensuring the efficiency of determining the conversation parameters and scenario and avoiding additional consumption of resources.

[0079] In one embodiment, the specific processing of the step of "determining target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data" includes:

[0080] If the conversation turn corresponding to the user question data is the initial conversation turn, the target conversation parameters are obtained from the information library, and the user question data is processed based on the intent recognition model to obtain the target intent and determine the scenario corresponding to the target intent.

[0081] The information library may be an API information library, which may store a variety of dialogue parameter information; the intent recognition model may be a trained cyclic GRU model, which is used to recognize user intent based on user input data.

[0082] Specifically, if the terminal device determines that the conversation turn of the user question data is an initial conversation turn, that is, it determines that the user inputs the user question data to the terminal device through a new conversation between the terminal device, then the terminal device can extract the target dialogue parameters from the preset API information library. The terminal device can also perform intent recognition on the user question data based on the recurrent GRU model to obtain the user intent represented by the user question data, that is, the target intent; based on this, the terminal can determine the scene represented by the target intent, or the scene where the target intent is located.

[0083] In this embodiment, by determining the conversation turn of the user's question data, when the conversation turn of the user's question data is the initial conversation, the user's question data can be processed based on the intention recognition model, so as to accurately identify the user's intention and further improve the recognition efficiency and the output efficiency of the answer information.

[0084] In one embodiment, Figure 2 As shown in FIG. 1 , the specific processing process of the step “processing the user question data based on the intent recognition model to obtain the target intent” includes:

[0085] Step 202: Encode each word in the user's question data through a recurrent encoding module in the intention recognition model to obtain an intention vector corresponding to each word.

[0086] Among them, the cyclic encoding module includes a word-by-word encoding unit, a sentence encoding unit and a semantic slot encoding unit.

[0087] Specifically, the terminal device can determine multiple words based on the user question data, and encode the multiple words corresponding to the user question data through the recursive coding module in the intention recognition model to obtain the coding results corresponding to each word, and determine the coding results as the intention vectors corresponding to each word.

[0088] In one example, the terminal device can encode each word in the user's question data through the word-by-word encoding unit in the circular encoding module to obtain a first encoding result, and input the first encoding result into the sentence encoding unit for encoding to obtain a second encoding result, and input the second encoding result into the semantic slot encoding unit to obtain a third encoding result, and use the third encoding result to determine the intention vector corresponding to the user's question data.

[0089] In another example, the terminal device may process each word in the user's question data separately through one or more units of a word-by-word encoding unit, a sentence encoding unit, and a semantic slot encoding unit to obtain the intention vector corresponding to each word.

[0090] Among them, the intention vector output by the recurrent encoding module is a hidden vector containing semantic information. The intention vector is obtained after capturing the overall meaning of the sentence in the user's question data. The intention vector can also represent the word details and contextual relationships in the user's question data. The initial position of the intention vector can be the [CLS] tag.

[0091] Step 204: weight the intention vectors corresponding to the respective words to obtain a target intention vector.

[0092] Specifically, the terminal device can filter the intent vectors corresponding to each word through the weight screening model, and determine the intent vector with the largest weight as the target intent vector. The terminal device can perform weight screening through the following formula:

[0093] y i =σ(W(Relu(W i x i +b i )))

[0094] Among them, W and W i is the weight matrix, σ is the activation function, b i is the bias matrix, y i is the intention vector obtained for the i-th word.

[0095] That is to say, the terminal device can calculate the weights corresponding to the intention vectors of each word, and sort the intention vectors in descending order based on the weights corresponding to each intention vector, and filter out the intention vector with the largest weight, and determine the intention vector with the largest weight as the target intention vector.

[0096] Step 206: Input the target intent vector and the pre-configured slot vector into the slot filling network for iterative update to obtain the slot filling vectors corresponding to each word. Input the target intent vector and the pre-configured slot vector into the intent classification network for iterative update to obtain the intent classification vector.

