Conversation-based intention recognition method, system and equipment and storage medium

Abstract

The embodiment of the invention discloses a dialogue-based intention recognition method, system and device and a storage medium, and relates to the technical field of data processing, and the method comprises the steps: determining a first embedded vector and a second embedded vector in response to collected current dialogue information and historical dialogue information of a user; calculating the similarity between the first embedding vector and each knowledge embedding vector in the M knowledge embedding vectors to obtain M first similarities; calculating the similarity between the second embedding vector and each knowledge embedding vector in the M knowledge embedding vectors to obtain M second similarities; determining N mixed similarities according to the M first similarities and the M second similarities; determining N knowledge entries according to the N mixing similarities; and inputting the N knowledge entries into the pre-configured large language model, and outputting the intention recognition result of the user, so that the intention recognition accuracy of the user can be improved, and the progress of various businesses in financial or medical business scenes can be further promoted.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of data processing, and particularly relates to a dialogue-based intent recognition method, system, device and storage medium. BACKGROUND

[0002] In artificial intelligence interaction, intent recognition is one of the core contents of accurate service, and its importance in many business scenarios such as finance and medical treatment is self-evident.

[0003] However, in the related art, it often depends on a rule template or a single model, and it is difficult to identify the real intent of the user. For example, in a financial business scenario, it is difficult to identify the real intent of the user that the dialogue content of the user is to query financial income or deposit income. For another example, in a medical business scenario, the dialogue content of the user "abdominal pain during pregnancy" may be misidentified as a prenatal examination appointment.

[0004] It can be seen that the prior art has the problem of low accuracy of intent recognition in the process of intent recognition of the user.

[0005] SUMMARY

[0006] Therefore, one of the purposes of the present application is to provide a dialogue-based intent recognition method, system, device and storage medium, which can improve the accuracy of intent recognition of the user.

[0007] In a first aspect, an embodiment of the present application provides a dialogue-based intent recognition method, which comprises:

[0008] In response to the current dialogue information and the historical dialogue information of the user collected, determining a first embedding vector corresponding to the current dialogue information and a second embedding vector corresponding to the historical dialogue information;

[0009] Calculating the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scenario knowledge base, to obtain M first similarities, the M knowledge embedding vectors being pre-configured based on M knowledge items, and the pre-configured scenario knowledge base comprising the M knowledge items;

[0010] Calculating the similarity between the second embedding vector and each of the M knowledge embedding vectors, to obtain M second similarities;

[0011] Determining N mixed similarities according to the M first similarities and the M second similarities, N being a positive integer and N being less than M;

[0012] Determining N knowledge items from the pre-configured scenario knowledge base according to the N mixed similarities;

[0013] The N knowledge entries are input into a pre-configured large language model, and an intention recognition result of the user is output by using the pre-configured large language model, and the pre-configured large language model is trained based on a plurality of historical knowledge entries and corresponding user intentions.

[0014] In a possible implementation, before the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scenario knowledge base is calculated, the method further includes:

[0015] In response to the collected current dialogue information and historical dialogue information, a user scenario in which the user is located is determined.

[0016] According to the user scenario, a pre-configured scenario knowledge base is determined.

[0017] In a possible implementation, in response to the collected current dialogue information and historical dialogue information, the user scenario in which the user is located is determined, including:

[0018] According to the current dialogue information and the historical dialogue information, the user scenario is determined.

[0019] In a possible implementation, in response to the collected current dialogue information and historical dialogue information, the user scenario in which the user is located is determined, including:

[0020] According to the current dialogue information and the historical dialogue information, unique identification information and scenario feature information of the user are determined.

[0021] From a pre-configured user portrait set, a target user portrait corresponding to the unique identification information is determined, and the user portrait set includes a plurality of user portraits of users.

[0022] According to the target user portrait and the scenario feature information, the user scenario is determined.

[0023] In a possible implementation, before the similarity between the second embedding vector and each of the M knowledge embedding vectors is calculated to obtain the M second similarities, the method further includes:

[0024] A first time stamp of the current dialogue information and a second time stamp of the historical dialogue information are obtained.

[0025] According to the first time stamp and the second time stamp, a dialogue interval duration corresponding to the historical dialogue information is determined.

[0026] According to the dialogue interval duration, a time decay factor is determined.

[0027] The similarity between the second embedding vector and each of the M knowledge embedding vectors is calculated to obtain the M second similarities, including:

[0028] Based on the time decay factor, the similarity between the second embedding vector and each of the M knowledge embedding vectors is calculated respectively to obtain M second similarities.

[0029] In a possible implementation, according to the M first similarities and the M second similarities, N mixed similarities are determined, including:

[0030] The M first similarities are arranged in descending order of values, and the N first similarities are obtained from the M first similarities.

[0031] According to the N first similarities and the N second similarities, N mixed similarities are determined.

[0032] In a possible implementation, the first similarity and the second similarity are cosine similarities.

[0033] In a second aspect, an embodiment of the present application provides a dialog-based intent recognition system, which comprises:

[0034] A first determination module is configured to determine, in response to the collected current dialog information and historical dialog information of a user, a first embedding vector corresponding to the current dialog information and a second embedding vector corresponding to the historical dialog information.

[0035] A first calculation module is configured to calculate the similarity between the first embedding vector and each of M knowledge embedding vectors in a preconfigured scenario knowledge base, to obtain M first similarities, the M knowledge embedding vectors being obtained based on M knowledge items, and the preconfigured scenario knowledge base comprising the M knowledge items.

[0036] A second calculation module is configured to calculate the similarity between the second embedding vector and each of the M knowledge embedding vectors to obtain M second similarities.

[0037] A second determination module is configured to determine, according to the M first similarities and the M second similarities, N mixed similarities, N being a positive integer and N being less than M.

[0038] A third determination module is configured to determine, according to the N mixed similarities, N knowledge items from the preconfigured scenario knowledge base.

