User intention identification method and device based on multiple models, equipment and medium

Abstract

The invention provides a user intention recognition method and device based on multiple models, equipment and a medium, and relates to the field of artificial intelligence, in particular to the field of intention recognition. According to the specific implementation scheme, in response to current request information of a user, intention recognition processing is conducted on the current request information based on a preset first model, and a recognition result is obtained; wherein the identification result represents whether a preset first model outputs a user intention corresponding to the current request information; if the identification result represents that the user intention corresponding to the current request information is not obtained, intention identification processing is carried out on the current request information based on a preset second model, and the user intention corresponding to the current request information is obtained; wherein the parameter quantity of the first model is smaller than the parameter quantity of the second model. Two models are adopted, the recall rate of intention recognition is increased, and the user experience is improved.

Description

Technical Field

[0001] The present disclosure relates to the field of intention recognition in the field of artificial intelligence, and in particular to a method, device, equipment and medium for recognizing user intention based on multiple models. Background Art

[0002] In the customer service scenario of intelligent robots, understanding the user's intention plays a vital role in improving the user experience. By understanding the user's intention, it can save the user's time and provide efficient services to the user.

[0003] However, user language is full of colloquial expressions, and the same intention can be expressed in many different ways. This ambiguity and diversity makes it difficult for robot customer service to accurately understand the user's true intentions, affecting the user experience. Summary of the invention

[0004] The present disclosure provides a method, apparatus, device and medium for identifying user intent based on multiple models.

[0005] According to a first aspect of the present disclosure, a method for identifying user intent based on multiple models is provided, comprising:

[0006] In response to the user's current request information, based on the preset first model, the current request information is subjected to intent recognition processing to obtain a recognition result; wherein the recognition result indicates whether the preset first model outputs the user intent corresponding to the current request information;

[0007] If the recognition result indicates that the user intention corresponding to the current request information is not obtained, then based on a preset second model, intent recognition processing is performed on the current request information to obtain the user intention corresponding to the current request information; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

[0008] According to a second aspect of the present disclosure, a multi-model based user intention recognition device is provided, comprising:

[0009] A first recognition unit is used to respond to the user's current request information and perform intent recognition processing on the current request information based on a preset first model to obtain a recognition result; wherein the recognition result indicates whether the preset first model outputs the user intention corresponding to the current request information;

[0010] A second identification unit is used to perform intent identification processing on the current request information based on a preset second model to obtain the user intention corresponding to the current request information if the identification result indicates that the user intention corresponding to the current request information is not obtained; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

[0011] According to a third aspect of the present disclosure, there is provided an electronic device, including:

[0012] at least one processor; and

[0013] a memory communicatively coupled to the at least one processor;

[0014] The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect of the present disclosure.

[0015] According to a fourth aspect of the present disclosure, a non-transitory computer-readable storage medium storing computer instructions is provided, wherein the computer instructions are used to cause the computer to execute the method according to the first aspect of the present disclosure.

[0016] According to a fifth aspect of the present disclosure, a computer program product is provided, comprising a computer program, which implements the steps of the method described in the first aspect of the present disclosure when executed by a processor.

[0017] According to the technology disclosed in the present invention, the accuracy of intent recognition is improved and the user experience is enhanced.

[0018] It should be understood that the content described in this section is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it intended to limit the scope of the present disclosure. Other features of the present disclosure will become easily understood through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0019] The accompanying drawings are used to better understand the present solution and do not constitute a limitation of the present disclosure.

[0020] Figure 1 is a flowchart of a method for identifying user intent based on multiple models provided according to an embodiment of the present disclosure;

[0021] Figure 2 is a schematic diagram of a process of mining training data provided according to an embodiment of the present disclosure;

[0022] Figure 3 is a flowchart of a method for identifying user intention based on multiple models provided according to an embodiment of the present disclosure;

[0023] Figure 4 is a flowchart of a method for identifying user intention based on multiple models provided according to an embodiment of the present disclosure;

[0024] Figure 5 is a flowchart of a method for identifying user intention based on multiple models provided according to an embodiment of the present disclosure;

[0025] Figure 6 is a structural block diagram of a multi-model based user intention recognition device provided according to an embodiment of the present disclosure;

[0026] Figure 7 is a structural block diagram of a multi-model based user intention recognition device provided according to an embodiment of the present disclosure;

[0027] Figure 8 is a block diagram of an electronic device for implementing the method for identifying user intent based on multiple models according to an embodiment of the present disclosure;

[0028] Fig. 9 The block diagram is a block diagram of an electronic device used to implement the method for identifying user intent based on multiple models according to an embodiment of the present disclosure. DETAILED DESCRIPTION

[0029] The following is a description of exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, including various details of the embodiments of the present disclosure to facilitate understanding, which should be considered as merely exemplary. Therefore, it should be recognized by those of ordinary skill in the art that various changes and modifications may be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, descriptions of well-known functions and structures are omitted in the following description.

[0030] Intelligent robots are used in various fields to bring convenience to people's lives. In the customer service scenarios of intelligent robots, understanding the user's intention plays a vital role in improving the user experience. However, user language is full of colloquial expressions, and the same intention can be expressed in many different ways. This ambiguity and diversity make it difficult for robots to accurately understand the user's true intention.

[0031] The traditional way to identify user intent is to perform text classification through BERT (Bidirectional Encoder Representation of Transformer, a pre-trained model) to identify the user's specific intent. The main problem with this method is that the directions covered by user language are diverse, and the BERT-based classification model is an intelligent customer service robot for a specific vertical category, which requires a large amount of training data. However, the amount of data for training downstream tasks is limited, the generalization ability of the model is poor, and it cannot cope with a wide variety of user questions. For user questions that are beyond the training set, the effect of BERT intent recognition is poor. That is, when the user's question does not appear in the data of the training set, the generalization ability of the model is limited, and it is often difficult to deal with the user's "out-of-scope" questions, resulting in low recognition accuracy of user intent, affecting user experience.

[0032] The present disclosure provides a method, apparatus, device and medium for identifying user intent based on multiple models, which are applied to the field of intent recognition in the field of artificial intelligence to improve the recall rate and accuracy of intent recognition.

[0033] It should be noted that the model in this embodiment is not targeted at a specific user and cannot reflect the personal information of a specific user. It should be noted that the data in this embodiment comes from a public data set.