[0097] Step 208, using the conditional random field model, perform intent recognition and intent classification on the slot filling vector and the intent classification vector to obtain the target intent.

[0098] Specifically, the terminal device can input the target intent vector and the pre-configured slot vector into the slot filling network to obtain the slot filling vectors of each word. The terminal device can also input the target intent vector and the pre-configured slot vector into the intent classification network to obtain the intent classification vector; based on this, the terminal device can perform intent recognition and intent classification on the slot filling vector and the intent classification vector through the Conditional Random Fields (CRF) model to obtain the target intent.

[0099] In this embodiment, by processing the user question data based on the intention recognition model, the user intention can be accurately identified, further improving the recognition efficiency and the output efficiency of the answer information.

[0100] In one embodiment, Figure 3 As shown, the specific processing process of the step "obtaining dynamic scene information through target dialogue parameters" includes:

[0101] Step 302: If the target dialogue parameters meet the preset complete conditions, the target dialogue parameters are instantiated to obtain the target capability.

[0102] Specifically, the terminal device can determine whether the currently extracted target dialogue parameter is complete. If it is determined that the target dialogue parameter is complete, the target dialogue parameter can be instantiated by API to obtain the target capability. The specific process of instantiation can be to convert the target dialogue parameter into a code file, determine the code file as the instance capability of the target dialogue parameter, that is, the target capability, and in the subsequent process, the instance capability can be directly executed to call the step of executing the instance capability to obtain the capability execution result. The target capability can be called by other devices.

[0103] Step 304, execute the target capability, obtain the execution result, and obtain dynamic scene information based on the execution result.

[0104] Specifically, the terminal device may execute the target capability corresponding to the target dialogue parameter and obtain the execution result of the target capability; the terminal device may determine the execution result as dynamic scene information.

[0105] In this embodiment, environmental information is collected based on intelligent agent technology, and learning and optimization are performed to gradually improve the ability to perceive and understand the environment, thereby deeply understanding user needs, performing intelligent arrangement and execution, and improving the match between the output answer information and the user's question data.

[0106] In one embodiment, Figure 4 As shown, the specific processing process of the step "arranging the initial answer information through dynamic scene information to obtain the target answer information" includes:

[0107] Step 402, the initial answer information is arranged according to the dynamic scene information to obtain the arranged initial answer information, and if the arranged initial answer information meets the compliance condition, the arranged initial answer information is determined to be the target answer information. Or,

[0108] Among them, the compliance condition can be used to determine whether the format complies with the specification and whether there are prohibited words. The content of the compliance condition is similar to the compliance condition in the above embodiment and will not be repeated here.

[0109] Specifically, the terminal device can adjust the initial answer information through dynamic scene information. For example, the content of the initial answer information can be dynamically adjusted based on the needs and scenes of the user represented by the dynamic scene information to obtain the arranged answer information. The dependency and interaction method between the initial answer information and the scene can also be defined, and the initial answer information can be adjusted based on the context of the conversation corresponding to the user's question data to obtain the arranged answer information. The terminal device can determine whether the arranged initial answer information meets the compliance conditions. If the terminal device determines that the arranged initial answer information is compliant, the terminal device can determine that the arranged initial answer information is the target answer information.

[0110] Step 404, the initial answer information is arranged by using the dynamic scene information to obtain the target answer information, and if the target answer information meets the compliance conditions, the target answer information is output.

[0111] Specifically, the terminal device can adjust the initial answer information through dynamic scene information. For example, the content of the initial answer information can be dynamically adjusted based on the user's needs and scenes represented by the dynamic scene information to obtain the target answer information. The dependency and interaction method between the initial answer information and the scene can also be defined, and the initial answer information can be adjusted based on the context of the conversation corresponding to the user's question data to obtain the target answer information. The terminal device can determine whether the target answer information meets the compliance conditions. If the terminal device determines that the target initial answer information is compliant, the terminal device directly outputs the target answer information to prompt the user device that inputs the user's question data.