[0039] A processing module is configured to input the N knowledge items into a preconfigured large language model, and input a user intent recognition result by using the preconfigured large language model, the preconfigured large language model being trained based on a plurality of historical knowledge items and corresponding user intents.

[0040] In a third aspect, an electronic device is provided, which includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, the method for recognizing an intent based on a dialogue according to the first aspect is implemented.

[0041] In a fourth aspect, a computer readable storage medium is provided, which stores a computer program. When the computer program is executed by one or more processors, the method for recognizing an intent based on a dialogue according to the first aspect is implemented.

[0042] The method for recognizing an intent based on a dialogue provided by the embodiments of the present application can determine a first embedding vector corresponding to current dialogue information and a second embedding vector corresponding to historical dialogue information in response to the current dialogue information and the historical dialogue information collected by the user. Then, the similarities between the first embedding vector and each of the M knowledge embedding vectors in the preconfigured scenario knowledge base are calculated to obtain M first similarities. Subsequently, the similarities between the second embedding vector and each of the M knowledge embedding vectors are calculated to obtain M second similarities, and N mixed similarities are determined according to the M first similarities and the M second similarities. Finally, N knowledge items are determined from the preconfigured scenario knowledge base according to the N mixed similarities, and the N knowledge items are input into the preconfigured large language model, and the preconfigured large language model is used to output the intent recognition result of the user, which can improve the accuracy of the intent recognition of the user, and further promote the progress of various businesses in the financial or medical business scenarios. BRIEF DESCRIPTION OF DRAWINGS

[0043] In order to more clearly illustrate the technical solutions of the embodiments of the present application or the prior art, the drawings needed in the embodiments or the prior art description will be briefly introduced below. It should be understood that the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative labor on the basis of these drawings.

[0044] Figure 1 A flowchart of a method for recognizing an intent based on a dialogue provided by the embodiments of the present application;

[0045] Figure 2 A flowchart of determining a user scenario included in a method for recognizing an intent based on a dialogue provided by the embodiments of the present application;

[0046] Figure 3 A method for recognizing an intent based on a dialogue provided by the embodiments of the present application;

[0047] Figure 4A functional module schematic diagram of a dialogue-based intent recognition system provided by an embodiment of the present application;

[0048] Figure 5 A hardware structure schematic diagram of an electronic device provided by an embodiment of the present application.

[0049] Explanation of reference signs:

[0050] 400, a dialogue-based intent recognition system;

[0051] 410, a first determination module;

[0052] 420, a first calculation module;

[0053] 430, a second calculation module;

[0054] 440, a second determination module;

[0055] 450, a third determination module;

[0056] 460, a processing module;

[0057] 501, a processor;

[0058] 502, a memory;

[0059] 503, a communication interface;

[0060] 510, a bus. DETAILED DESCRIPTION

[0061] In order to make the objectives, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, rather than all the embodiments of the present application. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0062] Therefore, the detailed description of the embodiments of the present application provided below in the drawings is not intended to limit the scope of the claimed present application, but only represents selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present application.

[0063] It should be noted that: similar reference signs and letters represent similar items in the following drawings, therefore, once an item is defined in one drawing, it does not need to be further defined and explained in the subsequent drawings.

[0064] In various embodiments of the present application, the expression "or" or "at least one of A or / and B" includes any combination of the listed terms or all combinations of the listed terms. For example, the expression "A or B" or "at least one of A or / and B" can include A, can include B, or can include both A and B.

[0065] In the description of the present application, it should be noted that if the terms "upper", "lower", "inner", "outer" and the like indicating the orientation or positional relationship are based on the orientation or positional relationship shown in the drawings, or the orientation or positional relationship in which the product of the present application is usually placed, and are only for the convenience of describing the present application and simplifying the description, and therefore cannot be understood as indicating or implying that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as limiting the present application.

[0066] In addition, if the terms "first", "second" and the like appear, they are only used to distinguish the description and cannot be understood as indicating or implying relative importance.

[0067] It should be noted that the features in the embodiments of the present application can be combined with each other without conflict.

[0068] Also, in the embodiments of the present application, the term "connection" can mean "electrical connection", and can also mean "direct connection". "Electrical connection" can mean that two components are directly electrically connected, or can mean that two components are electrically connected via one or more other components such as normally open tubes.

[0069] In order to facilitate better understanding of the scheme of the embodiments of the present application, the related art will be introduced first as follows.

[0070] Artificial intelligence (AI): is a new technical science of studying, developing, simulating, extending and expanding human intelligence, and is a branch of computer science. Artificial intelligence aims to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. The research in this field includes robots, language recognition, image recognition, natural language processing and expert systems. Artificial intelligence can simulate the information process of human consciousness and thinking. Artificial intelligence is also the theory, method, technology and application system of using digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, to perceive the environment, acquire knowledge and use knowledge to obtain the best results.

[0071] Natural language processing (NLP): NLP uses computers to process, understand and use human languages (such as Chinese, English, etc.), and NLP is a branch of artificial intelligence, an interdisciplinary subject of computer science and linguistics, and is also commonly referred to as computational linguistics. Natural language processing includes syntax analysis, semantic analysis, discourse understanding, etc. Natural language processing is commonly used in technical fields such as machine translation, handwritten and printed character recognition, speech recognition and text-to-speech conversion, information intent recognition, information extraction and filtering, text classification and clustering, public opinion analysis and opinion mining, etc. It involves data mining, machine learning, knowledge acquisition, knowledge engineering, artificial intelligence research related to language processing, and linguistic research related to language computing.

[0072] Information extraction: A text processing technology that extracts specified types of entity, relationship, event, etc. from natural language text, and forms structured data output. Information extraction is a technology for extracting specific information from text data. Text data is composed of specific units such as sentences, paragraphs, and chapters, and text information is composed of specific units such as words, phrases, sentences, paragraphs, or combinations of these specific units. Extracting noun phrases, names, and places from text data is text information extraction, and of course, the information extracted by the text information extraction technology can be various types of information.

