WO2024093403A1 - 一种意图识别方法、装置、设备及可读存储介质 - Google Patents
一种意图识别方法、装置、设备及可读存储介质 Download PDFInfo
- Publication number
- WO2024093403A1 WO2024093403A1 PCT/CN2023/110632 CN2023110632W WO2024093403A1 WO 2024093403 A1 WO2024093403 A1 WO 2024093403A1 CN 2023110632 W CN2023110632 W CN 2023110632W WO 2024093403 A1 WO2024093403 A1 WO 2024093403A1
- Authority
- WO
- WIPO (PCT)
- Prior art keywords
- information
- historical
- output information
- reply
- reply information
- Prior art date
Links
- 238000000034 method Methods 0.000 title claims abstract description 79
- 238000012549 training Methods 0.000 claims abstract description 69
- 230000004044 response Effects 0.000 claims description 31
- 238000013145 classification model Methods 0.000 claims description 26
- 238000004590 computer program Methods 0.000 claims description 18
- 206010000117 Abnormal behaviour Diseases 0.000 claims description 15
- 238000001914 filtration Methods 0.000 claims description 2
- 230000008569 process Effects 0.000 description 13
- 238000010586 diagram Methods 0.000 description 10
- 230000006870 function Effects 0.000 description 9
- 230000006872 improvement Effects 0.000 description 9
- 230000001680 brushing effect Effects 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 6
- 238000012545 processing Methods 0.000 description 6
- 238000002372 labelling Methods 0.000 description 4
- 230000003287 optical effect Effects 0.000 description 3
- 101001121408 Homo sapiens L-amino-acid oxidase Proteins 0.000 description 2
- 102100026388 L-amino-acid oxidase Human genes 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 101100012902 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) FIG2 gene Proteins 0.000 description 1
- 101100233916 Saccharomyces cerevisiae (strain ATCC 204508 / S288c) KAR5 gene Proteins 0.000 description 1
- 238000013475 authorization Methods 0.000 description 1
- 230000006399 behavior Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000005540 biological transmission Effects 0.000 description 1
- 239000006227 byproduct Substances 0.000 description 1
- 230000001413 cellular effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 238000004891 communication Methods 0.000 description 1
- 238000013461 design Methods 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000003993 interaction Effects 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 229920001296 polysiloxane Polymers 0.000 description 1
- 239000000047 product Substances 0.000 description 1
- 230000000750 progressive effect Effects 0.000 description 1
- 239000010979 ruby Substances 0.000 description 1
- 229910001750 ruby Inorganic materials 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F40/00—Handling natural language data
- G06F40/30—Semantic analysis
- G06F40/35—Discourse or dialogue representation
Definitions
- the present invention relates to the field of computer technology, and in particular to an intent recognition method, device, equipment and readable storage medium.
- intelligent dialogue systems can identify the user's intention through the dialogue data input by the user, and then output intelligent dialogue data that meets the user's intention, so as to communicate smoothly with the user. Therefore, improving the accuracy of identifying user intentions has become an urgent problem to be solved.
- this specification provides an intent recognition method.
- This specification provides an intent recognition method, device, equipment and readable storage medium to partially solve the above-mentioned problems existing in the related art.
- the present specification adopts the following technical solution:
- the present specification provides an intention recognition method, comprising: receiving an intention recognition request, determining the reply information input by the target user to the output information of the intelligent dialogue system as the information to be recognized; matching the information to be recognized with each preset keyword to obtain the target keyword matched by the information to be recognized; searching for reference output information corresponding to the target keyword and reference reply information input by the target user to the reference output information from the historical dialogue records of the target user according to the correspondence between the target keyword and the preset historical output information and the keyword; inputting the information to be recognized and the reference reply information as input into a pre-trained intention recognition model to obtain the intention output by the intention recognition model as the target intention of the information to be recognized; wherein the intention recognition model is trained by the following method: determining a first training sample according to the historical output information output by the intelligent dialogue system and the historical reply information input by the reference user to the historical output information in a specified round of historical dialogue between the intelligent dialogue system and the reference user, determining a label of the first training sample
- This specification provides an intention recognition device, including: a receiving module, configured to receive an intention recognition request, confirm The reply information input by the target user in response to the output information of the intelligent dialogue system is used as the information to be identified; a matching module is used to match the information to be identified with each preset keyword to obtain the target keyword matched by the information to be identified; a search module is used to search for reference output information corresponding to the target keyword and reference reply information input by the target user in response to the reference output information from the historical dialogue records of the target user according to the correspondence between the target keyword and the preset historical output information and the keyword; an identification module is used to input the information to be identified and the reference reply information as input into a pre-trained intent recognition model to obtain the intent output by the intent recognition model as the target intent of the information to be identified.
- This specification provides a computer-readable storage medium, which stores a computer program.
- the computer program is executed by a processor, the above-mentioned intention recognition method is implemented.
- This specification provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above-mentioned intention recognition method when executing the program.
- At least one of the above technical solutions adopted in this specification can achieve the following beneficial effects:
- the intent recognition method provided in this specification by matching the information to be recognized with each preset keyword, the target keyword matching the information to be recognized is obtained, and according to the target keyword, the reference output information corresponding to the target keyword and the reference reply information input by the target user for the reference output information are searched from the historical conversation records of the target user, and then the information to be recognized and the reference reply information are input into the pre-trained intent recognition model, and the intent output by the model is obtained as the target intent of the information to be recognized.
- the method of searching for reference output information and reference reply information according to the target keyword matching the information to be recognized does not require adjustments to the training method and model structure of the existing intent recognition model, and can also be combined with multiple rounds of conversations for intent recognition, thereby improving the accuracy of intent recognition.
- FIG1 is a flow chart of an intention recognition method in this specification
- FIG2 is a flow chart of an intention recognition method in this specification.
- FIG3 is a schematic diagram of an intention recognition device provided in this specification.
- FIG. 4 is a schematic diagram of an electronic device corresponding to FIG. 1 provided in this specification.
- human-computer dialogue users and intelligent dialogue systems use a certain dialogue language and interact in a certain way to complete the information exchange between people and intelligent dialogue systems.
- the intelligent dialogue system usually recognizes the intention of the user's input voice or text information to obtain the user's intention, and then obtains the output result based on the intention to complete the dialogue with the user.
- this specification provides an intent recognition method, which uses the intent predicted jointly by the information to be recognized and the reference response information as the target intent corresponding to the information to be recognized. There is no need to adjust the training method and model structure of the existing intent recognition model. Intent recognition can also be performed in combination with multiple rounds of conversations to improve the accuracy of intent recognition.
- FIG1 is a flow chart of an intent recognition method provided in this specification.
- S100 receiving an intent recognition request, and determining reply information input by a target user in response to output information of an intelligent dialogue system as information to be recognized.
- the business platform can use outbound calls or proactively initiate conversations to exchange information with the user through multiple rounds of conversations, thereby identifying whether the transaction performed by the user is risky, so as to detect risks and alert the user in time to avoid infringement of the user's interests.
- each round of dialogue between the intelligent dialogue system and the target user may include historical output information output by the intelligent dialogue system and historical reply information input by the user.
- the intelligent dialogue system generally asks questions and the user inputs answers to the questions, so that the historical output information output by the intelligent dialogue system may be historical questions, and the historical reply information input by the target user may be historical answers to the historical questions.
- an intent recognition request is received, and the intent recognition request may carry the reply information input by the target user to the output information of the intelligent dialogue system, and the reply information may be used as the information to be recognized.
- the specific mode of the human-computer dialogue between the target user and the intelligent dialogue system may be any existing human-computer dialogue mode such as voice, text, and picture, and this specification does not limit this.
- S102 Match the information to be identified with preset keywords to obtain target keywords that match the information to be identified.
- the intelligent dialogue system may repeatedly ask the user the same type of questions.
- the target user may give completely different responses to the same type of questions. For example, "Did you just make a transaction for a fake order task?" and "Are you doing fake orders on the shopping platform?”