[0112] It should be noted that the present disclosure does not limit the execution order of step 402 and step 404. In one example, only step 402 may be executed, and in another example, only step 404 may be executed.

[0113] In this embodiment, dynamic scene information is determined by perceiving the environment, and the content of the initial answer information is arranged based on the dynamic scene information. This allows the output answer information to be highly matched with the user's real-time intention, achieves a high degree of customization and targeting of the answer information, and improves the intelligence and accuracy of information transmission.

[0114] The specific implementation process of the above-mentioned data processing method based on intent recognition is described in detail below in conjunction with a specific embodiment:

[0115] With the continuous development of communication technology and the growing needs of users, operators are facing unprecedented challenges. The expansion of network scale, the increase in business types, and the surge in the number of users have put forward higher requirements for operators' operation and management. The scheduling system can realize the automatic scheduling and intelligent scheduling of tasks, helping operators to better manage and utilize resources and improve network performance and service quality.

[0116] The scheduling methods in related technologies are based on simple priority rules or static optimization models, and only rely on preset rules and static algorithms to process tasks for scheduling, which limits the response speed when facing environmental changes or unforeseen situations. When external conditions change rapidly, it is difficult to adjust strategies in time, resulting in inefficiency or failure to correctly and deeply understand the user's intentions, resulting in misjudgment and erroneous scheduling, making the system's scheduling accuracy low and reliability weak; moreover, the traditional intent recognition model uses two independent models, intent classification and semantic slot filling, to process tasks, and the two independent models have problems such as error accumulation.

[0117] The data processing method based on intent recognition provided in this embodiment is based on intelligent agent technology, artificial intelligence, machine learning, and deep learning. It can extract useful information from massive data through learning algorithms and data analysis to form its own knowledge base. In the decision-making process, the intelligent agent can comprehensively consider various factors, use logical reasoning, probability statistics and other methods to make the best decision, and has significant advantages in solving complex problems. By collecting environmental information and using algorithms for learning and optimization, the ability to perceive and understand the environment is gradually improved, so that complex tasks can be completed autonomously and the ability to understand and identify user intentions can be gradually improved, achieving a higher level of intelligent scheduling. The data processing method based on intent recognition provided in this embodiment is an intelligent agent expert orchestration execution method that can deeply understand user needs. It can make autonomous compliance judgments on the content input by the user (user question data), and judge whether the session established by the user corresponding to the user question data is a new conversation. According to the scenario, by sharing coding information, the two tasks of intent recognition and semantic slot filling are processed simultaneously, thereby making full use of the correlation between intent and slot value, performing functions such as intent recognition on user intent, and gradually improving the ability to perceive and understand the environment by collecting environmental information and using algorithms for learning and optimization, so as to autonomously complete complex tasks, achieve global optimal resource allocation, avoid waste of computing resources and avoid delays in key tasks; improve the response speed and flexibility to user question data input by the user, and when faced with environmental changes or unforeseen situations, it can quickly adapt and react in time, output corresponding answer information, and improve user experience. Specifically, the data processing method based on intent recognition provided by this embodiment has great improvements in response speed, flexibility, resource allocation optimization, intent recognition accuracy, etc.; more specifically, in terms of improving the response speed and flexibility of the scheduling system, the method provided by this embodiment introduces intelligent agent technology, so that the system can collect environmental information in real time, use algorithms for learning and optimization, so that it can quickly adjust strategies in the face of environmental changes or unforeseen situations, and improve scheduling efficiency and accuracy. In terms of deepening the understanding and recognition of user intentions, the method provided by this embodiment shares coding information and processes two tasks, intent recognition and semantic slot filling, at the same time, thereby making full use of the correlation between intent and slot value, and improving the recognition accuracy and reliability of user intentions. In terms of improving the in-depth understanding of user needs, the method provided by this embodiment can gradually improve the ability to perceive and understand the environment through continuous learning and optimization, so that it can understand user needs more deeply and achieve a higher level of intelligent scheduling.