[0034] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision and disclosure of user personal information involved are in compliance with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0035] In order to enable readers to more deeply understand the implementation principle of the present disclosure, the following Figure 1-Figure 9 The embodiment is further refined.

[0036] Figure 1 FIG. 1 is a flow chart of a method for identifying user intention based on multiple models according to an embodiment of the present disclosure. The method can be performed by a device for identifying user intention based on multiple models. Figure 1 As shown, the method comprises the following steps:

[0037] S101. In response to the user's current request information, based on a preset first model, perform intent recognition processing on the current request information to obtain a recognition result; wherein the recognition result indicates whether the preset first model outputs the user's intent corresponding to the current request information.

[0038] For example, the user can input his request information on the terminal device at any time according to the needs, for example, the user can input the request information in the form of text or voice. The terminal device can be installed with an intelligent robot customer service, which can be an AI model that can recognize the intent of the user's request information and thus communicate with the user.

[0039] The user's request information can be obtained in real time, that is, the current request information is obtained. For example, the current request information is "How much does a model A car cost now?" In response to the user's current request information, the preset first model is called. The first model is a small model. For example, the first model is a BERT model, and the pre-trained small model is applied to a specific vertical intelligent customer service robot. Vertical category refers to the strong binding and strong correlation between intent classification and business logic in a specific field. For example, in the intelligent robot customer service scenario in the automotive field, most of the questions raised by users are related to cars. If the user asks "How much can it be done after landing?", then this is asking about the landing price of the car, which is a price-related intention. The BERT model can answer in the field of automobiles and improve the accuracy of the answer. In this embodiment, the model structure of the first model is not specifically limited.

[0040] The BERT model is a pre-trained model, and downstream tasks are fine-tuned on a small amount of data to adapt to specific tasks. The BERT model uses a bidirectional attention mechanism to capture contextual information and is widely used in natural language processing tasks. The current request information entered by the user extracts sentence-level features through the BERT network and is sent to the classifier for classification. The number of intent types is pre-set. For example, there are three preset intent types, namely inquiry type, discount type, and test drive type. If the current request information is "How much is this car? Is it discounted now?", it can be determined that the user's intention is inquiry type and discount type.

[0041] The amount of training data for the first model is limited, and the current request information may not be within the scope of the training data, that is, the current request information is an "out-of-scope" question. For user questions that are beyond the training data, the first model may not be able to obtain the user's intent. After calling the first model, it can be determined whether the first model can output the user's intent corresponding to the current request information. Based on the recognition result of the first model, it can be determined whether the first model obtains the user's intent corresponding to the current request information. For example, a recognition result of 0 indicates that the user's intent corresponding to the current request information is not obtained; a recognition result of 1 indicates that the user's intent corresponding to the current request information is obtained. If the first model obtains the user's intent corresponding to the current request information, the recognition of the user's intent is completed, and communication with the user can be carried out based on the identified user's intent.

[0042] S102. If the recognition result indicates that the user intention corresponding to the current request information is not obtained, then based on a preset second model, the current request information is subjected to intent recognition processing to obtain the user intention corresponding to the current request information; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

[0043] Exemplarily, if it is determined based on the recognition result that the first model has not obtained the user intent corresponding to the current request information, the preset second model is called again. The second model is used to perform intent recognition processing on the current request information to obtain the user intent corresponding to the current request information. That is, the current request information is input into the second model, and the category of the user intent to which the current request information belongs is output. In this embodiment, the second model can be a large model, and the number of parameters of the second model is much larger than the number of parameters of the first model. In this embodiment, the model structure of the second model is not specifically limited.

[0044] The number of model parameters of BERT is about 100 million, while the number of parameters of the large model is now as high as about 175 billion. The pre-training data of the large model is much larger than that of BERT. The large model stores a lot of knowledge and its ability to understand text far exceeds that of small models such as BERT. However, the pre-training data of the large model does not include private domain data in a specific field, and its accuracy in intent recognition is often not as good as that of the small model. In addition, the intelligent customer service robot has relatively high requirements for reasoning time, and the reasoning speed of the large model is slow. Therefore, the small model, that is, the first model, is preferred for online intent prediction. When the small model has no prediction results, the large model will be used as a backup. 90% of user intentions are undertaken by the small model. Without affecting the average online time consumption, the knowledge of the large model is fully utilized to improve the coverage of intent understanding.

[0045] In this embodiment, the method also includes: obtaining historical request information and a first intention of the historical request information; wherein the historical request information represents request information within a preset historical time period, and the first intention is a user intention corresponding to the historical request information determined by a first model; based on a preset second model, performing intent recognition processing on the historical request information to obtain a second intention of the historical request information; if the first intention and the second intention are inconsistent, determining the historical request information as training data; wherein the training data is used to train the first model.

[0046] Specifically, during the working process of the intelligent robot customer service, request information within a preset historical time period can be obtained in real time or periodically as historical request information. The historical request information can be request information issued by a single user or multiple users in the past period of time. When the user issues the historical request information, the first model will perform intent recognition on the historical request information at that time to obtain the user intent corresponding to the historical request information. The identified user intent corresponding to each request information can be stored in real time. That is, the intent recognized by the first model for the historical request information can be directly obtained as the first intent.

[0047] Alternatively, the historical request information may be obtained first, and then the obtained historical request information may be input into a preset first model in real time, and the first model may perform intent recognition on the historical request information to obtain the first intent of the historical request information.

[0048] After obtaining the historical request information, the historical request information can be input into the second model, and the second model performs intent recognition processing on the historical request information, and determines the recognized user intent as the second intent.

[0049] For each historical request information, the first intention is compared with the second intention to determine whether there is a difference between the first intention and the second intention. If so, the historical request information is used as training data for the first model for training the first model; if not, it means that the first model can accurately identify the intent of the historical request information, and the historical request information is not used as new training data.

[0050] Figure 2 Schematic diagram of the process of mining training data. Figure 2 In the model, query is the request information raised by the user, the first model is the BERT model, and the second model is the LLM (Large Language Model), i.e., the large model. Historical online queries are regularly retrieved, and the intent of the query is predicted with the help of the knowledge of the large model itself. Historical queries with different predictions between the first model and the second model are selected, and after manual review and annotation, these queries are added to the training data, and the training data is continuously expanded, so that the first model can be iteratively optimized based on the new training data.