- the above two questions are not exactly the same, but they are of the same type, both used to ask the target user whether he has the abnormal behavior of fake orders.
- the historical output information in the historical conversation records can be classified in advance to obtain keywords corresponding to the historical output information, so as to determine the type of output information in response to the information to be identified, and find multiple reply information input by the target user for the same type of output information. Then, when determining the target intention corresponding to the information to be identified, the multiple reply information given by the target user for the same type of output information in multiple rounds of conversations are combined to obtain the information to be identified.
- S104 According to the target keyword and the correspondence between the preset historical output information and the keyword, searching for reference output information corresponding to the target keyword from the historical conversation records of the target user, and The reference reply information input by the target user in response to the reference output information.
- keywords of historical output information can be obtained through manual annotation, and the historical output information can be classified.
- the manual annotation method requires the annotation personnel to have a high level of professional skills.
- the correspondence between historical output information and keywords can be established through a pre-trained classification model based on multiple rounds of historical conversation records, and then the historical output information can be classified.
- one historical output information can correspond to multiple keywords, and at the same time, one keyword can correspond to multiple historical output information.
- the historical dialogue records may be historical dialogue records generated by the human-computer dialogue between the target user and the intelligent dialogue system within a specified time period.
- the specific length of the specified time period may be determined according to the specific application scenario and is not limited in this specification.
- reference output information of the same type as the output information in response to the information to be identified can be obtained, and then reference reply information input by the target user for multiple reference output information of the same type in the human-computer dialogue between the target user and the intelligent dialogue system within a specified period of time can be found.
- the purpose of identifying the intention of the target user by combining multiple rounds of dialogue can be achieved without changing the model structure and training method of the existing intention recognition model, which not only achieves the advantages of convenient labeling, small video memory usage, and fast prediction speed of single-round dialogue prediction intention, but also achieves the effect of jointly predicting intention in multiple rounds of dialogue.
- S106 The information to be identified and the reference reply information are input into a pre-trained intent recognition model to obtain the intent output by the intent recognition model as the target intent of the information to be identified.
- the target user may input different reply information for the same output information.
- the reference output information is searched through the target keywords of the information to be recognized, and then the reference reply information input by the user for the same type of reference output information is determined.
- the information to be identified is spliced with the reference reply information, and the spliced information is input into a pre-trained intent recognition model. If the reference reply information is consistent with the intent of the information to be identified, the reference reply information and the information to be identified are input into the intent recognition model, and the intent output by the model may be the same as or similar to the intent represented by the reference reply information. If the intent represented by the reference reply information is inconsistent with the intent represented by the information to be identified, the reference reply information and the information to be identified are input into the intent recognition model, and the intent output by the model may be opposite to the intent represented by the reference reply information.
- the intent recognition model can be trained using an existing training method. Specifically, the intent recognition model is trained using the following method:
- a first training sample is determined based on historical output information output by the intelligent dialogue system and historical reply information input by the reference user in a specified round of historical dialogue between the intelligent dialogue system and a reference user, a label of the first training sample is determined based on the intention of the reference user corresponding to the specified round of historical dialogue, and the intent recognition model is trained based on the first training sample and the label of the first training sample.
- a multi-round historical dialogue record is obtained when the intelligent dialogue system and the reference user have a multi-round historical dialogue, and then at least one round of historical dialogue is selected from the multi-round historical dialogue record as a designated round of historical dialogue, wherein the number of rounds and the dialogue time of the designated round of historical dialogue are not limited in this specification.
- the intent recognition model can use the historical output information and historical reply information in a single round of historical dialogue as the first training sample for training.
- multiple historical output information and multiple historical reply information in multiple rounds of historical dialogue can also be used as the first training sample for training.
- the training samples of the intent recognition model can be the historical output information and historical reply information in a single round of historical conversation records, and the labels of the training samples can be the user intent represented by the single round of historical conversation records, wherein the labels of the training samples can be obtained by any existing method, such as manual labeling. That is to say, in one or more embodiments of this specification, the intent recognition model used can also take into account the advantages of convenient labeling, small memory usage, and fast prediction speed of the single round conversation prediction intent.
- the training samples of the intent recognition model and the labels of the training samples, the model structure, the training method, etc. are not changed.
- the target keyword corresponding to the information to be identified is obtained to search for the reference output information corresponding to the information to be identified, so as to determine multiple reference reply information of the target user for the output information corresponding to the same keyword in multiple rounds of conversations, and comprehensively identify the intent of the target user based on the information to be identified and the multiple reference reply information, so that when the target user gives different reply information, the intent of the target user can be obtained in combination with multiple rounds of conversations.
- the target keyword matching the information to be recognized is obtained by matching the information to be recognized with each preset keyword, and the reference output information corresponding to the target keyword and the reference reply information input by the target user for the reference output information are searched from the historical conversation records of the target user according to the target keyword, and then the information to be recognized and the reference reply information are input into the pre-trained intent recognition model to obtain the intent output by the model as the target intent of the information to be recognized.
- the method of searching for reference output information and reference reply information according to the target keyword matching the information to be recognized does not require adjustments to the training method and model structure of the existing intent recognition model, and can also be combined with multiple rounds of conversations for intent recognition, thereby improving the accuracy of intent recognition.
- step S102 of FIG. 1 the information to be identified is matched with each preset keyword.
- the correspondence between each keyword and the historical output information can be determined based on the historical dialogue records of multiple rounds of dialogue between the reference user and the intelligent dialogue system. This is specifically achieved through the following steps S200 to S206, as shown in FIG. 2 .
- S200 Obtain historical dialogue records of multiple rounds of dialogues between a reference user and an intelligent dialogue system, wherein each round of historical dialogue records includes historical output information output by the intelligent dialogue system and historical reply information input by the reference user in response to the historical output information.
- the correspondence between the preset historical output information and the keywords can be manually annotated.
- at least one keyword of the historical output information is extracted from the text of the historical output information based on the semantics of the historical output information.
- this method requires too high a professional ability of the annotator, and may result in incomplete or incorrect annotation. Therefore, in the embodiment of this specification, the correspondence between each historical output information and each keyword is automatically established by extracting keywords from the exclusive reply information of the historical output information.
- the historical conversation records of multiple rounds of conversations between the reference user and the intelligent conversation system may be historical conversation records of multiple rounds of conversations between multiple different reference users and the intelligent conversation system within a preset time period.
- Each reference user may include the target user, which is not limited in this specification.
- the intention recognition method provided in one or more embodiments of the present specification can be applied to the scenario of identifying abnormal behavior of users in an intelligent anti-fraud system
- the historical output information output by the intelligent dialogue system can be recorded in each round of historical dialogue records, and the historical reply information input by the user in response to the historical output information can be referenced.
- the historical reply information is input into a pre-trained classification model to obtain the probability that the historical reply information output by the classification model corresponds to each historical output information.
- the historical reply information input by the user in response to the historical output information can be in two forms.
- One is general reply information, which means that the reply information can be used to respond to various types of output information, such as "um”, “yes” and “no".
- the second is exclusive reply information, which means that the reply information can only be used to respond to the output information corresponding to the reply information, such as "I did not swiped the order" or "I made a purchase”.
- the exclusive reply information may include the type or subject of the output information to which the exclusive reply information responds. Therefore, the exclusive reply information can be removed by removing the general reply information, retaining the exclusive reply information, and extracting keywords based on the exclusive reply information to determine the historical information corresponding to the exclusive reply information. Keywords for output information.
- the classification model is trained by the following method: Step 1: For each round of historical conversation records, the historical reply information in the round of historical conversation records is used as the second training sample, and the historical output information in the round of historical conversation records is used as the label of the second training sample.
- Step 2 Input the second training sample into the classification model to be trained to obtain prediction information of the second training sample output by the classification model.
- Step 3 Train the classification model with minimizing the difference between the prediction information of the second training sample and the label of the second training sample as the training goal.
- the classification model can predict the probability that the historical reply information corresponds to each historical output information based on the historical reply information.