[0118] like Figure 5 As shown, it can be a structural diagram of the GRU model:

[0119] Specifically, the GRU model can effectively capture and store long-term dependencies in sequence data by introducing gating mechanisms (including update gates and reset gates). This is crucial for understanding the continuity and contextual relevance of user intent. The recurrent GRU structure enables the model to not only focus on the current intent when processing user input, but also combine previous intent information to more accurately understand the user's overall intent, so that GRU units at different levels can be interdependent, highlighting the important features of user data.

[0120] The following is the calculation process of GRU:

[0121] z t =σ(w z ·concat(h t-1 , x t )+b z )

[0122] r t =σ(w r ·concat(h t-1 , x t )+b r )

[0123]

[0124]

[0125] Among them, x t is the input data of the GRU model, that is, the input information at the current moment, h t-1 is the hidden state of the previous moment, which can act as a neural network and contains the data information passed by the GRU model at the previous moment; h t is the hidden state at the current moment, that is, the hidden state that can be passed to the next moment, is a candidate hidden state; r t is the reset gate, zt is the update gate, σ() is the sigmoid function, which can convert data into values ​​in the range of 0-1; through the tanh() function, data can be converted into values ​​in the range of [-1, 1]. · represents dot product, ⊙ represents dot product. In other words, the GRU model can be transformed into a numerical value in the range of [-1, 1] by inputting data x t And the hidden state h of the GRU model at the previous moment t-1 Processing is performed to obtain the output yt of the GRU model at the current moment and the hidden state h that can be passed to the next moment t . b z , b r , b h Can be a pre-configured weight parameter.

[0126] like Figure 6 As shown in the figure, it can be a schematic diagram of the joint slot filling and intent classification model used to obtain the target intent. The overall operation process is as follows: First, the intent input by the user is encoded through the recurrent GRU structure. The encoding layer includes word-by-word encoding, sentence encoding, and semantic slot encoding. This process converts the vector input by the user into a hidden vector rich in semantic information. This vector not only captures the overall meaning of the sentence, but also retains the details and contextual relationships therein, and adds the [CLS] tag before the sequence.

[0127] The input information is X = {[CLS], X 0 , ..., X n}, the output is in

[0128] V = CycleGRU(X i )

[0129] The terminal device can use the intention vector output by CycleGRU to select the intention vector with a large weight through the weight screening model to obtain the target intention vector:

[0130] y i =σ(W(Relu(W i x i +b i )))

[0131] Among them, W and W i is the weight matrix, σ is the activation function, b i is the bias matrix, y i It is the intention result obtained by the i-th word and is sorted.

[0132] The intent vector and slot vector output after encoding are input into the slot filling network and the intent classification network for iterative update to generate new vectors. Finally, the Conditional Random Fields (CRF) model is used to perform intent recognition and classification to obtain the target intent.

[0133] Specifically, the terminal device may input information X={[CLS], X 0 , ..., X n} are input into the multi-layer attention mechanism module respectively. The input information X can contain multiple words, for example, it can be X 0 , ..., X n The terminal device obtains the attention results corresponding to each word through the multi-layer attention mechanism module, and inputs the attention results corresponding to each word into different GRU models respectively. Through the multi-layer recurrent GRU model, the intention vector output by the GRU model can be obtained. The intent vector output by the GRU model is input into the slot filling network and the intent classification network respectively. slot And the intent E output by the intent classification network intent vector Iterate and update the slot filling network and the intent classification network to obtain the output information y of the slot filling network. 1 ,y 2 ,……,y n , and the output information y of the intent classification network intent vector ; Output information y of slot filling network through CRF model 1 ,y 2 ,……,y nn Processing is performed to obtain multiple keywords, such as "xx province", "5G" and "user"; the output information y of the intent classification network intent vector Processing is performed to obtain output text information, such as "how many 5G users are there in xx province". For example, xx province may be Anhui Province.