[0051] The beneficial effect of this setting is that it can dig out the "out-of-scope" problems of online users, continuously improve the coverage of the first model's understanding of intent, and form a closed-loop iterative update of the first model. When the distribution of request information changes, it can also be perceived in the first place, improving the recognition accuracy of user intent and improving user experience.

[0052] In this embodiment, obtaining historical request information and the first intention of the historical request information includes: obtaining historical request information; deduplicating the historical request information to obtain deduplicated historical request information; and obtaining the first intention of the deduplicated historical request information.

[0053] Specifically, there may be duplicate request information in the historical request information. After obtaining a batch of historical request information, the batch of historical request information can be deduplicated, that is, the duplicate historical request information can be removed. For example, the number of words in the same text in every two historical request information can be determined. If the number of words in the same text exceeds a preset word count threshold, the two historical request information are considered to be duplicate historical request information. In this embodiment, the deduplication process is not specifically limited.

[0054] For the historical request information retained after deduplication, the first intent of the historical request information is determined. The first intent can be directly obtained from the record of historical intent recognition, or the historical request information retained after deduplication can be input into the first model to obtain the corresponding first intent.

[0055] The beneficial effect of this setting is that by deduplicating historical request information, data redundancy is reduced and the efficiency of training data mining is improved.

[0056] In this embodiment, the method further includes: if it is determined according to the recognition result that the user intention corresponding to the current request information is not obtained, then the current request information is determined as training data.

[0057] Specifically, the training data of the first model can also be mined at any time during the application of the intelligent robot customer service. For example, for the real-time current request information, determine whether the first model can output the corresponding user intent. If so, store the current request information as historical request information, and record the user intent corresponding to the historical request information output by the first model as the first intent, to facilitate subsequent mining; if not, the current request information can be directly determined as the training data of the first model for subsequent update training of the first model.

[0058] The beneficial effect of this setting is that it can timely mine the data that the small model lacks, improve the mining accuracy and efficiency of training data, enable the small model to cover the diverse expressions of users, and improve the accuracy and efficiency of intent recognition.

[0059] In the disclosed embodiment, when using an intelligent robot for human-computer interaction, the current request information sent by the user can be responded to, and the first model can be used first for intent recognition. When the first model has no prediction results, the second model will be used as a backup to improve the recall rate of intent recognition and ensure that the user's request information can be understood. In actual applications, 90% of user intentions are taken over by the first model. Without affecting the average online time consumption, the knowledge of the second model is fully utilized to increase the coverage of intent understanding, improve the accuracy and efficiency of intent recognition, and enhance the user's human-computer interaction experience.

[0060] Figure 3 A flowchart of a method for identifying user intent based on multiple models is provided in an embodiment of the present disclosure.

[0061] In this embodiment, based on the preset first model, the current request information is processed for intent recognition to obtain a recognition result, including: determining the current dialogue state based on the current request information; wherein the current dialogue state represents that the current request information is in a single-round dialogue or in multiple-round dialogues; based on the preset first model and the current dialogue state, the current request information is processed for intent recognition to obtain a recognition result.

[0062] This embodiment is based on the above embodiment. Figure 3 As shown, the method comprises the following steps:

[0063] S301. In response to current request information of a user, determine a current dialog state according to the current request information; wherein the current dialog state indicates that the current request information is in a single-round dialog or in multiple-round dialogs.

[0064] Exemplarily, in real-time response to the user's current request information, the preset first model is first called to perform intent recognition processing on the current request information. When performing intent recognition processing through the first model, the dialogue state of the current request information can be first determined as the current dialogue state.

[0065] The current conversation state can indicate whether the current request information is in a single-round conversation or in multiple-round conversations. For example, if the current request information is the first request information in a period of time, then the current request information is in a single-round conversation; although the current request information is not the first request information, it is not related to the previous request information, that is, the user has started a new conversation, then the current request information is also in a single-round conversation. If the current request information has a related previous context, then the current request information is in multiple-round conversations.

[0066] In this embodiment, determining the current dialog state according to the current request information includes: obtaining context information of the current request information; and determining the current dialog state according to the context information and the current request information.

[0067] Specifically, the previous information of the current request information refers to the request information issued before the current request information, and the previous information issued within a preset time period can be obtained, or a preset number of previous information can be obtained. For example, 5 pieces of information before the current request information are obtained as the previous information of the current request information.

[0068] Combine the current request information and the previous information to determine whether the current conversation is in a single-round conversation or a multi-round conversation, that is, determine the current conversation state. For example, it can be determined whether there is a semantic association between the previous information and the current request information. If so, it is determined that the conversation is in a multi-round conversation; if not, it is determined that the conversation is in a single-round conversation.

[0069] The beneficial effect of such a setting is that, by combining the above information, the dialogue state is determined and the accuracy of identifying the intention of the current request for information is improved.

[0070] In this embodiment, the current conversation state is determined based on the above information and the current request information, including: performing semantic recognition processing on the above information and the current request information respectively to obtain semantic information corresponding to the above information and semantic information corresponding to the current request information; determining the similarity between the semantic information corresponding to the above information and the semantic information corresponding to the current request information; wherein the similarity represents the degree of association between the semantic information; and determining the current conversation state based on the similarity.

[0071] Specifically, a semantic recognition algorithm is pre-set, and according to the preset semantic recognition algorithm, semantic recognition processing is performed on the above information to obtain semantic information corresponding to the above information, and semantic recognition processing is performed on the current request information to obtain semantic information corresponding to the current request information. In this embodiment, the preset semantic recognition algorithm is not specifically limited.

[0072] Determine the similarity between the semantic information corresponding to the above information and the semantic information corresponding to the current request information. The similarity can represent the degree of association between the semantic information. For example, the similarity can be calculated using a cosine similarity calculation method. If the similarity is equal to or greater than a preset similarity threshold, the current dialogue state is determined to be in a multi-round dialogue; if the similarity is less than the preset similarity threshold, the current dialogue state is determined to be in a single-round dialogue.

[0073] The beneficial effect of such a setting is that it determines whether the above information and the current request information are related, thereby determining what information is needed to determine the intent recognition, thereby improving the accuracy of the intent recognition.

[0074] S302: Perform intent recognition processing on the current request information according to the preset first model and the current dialog state to obtain a recognition result.