- the historical output information with the highest probability is the historical output information responded by the historical reply information.
- the general reply information is not eliminated in the training samples of the classification model. This results in that when the general reply information is input into the classification model, the probability that the general reply information output by the model corresponds to each historical output information is usually similar. Therefore, further, the difference between the probabilities of the historical reply information corresponding to each historical output information is determined, and each historical reply information with a difference higher than the preset difference threshold is used as each candidate information.
- the designated reply information corresponding to the historical output information is screened out from the historical reply information according to the probability that the candidate reply information corresponds to the historical output information.
- S204 Determine at least one keyword corresponding to the historical output information according to each designated reply information corresponding to the historical output information.
- each designated reply information corresponding to the historical output information is segmented to obtain candidate words, and the candidate words are input into a pre-trained intent recognition model, and at least one keyword corresponding to the historical output information is filtered out from the candidate words according to the output of the intent recognition model.
- the historical output information is "Did you just make a transaction for a fake order task?"
- the specified reply information of the historical output information is "I did not make fake orders.”
- "I did not make fake orders” can be segmented to obtain single candidate words "I", "no” and "fake orders”.
- these three candidate words are respectively input into the pre-trained intent recognition model. Since the two keywords "I” and “no” cannot represent any user intent, the probability of each intent output by the model for these two keywords is low, and the keyword "fake orders" can hit the abnormal behavior of "fake orders". Therefore, the candidate words are obtained by segmenting the specified reply information, and then the keywords corresponding to the historical output information are screened out from the candidate words.
- one historical output information may correspond to at least one keyword, and one keyword may also correspond to at least one historical output information.
- the designated reply information corresponding to each historical output information is obtained, and then a keyword corresponding to the historical output information is determined from the designated reply information, thereby establishing the corresponding relationship between each historical output information and each keyword. It can be seen that the non-manual method of establishing the corresponding relationship between the historical output information and the keyword avoids the problem of incomplete or incorrect annotation that may occur during manual annotation.
- the question in response to the information to be identified is: a question for inquiring whether the target user has abnormal behavior;
- the information to be identified contains a negative word.
- the intelligent dialogue system can usually output a question asking the target user whether the target user has abnormal behavior. If the target user has given a positive response to the above-mentioned type of question in the previous rounds of dialogue, but gives a negative response in the current round of dialogue, it means that the target user's dialogue intention has changed. It is necessary to re-identify the target user's intention in response to this situation, and then correct the target user's intention in the previous rounds of dialogue.
- the intent of the target user corresponding to the reference reply information can also be obtained, and the intent of the target user corresponding to the reference reply information can be adjusted according to the target intent, so as to determine whether the target user has abnormal behavior according to the adjusted intent.
- the intent of the target user corresponding to the reference reply information can be covered, adjusted, etc. according to the target intent output by the spliced reply information according to the intent recognition model.
- the output information of the intelligent dialogue system is "Did you just make a transaction for the purpose of brushing orders?", and the reply information input by the target user is "Yeah”.
- the target user's intention can be identified as "brushing orders”.
- the output of the intelligent dialogue system is "Are you brushing orders on the shopping platform?”, and the information to be identified input by the target user is "I didn't brush orders”.
- "I didn't brush orders" and "Yeah” can be concatenated and input into the pre-trained intent recognition model, and the output of the intent recognition model is that the target user's intention is "no brushing orders”.
- the intention of "brushing orders" in the historical dialogue record can be corrected according to the intention of "no brushing orders", and the user corresponding to the corrected historical dialogue can be corrected.
- the behavior is redefined to improve the accuracy of judging whether the target user has abnormal behavior.
- Figure 3 is a schematic diagram of an intention recognition device provided in this specification, which specifically includes: a receiving module 300, which is used to receive an intention recognition request and determine the reply information input by the target user to the output information of the intelligent dialogue system as the information to be recognized; a matching module 302, which is used to match the information to be recognized with each preset keyword to obtain the target keyword matched by the information to be recognized; a search module 304, which is used to search for reference output information corresponding to the target keyword and reference reply information input by the target user to the reference output information from the historical dialogue records of the target user according to the correspondence between the target keyword and the preset historical output information and the keyword; an identification module 306, which is used to input the information to be recognized and the reference reply information as input into a pre-trained intention recognition model to obtain the intent output by the intention recognition model as the target intent of the information to be recognized.
- a receiving module 300 which is used to receive an intention recognition request and determine the reply information input by the target user to the output information of the intelligent dialogue system as the information to be
- the device also includes: a correspondence determination module 308, which is specifically used to obtain historical dialogue records of multiple rounds of dialogues between a reference user and an intelligent dialogue system, wherein each round of historical dialogue records contains historical output information output by the intelligent dialogue system, and historical reply information input by the reference user for the historical output information; for each historical output information, filtering out each designated reply information corresponding to the historical output information from each historical reply information; determining at least one keyword corresponding to the historical output information based on each designated reply information corresponding to the historical output information; and determining the correspondence between the historical output information and the keyword based on the keyword corresponding to each historical output information.
- a correspondence determination module 308 is specifically used to obtain historical dialogue records of multiple rounds of dialogues between a reference user and an intelligent dialogue system, wherein each round of historical dialogue records contains historical output information output by the intelligent dialogue system, and historical reply information input by the reference user for the historical output information; for each historical output information, filtering out each designated reply information corresponding to the historical output information from each historical reply information; determining at least one keyword corresponding
- the correspondence determination module 308 is specifically used to, for each historical reply information, input the historical reply information into a pre-trained classification model to obtain the probability that the historical reply information output by the classification model corresponds to each historical output information; determine the difference between the probabilities that the historical reply information corresponds to each historical output information, and use each historical reply information whose difference is higher than a preset difference threshold as each candidate reply information; for each historical output information, filter out each designated reply information corresponding to the historical output information from the each historical reply information according to the probability that each candidate reply information corresponds to the historical output information.
- the device also includes: a training module 310, which is specifically used to pre-target each round of historical conversation records, using the historical reply information input by the reference user in that round of historical conversation records as the second training sample, and using the historical output information in response to the historical reply information in that round of historical conversation records as the label of the second training sample; inputting the second training sample into the classification model to be trained to obtain the prediction information of the second training sample output by the classification model; and training the classification model with minimizing the difference between the prediction information of the second training sample and the label of the second training sample as the training goal.
- the corresponding relationship determination module 308 is specifically configured to determine the corresponding responses of the historical output information.
- the complex information is segmented respectively to obtain candidate words; the candidate words are input into a pre-trained intention recognition model respectively, and at least one keyword corresponding to the historical output information is screened out from the candidate words according to the output of the intention recognition model.
- the output information of the intelligent dialogue system in response to the information to be identified is: a question used to ask the target user whether there is abnormal behavior; optionally, the search module 304 is also used to determine whether the information to be identified contains negative words before the search module 304 searches for reference output information corresponding to the target keyword and the target user inputs reference reply information for the reference output information.
- the device also includes: an adjustment module 312, which is specifically used to obtain the intention of the target user corresponding to the reference reply information; according to the target intention, adjust the intention of the target user corresponding to the reference reply information, so as to determine whether the target user has abnormal behavior based on the adjusted intention.
- an adjustment module 312 which is specifically used to obtain the intention of the target user corresponding to the reference reply information; according to the target intention, adjust the intention of the target user corresponding to the reference reply information, so as to determine whether the target user has abnormal behavior based on the adjusted intention.
- This specification also provides a computer-readable storage medium, which stores a computer program.
- the computer program can be used to execute the intent recognition method shown in FIG. 1 above.
- the electronic device includes a processor, an internal bus, a network interface, a memory, and a non-volatile memory, and of course may also include hardware required for other services.
- the processor reads the corresponding computer program from the non-volatile memory into the memory and then runs it to implement the intent recognition method shown in Figure 1 above.
- this specification does not exclude other implementation methods, such as logic devices or a combination of software and hardware, etc., that is to say, the execution subject of the following processing flow is not limited to each logic unit, but can also be hardware or logic devices.