[0134] like Figure 7 As shown in FIG. 1 , the execution process of sorting and obtaining the target intent is as follows: the user input data and the output result of the cyclic connection GRU model are processed by the prompt model to obtain the processed user input data, and the processed user input data is processed by the general model to obtain multiple user intents. For example, the user intent Figure 1 、User opinion Figure 2 、User opinion Figure 3 And user intention Figure 4 ; Sort each user intention through the output results of the weight screening model and the recurrent connection GRU model, determine the user intention with the largest weight as the target intention, and output the target intention.

[0135] like Figure 8As shown, it can be a specific implementation process of a specific embodiment: User question: The user can express his intention in a variety of ways, such as text, voice, graphical interface, etc. Automatically determine whether the question is compliant: The content security module reviews the question based on the three indicators of "question + user attribute + parameter". If the question is not compliant, the question will be sent to the user side. Conversation round judgment: The memory module makes a judgment based on the current conversation round. If it is a new conversation, the intelligent body will save the conversation, and the intention recognition module will try to obtain the scene information. If the scene information is successfully obtained, the API group will obtain the API information. If it is an existing conversation, parameter extraction is performed. By creating an API instance, the developer can call its method to send requests and receive responses. Parameter extraction: The API group extracts the parameters of the existing conversation based on the API information. Large model reasoning: The process of taking the scene content as input and performing reasoning or prediction on a large machine learning model with a large number of parameters and complex structures. The large model arranges the content of the general answer through reasoning. Content arrangement: Take the dynamic scene content and the general answer as input, and define the dependency and interaction method between them. Content compliance judgment: Determine whether the input content is compliant, and return it to the user side if it is not in compliance with the rules.

[0136] The above embodiments describe in detail how the agent handles user questions, including steps such as identifying questions, extracting parameters, and executing capabilities. In this way, the agent can provide accurate and efficient answers to meet user needs.

[0137] like Figure 9aAs shown, the user can trigger the AI ​​interface, and the interface can display a first prompt message, which is used to prompt the user to input questions, such as "Hello, I am your home broadband smart assistant Xiaox, I am happy to serve you! What problems have you encountered with your home broadband network? You can tell me and I will help you solve it~"; the user question data entered by the user can be, for example, "My home broadband cannot access the Internet", and the intelligent agent in the embodiment provided in this embodiment can perform identity recognition and intention recognition, and the interface can display a second prompt message, which is used to comfort the user, for example, "Dear, I am very sorry to bring you a bad experience", and the terminal device can obtain the user identity information in the background to confirm the user identity. After determining that the user has multiple broadbands to be processed, the third prompt message and multiple broadband options can be displayed on the interface, such as broadband a and broadband b; the terminal can determine the target broadband based on the user's selection operation of the broadband option, such as broadband a; based on this, the terminal can perform intention recognition and scene confirmation, identify the user's intention obstacle phenomenon, and call the large model knowledge base to obtain possible reasons. Secondly, perform obstacle diagnosis, query the broadband account and installation status, and determine whether there are arrears, authentication status, group obstacles, etc. Then it interacts with the user and displays a check prompt message, for example, "OK, the broadband you applied for is: broadband a; please wait, we will check it for you right away, the whole process will take about 30 seconds." After the fault problem is determined, the cause of the fault and the solution strategy are displayed, for example, it can be displayed on the AI ​​interface, "Dear, sorry to have kept you waiting. Your account is in arrears, the amount in arrears is xx yuan, please go to xx method to pay." and display the options that the user has operated, for example, it can be "paid". In response to the triggering operation of the user's already operated option, such as Figure 9b As shown, other fault prompt information can be displayed, for example, "It has been recognized that you have withdrawn the fee, and you currently have no outstanding fees. Your account status has returned to normal." The interface can display the cause of the fault and solution prompt options, for example, "Your password is wrong. The terminal dial-up password needs to be synchronized. Please confirm." The solution prompt option can be a confirmation option and a cancel option. Based on the user's triggering operation on the solution prompt option, solution prompt information and solutions can be displayed. The solution prompt information can be "Sorry, you chose to cancel the synchronization password. You can enter the router yourself, change the dial-up password and try again. The following is a common routing configuration manual. Please select a brand: Brand 1,..., Brand n", and the solution can be a prompt picture. The AI ​​interface can also display result feedback and a feedback information input area for users to enter feedback information. The terminal device can follow up based on the feedback information entered by the user to further improve the user experience.