[0073] To solve the technical problems in the background art, the embodiments of the present application provide a dialogue-based intent recognition method, a dialogue-based intent recognition system, an electronic device and a computer readable storage medium. First, the dialogue-based intent recognition method provided by the embodiments of the present application will be introduced.

[0074] Please refer to Figure 1 , Figure 1 A flowchart of a dialogue-based intent recognition method provided by the embodiments of the present application, which can be applied to the dialogue-based intent recognition system or the electronic device in the following embodiments, wherein the electronic device includes personal computers, servers, mobile devices, cloud computing platforms and supercomputers, etc. The dialogue-based intent recognition method will be introduced from the application in the electronic device.

[0075] As shown in the dialogue-based intent recognition method Figure 1 specifically includes the following steps:

[0076] Step 110, in response to the collected current dialogue information and historical dialogue information of the user, determining a first embedding vector corresponding to the current dialogue information and a second embedding vector corresponding to the historical dialogue information.

[0077] In step 120, similarities between the first embedding vector and each of M knowledge embedding vectors in the pre-configured scenario knowledge base are calculated, to obtain M first similarities, the M knowledge embedding vectors being pre-configured based on M knowledge items, and the pre-configured scenario knowledge base including the M knowledge items.

[0078] In step 130, similarities between the second embedding vector and each of the M knowledge embedding vectors are calculated, to obtain M second similarities.

[0079] In step 140, N mixed similarities are determined according to the M first similarities and the M second similarities, N being a positive integer and smaller than M.

[0080] In step 150, N knowledge items are determined from the pre-configured scenario knowledge base according to the N mixed similarities.

[0081] In step 160, the N knowledge items are input into a pre-configured large language model, and an intention recognition result of the user is output by the pre-configured large language model, the pre-configured large language model being trained based on a plurality of historical knowledge items and corresponding user intentions.

[0082] The dialog-based intention recognition method provided by the embodiments of the present application can determine a first embedding vector corresponding to current dialog information of a user and a second embedding vector corresponding to historical dialog information of the user in response to the current dialog information and the historical dialog information collected. Then, similarities between the first embedding vector and each of M knowledge embedding vectors in a pre-configured scenario knowledge base are calculated, to obtain M first similarities. Subsequently, similarities between the second embedding vector and each of the M knowledge embedding vectors are calculated, to obtain M second similarities, and N mixed similarities are determined according to the M first similarities and the M second similarities. Finally, N knowledge items are determined from the pre-configured scenario knowledge base according to the N mixed similarities, and the N knowledge items are input into a pre-configured large language model, and an intention recognition result of the user is output by the pre-configured large language model, which can improve the intention recognition accuracy of the user and further promote the progress of various businesses in financial or medical business scenarios.

[0083] The embodiments of the present application can acquire and process related data based on artificial intelligence technology. AI is a theory, method, technology and application system for simulating, extending and expanding human intelligence by using a digital computer or a machine controlled by a digital computer, perceiving an environment, acquiring knowledge and using the knowledge to obtain optimal results.

[0084] The basic technologies of artificial intelligence generally include technologies such as sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation / interaction systems, mechatronics, and the like. The software technologies of artificial intelligence mainly include computer vision technologies, robot technologies, biometric identification technologies, speech processing technologies, natural language processing technologies, and machine learning / deep learning, and the like.

[0085] The present application can be used in a variety of general-purpose or special-purpose computer system environments or configurations. For example: personal computers, server computers, handheld or portable devices, tablet devices, multiprocessor systems, microprocessor-based systems, set-top boxes, programmable consumer electronics, network PCs, minicomputers, mainframe computers, distributed computing environments that include any of the above systems or devices, and the like. The present application can be described in the general context of computer-executable instructions executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, and the like that perform specific tasks or implement specific abstract data types. The present application can also be practiced in a distributed computing environment, in which tasks are performed by remote processing devices connected by a communication network. In a distributed computing environment, program modules can be located in local and remote computer storage media, including storage devices.

[0086] It should be noted that in each specific embodiment of the present application, when it is necessary to process relevant data related to the identity or characteristics of the user according to user information, user behavior data, user history data, and user location information, etc., the user's permission or consent will be obtained first, and the collection, use and processing of these data will comply with relevant laws, regulations and standards. In addition, when the embodiments of the present application need to obtain sensitive personal information of the user, the separate permission or separate consent of the user will be obtained through a pop-up window or a jump to a confirmation page, and after obtaining the separate permission or separate consent of the user, the necessary user-related data for the normal operation of the embodiments of the present application will be obtained.

[0087] The following will be described in detail as Figure 1 The steps of the method will be described in detail.

[0088] In step 110, the electronic device can determine a first embedding vector corresponding to the current conversation information of the user and a second embedding vector corresponding to the historical conversation information of the user in response to the current conversation information and the historical conversation information of the user collected. In other words, the electronic device converts the current conversation information and the historical conversation information of the user into vector representations, which facilitates the fast calculation of the similarity in the subsequent steps.

[0089] The current dialogue information is information input by the user last time. The historical dialogue information is a multi-round dialogue record before the current dialogue information of the user.

[0090] The form of the current dialogue information and the historical dialogue information includes but is not limited to text information and voice information input by the user, and the embodiments of the present application do not specifically limit the form of the current dialogue information and the historical dialogue information.

[0091] Taking a financial service scenario as an example, the current dialogue information can be "my car was hit and the rear bumper was damaged", and the historical dialogue information can be "I have insured the car damage insurance".

[0092] Taking a medical service scenario as an example, the current dialogue information can be "continuous cough for several days", and the historical dialogue information can be "I have a history of allergies".

[0093] The first embedding vector and the second embedding vector are both semantic vectors, the first embedding vector can be used to represent the semantic features of the current dialogue information, and the second embedding vector can be used to represent the semantic context of the historical dialogue information.

[0094] In some embodiments, the electronic device can convert the current dialogue information into the first embedding vector and the historical dialogue information into the second embedding vector by using an open source general vector model BGE.