- a programmable logic device such as a field programmable gate array (FPGA)
- FPGA field programmable gate array
- HDL Hardware Description Language
- ABEL Advanced Boolean Expression Language
- AHDL Altera Hardware Description Language
- HDCal JHDL
- Lava Lava
- Lola MyHDL
- PALASM RHDL
- VHDL Very-High-Speed Integrated Circuit Hardware Description Language
- Verilog Verilog
- the controller may be implemented in any suitable manner, for example, the controller may take the form of a microprocessor or processor and a computer readable medium storing a computer readable program code (e.g., software or firmware) executable by the (micro)processor, a logic gate, a switch, an application specific integrated circuit (ASIC), a programmable logic controller, and an embedded microcontroller, examples of which include but are not limited to the following microcontrollers: ARC 625D, Atmel AT91SAM, Microchip PIC18F26K20, and Silicone Labs C8051F320, and the memory controller may also be implemented as part of the control logic of the memory.
- a computer readable program code e.g., software or firmware
- the controller may be implemented in the form of a logic gate, a switch, an application specific integrated circuit, a programmable logic controller, and an embedded microcontroller by logically programming the method steps. Therefore, such a controller may be considered as a hardware component, and the means for implementing various functions included therein may also be considered as a structure within the hardware component. Or even, the means for implementing various functions may be considered as both a software module for implementing the method and a structure within the hardware component.
- a typical implementation device is a computer.
- the computer may be, for example, a personal computer, a laptop computer, a cellular phone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.
- the embodiments of the present application may be provided as methods, systems, or computer program products. Therefore, the present application may adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the present application may adopt the form of a computer program product implemented in one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.
- a computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
- These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer-readable memory produce a manufactured product including an instruction device that implements the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.
- These computer program instructions may also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, whereby the instructions executed on the computer or other programmable device provide steps for implementing the functions specified in one or more processes in the flowchart and/or one or more boxes in the block diagram.
- a computing device includes one or more processors (CPU), input/output interfaces, network interfaces, and memory.
- processors CPU
- input/output interfaces network interfaces
- memory volatile and non-volatile memory
- Memory may include non-permanent storage in a computer-readable medium, in the form of random access memory (RAM) and/or non-volatile memory, such as read-only memory (ROM) or flash memory (flash RAM). Memory is an example of a computer-readable medium.
- RAM random access memory
- ROM read-only memory
- flash RAM flash memory
- Computer readable media include permanent and non-permanent, removable and non-removable media that can be implemented by any method or technology to store information.
- Information can be computer readable instructions, data structures, program modules or other data.
- Examples of computer storage media include, but are not limited to, phase change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technology, compact disk read-only memory (CD-ROM), digital versatile disk (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices or any other non-transmission media that can be used to store information that can be accessed by a computing device.
- computer readable media does not include temporary computer readable media (transitory media), such as modulated data signals and carrier waves.