[0075] Exemplarily, for different current dialogue states, the first model can perform different processing procedures on the current request information. For example, the association relationship between different dialogue states and processing procedures can be preset, and the corresponding processing procedure can be determined according to the current dialogue state. Through the first model, the current request information is processed in a corresponding manner, and the recognition result is output. For example, for a single-round dialogue, the first model can extract keywords from the current request information, determine whether the category of the user's intention can be obtained based on the keywords, and thus obtain the recognition result; for multiple-round dialogues, the first model can combine the above information and the current request information, perform feature extraction on the combined information, and obtain the recognition result.

[0076] In this embodiment, the current dialog state is determined, and targeted processing of the first model is performed according to the current dialog state to improve the accuracy of intent recognition and enhance user experience.

[0077] In this embodiment, according to the preset first model and the current dialogue state, the current request information is processed for intent recognition to obtain a recognition result, including: if the current dialogue state represents that the current request information is in a single-round dialogue, then according to the single-round intention prediction sub-model in the preset first model, the current request information is processed for intent recognition to obtain a recognition result; wherein the single-round intention prediction sub-model is a branch in the preset first model.

[0078] Specifically, different model branches may be provided in the first model for processing request information in different dialogue states. For example, one branch in the first model may be a single-round intent prediction sub-model, which is used to identify the intent of request information in a single-round dialogue. If it is determined that the current request information is in a single-round dialogue according to the current dialogue state, the single-round intent prediction sub-model in the first model is called to perform intent recognition processing on the current request information to obtain a recognition result. The single-round intent prediction sub-model may contain network layers such as convolutional layers. In this embodiment, the model structure of the single-round intent prediction sub-model is not specifically limited.

[0079] For example, the first model first determines the conversation turn represented by the current conversation state, and then calls the corresponding sub-model branch to process the current request information. If the user's intention cannot be classified, the second model is used as a backup.

[0080] The beneficial effect of such a setting is that the user intention can be predicted for a single-round dialogue, thereby avoiding confusion between the processing of a single-round dialogue and multiple-round dialogues, and effectively improving the prediction accuracy of the user intention.

[0081] In this embodiment, according to the preset first model and the current dialogue state, the current request information is processed for intent recognition to obtain a recognition result, including: if the current dialogue state represents that the current request information is located in a multi-round dialogue, then according to the multi-round intention prediction sub-model in the preset first model, the current request information and the above information are processed for intent recognition to obtain a recognition result; wherein the multi-round intention prediction sub-model is a branch in the preset first model.

[0082] Specifically, different model branches may be provided in the first model for processing request information under different dialogue states. For example, one branch in the first model may be a multi-round intent prediction sub-model, which is used to identify the intent of request information under multi-round dialogues. If it is determined that the current request information is in a multi-round dialogue according to the current dialogue state, the multi-round intent prediction sub-model in the first model is called to perform intent recognition processing on the current request information to obtain a recognition result. For example, the multi-round intent prediction sub-model may fuse the above information and the current request information, and perform intent recognition processing on the fused information. The multi-round intent prediction sub-model may contain network layers such as convolutional layers. In this embodiment, the model structure of the multi-round intent prediction sub-model is not specifically limited.

[0083] The beneficial effect of such a setting is that the user intention can be predicted for multiple rounds of dialogue, thus avoiding confusion between the processing of single-round dialogue and multiple rounds of dialogue, and effectively improving the prediction accuracy of user intention.

[0084] S303. If the recognition result indicates that the user intention corresponding to the current request information is not obtained, the current request information is subjected to intent recognition processing based on a preset second model to obtain the user intention corresponding to the current request information; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

[0085] Exemplarily, this step may refer to the above-mentioned step S102 and will not be described in detail.

[0086] In the disclosed embodiment, when using an intelligent robot for human-computer interaction, the current request information sent by the user can be responded to, and the first model can be used first for intent recognition. When the first model has no prediction results, the second model will be used as a backup to improve the recall rate of intent recognition and ensure that the user's request information can be understood. In actual applications, 90% of user intentions are taken over by the first model. Without affecting the average online time consumption, the knowledge of the second model is fully utilized to increase the coverage of intent understanding, improve the accuracy and efficiency of intent recognition, and enhance the user's human-computer interaction experience.

[0087] Figure 4 A flowchart of a method for identifying user intent based on multiple models is provided in an embodiment of the present disclosure.

[0088] In this embodiment, based on a preset first model, intent recognition processing is performed on the current request information to obtain a recognition result, including: semantic recognition processing is performed on the current request information to obtain the ambiguity of the current request information; wherein the ambiguity represents the clarity of the expression of the current request information; if the ambiguity is less than a preset ambiguity threshold, a clarification prompt message is sent to the user; wherein the clarification prompt message is used to prompt the user to send new request information; new request information is obtained, and a recognition result is determined based on the preset first model, the new request information, and the current request information.

[0089] This embodiment is based on the above embodiment. Figure 4 As shown, the method comprises the following steps:

[0090] S401. In response to the user's current request information, semantic recognition processing is performed on the current request information to obtain the fuzziness of the current request information; wherein the fuzziness represents the clarity of expression of the current request information.

[0091] Exemplarily, the user's current request information is obtained, and semantic recognition processing is performed on the current request information to determine whether the semantics of the current request information is clearly described, and the fuzziness of the current request information is obtained. The fuzziness can represent the clarity of the expression of the current request information. The higher the fuzziness, the less clear the expression of the current request information is.

[0092] For example, the current request information sent by the user is "how much does a car cost?", and through semantic recognition, it can be determined that the user has not specified the brand or model of the car, the description is unclear, and the ambiguity is high.

[0093] S402: If the ambiguity is less than a preset ambiguity threshold, a clarification prompt message is sent to the user; wherein the clarification prompt message is used to prompt the user to send a new request message.

[0094] Exemplarily, a fuzziness threshold is pre-set, and the determined fuzziness is compared with the fuzziness threshold. If the fuzziness is less than the preset fuzziness threshold, it means that the current request information is clearly expressed and does not need to be clarified. The current request information can be directly identified for intent, or the current dialogue state of the current request information can be determined, and then the intent can be identified based on the current dialogue state. If the fuzziness is equal to or greater than the preset fuzziness threshold, it means that the current request information is not clearly expressed, and the user's intent cannot be identified based on the current request information. It is necessary to send a clarification prompt message to the user. The clarification prompt message can prompt the user to send a new request message, that is, prompt the user to clarify the current request message. For example, through human-computer interaction, a message can be sent to the user to prompt the user to re-describe the current request information. The user sends a new request message to describe or supplement the current request message.