[0095] In some embodiments, the electronic device can detect whether there is a new interaction event by monitoring the user input interface in real time, and if there is a new interaction event, the collection function can be triggered to collect the current dialogue information and the historical dialogue information of the user.

[0096] In some embodiments, the electronic device can trigger the collection function to collect the current dialogue information and the historical dialogue information of the user in response to satisfying a preset triggering condition.

[0097] The preset condition includes at least one of the following:

[0098] It is monitored that the user clicks the "send" button;

[0099] It is monitored that the user clicks the "voice input" button and the voice input duration exceeds a preset duration.

[0100] In some embodiments, if the current dialogue information or the historical dialogue information of the user is voice information, the voice information can be converted into text information by automatic speech recognition (ASR) technology, and then the text information is converted into a semantic vector form by using an open source general vector model BGE.

[0101] In step 120, the electronic device calculates similarities between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scenario knowledge base, respectively, to obtain M first similarities, and the semantic correlation between the current dialogue information and all knowledge in the scenario knowledge base can be quantified by the M first similarities.

[0102] The pre-configured scenario knowledge base described above can be regarded as a domain-specific structured knowledge set. For example, in a financial business scenario, the pre-configured scenario knowledge base can include a vehicle insurance clause library. For another example, in a medical business scenario, the pre-configured scenario knowledge base can include a disease type library.

[0103] The knowledge embedding vector described above is generated based on a pre-configured knowledge item. Specifically, the electronic device can convert the knowledge item by using an open-source general vector model (BGE) to obtain the knowledge embedding vector.

[0104] The M knowledge items correspond to the M knowledge embedding vectors, and it should be noted that the knowledge item and the knowledge embedding vector are one-to-one.

[0105] Exemplarily, if the pre-configured scenario knowledge base is a vehicle insurance clause library, the corresponding knowledge items include but are not limited to the scope of liability, the exemption clause, the compensation limit, and the claim materials.

[0106] For example, the scope of liability includes "collision, overturning, fire, and explosion caused by the vehicle direct loss". The exemption clause includes "driving after drinking using the insured motor vehicle is exempted". The compensation limit includes "single vehicle accident absolute exemption amount 5000 yuan". The claim materials include "driving license, accident identification, and maintenance invoice original".

[0107] In step 130, the electronic device can calculate similarities between the second embedding vector and each of the M knowledge embedding vectors, respectively, to obtain M second similarities, and the semantic correlation between the historical dialogue information and all knowledge in the scenario knowledge base can be quantified by the M second similarities.

[0108] In a possible implementation, the first similarity and the second similarity in steps 120 and 130 described above are cosine similarities.

[0109] In step 140, the electronic device can determine N hybrid similarities according to the M first similarities and the M second similarities determined in the foregoing steps, and based on the N hybrid similarities, the corresponding N knowledge items in the subsequent steps can be determined, which can avoid the phenomenon of context missing in a single round of dialogue and further improve the accuracy of the user's intent recognition.

[0110] In some embodiments, step 140 of determining the N hybrid similarities according to the M first similarities and the M second similarities further includes:

[0111] The M first similarities and the M second similarities are weighted and fused to obtain N mixed similarities.

[0112] In step 150, the electronic device can determine N knowledge items with high accuracy and high reliability from the pre-configured scenario knowledge base based on the N mixed similarities determined in the foregoing steps. Based on this, it can be understood that the electronic device can improve the accurate semantic context for the pre-configured large language model, thereby improving the accuracy of the user's intent recognition.

[0113] The N knowledge items can be understood as knowledge with high relevance to the current dialogue information and the historical dialogue information.

[0114] In step 160, the electronic device inputs the N knowledge items into the pre-configured large language model, and outputs the user's intent recognition result by using the pre-configured large language model.

[0115] In some embodiments, the intent recognition result is an intent identifier, which can be used to represent the real intent of the user.

[0116] Taking the financial business field as an example, the intent identifier is a character identifier such as "loan application" or "vehicle insurance claim", or a digital identifier such as 001 (corresponding to loan application) or 002 (corresponding to vehicle insurance claim).

[0117] Taking the medical business field as an example, the intent identifier is a character identifier such as "allergy consultation" or "registration and treatment", or a digital identifier such as 003 (corresponding to allergy consultation) or 004 (corresponding to registration and treatment).

[0118] In a possible implementation, before calculating the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scenario knowledge base, the method further includes:

[0119] In response to the collected current dialogue information and historical dialogue information, determining a user scenario in which the user is located;

[0120] According to the user scenario, determining the pre-configured scenario knowledge base.

[0121] In the embodiments of the present application, the user scenario in which the user is located is first determined, and then the pre-configured scenario knowledge base is determined based on the user scenario. The pre-configured scenario knowledge base can be selected in a targeted manner, the accuracy and reliability of the M knowledge items determined are improved, and the pre-configured large language model outputs the intent recognition result with high accuracy.

[0122] Taking financial business scenarios as an example, the aforementioned user scenarios may include, but are not limited to, auto insurance damage assessment and underwriting scenarios. If the user scenario is an auto insurance damage assessment scenario, the corresponding scenario knowledge base could be a repair parts price list database.

[0123] Taking a healthcare-related business scenario as an example, the aforementioned user scenarios may include, but are not limited to, drug consultation scenarios. If the user scenario is a drug consultation scenario, the corresponding scenario knowledge base could be a drug information database.

[0124] In some embodiments, a pre-configured scenario knowledge base is determined based on the user scenario, including:

[0125] Determine the target scenario identifier based on the user scenario;

[0126] Based on the target scene identifier, determine the scene knowledge base that has a mapping relationship with the scene identifier from several scene knowledge bases;

[0127] The scene knowledge base that has a mapping relationship with the scene identifier is identified as the pre-configured scene knowledge base mentioned above.

[0128] It should be noted that the aforementioned scenario knowledge bases are all pre-configured and stored in electronic devices. In this way, electronic devices can quickly determine the pre-configured scenario knowledge bases based on the target scenario identifier.