[0138] The method provided in this embodiment can be based on the user intention sorting algorithm of the cyclic GRU structure: by utilizing the time series propagation characteristics and gating mechanism of the cyclic GRU, the user's intention can be more accurately identified and sorted, which can further improve the accuracy and efficiency of the algorithm. Based on the intelligent agent technology, the intelligent agent collects environmental information, uses algorithms for learning and optimization, and gradually improves its ability to perceive and understand the environment, thereby deeply understanding user needs and performing intelligent arrangement and execution.

[0139] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0140] Based on the same inventive concept, the embodiment of the present application also provides a data processing device based on intent recognition for implementing the data processing method based on intent recognition involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more embodiments of the data processing device based on intent recognition provided below can refer to the limitations of the data processing method based on intent recognition above, and will not be repeated here.

[0141] In one embodiment, Fig.10 As shown, a data processing device 1000 based on intention recognition is provided, comprising:

[0142] A first determination module 1002, configured to determine target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data;

[0143] The first processing module 1004 is used to process the user question data and the scenario through the answer reasoning model to obtain initial answer information, and obtain dynamic scenario information through the target dialogue parameter;

[0144] The arrangement module 1006 is used to arrange the initial answer information according to the dynamic scene information, obtain target answer information, and output the target answer information.

[0145] In one embodiment, the first determining module is specifically used to:

[0146] If the user question data meets the compliance conditions, the target dialogue parameters and scenarios are determined based on the conversation turn corresponding to the user question data, and the user question data includes one or more of text data, voice data and image data.

[0147] In one embodiment, the first determining module is further specifically configured to:

[0148] If the conversation turn corresponding to the user question data is not the initial conversation turn, the target dialogue parameters and scenes corresponding to the user question data are extracted locally.

[0149] In one embodiment, the first determining module is further specifically configured to:

[0150] If the conversation turn corresponding to the user question data is an initial conversation turn, the target dialogue parameters are obtained from the information library, and the user question data is processed based on the intention recognition model to obtain the target intention, and the scenario corresponding to the target intention is determined.

[0151] In one embodiment, the first determining module is further specifically configured to:

[0152] Encode each word in the user's question data through a cyclic encoding module in the intention recognition model to obtain the intention vector corresponding to each word, wherein the cyclic encoding module includes a word-by-word encoding unit, a sentence encoding unit and a semantic slot encoding unit;

[0153] Performing weight screening on the intention vectors corresponding to the respective words to obtain a target intention vector;

[0154] Inputting the target intent vector and the pre-configured slot vector into a slot filling network for iterative updating to obtain the slot filling vectors corresponding to each of the words; inputting the target intent vector and the pre-configured slot vector into an intent classification network for iterative updating to obtain the intent classification vector;

[0155] Through the conditional random field model, the word slot filling vector and the intent classification vector are subjected to intent recognition and intent classification to obtain the target intent.

[0156] In one embodiment, the first processing module is specifically used for:

[0157] If the target dialogue parameter satisfies the preset complete condition, instantiating the target dialogue parameter to obtain the target capability;

[0158] Execute the target capability, obtain the execution result, and obtain dynamic scene information based on the execution result.

[0159] In one embodiment, the orchestration module is specifically used to:

[0160] Arranging the initial answer information according to the dynamic scene information to obtain arranged initial answer information, and if the arranged initial answer information meets the compliance condition, determining the arranged initial answer information as the target answer information; or,

[0161] The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and if the target answer information meets the compliance condition, the target answer information is output.