- this specification may be provided as methods, systems or computer program products. Therefore, this specification may take the form of a complete hardware embodiment, a complete software embodiment or an embodiment combining software and hardware. Moreover, this specification may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
- computer-usable storage media including but not limited to disk storage, CD-ROM, optical storage, etc.
- program modules include routines, programs, objects, components, data structures, etc. that perform specific tasks or implement specific abstract data types.
- This specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices connected through a communication network.
- program modules may be located in local and remote computer storage media including storage devices.
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Health & Medical Sciences (AREA)
- Artificial Intelligence (AREA)
- Audiology, Speech & Language Pathology (AREA)
- Computational Linguistics (AREA)
- General Health & Medical Sciences (AREA)
- Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Information Retrieval, Db Structures And Fs Structures Therefor (AREA)
Abstract
一种意图识别方法、装置、设备及可读存储介质，通过将待识别信息与预设的各关键词进行匹配，得到待识别信息匹配的目标关键词，根据目标关键词，从目标用户的历史对话记录中，查找与目标关键词对应的参考输出信息，以及目标用户针对参考输出信息输入的参考回复信息，进而将待识别信息和参考回复信息输入预训练的意图识别模型，得到模型输出的意图作为待识别信息的目标意图。可见，根据待识别信息匹配的目标关键词，查找参考输出信息和参考回复信息的方式，不需要对现有的意图识别模型的训练方式和模型结构做出调整，也可以结合多轮对话进行意图识别，提高了意图识别的准确性以及隐私信息的安全性。
Description
本说明书涉及计算机技术领域，尤其涉及一种意图识别方法、装置、设备及可读存储介质。
随着人们对隐私数据关注度的提高，人机交互领域也受到了广泛的关注。当前，智能对话系统可通过用户输入的对话数据识别用户的意图，从而输出符合用户意图的智能对话数据，以便与用户进行流畅的沟通。因此，提高识别用户意图的准确性成为亟待解决的问题。
基于此，本说明书提供一种意图识别方法。
发明内容
本说明书提供一种意图识别方法、装置、设备及可读存储介质，以部分的解决相关技术存在的上述问题。
本说明书采用下述技术方案：本说明书提供了一种意图识别方法，包括：接收意图识别请求，确定目标用户针对智能对话系统的输出信息输入的回复信息，作为待识别信息；将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词；根据所述目标关键词以及预设的历史输出信息与关键词之间的对应关系，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息；将所述待识别信息与所述参考回复信息作为输入，输入到预先训练的意图识别模型，得到所述意图识别模型输出的意图，作为所述待识别信息的目标意图；其中，采用下述方法训练所述意图识别模型：根据智能对话系统与参考用户的指定轮历史对话中，所述智能对话系统输出的历史输出信息以及所述参考用户针对所述历史输出信息输入的历史回复信息，确定第一训练样本，根据所述指定轮历史对话对应的所述参考用户的意图确定所述第一训练样本的标签，根据所述第一训练样本和所述第一训练样本的标签训练所述意图识别模型。
本说明书提供了一种意图识别装置，包括：接收模块，用于接收意图识别请求，确
定目标用户针对智能对话系统的输出信息输入的回复信息，作为待识别信息；匹配模块，用于将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词；查找模块，用于根据所述目标关键词以及预设的历史输出信息与关键词之间的对应关系，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息；识别模块，用于将所述待识别信息与所述参考回复信息作为输入，输入到预先训练的意图识别模型，得到所述意图识别模型输出的意图，作为所述待识别信息的目标意图。
本说明书提供了一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序被处理器执行时实现上述意图识别方法。
本说明书提供了一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现上述意图识别方法。
本说明书采用的上述至少一个技术方案能够达到以下有益效果：本说明提供的意图识别方法中，通过将待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词，根据所述目标关键词，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及目标用户针对参考输出信息输入的参考回复信息，进而将待识别信息和参考回复信息输入预训练的意图识别模型，得到模型输出的意图作为待识别信息的目标意图。可见，根据待识别信息匹配的目标关键词，查找参考输出信息和参考回复信息的方式，不需要对现有的意图识别模型的训练方式和模型结构做出调整，也可以结合多轮对话进行意图识别，提高了意图识别的准确性。
此处所说明的附图用来提供对本说明书的进一步理解，构成本说明书的一部分，本说明书的示意性实施例及其说明用于解释本说明书，并不构成对本说明书的不当限定。在附图中：
图1为本说明书中一种意图识别方法的流程示意图；
图2为本说明书中一种意图识别方法的流程示意图；
图3为本说明书提供的一种意图识别装置的示意图；
图4为本说明书提供的对应于图1的电子设备示意图。
为使本说明书的目的、技术方案和优点更加清楚，下面将结合本说明书具体实施例及相应的附图对本说明书技术方案进行清楚、完整地描述。显然，所描述的实施例仅是本说明书一部分实施例，而不是全部的实施例。基于本说明书中的实施例，本领域普通技术人员在没有做出创造性劳动前提下所获得的所有其他实施例，都属于本说明书保护的范围。
另外，需要说明的是，本申请中所有获取信号、信息或数据的动作都是在遵照所在地国家相应的数据保护法规政策的前提下，并获得由相应装置所有者给予授权的情况下进行的。
在人机对话中，用户与智能对话系统之间使用某种对话语言，通过一定的交互方式，完成人与智能对话系统之间的信息交换。在人机对话过程中，智能对话系统通常对用户输入的语音或文本信息进行意图识别得到用户的意图，然后根据意图得到输出结果以完成与用户之间的对话。
但在实际应用中，可能会出现多轮对话中用户在某一轮对话中否认了之前的回答的情况。此时，如果意图识别模型是以单轮对话的对话数据为训练样本，以单轮对话的用户意图为训练样本的标签训练得到的话，则无法针对上述情况输出用户在多轮对话中的意图。而采用多轮对话的对话数据为训练样本，以多轮对话的用户意图为标签训练意图识别模型，在训练过程中存在人工打标困难，模型训练占用计算资源高，预测效率低的问题。
基于此，本说明书提供一种意图识别方法，基于待识别信息和与参考回复信息共同预测出的意图，作为待识别信息对应的目标意图，不需要对现有的意图识别模型的训练方式和模型结构做出调整，也可以结合多轮对话进行意图识别，以提高意图识别的准确性。
以下结合附图，详细说明本说明书各实施例提供的技术方案。
图1为本说明书提供的一种意图识别方法的流程示意图。
S100：接收意图识别请求，确定目标用户针对智能对话系统的输出信息输入的回复信息，作为待识别信息。
随着互联网技术的快速发展，越来越多的用户通过业务平台进行线上交易业务。同
时，可能存在用户遭遇网络诈骗从而进行具有交易风险的线上交易的情况。针对上述情况，业务平台可采用外呼或主动发起对话等方式，通过多轮对话与用户进行信息交换，从而识别用户执行的交易业务是否存在风险，以便即使发现风险并提示用户，避免用户利益收到侵害。
在本说明书实施例中，智能对话系统与目标用户之间的每一轮对话中，可以包含智能对话系统输出的历史输出信息和用户输入的历史回复信息。可选的，通过人机对话的方式确认目标用户是否存在异常行为的场景中，一般由智能对话系统提出问题，由用户输入问题的回答，由此，智能对话系统输出的历史输出信息可以是历史提问，目标用户输入的历史回复信息可以是针对历史提问的历史回答。
具体的，接收意图识别请求，意图识别请求中可以携带目标用户针对智能对话系统的输出信息输入的回复信息，可将回复信息作为待识别信息。其中，目标用户与智能对话系统之间的人机对话的具体方式可以语音、文字、图片等任意现有的人机对话方式，本说明书对此不做限定。