[0129] In one possible implementation, in response to the collected current dialogue information and historical dialogue information, the user's current user scenario is determined, including:

[0130] Determine the user scenario based on current and historical dialogue information.

[0131] In this embodiment, the user scenario can be determined directly based on the current dialogue information and historical dialogue information, enabling rapid determination of the user scenario.

[0132] In some embodiments, determining the user scenario based on current dialogue information and historical dialogue information includes:

[0133] Extract several keywords from the current dialogue information and historical dialogue information;

[0134] Determine the frequency of each keyword among a set of keywords;

[0135] Keywords whose frequency exceeds a preset frequency threshold are identified as target keywords.

[0136] Determine the user scenario based on the target keywords.

[0137] Taking a financial service scenario as an example, if the target keyword determined by the electronic device based on the current dialogue information and the historical dialogue information includes a policy number, a loss assessment, full liability, and compulsory insurance, the electronic device can determine that the user scenario is a vehicle loss assessment scenario based on the target keyword.

[0138] Taking a medical service scenario as an example, if the target keyword determined by the electronic device based on the current dialogue information and the historical dialogue information includes a cold, frequency, time, taking medicine, precautions, before meals, and after meals, the electronic device can determine that the user scenario is a medication consultation scenario based on the target keyword.

[0139] In some embodiments, in response to the collected current dialogue information and historical dialogue information, the user scenario in which the user is located is determined, including:

[0140] According to the current dialogue information, a first scenario is determined.

[0141] According to the historical dialogue information, a second scenario is determined, the user scenario includes the first scenario and the second scenario, and the first scenario and the second scenario are different scenarios.

[0142] See Figure 2 , Figure 2 A flowchart for determining a user scenario included in an intent recognition method based on a dialogue provided by an embodiment of the present application.

[0143] In a possible implementation, in response to the collected current dialogue information and historical dialogue information, the user scenario in which the user is located is determined, including but not limited to the following steps 210 to 230:

[0144] Step 210, according to the current dialogue information and the historical dialogue information, the unique identification information and the scene feature information of the user are determined.

[0145] Step 220, a target user portrait corresponding to the unique identification information is determined from a pre-configured user portrait set, and the user portrait set includes user portraits of several users.

[0146] Step 230, according to the target user portrait and the scene feature information, the user scenario is determined.

[0147] In the embodiment of the present application, by determining the unique identification information of the user, and determining the target user portrait based on the unique identification information, finally the target user portrait and the scene feature information can be comprehensively determined to determine the user scenario.

[0148] The scenario feature information includes one or more of the keywords, the sentiment feature, the dialogue turn, the dialogue channel, and the time interval in the foregoing embodiments. The keywords can refer to the description of the keywords in the foregoing embodiments, and the sentiment feature can be represented by the keywords (for example, the keywords such as “quickly” and “a little faster” can represent the anxious and anxious emotions of the user).

[0149] The user portraits of the plurality of users in the user portrait set are preconfigured and stored in the electronic device.

[0150] Taking a financial service scenario as an example, the target user portrait includes basic information (for example, age, occupation, and historical insurance policy) and behavior information (for example, commonly used service type, risk preference, and consultation frequency) of the user. Taking a medical service scenario as an example, the target user portrait includes basic information (for example, age, occupation, and past medical history) and behavior information (for example, diet and exercise) of the user.

[0151] Still taking the financial service scenario as an example, if the electronic device determines, based on the target user portrait, that the consultation frequency (which is different from the dialogue turn described above, and can be the historical consultation frequency of the user) of the user is greater than a preset frequency threshold; and the electronic device determines that the scenario feature information includes the dialogue channel (VIP hotline) and the keywords (including loss determination, full liability, and compulsory insurance). Then, the electronic device can determine, based on the consultation frequency and the VIP hotline, that the user scenario is a quick vehicle insurance loss determination scenario. In the case of the quick vehicle insurance loss determination scenario, a dedicated staff can be assigned to handle it one-on-one.

[0152] In a possible implementation, before the similarity between the second embedding vector and each of the M knowledge embedding vectors is calculated to obtain the M second similarities, the method further includes:

[0153] obtaining a first timestamp of the current dialogue information and a second timestamp of the historical dialogue information;

[0154] determining a dialogue interval duration corresponding to the historical dialogue information according to the first timestamp and the second timestamp;

[0155] determining a time decay factor according to the dialogue interval duration;

[0156] calculating the similarity between the second embedding vector and each of the M knowledge embedding vectors to obtain the M second similarities, including:

[0157] calculating the similarity between the second embedding vector and each of the M knowledge embedding vectors based on the time decay factor to obtain the M second similarities.

[0158] In the embodiments of the present application, the time decay factor can be determined based on the determined conversation interval duration, and the accuracy and reliability of the determined M second similarities can be further improved based on the time decay factor.

[0159] In some embodiments, the longer the conversation interval duration, the smaller the value of the corresponding time decay factor.

[0160] In some embodiments, when the electronic device determines the conversation interval duration, the time decay factor corresponding to the conversation interval duration can be determined through a time decay factor mapping table. The time decay factor mapping table in the embodiments of the present application includes the association between the conversation interval duration and the time decay factor.

[0161] For example, if the conversation interval duration is greater than 1 day, the corresponding time decay factor is 0.5; if the conversation interval duration is less than 10 minutes, the corresponding time decay factor is 0.9.

[0162] Although the values of the conversation interval duration and the corresponding time decay factor are shown for illustrative purposes, more different values of the conversation interval duration and the corresponding time decay factor can be set according to actual needs, which are all within the protection scope of the embodiments of the present application.

[0163] In a possible implementation, the N hybrid similarities are determined according to the M first similarities and the M second similarities, including:

[0164] determining N first similarities in the M first similarities arranged in descending order of values, and N second similarities in the M second similarities arranged in descending order of values;

[0165] determining the N hybrid similarities according to the N first similarities and the N second similarities.