[0095] S403: Obtain new request information, and determine a recognition result according to the preset first model, the new request information, and the current request information.

[0096] Exemplarily, the request information sent by the user is obtained in real time, that is, the new request information sent by the user can be obtained in real time. The first model is used to perform intent recognition on the current request information and the new request information to obtain a recognition result. That is, the current request information and the new request information are both input into the first model, and the recognition result is output. The multi-round intent prediction sub-model in the first model can be used for intent recognition.

[0097] In this embodiment, for unclearly expressed current request information, the user is prompted to clarify so that the user's expression can represent his or her intention, which facilitates subsequent intention recognition and improves the accuracy of intention recognition.

[0098] In this embodiment, the recognition result is determined based on the preset first model, the new request information, and the current request information, including: fusing the new request information and the current request information to obtain the information to be processed; performing intent recognition processing on the information to be processed according to the preset first model to obtain the recognition result.

[0099] Specifically, new request information is obtained, the new request information is merged with the current request information, and the merged information is determined as the information to be processed. For example, the new request information and the current request information can be listed in a data set, and the data set is determined as the information to be processed.

[0100] The information to be processed is input into the first model, and the first model is used to perform intent recognition processing on the information to be processed. For example, the first model is a BERT model, and the information to be processed is subjected to sentence-level features extracted through the BERT network and sent to the classifier for classification, thereby obtaining a recognition result.

[0101] The beneficial effect of such a setting is that, through multiple rounds of clarification, the user's intention is clarified, context information is combined for intent recognition, the intent recognition accuracy of the first model is improved, and the user experience is enhanced.

[0102] S404. If the recognition result indicates that the user intention corresponding to the current request information is not obtained, the current request information is subjected to intent recognition processing based on a preset second model to obtain the user intention corresponding to the current request information; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

[0103] Exemplarily, this step may refer to the above-mentioned step S102 and will not be described in detail.

[0104] In the disclosed embodiment, when using an intelligent robot for human-computer interaction, the current request information sent by the user can be responded to, and the first model can be used first for intent recognition. When the first model has no prediction results, the second model will be used as a backup to improve the recall rate of intent recognition and ensure that the user's request information can be understood. In actual applications, 90% of user intentions are taken over by the first model. Without affecting the average online time consumption, the knowledge of the second model is fully utilized to increase the coverage of intent understanding, improve the accuracy and efficiency of intent recognition, and enhance the user's human-computer interaction experience.

[0105] Figure 5 A flowchart of a method for identifying user intent based on multiple models is provided in an embodiment of the present disclosure.

[0106] In this embodiment, based on the preset second model, the current request information is subjected to intent recognition processing to obtain the user intent corresponding to the current request information, including: generating prompt words based on a preset prompt word template according to the current request information; wherein the prompt words are data in text format and represent the current request information; the prompt words are input into the preset second model to obtain the user intent corresponding to the current request information; wherein the preset second model is a large language model.

[0107] This embodiment is based on the above embodiment. Figure 5 As shown, the method comprises the following steps:

[0108] S501. In response to the user's current request information, based on a preset first model, perform intent recognition processing on the current request information to obtain a recognition result; wherein the recognition result represents whether the preset first model outputs the user's intent corresponding to the current request information.

[0109] Exemplarily, this step may refer to the above-mentioned step S101 and will not be described in detail.

[0110] S502: If the recognition result indicates that the user intention corresponding to the current request information is not obtained, a prompt word is generated according to the current request information based on a preset prompt word template; wherein the prompt word is data in a text format, representing the current request information.

[0111] Exemplarily, if it is determined that the recognition result represents the user intention that does not correspond to the current request information, the second model can be used as a backup to ensure that the user intention can be obtained. The second model can be a large model. During the application of the large model, prompt (prompt word) can be used to assist in understanding the large model. Prompt is data in text form, and a prompt template is pre-set. According to the current request information, the prompt template is supplemented to obtain a complete prompt, that is, prompt can represent the current request information.

[0112] For example, the preset prompt word template can be:

[0113] "You are a professional salesperson in the automotive field and need to classify user messages by intent. Below are the names and definitions of [User Intent]. Please read them carefully and classify the intent of user messages according to the definitions.

[0114] Please note:

[0115] 1. If the user's intent is unclear, please answer "No clear intent".

[0116] 2. The answer can only output [Intent Name].

[0117] 3. If the user has multiple intentions, multiple intention names can be output, separated by -.

[0118]

Intent definition

[0119] [Price]: Query the price of the car, including the bare car price, landing price, etc.

[0120] [Parameters]: Query the configuration of the vehicle model, including configuration parameters, fuel consumption, vehicle width and other parameter values

[0121] [Discount]: Query the discount activities of car models, including price reduction, car purchase discounts, etc.

[0122]

User message 1

[0123] How much is the down payment

[0124]

Intent Classification 1

[0125] Staging Plan

[0126]

User message 2

[0127] What is the price of this car? Is there any discount?

[0128]

Intent Classification 2

[0129] Price-Discount

[0130]

User message

[0131] {question}

[0132] "Intent Classification"

[0133] Among them, "User message 1" is an example in the template, "Intent category 1" is the category of the user intent corresponding to the example; "User message 2" is another example in the template, and "Intent category 2" is the category of the user intent corresponding to the example. A template can include multiple examples. "{question}" is used to supplement the user's current request information.

[0134] S503: Input the prompt word into a preset second model to obtain the user intention corresponding to the current request information; wherein the preset second model is a large language model.

[0135] For example, the large model stores a lot of knowledge and has a much better understanding of text than a small model like BERT. The current request information is passed into the prompt template, and then the prompt is input into the large model, which then outputs the user intent corresponding to the current request information.

[0136] In this embodiment, the advantage of the large model is that it has better generalization ability for users' "out-of-scope" questions. When users' questions exceed the training set, BERT often has difficulty coping. As a backup model for BERT, the large model can expand the recall rate of intent understanding, that is, make full use of the large amount of knowledge possessed by the large model to improve the generalization ability of intent recognition. At the same time, offline, by screening the request information that is different from the prediction results of the large model and BERT, after manual review, the data that the small model lacks can be continuously excavated, the coverage of the small model can be improved, and the closed-loop iteration of the small model can be achieved, thereby improving the accuracy and efficiency of intent recognition.