S102：将将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词。
在智能对话系统与目标用户的人机对话过程中，智能对话系统可能会重复询问用户同一类型的问题。但是在实际应用中，目标用户可能会针对同一类型的提问给出截然不同的回复。例如，“您刚才的交易是为了做刷单任务吗？”和“你是在购物平台上刷单的吗？”，上述两个问题并不完全一致，但类型相同，都是用于询问目标用户是否存在刷单这一异常行为。基于上述实际情况，仅根据单轮对话数据为训练样本训练的意图识别模型，难以从目标用户针对同一类型的提问给出的截然不同的回复中准确识别目标用户的真实意图。
因此，可以预先对历史对话记录中的历史输出信息进行分类，得到历史输出信息对应的关键词，以便确定待识别信息回应的输出信息的类型，并查找到针对同一类型的输出信息目标用户输入的多个回复信息，进而在确定待识别信息对应的目标意图时，结合多轮对话中目标用户针对同一类型的输出信息给出的多个回复信息综合得到待识别信息。
S104：根据所述目标关键词以及预设的历史输出信息与关键词之间的对应关系，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及所
述目标用户针对所述参考输出信息输入的参考回复信息。
具体的，可以通过人工标注的方式，得到历史输出信息的关键词，并对历史输出信息进行分类。当然，人工标注的方式要求标注人员的业务水平较高，基于此，还可以根据多轮历史对话记录通过预先训练的分类模型，建立历史输出信息与关键词之间的对应关系，进而对历史输出信息进行分类。其中，一个历史输出信息可以对应于多个关键词，同时，一个关键词可以对应于多个历史输出信息。
进一步的，从目标用户与智能对话系统的历史对话记录中，查找与目标关键词对应的参考输出信息。其中，历史对话记录可以是目标用户与智能对话系统在指定时间段内进行人机对话生成的历史对话记录，指定时间段的具体时长可根据具体的应用场景确定，本说明书不做限定。
于是，通过查找待识别信息匹配的关键词，可以得到与待识别信息回应的输出信息相同类型的参考输出信息，进而可以查找到目标用户在指定时段内与智能对话系统的人机对话中，针对多个相同类型的参考输出信息输入的参考回复信息。通过本说明书实施例中结合历史对话记录中多个参考回复信息识别目标用户的意图的方法，在不改变现有的意图识别模型的模型结构以及训练方法的前提下，就可以实现结合多轮对话识别目标用户意图的目的，既实现了单轮对话预测意图的打标方便、显存占用小、预测速度快的优点，又达到了多轮对话共同预测意图的效果。
S106：将所述待识别信息与所述参考回复信息作为输入，输入到预先训练的意图识别模型，得到所述意图识别模型输出的意图，作为所述待识别信息的目标意图。
在实际应用中，目标用户与智能对话系统之间的人机对话过程中，可能存在目标用户针对同一输出信息的输入不同的回复信息的情况。为了在上述情况下提高意图识别的准确性，通过待识别信息的目标关键词查找参考输出信息，进而确定用户针对同一类型的参考输出信息输入的参考回复信息。
具体的，在本说明书实施例中，将待识别信息与参考回复信息拼接，并将拼接后的信息输入到预先训练的意图识别模型。如果参考回复信息与待识别信息的意图一致，则将参考回复信息与待识别信息输入意图识别模型，模型输出的意图可以与参考回复信息表征的意图相同或相近。如果参考回复信息表征的意图与待识别信息的表征的意图不一致，则将参考回复信息与待识别信息输入意图识别模型，模型输出的意图可能与参考回复信息所表征的意图相反。
在本说明书实施例中，意图识别模型可以采用现有的训练方式进行训练，具体的，采用下述方法训练所述意图识别模型：
根据智能对话系统与参考用户的指定轮历史对话中，所述智能对话系统输出的历史输出信息以及所述参考用户针对所述历史输出信息输入的历史回复信息，确定第一训练样本，根据所述指定轮历史对话对应的所述参考用户的意图确定所述第一训练样本的标签，根据所述第一训练样本和所述第一训练样本的标签训练所述意图识别模型。
可选的，在本说明书实施例中，首先获取智能对话系统和参考用户进行多轮历史对话时的多轮历史对话记录，然后，在上述多轮历史对话记录中选择至少一轮历史对话作为指定轮历史对话，其中，指定轮历史对话的轮数和对话时间本说明书不做限定。也就是说，在本说明书实施例中，意图识别模型可以采用单轮历史对话中的历史输出信息和历史回复信息作为第一训练样本进行训练。当然，也可以采用多轮历史对话中的多个历史输出信息和多个历史回复信息作为第一训练样本进行训练。
于是，意图识别模型的训练样本可以是单轮历史对话记录中的历史输出信息和历史回复信息，训练样本的标签可以是单轮历史对话记录所表征的用户意图，其中，训练样本的标签可以采用现有的任意方式获得，如人工标注。也就是说，本说明书一个或多个实施例中，采用的意图识别模型也可以兼顾单轮对话预测意图的打标方便、显存占用小、预测速度快的优点。
由于在本说明书实施例中，并未对意图识别模型的训练样本及训练样本的标签、模型结构、训练方式等进行更改，而是通过获取待识别信息对应的目标关键词查找待识别信息对应的参考输出信息的方式，确定目标用户在多轮对话中针对同一关键词对应的输出信息的多个参考回复信息，并基于待识别信息与多个参考回复信息综合识别目标用户的意图，以便在目标用户给出不同回复信息时能够结合多轮对话得到目标用户的意图。
本说明提供的意图识别方法中，通过将待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词，根据所述目标关键词，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及目标用户针对参考输出信息输入的参考回复信息，进而将待识别信息和参考回复信息输入预训练的意图识别模型，得到模型输出的意图作为待识别信息的目标意图。可见，根据待识别信息匹配的目标关键词，查找参考输出信息和参考回复信息的方式，不需要对现有的意图识别模型的训练方式和模型结构做出调整，也可以结合多轮对话进行意图识别，提高了意图识别的准确性。
在本说明书一个或多个实施例中，如图1步骤S102所示将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词之前，可以基于参考用户与智能对话系统的多轮对话的历史对话记录，确定各关键词与历史输出信息之间的对应关系，具体通过以下步骤S200至S206实现，如图2所示。
S200：获取参考用户与智能对话系统多轮对话的历史对话记录，其中，每轮历史对话记录中包含所述智能对话系统输出的历史输出信息，以及所述参考用户针对所述历史输出信息输入的历史回复信息。
在实际应用中，预设的历史输出信息与关键词之间的对应关系可以通过人工标注的方式，人工针对每个历史输出信息，从该历史输出信息的文本中，基于该历史输出信息的语义，提取该历史输出信息的至少一个关键词。然而，这种方法对标注人员的业务能力要求过高，可能会出现标注不全或标注错误的问题。由此，本说明书实施例中，通过从历史输出信息的专属回复信息中提取关键词的方式，自动建立各历史输出信息与各关键词之间的对应关系。
其中，参考用户与智能对话系统的多轮对话的历史对话记录，可以是在预设时间段内，多个不同参考用户分别与智能对话系统之间的多轮对话的历史对话记录。各参考用户中可以包含上述目标用户，本说明书对此不做限定。
由于本说明书一个或多个实施例中提供的意图识别方法可以应用在智能反诈系统中识别用户的异常行为的场景中，由此，在每轮历史对话记录中，可以记录智能对话系统输出的历史输出信息，以及参考用户针对历史输出信息输入的历史回复信息。
S202：针对每个历史输出信息，从各历史回复信息中筛选出该历史输出信息对应的各指定回复信息。
具体的，针对每个历史回复信息，将该历史回复信息输入到预训练的分类模型，得到所述分类模型输出的该历史回复信息对应于各历史输出信息的概率。
在人机对话过程中，参考用户针对历史输出信息输入的历史回复信息可以有两种形式，其一为通用回复信息，指该回复信息可用于回应各种类型的输出信息，如“嗯”、“是”和“不是”。其二为专属回复信息，值该回复信息仅可用于回应对应于该回复信息的输出信息，如“我没有刷单”或“我购物了”。通常，专属回复信息可以包含专属回复信息回应的输出信息的类型或者主题，因此，可以通过将通用回复信息剔除，保留专属回复信息，并基于专属回复信息提取关键词的方式，确定专属回复信息对应的历史
输出信息的关键词。
其中，分类模型是采用下述方法训练得到的：第一步：预先针对每轮历史对话记录，以该轮历史对话记录中的历史回复信息为第二训练样本，以该轮历史对话记录中的历史输出信息为所述第二训练样本的标签。
第二步：将所述第二训练样本输入到待训练的分类模型中，得到所述分类模型输出的所述第二训练样本的预测信息。
第三步：以所述第二训练样本的预测信息与所述第二训练样本的标签之间的差异最小化为训练目标，训练所述分类模型。
由此，分类模型可以根据历史回复信息预测该历史回复信息对应于各历史输出信息的概率，通常概率最大的历史输出信息即为该历史回复信息回应的历史输出信息，然而在实际应用中，历史对话记录中存在一定数量的通用回复信息，在分类模型的训练样本中并未剔除通用的回复信息，这就导致，通用的回复信息输入到分类模型中，模型输出的通用的回复信息对应于各历史输出信息的概率通常相近。由此，进一步的，确定该历史回复信息对应于各历史输出信息的概率之间的差异，将差异高于预设差异阈值的各历史回复信息作为各候选信息。
于是，针对每个历史输出信息，根据所述各候选回复信息分别对应于该历史输出信息的概率，从所述各历史回复信息中筛选出该历史输出信息对应的各指定回复信息。
S204：根据该历史输出信息对应的各指定回复信息，确定该历史输出信息对应的至少一个关键词。