[0166] In the embodiments of the present application, the N first similarities and the N second similarities with higher similarity values can be determined, and the N hybrid similarities can be determined based on the N first similarities and the N second similarities. By retaining the N hybrid similarities with higher values, the N knowledge items can be quickly determined, and the speed of the pre-configured large language model outputting the user's intent recognition result can be indirectly improved.

[0167] In some embodiments, N = 10.

[0168] Please refer to Figure 3 , Figure 3 A flowchart of a dialogue-based intent recognition method provided in the embodiments of the present application is involved. Figure 3 In the embodiments of the present application, the N hybrid similarities are determined according to the M first similarities and the M second similarities, including:

[0169] The current user input input corresponds to the above-mentioned current dialogue information;

[0170] The current user communicates the above-mentioned historical dialogue information;

[0171] Input enhanced retrieval and History enhanced retrieval correspond to the process of determining the first embedding vector, the second embedding vector, and determining the first similarity and the second similarity;

[0172] Mixed similarity sorting corresponds to the process of determining N mixed similarities;

[0173] Top10 knowledge, entity, intent corresponds to the determination of N(N=10) knowledge items;

[0174] LLM extracts intent and information, LLM standardizes intent and information, and output output corresponds to the above-mentioned use of a pre-configured large language model to output the user's intent recognition result.

[0175] The following will clearly show the implementation process of the above-mentioned embodiments through examples, specifically including the following processes:

[0176] ① After accessing the user line communication, determine the user's scene label (used to represent the above-mentioned user scene).

[0177] ② Use the open source general vector model BGE to jointly retrieve and enhance the current user information (corresponding to the above-mentioned current dialogue information) and the above-mentioned information (corresponding to the above-mentioned historical dialogue information) in the corresponding user scene (i.e. the above-mentioned current dialogue information corresponds to the first scene, and the above-mentioned historical dialogue information corresponds to the second scene). 10 similar entities and 10 similar intents, 10 similar knowledge (corresponding to the above-mentioned N knowledge items, where N=10) are obtained.

[0178] Step 1, semantic embedding is performed on the current user information (corresponding to the above-mentioned current dialogue information) and the current context information (corresponding to the above-mentioned historical dialogue information).

[0179] input embedding =f θ (input) (1);

[0180] input_history embedding =f θ (input_history) (2);

[0181] Wherein:

[0182] f θ () represents the function of the embedding model used and its corresponding parameters, and an embedding model in the related art can be used;

[0183] input embedding denotes the first embedding vector;

[0184] input_history embedding denotes the second embedding vector.

[0185] Step2, calculate the cosine similarity (corresponding to the M first similarities) between the current input embedding (i.e., the first embedding vector) and the knowledge base input embedding (corresponding to the M knowledge embedding vectors).

[0186]

[0187]

[0188] wherein:

[0189]

[0190] similarity;

[0191] TOP_10_similarity(input) denotes the top 10 first similarities in the M first similarities sorted in descending order of similarity values.

[0192] Step3, calculate the cosine similarity (corresponding to the M second similarities) between the current context embedding (corresponding to the second embedding vector) and the knowledge base context embedding (corresponding to the M knowledge embedding vectors), and weighted sum the input similarity (corresponding to the M first similarities) to obtain the hybrid similarity (corresponding to the N hybrid similarities).

[0193]

[0194]

[0195] wherein:

[0196] denotes the i-th similarity in the M second similarities;

[0197] TOP_10_hybrid_similarity(input) denotes the N hybrid similarities, N = 10.

[0198] The 0.6 and 0.4 in the above formula (6) are weight values, which can be assigned by the electronic device based on actual needs.

[0199] Step4, obtain 10 final relevant knowledge (corresponding to the N knowledge items, N = 10) according to the input similarity (corresponding to the first similarity) and the hybrid similarity.

[0200] UTOP_10_KB(input) = nique_Top_10 [TOP_10_similarity(input) U TOP_10_hybrid_similarity(input)] (7);

[0202] ③Combined with similar knowledge (corresponding to the above knowledge items), use the large model prompt engineering (corresponding to the above pre-configured large language model) to extract the intent and information in the user dialogue.

[0203] ④Combined with similar entities and intents (corresponding to the above knowledge items), use the large model prompt engineering to normalize and standardize the obtained event.

[0204] ⑤Output the user intent and information (corresponding to the above intent recognition result).

[0205] In some embodiments, if it is determined based on user feedback that the final obtained intent is incorrect or biased, the scene knowledge base can be updated. When dialoguing with the user again, the updated scene knowledge base can be used.

[0206] In some embodiments, if it is determined based on user feedback that the final obtained intent is incorrect or biased, the pre-configured large language model can be updated. When dialoguing with the user again, the updated large language model can be used.

[0207] Corresponding to the above method embodiment, the embodiment of the present application also provides a dialogue-based intent recognition system, please see Figure 4 , Figure 4 The function module schematic diagram of the dialogue-based intent recognition system provided by the embodiment of the present application, wherein the dialogue-based intent recognition system 400 comprises:

[0208] The first determination module 410 is configured to determine a first embedding vector corresponding to the current dialogue information and a second embedding vector corresponding to the historical dialogue information in response to the collected current dialogue information and the historical dialogue information of the user.

[0209] The first calculation module 420 is configured to calculate the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scene knowledge base, to obtain M first similarities, the M knowledge embedding vectors being pre-configured based on M knowledge items, and the pre-configured scene knowledge base comprising the M knowledge items.

[0210] The second calculation module 430 is configured to calculate the similarity between the second embedding vector and each of the M knowledge embedding vectors, to obtain M second similarities.

[0211] The second determining module 440 is configured to determine N mixed similarities according to the M first similarities and the M second similarities, where N is a positive integer and N is less than M.

[0212] The third determining module 450 is configured to determine N knowledge items from the preconfigured scene knowledge base according to the N mixed similarities.

[0213] The processing module 460 is configured to input the N knowledge items into a preconfigured large language model, and input the user's intent recognition result into the preconfigured large language model, where the preconfigured large language model is trained based on a plurality of historical knowledge items and corresponding user intents.