[0137] In the disclosed embodiment, when using an intelligent robot for human-computer interaction, the current request information sent by the user can be responded to, and the first model can be used first for intent recognition. When the first model has no prediction results, the second model will be used as a backup to improve the recall rate of intent recognition and ensure that the user's request information can be understood. In actual applications, 90% of user intentions are taken over by the first model. Without affecting the average online time consumption, the knowledge of the second model is fully utilized to increase the coverage of intent understanding, improve the accuracy and efficiency of intent recognition, and enhance the user's human-computer interaction experience.

[0138] Figure 6 The structure block diagram of a multi-model based user intention recognition device provided by the embodiment of the present disclosure. For the convenience of explanation, only the part related to the embodiment of the present disclosure is shown. Figure 6 , a user intention recognition device 600 based on multiple models includes: a first recognition unit 601 and a second recognition unit 602 .

[0139] The first recognition unit 601 is used to respond to the user's current request information and perform intent recognition processing on the current request information based on a preset first model to obtain a recognition result; wherein the recognition result indicates whether the preset first model outputs the user intent corresponding to the current request information;

[0140] The second recognition unit 602 is used to perform intent recognition processing on the current request information based on a preset second model to obtain the user intention corresponding to the current request information if the recognition result indicates that the user intention corresponding to the current request information is not obtained; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

[0141] Figure 7 A structural block diagram of a multi-model based user intention recognition device provided in an embodiment of the present disclosure, such as Figure 7 As shown, the multi-model based user intention recognition device 700 includes a first recognition unit 701 and a second recognition unit 702 , wherein the first recognition unit 701 includes a state determination module 7011 and a result acquisition module 7012 .

[0142] The state determination module 7011 is used to determine the current dialogue state according to the current request information; wherein the current dialogue state indicates that the current request information is in a single-round dialogue or in multiple-round dialogues;

[0143] The result obtaining module 7012 is used to perform intent recognition processing on the current request information according to the preset first model and the current dialogue state to obtain the recognition result.

[0144] In one example, the state determination module 7011 includes:

[0145] The above-text acquisition submodule is used to obtain the above-text information of the current request information;

[0146] The state determination submodule is used to determine the current dialog state according to the above information and the current request information.

[0147] In one example, the state determination submodule is specifically used to:

[0148] Performing semantic recognition processing on the above information and the current request information respectively to obtain semantic information corresponding to the above information and semantic information corresponding to the current request information;

[0149] Determine the similarity between the semantic information corresponding to the above information and the semantic information corresponding to the current request information; wherein the similarity represents the degree of association between the semantic information;

[0150] The current dialog state is determined according to the similarity.

[0151] In one example, the result is module 7012, including:

[0152] A single-round identification submodule is used to perform intent identification processing on the current request information to obtain the identification result according to the single-round intention prediction submodel in the preset first model if the current dialogue state represents that the current request information is in a single-round dialogue; wherein the single-round intention prediction submodel is a branch in the preset first model.

[0153] In one example, the result is module 7012, including:

[0154] A multi-round identification submodule is used to perform intent identification processing on the current request information and the above information according to the multi-round intention prediction submodel in the preset first model to obtain the identification result if the current dialogue state represents that the current request information is located in a multi-round dialogue; wherein the multi-round intention prediction submodel is a branch in the preset first model.

[0155] In one example, the first identification unit 701 includes:

[0156] A fuzziness determination module, used to perform semantic recognition processing on the current request information to obtain the fuzziness of the current request information; wherein the fuzziness represents the clarity of the expression of the current request information;

[0157] a prompt module, configured to send a clarification prompt message to the user if the ambiguity is less than a preset ambiguity threshold; wherein the clarification prompt message is used to prompt the user to send a new request message;

[0158] The result determination module is used to obtain new request information and determine the recognition result according to the preset first model, the new request information, and the current request information.

[0159] In one example, the result determination module includes:

[0160] A fusion submodule, used to fuse the new request information with the current request information to obtain information to be processed;

[0161] The recognition submodule is used to perform intent recognition processing on the information to be processed according to the preset first model to obtain the recognition result.

[0162] In one example, the second identification unit 702 includes:

[0163] A prompt word generation module, used to generate a prompt word according to the current request information and based on a preset prompt word template; wherein the prompt word is data in a text format, representing the current request information;

[0164] The intention obtaining module is used to input the prompt word into the preset second model to obtain the user intention corresponding to the current request information; wherein the preset second model is a large language model.

[0165] In one example, it also includes:

[0166] A history acquisition unit, used to acquire historical request information and a first intention of the historical request information; wherein the historical request information represents request information within a preset historical time period, and the first intention is a user intention corresponding to the historical request information determined by a first model;

[0167] A history recognition unit, configured to perform intent recognition processing on the history request information based on a preset second model to obtain a second intent of the history request information;

[0168] An intention comparison unit is used to determine the historical request information as training data if the first intention and the second intention are inconsistent; wherein the training data is used to train the first model.

[0169] In one example, the history acquisition unit includes:

[0170] History acquisition module, used to obtain historical request information;

[0171] A deduplication module, used for performing deduplication processing on the historical request information to obtain deduplicated historical request information;

[0172] An intent acquisition module is used to acquire the first intent of the deduplicated historical request information.

[0173] In one example, it also includes:

[0174] A data determination unit is used to determine the current request information as the training data if it is determined according to the recognition result that the user intention corresponding to the current request information is not obtained.

[0175] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device.

[0176] Figure 8 A structural block diagram of an electronic device provided in an embodiment of the present disclosure, such as Figure 8As shown, the electronic device 800 includes: at least one processor 802; and a memory 801 communicatively connected to the at least one processor 802; wherein the memory stores instructions executable by the at least one processor 802, and the instructions are executed by the at least one processor 802 so that the at least one processor 802 can execute the multi-model-based user intent recognition method disclosed in the present invention.

[0177] The electronic device 800 further includes a receiver 803 and a transmitter 804. The receiver 803 is used to receive instructions and data sent by other devices, and the transmitter 804 is used to send instructions and data to external devices.

[0178] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, a readable storage medium and a computer program product.

[0179] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which includes: a computer program, the computer program is stored in a readable storage medium, at least one processor of an electronic device can read the computer program from the readable storage medium, and at least one processor executes the computer program so that the electronic device executes the solution provided by any of the above embodiments.