具体的，对该历史输出信息对应的各指定回复信息分别进行分词，得到各候选词，将所述各候选词分别输入预训练的意图识别模型，根据所述意图识别模型的输出从所述各候选词中筛选出该历史输出信息对应的至少一个关键词。
例如，历史输出信息为“您刚才的交易是为了做刷单任务吗？”，该历史输出信息的指定回复信息为“我没有刷单。”，则可以将“我没有刷单”分词得到“我”、“没有”和“刷单”单个候选词，进一步的，将这三个候选词分别输入预训练的意图识别模型中，“我”、“没有”这两个关键词由于并不能表征任何用户意图，因此模型针对这两个关键词输出的各意图的概率均较低，而“刷单”这一关键词可以命中“刷单”的异常行为，因此，通过对指定回复信息分词得到候选词，进而从各候选词中筛选出该历史输出信息对应的关键词。
其中，一个历史输出信息可以对应至少一个关键词，一个关键词也可对应于至少一个历史输出信息。
S206：根据各历史输出信息对应的关键词，确定历史输出信息与关键词之间的对应关系。
基于上述方案，通过从参考用户针对各历史输出信息输入的各历史回复信息中通用的历史回复信息剔除，得到各历史输出信息分别对应的指定回复信息，进而从指定回复信息中确定历史输出信息对应的知道一个关键词，从而，建立各历史输出信息与各关键词之间的对应关系。可见，采用非人工建立历史输出信息与关键词之间的对应关系的方式，避免了人工标注时可能会出现的标注不全或标注错误的问题。
在本说明书一个或多个实施例中，所述待识别信息回应的问题为：用于询问所述目标用户是否存在异常行为的问题；
查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息之前，还可以确定所述待识别信息包含否定词。
在通过智能对话系统外呼与目标用户进行人机对话确定目标用户是否存在异常行为的场景中，通常情况下，智能对话系统可以输出询问目标用户是否存在异常行为的问题，如果目标用户在多轮对话中的前几轮对话中均针对上述类型的问题给出确定的回复，而在本轮对话中给出了否定的回复，则说明目标用户的对话意图发生的改变，需要针对这种情况重新识别目标用户的意图，进而对目标用户在前几轮对话中的意图进行修正。
在本说明书一个或多个实施例中，在如图1步骤S106所示得到待识别信息的目标意图之后，还可以获取所述参考回复信息对应的所述目标用户的意图，根据所述目标意图，调整所述参考回复信息对应的所述目标用户的意图，以根据调整后的意图判断所述目标用户是否存在异常行为。具体的，可以根据意图识别模型根据拼接后的回复信息输出的目标意图对参考回复信息对应的目标用户的意图进行覆盖、调整等修正。
例如，智能对话系统的输出信息为“您刚才的交易是为了做刷单任务吗？”，目标用户输入的回复信息为“嗯”。此时，目标用户的意图可以被识别为“刷单”。智能对话系统输出“你是在购物平台上刷单的吗？”，目标用户输入的待识别信息为“我没有刷单”。此时，可将“我没有刷单”和“嗯”拼接，并输入到预训练的意图识别模型中，得到意图识别模型的输出的目标用户的意图为“没有刷单”。进而，可以根据意图为“没有刷单”修正历史对话记录中意图为“刷单”的情况，并对修正的历史对话对应的用户
的行为重新定义，以便提高判断目标用户是否存在异常行为的准确率。
图3为本说明书提供的一种意图识别装置示意图，具体包括：接收模块300，用于接收意图识别请求，确定目标用户针对智能对话系统的输出信息输入的回复信息，作为待识别信息；匹配模块302，用于将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词；查找模块304，用于根据所述目标关键词以及预设的历史输出信息与关键词之间的对应关系，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息；识别模块306，用于将将所述待识别信息与所述参考回复信息作为输入，输入到预先训练的意图识别模型，得到所述意图识别模型输出的意图，作为所述待识别信息的目标意图。
可选地，所述装置还包括：对应关系确定模块308，具体用于获取参考用户与智能对话系统多轮对话的历史对话记录，其中，每轮历史对话记录中包含所述智能对话系统输出的历史输出信息，以及所述参考用户针对所述历史输出信息输入的历史回复信息；针对每个历史输出信息，从各历史回复信息中筛选出该历史输出信息对应的各指定回复信息；根据该历史输出信息对应的各指定回复信息，确定该历史输出信息对应的至少一个关键词；根据各历史输出信息对应的关键词，确定历史输出信息与关键词之间的对应关系。
可选地，所述对应关系确定模块308，具体用于针对每个历史回复信息，将该历史回复信息输入到预训练的分类模型，得到所述分类模型输出的该历史回复信息对应于各历史输出信息的概率；确定该历史回复信息对应于各历史输出信息的概率之间的差异，将差异高于预设差异阈值的各历史回复信息作为各候选回复信息；针对每个历史输出信息，根据所述各候选回复信息分别对应于该历史输出信息的概率，从所述各历史回复信息中筛选出该历史输出信息对应的各指定回复信息。
可选地，所述装置还包括：训练模块310，具体用于预先针对每轮历史对话记录，以该轮历史对话记录中的所述参考用户输入的历史回复信息为第二训练样本，以该轮历史对话记录中所述历史回复信息回应的历史输出信息为所述第二训练样本的标签；将所述第二训练样本输入到待训练的分类模型中，得到所述分类模型输出的所述第二训练样本的预测信息；以所述第二训练样本的预测信息与所述第二训练样本的标签之间的差异最小化为训练目标，训练所述分类模型。
可选地，所述对应关系确定模块308，具体用于对该历史输出信息对应的各指定回
复信息分别进行分词，得到各候选词；将所述各候选词分别输入预训练的意图识别模型，根据所述意图识别模型的输出从所述各候选词中筛选出该历史输出信息对应的至少一个关键词。
可选地，所述待识别信息回应的所述智能对话系统的输出信息为：用于询问所述目标用户是否存在异常行为的问题；可选地，所述查找模块304在所述查找模块304查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息之前，还用于确定所述待识别信息包含否定词。
可选地，所述装置还包括：调整模块312，具体用于获取所述参考回复信息对应的所述目标用户的意图；根据所述目标意图，调整所述参考回复信息对应的所述目标用户的意图，以根据调整后的意图判断所述目标用户是否存在异常行为。
本说明书还提供了一种计算机可读存储介质，该存储介质存储有计算机程序，计算机程序可用于执行上述图1所示的意图识别方法。
本说明书还提供了图4所示的电子设备的示意结构图。如图4所述，在硬件层面，该电子设备包括处理器、内部总线、网络接口、内存以及非易失性存储器，当然还可能包括其他业务所需要的硬件。处理器从非易失性存储器中读取对应的计算机程序到内存中然后运行，以实现上述图1所示的意图识别方法。当然，除了软件实现方式之外，本说明书并不排除其他实现方式，比如逻辑器件抑或软硬件结合的方式等等，也就是说以下处理流程的执行主体并不限定于各个逻辑单元，也可以是硬件或逻辑器件。
在20世纪90年代，对于一个技术的改进可以很明显地区分是硬件上的改进(例如，对二极管、晶体管、开关等电路结构的改进)还是软件上的改进(对于方法流程的改进)。然而，随着技术的发展，当今的很多方法流程的改进已经可以视为硬件电路结构的直接改进。设计人员几乎都通过将改进的方法流程编程到硬件电路中来得到相应的硬件电路结构。因此，不能说一个方法流程的改进就不能用硬件实体模块来实现。例如，可编程逻辑器件(Programmable Logic Device,PLD)(例如现场可编程门阵列(Field Programmable Gate Array，FPGA))就是这样一种集成电路，其逻辑功能由用户对器件编程来确定。由设计人员自行编程来把一个数字系统“集成”在一片PLD上，而不需要请芯片制造厂商来设计和制作专用的集成电路芯片。而且，如今，取代手工地制作集成电路芯片，这种编程也多半改用“逻辑编译器(logic compiler)”软件来实现，它与程序开发撰写时所用的软件编译器相类似，而要编译之前的原始代码也得用特定的编程语言来撰写，此称之为硬件描述语言(Hardware Description Language，HDL)，而HDL也并
非仅有一种，而是有许多种，如ABEL(Advanced Boolean Expression Language)、AHDL(Altera Hardware Description Language)、Confluence、CUPL(Cornell University Programming Language)、HDCal、JHDL(Java Hardware Description Language)、Lava、Lola、MyHDL、PALASM、RHDL(Ruby Hardware Description Language)等，目前最普遍使用的是VHDL(Very-High-Speed Integrated Circuit Hardware Description Language)与Verilog。本领域技术人员也应该清楚，只需要将方法流程用上述几种硬件描述语言稍作逻辑编程并编程到集成电路中，就可以很容易得到实现该逻辑方法流程的硬件电路。
控制器可以按任何适当的方式实现，例如，控制器可以采取例如微处理器或处理器以及存储可由该(微)处理器执行的计算机可读程序代码(例如软件或固件)的计算机可读介质、逻辑门、开关、专用集成电路(Application Specific Integrated Circuit，ASIC)、可编程逻辑控制器和嵌入微控制器的形式，控制器的例子包括但不限于以下微控制器：ARC 625D、Atmel AT91SAM、Microchip PIC18F26K20以及Silicone Labs C8051F320，存储器控制器还可以被实现为存储器的控制逻辑的一部分。本领域技术人员也知道，除了以纯计算机可读程序代码方式实现控制器以外，完全可以通过将方法步骤进行逻辑编程来使得控制器以逻辑门、开关、专用集成电路、可编程逻辑控制器和嵌入微控制器等的形式来实现相同功能。因此这种控制器可以被认为是一种硬件部件，而对其内包括的用于实现各种功能的装置也可以视为硬件部件内的结构。或者甚至，可以将用于实现各种功能的装置视为既可以是实现方法的软件模块又可以是硬件部件内的结构。