[0214] The dialog-based intent recognition method provided by the embodiments of the present application can determine the first embedding vector corresponding to the current dialogue information and the second embedding vector corresponding to the historical dialogue information through the first determining module in response to the collected current dialogue information and historical dialogue information. Then, the first computing module is used to calculate the similarity between the first embedding vector and each knowledge embedding vector in the M knowledge embedding vectors in the preconfigured scene knowledge base, to obtain M first similarities. Subsequently, the second computing module is used to calculate the similarity between the second embedding vector and each knowledge embedding vector in the M knowledge embedding vectors, to obtain M second similarities. The second determining module is used to determine N mixed similarities according to the M first similarities and the M second similarities. Finally, the third determining module is used to determine N knowledge items from the preconfigured scene knowledge base according to the N mixed similarities, and the processing module is used to input the N knowledge items into a preconfigured large language model, and output the user's intent recognition result by using the preconfigured large language model, which can improve the accuracy of user intent recognition.

[0215] In a possible implementation, the dialog-based intent recognition system 400 further includes a fourth determining module, which can be configured to:

[0216] determine a user scenario of the user in response to the collected current dialogue information and historical dialogue information;

[0217] determine a preconfigured scene knowledge base according to the user scenario.

[0218] In a possible implementation, the fourth determining module is further specifically configured to:

[0219] determine the user scenario according to the current dialogue information and the historical dialogue information.

[0220] In a possible implementation, the fourth determining module is further specifically configured to:

[0221] According to the current dialogue information and the historical dialogue information, unique identification information and scene feature information of the user are determined;

[0222] A target user portrait corresponding to the unique identification information is determined from a pre-configured user portrait set, the user portrait set including user portraits of a plurality of users;

[0223] According to the target user portrait and the scene feature information, a user scene is determined.

[0224] In a possible implementation, the dialogue-based intent recognition system 400 further includes an acquisition module, which can be used to:

[0225] A first timestamp of the current dialogue information and a second timestamp of the historical dialogue information are acquired;

[0226] According to the first timestamp and the second timestamp, a dialogue interval duration corresponding to the historical dialogue information is determined;

[0227] According to the dialogue interval duration, a time decay factor is determined;

[0228] The second calculation module 430 is further specifically used to:

[0229] Based on the time decay factor, similarities between the second embedding vector and each of the M knowledge embedding vectors are calculated, to obtain M second similarities.

[0230] In a possible implementation, the second determination module 440 is further specifically used to:

[0231] N first similarities in the M first similarities arranged in descending order of values and N second similarities in the M second similarities arranged in descending order of values are determined;

[0232] According to the N first similarities and the N second similarities, N mixed similarities are determined.

[0233] In a possible implementation, the first similarity and the second similarity are cosine similarities.

[0234] Figure 5 A hardware structure diagram of the xx provided by the embodiments of the present application is shown.

[0235] The electronic device can include a processor 501 and a memory 502 having computer program instructions stored therein.

[0236] In particular, the processor 501 can include a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or can be configured to implement one or more integrated circuits that embody the embodiments of the present application.

[0237] The memory 502 can include mass storage for data or instructions. By way of example, and not limitation, the memory 502 can include a hard disk drive (HDD), a floppy disk drive, flash memory, an optical disc (e.g., a CD or DVD), a tape drive, a USB drive, or a combination of two or more of these. The memory 502 can be removable and / or non-removable (or fixed) as appropriate. The memory 502 can be internal or external as appropriate. In certain embodiments, the memory 502 is non-volatile solid-state memory.

[0238] In some embodiments, the memory 502 can include read-only memory (ROM), random access memory (RAM), a disk storage medium device, an optical storage medium device, a flash memory device, an electrical, optical, or other physical / tangible memory storage device. Thus, generally, the memory includes one or more tangible (non-transitory) computer-readable storage media (e.g., memory devices) encoded with software that, when executed (e.g., by one or more processors), is operable to perform operations described with reference to the methods provided according to embodiments of the present application.

[0239] The processor 501 implements the methods provided in the above-described embodiments by reading and executing computer program instructions stored in the memory 502.

[0240] In one example, the electronic device can further include a communication interface 503 and a bus 510. The processor 501, the memory 502, and the communication interface 503 are connected through the bus 510 and complete communication therebetween.

[0241] The communication interface 503 is mainly used to realize the communication between the modules, devices, units, and / or equipment in the embodiments of the present application.

[0242] Bus 510 includes hardware, software, or both, to couple electronic devices to each other in a network. While Figure 5 illustrates a bus, other interconnects that are used to interconnect various hardware components can be utilized. Although Figure 5 is described and shown as including a particular number and configuration of components, this application contemplates any suitable number and configuration of components.

[0243] In addition, the method provided by the above-mentioned embodiments can be implemented by a computer-readable storage medium. The computer-readable storage medium stores computer program instructions. The computer program instructions are executed by a processor to implement any one of the above-mentioned embodiments.

[0244] In addition, the method provided by the above-mentioned embodiments can be implemented by a computer program product. The program product is stored in a storage medium. The program product is executed by at least one processor to implement each process of the embodiments of the method provided by the above-mentioned embodiments, and can achieve similar or the same technical effects. To avoid repetition, details are not described here.

[0245] It should be noted that the present application is not limited to the specific configurations and processes described above and shown in the drawings. For the sake of brevity, detailed descriptions of well-known methods are omitted here. In the above-mentioned embodiments, several specific steps are described and shown as examples. However, the method process of the present application is not limited to the specific steps described and shown. Those skilled in the art can make various changes, modifications and additions, or change the order between steps, after understanding the spirit of the present application.