[0162] Each module in the above-mentioned data processing device based on intent recognition can be implemented in whole or in part by software, hardware and a combination thereof. Each of the above-mentioned modules can be embedded in or independent of the processor in the computer device in the form of hardware, or can be stored in the memory of the computer device in the form of software, so that the processor can call and execute the operations corresponding to each of the above modules.

[0163] In one embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as follows: Fig.11 As shown. The computer device includes a processor, a memory and a network interface connected through a system bus. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store relevant data of the data processing method based on intent recognition. The network interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a data processing method based on intent recognition is implemented.

[0164] Those skilled in the art will understand that Fig.11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0165] In one embodiment, a computer device is further provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above method embodiments when executing the computer program.

[0166] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0167] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0168] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties.

[0169] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0170] The technical features of the above embodiments may be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0171] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A data processing method based on intention recognition, characterized in that: The method comprises: Determine the target conversation parameters and scenarios based on the conversation turns corresponding to the user's question data; Processing the user question data and the scenario through an answer reasoning model to obtain initial answer information, and obtaining dynamic scenario information through the target dialogue parameters; The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and the target answer information is output.

2. The method according to claim 1, characterized in that The determining of target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data includes: If the user question data meets the compliance conditions, the target dialogue parameters and scenarios are determined based on the conversation turn corresponding to the user question data, and the user question data includes one or more of text data, voice data and image data.

3. The method according to claim 1, characterized in that: The determining of target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data includes: If the conversation turn corresponding to the user question data is not the initial conversation turn, the target dialogue parameters and scenes corresponding to the user question data are extracted locally.

4. The method according to claim 1, characterized in that: The determining of target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data includes: If the conversation turn corresponding to the user question data is an initial conversation turn, the target dialogue parameters are obtained from the information library, and the user question data is processed based on the intention recognition model to obtain the target intention, and the scenario corresponding to the target intention is determined.

5. The method according to claim 4, characterized in that The processing of the user question data based on the intention recognition model to obtain the target intention includes: Encode each word in the user's question data through a cyclic encoding module in the intention recognition model to obtain the intention vector corresponding to each word, wherein the cyclic encoding module includes a word-by-word encoding unit, a sentence encoding unit and a semantic slot encoding unit; Performing weight screening on the intention vectors corresponding to the respective words to obtain a target intention vector; Inputting the target intention vector and the pre-configured slot vector into the slot filling network for iterative update to obtain the word slot filling vectors corresponding to each of the words; inputting the target intention vector and the pre-configured slot vector into the intention classification network for iterative update to obtain the intention classification vector; Through the conditional random field model, the word slot filling vector and the intent classification vector are subjected to intent recognition and intent classification to obtain the target intent.

6. The method according to claim 1, characterized in that The obtaining of dynamic scene information through the target dialogue parameter includes: If the target dialogue parameter satisfies the preset complete condition, instantiating the target dialogue parameter to obtain the target capability; Execute the target capability, obtain the execution result, and obtain dynamic scene information based on the execution result.

7. The method according to any one of claims 1 to 6, characterized in that: The step of arranging the initial answer information by using the dynamic scene information to obtain target answer information includes: Arranging the initial answer information by using the dynamic scene information to obtain arranged initial answer information, and if the arranged initial answer information meets the compliance condition, determining the arranged initial answer information as the target answer information; or, The initial answer information is arranged by using the dynamic scene information to obtain target answer information, and if the target answer information meets the compliance condition, the target answer information is output.

8. A data processing device based on intention recognition, characterized in that: The device comprises: A first determination module, used to determine target dialogue parameters and scenarios based on the conversation turn corresponding to the user question data; A first processing module is used to process the user question data and the scenario through an answer reasoning model to obtain initial answer information, and to obtain dynamic scenario information through the target dialogue parameter; The arrangement module is used to arrange the initial answer information through the dynamic scene information to obtain the target answer information and output the target answer information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

11. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images