[0180] Fig. 9 A schematic block diagram of an example electronic device 900 that can be used to implement an embodiment of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as personal digital processing, cellular phones, smart phones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely examples and are not intended to limit the implementation of the present disclosure described and / or required herein.

[0181] like Fig. 9 As shown, the device 900 includes a computing unit 901, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 902 or a computer program loaded from a storage unit 908 into a random access memory (RAM) 903. In the RAM 903, various programs and data required for the operation of the device 900 can also be stored. The computing unit 901, the ROM 902, and the RAM 903 are connected to each other via a bus 904. An input / output (I / O) interface 905 is also connected to the bus 904.

[0182] A number of components in the device 900 are connected to the I / O interface 905, including: an input unit 906, such as a keyboard, a mouse, etc.; an output unit 907, such as various types of displays, speakers, etc.; a storage unit 908, such as a disk, an optical disk, etc.; and a communication unit 909, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 909 allows the device 900 to exchange information / data with other devices through a computer network such as the Internet and / or various telecommunication networks.

[0183] The computing unit 901 may be a variety of general and / or special processing components with processing and computing capabilities. Some examples of the computing unit 901 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, digital signal processors (DSPs), and any appropriate processors, controllers, microcontrollers, etc. The computing unit 901 performs the various methods and processes described above, such as a method for identifying user intent based on multiple models. For example, in some embodiments, a method for identifying user intent based on multiple models may be implemented as a computer software program, which is tangibly included in a machine-readable medium, such as a storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on the device 900 via the ROM 902 and / or the communication unit 909. When the computer program is loaded into the RAM 903 and executed by the computing unit 901, one or more steps of the method for identifying user intent based on multiple models described above may be performed. Alternatively, in other embodiments, the computing unit 901 may be configured to execute the multi-model based user intention recognition method in any other appropriate manner (eg, by means of firmware).

[0184] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field programmable gate arrays (FPGAs), application specific integrated circuits (ASICs), application specific standard products (ASSPs), systems on chips (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various implementations can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a special purpose or general purpose programmable processor that can receive data and instructions from a storage system, at least one input device, and at least one output device, and transmit data and instructions to the storage system, the at least one input device, and the at least one output device.

[0185] The program code for implementing the method of the present disclosure may be written in any combination of one or more programming languages. These program codes may be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, so that the program code, when executed by the processor or controller, enables the functions / operations specified in the flow chart and / or block diagram to be implemented. The program code may be executed entirely on the machine, partially on the machine, partially on the machine and partially on a remote machine as a stand-alone software package, or entirely on a remote machine or server.

[0186] In the context of the present disclosure, a machine-readable medium may be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, device, or equipment. A machine-readable medium may be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium may include, but is not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or device, or any suitable combination of the foregoing. A more specific example of a machine-readable storage medium may include an electrical connection based on one or more lines, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0187] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user can provide input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0188] The systems and techniques described herein may be implemented in a computing system that includes back-end components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes front-end components (e.g., a user computer with a graphical user interface or a web browser through which a user can interact with implementations of the systems and techniques described herein), or a computing system that includes any combination of such back-end components, middleware components, or front-end components. The components of the system may be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0189] A computer system may include a client and a server. The client and the server are generally remote from each other and usually interact through a communication network. The relationship between the client and the server is generated by computer programs running on the corresponding computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or cloud host, which is a host product in the cloud computing service system to solve the defects of difficult management and weak business scalability in traditional physical hosts and VPS services ("Virtual Private Server", or "VPS" for short). The server may also be a server of a distributed system, or a server combined with a blockchain.

[0190] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps recorded in this disclosure can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solutions disclosed in this disclosure can be achieved, and this document does not limit this.

[0191] The above specific implementations do not constitute a limitation on the protection scope of the present disclosure. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present disclosure shall be included in the protection scope of the present disclosure.

Claims

1. A method for identifying user intention based on multiple models, comprising: In response to the user's current request information, based on the preset first model, the current request information is subjected to intent recognition processing to obtain a recognition result; wherein the recognition result indicates whether the preset first model outputs the user intent corresponding to the current request information; If the recognition result indicates that the user intention corresponding to the current request information is not obtained, then based on a preset second model, intent recognition processing is performed on the current request information to obtain the user intention corresponding to the current request information; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

2. The method according to claim 1, wherein: The performing intention recognition processing on the current request information based on the preset first model to obtain a recognition result includes: Determine the current dialogue state according to the current request information; wherein the current dialogue state indicates that the current request information is in a single-round dialogue or in multiple-round dialogues; According to the preset first model and the current dialog state, the current request information is processed for intention recognition to obtain the recognition result.

3. The method according to claim 2, wherein: The determining the current dialog state according to the current request information includes: Obtaining the context information of the current request information; The current dialog state is determined according to the above information and the current request information.

4. The method according to claim 3, wherein: The determining the current dialog state according to the above information and the current request information includes: Performing semantic recognition processing on the above information and the current request information respectively to obtain semantic information corresponding to the above information and semantic information corresponding to the current request information; Determine the similarity between the semantic information corresponding to the above information and the semantic information corresponding to the current request information; wherein the similarity represents the degree of association between the semantic information; The current dialog state is determined according to the similarity.

5. The method according to any one of claims 2 to 4, wherein: The performing intention recognition processing on the current request information according to the preset first model and the current dialog state to obtain the recognition result includes: If the current dialogue state represents that the current request information is in a single-round dialogue, the current request information is processed for intent recognition according to the single-round intention prediction sub-model in the preset first model to obtain the recognition result; wherein the single-round intention prediction sub-model is a branch in the preset first model.

6. The method according to claim 5, wherein: The performing intention recognition processing on the current request information according to the preset first model and the current dialog state to obtain the recognition result includes: If the current conversation state represents that the current request information is located in a multi-round conversation, then according to the multi-round intention prediction sub-model in the preset first model, the current request information and the above information are processed for intention recognition to obtain the recognition result; wherein the multi-round intention prediction sub-model is a branch in the preset first model.

7. The method according to any one of claims 1 to 6, wherein: The performing intention recognition processing on the current request information based on the preset first model to obtain a recognition result includes: Performing semantic recognition processing on the current request information to obtain the fuzziness of the current request information; wherein the fuzziness represents the clarity of expression of the current request information; If the ambiguity is less than a preset ambiguity threshold, a clarification prompt message is sent to the user; wherein the clarification prompt message is used to prompt the user to send a new request message; New request information is obtained, and the recognition result is determined according to the preset first model, the new request information, and the current request information.