[0246] The functions noted in the description of the structural block diagrams above can be implemented as hardware, software, firmware, or a combination thereof. When implemented in hardware, they can be, for example, electronic circuits, application specific integrated circuits (ASICs), appropriate firmware, plug-ins, function cards, and the like. When implemented in software, the elements of the present application are program or code segments that are used to perform the required tasks. The program or code segments can be stored in a machine-readable medium, or transmitted through a data signal carried in a carrier wave over a transmission medium or communication link. A "machine-readable medium" includes any medium that can store or transport information. Examples of machine-readable media include electronic circuits, semiconductor memory devices, ROM, flash memory, erasable ROM (EROM), floppy disks, CD-ROMs, optical disks, hard disks, fiber optic media, radio frequency (RF) links, and the like. The code segments can be downloaded via computer networks such as the Internet, intranets, and the like.

[0247] It is also important to note that the examples mentioned in the present application describe some methods or systems based on a series of steps or devices. However, the present application is not limited to the order of the steps mentioned above, that is, the steps can be performed in the order mentioned in the examples, or in an order different from the examples, or several steps can be performed simultaneously.

[0248] The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks. The computer program instructions can also be loaded onto a computer, other programmable data processing apparatus, or other processing device to cause a series of operational steps to be performed on the computer, other programmable apparatus or other processing device to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide processes for implementing the functions / acts specified in the flowchart and / or block diagram block or blocks. These computer program instructions can also be stored in a computer readable medium that can direct a computer, other programmable data processing apparatus, or other processing device to operate in a particular manner, such that the instructions stored in the computer readable medium produce an article of manufacture including instructions which implement the function / act specified in the flowchart and / or block diagram block or blocks.

[0249] The above merely describes a specific implementation of the present application. Those skilled in the art can clearly understand the specific working processes of the system, modules and units described above for the convenience and brevity of description, and can refer to the corresponding processes in the foregoing method embodiments, which will not be described herein again. It should be understood that the protection scope of the present application is not limited to this, and any person skilled in the art can easily think of various equivalent modifications or replacements within the technical range disclosed by the present application, and these modifications or replacements should be covered within the protection scope of the present application.

Claims

1. A dialogue-based intent recognition method, characterized in that, The method includes: In response to the collected current dialogue information and historical dialogue information of the user, a first embedding vector corresponding to the current dialogue information and a second embedding vector corresponding to the historical dialogue information are determined. Calculate the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scene knowledge base to obtain M first similarities. The M knowledge embedding vectors are pre-configured based on M knowledge items. The pre-configured scene knowledge base includes the M knowledge items. Calculate the similarity between the second embedding vector and each of the M knowledge embedding vectors to obtain M second similarities; Based on the M first similarities and the M second similarities, N mixed similarities are determined, where N is a positive integer and N is less than M; Based on the N mixed similarities, N knowledge entries are determined from the pre-configured scene knowledge base; The N knowledge items are input into a pre-configured large language model, and the pre-configured large language model is used to output the user's intent recognition result. The pre-configured large language model is trained based on several historical knowledge items and their corresponding user intents.

2. The method according to claim 1, characterized in that, Before calculating the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scene knowledge base, the method further includes: In response to the collected current dialogue information and the historical dialogue information, the user scenario in which the user is located is determined; Based on the user scenario, the pre-configured scenario knowledge base is determined.

3. The method according to claim 2, characterized in that, The process of determining the user's current scenario in response to the collected current dialogue information and historical dialogue information includes: The user scenario is determined based on the current dialogue information and the historical dialogue information.

4. The method according to claim 2, characterized in that, The process of determining the user's current scenario in response to the collected current dialogue information and historical dialogue information includes: Based on the current dialogue information and the historical dialogue information, determine the user's unique identifier information and scene feature information; The target user profile corresponding to the unique identifier information is determined from a pre-configured user profile set, wherein the user profile set includes user profiles of several users; The user scenario is determined based on the target user profile and the scenario feature information.

5. The method according to claim 1, characterized in that, Before calculating the similarity between the second embedding vector and each of the M knowledge embedding vectors to obtain the M second similarities, the method further includes: Obtain the first timestamp of the current dialogue information and the second timestamp of the historical dialogue information; Based on the first timestamp and the second timestamp, the dialogue interval duration corresponding to the historical dialogue information is determined; Determine the time decay factor based on the dialogue interval duration; The calculation of the similarity between the second embedding vector and each of the M knowledge embedding vectors yields M second similarities, including: Based on the time decay factor, the similarity between the second embedding vector and each of the M knowledge embedding vectors is calculated to obtain M second similarities.

6. The method according to claim 1, characterized in that, The step of determining N mixed similarities based on the M first similarities and the M second similarities includes: Determine N first similarities from the M first similarities arranged in descending order of numerical value, and N second similarities from the M second similarities arranged in descending order of numerical value; The N mixed similarities are determined based on the N first similarities and the N second similarities.

7. The method according to claim 1, characterized in that, The first similarity and the second similarity are cosine similarities.

8. A dialogue-based intent recognition system, characterized in that, The system includes: The first determining module is used to determine, in response to the collected current dialogue information and historical dialogue information of the user, a first embedding vector corresponding to the current dialogue information and a second embedding vector corresponding to the historical dialogue information. The first calculation module is used to calculate the similarity between the first embedding vector and each of the M knowledge embedding vectors in the pre-configured scene knowledge base to obtain M first similarities. The M knowledge embedding vectors are pre-configured based on M knowledge items, and the pre-configured scene knowledge base includes the M knowledge items. The second calculation module is used to calculate the similarity between the second embedding vector and each of the M knowledge embedding vectors to obtain M second similarities. The second determining module is used to determine N mixed similarities based on the M first similarities and the M second similarities, where N is a positive integer and N is less than M; The third determining module is used to determine N knowledge entries from the pre-configured scene knowledge base based on the N mixed similarities. The processing module is used to input the N knowledge items into a pre-configured large language model, and use the pre-configured large language model to input the user's intent recognition result. The pre-configured large language model is trained based on several historical knowledge items and their corresponding user intents.

9. An electronic device, characterized in that, The electronic device includes a memory and a processor, wherein the memory stores a computer program that, when executed by the processor, implements the dialogue-based intent recognition method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by one or more processors, implements the dialogue-based intent recognition method according to any one of claims 1 to 7.

Images