8. The method according to claim 7, wherein: The determining the recognition result according to the preset first model, the new request information, and the current request information includes: Merging the new request information with the current request information to obtain information to be processed; According to the preset first model, the information to be processed is processed for intent recognition to obtain the recognition result.

9. The method according to any one of claims 1 to 8, wherein: The performing intention recognition processing on the current request information based on the preset second model to obtain the user intention corresponding to the current request information includes: Generate prompt words according to the current request information and based on a preset prompt word template; wherein the prompt words are data in text format, representing the current request information; The prompt word is input into the preset second model to obtain the user intention corresponding to the current request information; wherein the preset second model is a large language model.

10. The method according to any one of claims 1 to 9, further comprising: Acquire historical request information and a first intention of the historical request information; wherein the historical request information represents request information within a preset historical time period, and the first intention is a user intention corresponding to the historical request information determined by a first model; Based on a preset second model, performing intent recognition processing on the historical request information to obtain a second intent of the historical request information; If the first intention and the second intention are inconsistent, the historical request information is determined as training data; wherein the training data is used to train the first model.

11. The method according to claim 10, wherein: The acquiring of the historical request information and the first intention of the historical request information include: Get historical request information; Deduplication processing is performed on the historical request information to obtain deduplication-free historical request information; The first intention is to obtain the deduplicated historical request information.

12. The method according to claim 10 or 11, further comprising: If it is determined according to the recognition result that the user intention corresponding to the current request information is not obtained, the current request information is determined as the training data.

13. A user intention recognition device based on multiple models, comprising: A first recognition unit is used to respond to the user's current request information and perform intent recognition processing on the current request information based on a preset first model to obtain a recognition result; wherein the recognition result indicates whether the preset first model outputs the user intention corresponding to the current request information; A second identification unit is used to perform intent identification processing on the current request information based on a preset second model to obtain the user intention corresponding to the current request information if the identification result indicates that the user intention corresponding to the current request information is not obtained; wherein the parameter amount of the first model is smaller than the parameter amount of the second model.

14. The device according to claim 13, wherein: The first identification unit includes: A state determination module, used to determine the current dialogue state according to the current request information; wherein the current dialogue state indicates that the current request information is in a single-round dialogue or in multiple-round dialogues; The result obtaining module is used to perform intention recognition processing on the current request information according to the preset first model and the current dialogue state to obtain the recognition result.

15. The device according to claim 14, wherein: The state determination module comprises: The above-text acquisition submodule is used to obtain the above-text information of the current request information; The state determination submodule is used to determine the current dialog state according to the above information and the current request information.

16. The device according to claim 15, wherein: The state determination submodule is specifically used for: Performing semantic recognition processing on the above information and the current request information respectively to obtain semantic information corresponding to the above information and semantic information corresponding to the current request information; Determine the similarity between the semantic information corresponding to the above information and the semantic information corresponding to the current request information; wherein the similarity represents the degree of association between the semantic information; The current dialog state is determined according to the similarity.

17. The device according to any one of claims 14 to 16, wherein: The result is a module including: A single-round identification submodule is used to perform intent identification processing on the current request information to obtain the identification result according to the single-round intention prediction submodel in the preset first model if the current dialogue state represents that the current request information is in a single-round dialogue; wherein the single-round intention prediction submodel is a branch in the preset first model.

18. The device according to claim 17, wherein: The result is a module including: A multi-round identification submodule is used to perform intent identification processing on the current request information and the above information according to the multi-round intention prediction submodel in the preset first model to obtain the identification result if the current dialogue state represents that the current request information is located in a multi-round dialogue; wherein the multi-round intention prediction submodel is a branch in the preset first model.

19. The device according to any one of claims 13 to 18, wherein: The first identification unit includes: A fuzziness determination module, used to perform semantic recognition processing on the current request information to obtain the fuzziness of the current request information; wherein the fuzziness represents the clarity of the expression of the current request information; a prompt module, configured to send a clarification prompt message to the user if the ambiguity is less than a preset ambiguity threshold; wherein the clarification prompt message is used to prompt the user to send a new request message; The result determination module is used to obtain new request information and determine the recognition result according to the preset first model, the new request information, and the current request information.

20. The device according to claim 19, wherein The result determination module comprises: A fusion submodule, used to fuse the new request information with the current request information to obtain information to be processed; The recognition submodule is used to perform intent recognition processing on the information to be processed according to the preset first model to obtain the recognition result.

21. The device according to any one of claims 13 to 20, wherein: The second identification unit includes: A prompt word generation module, used to generate a prompt word according to the current request information and based on a preset prompt word template; wherein the prompt word is data in a text format, representing the current request information; The intention obtaining module is used to input the prompt word into the preset second model to obtain the user intention corresponding to the current request information; wherein the preset second model is a large language model.

22. The apparatus according to any one of claims 13 to 21, further comprising: A history acquisition unit, used to acquire historical request information and a first intention of the historical request information; wherein the historical request information represents request information within a preset historical time period, and the first intention is a user intention corresponding to the historical request information determined by a first model; A history recognition unit, configured to perform intent recognition processing on the history request information based on a preset second model to obtain a second intent of the history request information; An intention comparison unit is used to determine the historical request information as training data if the first intention and the second intention are inconsistent; wherein the training data is used to train the first model.

23. The device according to claim 22, wherein: The history acquisition unit includes: History acquisition module, used to obtain historical request information; A deduplication module, used for performing deduplication processing on the historical request information to obtain deduplicated historical request information; An intent acquisition module is used to acquire the first intent of the deduplicated historical request information.

24. The apparatus according to claim 22 or 23, further comprising: A data determination unit is used to determine the current request information as the training data if it is determined according to the recognition result that the user intention corresponding to the current request information is not obtained.

25. An electronic device comprising: at least one processor; as well as a memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 12.

26. A non-transitory computer-readable storage medium storing computer instructions, wherein: The computer instructions are used to cause the computer to execute the method according to any one of claims 1-12.

27. A computer program product, wherein: The invention comprises a computer program, which implements the steps of the method according to any one of claims 1 to 12 when being executed by a processor.

Images