上述实施例阐明的系统、装置、模块或单元，具体可以由计算机芯片或实体实现，或者由具有某种功能的产品来实现。一种典型的实现设备为计算机。具体的，计算机例如可以为个人计算机、膝上型计算机、蜂窝电话、相机电话、智能电话、个人数字助理、媒体播放器、导航设备、电子邮件设备、游戏控制台、平板计算机、可穿戴设备或者这些设备中的任何设备的组合。
为了描述的方便，描述以上装置时以功能分为各种单元分别描述。当然，在实施本说明书时可以把各单元的功能在同一个或多个软件和/或硬件中实现。
本领域内的技术人员应明白，本申请的实施例可提供为方法、系统、或计算机程序产品。因此，本申请可采用完全硬件实施例、完全软件实施例、或结合软件和硬件方面的实施例的形式。而且，本申请可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本申请是参照根据本申请实施例的方法、设备(系统)、和计算机程序产品的流程图和/或方框图来描述的。应理解可由计算机程序指令实现流程图和/或方框图中的每一流程和/或方框、以及流程图和/或方框图中的流程和/或方框的结合。可提供这些计算机程序指令到通用计算机、专用计算机、嵌入式处理机或其他可编程数据处理设备的处理器以产生一个机器，使得通过计算机或其他可编程数据处理设备的处理器执行的指令产生用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的装置。
这些计算机程序指令也可存储在能引导计算机或其他可编程数据处理设备以特定方式工作的计算机可读存储器中，使得存储在该计算机可读存储器中的指令产生包括指令装置的制造品，该指令装置实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能。
这些计算机程序指令也可装载到计算机或其他可编程数据处理设备上，使得在计算机或其他可编程设备上执行一系列操作步骤以产生计算机实现的处理，从而在计算机或其他可编程设备上执行的指令提供用于实现在流程图一个流程或多个流程和/或方框图一个方框或多个方框中指定的功能的步骤。
在一个典型的配置中，计算设备包括一个或多个处理器(CPU)、输入/输出接口、网络接口和内存。
内存可能包括计算机可读介质中的非永久性存储器，随机存取存储器(RAM)和/或非易失性内存等形式，如只读存储器(ROM)或闪存(flash RAM)。内存是计算机可读介质的示例。
计算机可读介质包括永久性和非永久性、可移动和非可移动媒体可以由任何方法或技术来实现信息存储。信息可以是计算机可读指令、数据结构、程序的模块或其他数据。计算机的存储介质的例子包括，但不限于相变内存(PRAM)、静态随机存取存储器(SRAM)、动态随机存取存储器(DRAM)、其他类型的随机存取存储器(RAM)、只读存储器(ROM)、电可擦除可编程只读存储器(EEPROM)、快闪记忆体或其他内存技术、只读光盘只读存储器(CD-ROM)、数字多功能光盘(DVD)或其他光学存储、磁盒式磁带，磁带磁磁盘存储或其他磁性存储设备或任何其他非传输介质，可用于存储可以被计算设备访问的信息。按照本文中的界定，计算机可读介质不包括暂存电脑可读媒体(transitory media)，如调制的数据信号和载波。
还需要说明的是，术语“包括”、“包含”或者其任何其他变体意在涵盖非排他性的包含，从而使得包括一系列要素的过程、方法、商品或者设备不仅包括那些要素，而且还包括没有明确列出的其他要素，或者是还包括为这种过程、方法、商品或者设备所固有的要素。在没有更多限制的情况下，由语句“包括一个……”限定的要素，并不排除在包括所述要素的过程、方法、商品或者设备中还存在另外的相同要素。
本领域技术人员应明白，本说明书的实施例可提供为方法、系统或计算机程序产品。因此，本说明书可采用完全硬件实施例、完全软件实施例或结合软件和硬件方面的实施例的形式。而且，本说明书可采用在一个或多个其中包含有计算机可用程序代码的计算机可用存储介质(包括但不限于磁盘存储器、CD-ROM、光学存储器等)上实施的计算机程序产品的形式。
本说明书可以在由计算机执行的计算机可执行指令的一般上下文中描述，例如程序模块。一般地，程序模块包括执行特定任务或实现特定抽象数据类型的例程、程序、对象、组件、数据结构等等。也可以在分布式计算环境中实践本说明书，在这些分布式计算环境中，由通过通信网络而被连接的远程处理设备来执行任务。在分布式计算环境中，程序模块可以位于包括存储设备在内的本地和远程计算机存储介质中。
本说明书中的各个实施例均采用递进的方式描述，各个实施例之间相同相似的部分互相参见即可，每个实施例重点说明的都是与其他实施例的不同之处。尤其，对于系统实施例而言，由于其基本相似于方法实施例，所以描述的比较简单，相关之处参见方法实施例的部分说明即可。
以上所述仅为本说明书的实施例而已，并不用于限制本说明书。对于本领域技术人员来说，本说明书可以有各种更改和变化。凡在本说明书的精神和原理之内所作的任何修改、等同替换、改进等，均应包含在本说明书的权利要求范围之内。
Claims (16)
- 一种意图识别方法，所述方法包括：接收意图识别请求，确定目标用户针对智能对话系统的输出信息输入的回复信息，作为待识别信息；将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词；根据所述目标关键词以及预设的历史输出信息与关键词之间的对应关系，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息；将所述待识别信息与所述参考回复信息作为输入，输入到预先训练的意图识别模型，得到所述意图识别模型输出的意图，作为所述待识别信息的目标意图；其中，采用下述方法训练所述意图识别模型：根据智能对话系统与参考用户的指定轮历史对话中，所述智能对话系统输出的历史输出信息以及所述参考用户针对所述历史输出信息输入的历史回复信息，确定第一训练样本，根据所述指定轮历史对话对应的所述参考用户的意图确定所述第一训练样本的标签，根据所述第一训练样本和所述第一训练样本的标签训练所述意图识别模型。
- 如权利要求1所述的方法，采用下述方法确定历史输出信息与关键词之间的对应关系：获取参考用户与智能对话系统多轮对话的历史对话记录，其中，每轮历史对话记录中包含所述智能对话系统输出的历史输出信息，以及所述参考用户针对所述历史输出信息输入的历史回复信息；针对每个历史输出信息，从各历史回复信息中筛选出该历史输出信息对应的各指定回复信息；根据该历史输出信息对应的各指定回复信息，确定该历史输出信息对应的至少一个关键词；根据各历史输出信息对应的关键词，确定历史输出信息与关键词之间的对应关系。
- 如权利要求2所述的方法，针对每个历史输出信息，从各历史回复信息中筛选出该历史输出信息对应的各指定回复信息，包括：针对每个历史回复信息，将该历史回复信息输入到预训练的分类模型，得到所述分类模型输出的该历史回复信息对应于各历史输出信息的概率；确定该历史回复信息对应于各历史输出信息的概率之间的差异，将差异高于预设差 异阈值的各历史回复信息作为各候选回复信息；针对每个历史输出信息，根据所述各候选回复信息分别对应于该历史输出信息的概率，从所述各历史回复信息中筛选出该历史输出信息对应的各指定回复信息。
- 如权利要求3所述的方法，预先训练分类模型，包括：预先针对每轮历史对话记录，以该轮历史对话记录中的所述参考用户输入的历史回复信息为第二训练样本，以该轮历史对话记录中所述历史回复信息回应的历史输出信息为所述第二训练样本的标签；将所述第二训练样本输入到待训练的分类模型中，得到所述分类模型输出的所述第二训练样本的预测信息；以所述第二训练样本的预测信息与所述第二训练样本的标签之间的差异最小化为训练目标，训练所述分类模型。
- 如权利要求2所述的方法，根据该历史输出信息对应的各指定回复信息，确定该历史输出信息对应的至少一个关键词，包括：对该历史输出信息对应的各指定回复信息分别进行分词，得到各候选词；将所述各候选词分别输入预训练的意图识别模型，根据所述意图识别模型的输出从所述各候选词中筛选出该历史输出信息对应的至少一个关键词。
- 如权利要求1所述的方法，所述待识别信息回应的所述智能对话系统的输出信息为：用于询问所述目标用户是否存在异常行为的问题；查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息之前，所述方法还包括：确定所述待识别信息包含否定词。
- 如权利要求1所述的方法，所述方法还包括：获取所述参考回复信息对应的所述目标用户的意图；根据所述目标意图，调整所述参考回复信息对应的所述目标用户的意图，以根据调整后的意图判断所述目标用户是否存在异常行为。
- 一种意图识别装置，包括：接收模块，用于接收意图识别请求，确定目标用户针对智能对话系统的输出信息输入的回复信息，作为待识别信息；匹配模块，用于将所述待识别信息与预设的各关键词进行匹配，得到所述待识别信息匹配的目标关键词；查找模块，用于根据所述目标关键词以及预设的历史输出信息与关键词之间的对应 关系，从所述目标用户的历史对话记录中，查找与所述目标关键词对应的参考输出信息，以及所述目标用户针对所述参考输出信息输入的参考回复信息；识别模块，用于将所述待识别信息与所述参考回复信息作为输入，输入到预先训练的意图识别模型，得到所述意图识别模型输出的意图，作为所述待识别信息的目标意图。
- 如权利要求8所述的装置，所述装置还包括：对应关系确定模块，用于获取参考用户与智能对话系统多轮对话的历史对话记录，其中，每轮历史对话记录中包含所述智能对话系统输出的历史输出信息，以及所述参考用户针对所述历史输出信息输入的历史回复信息；针对每个历史输出信息，从各历史回复信息中筛选出该历史输出信息对应的各指定回复信息；根据该历史输出信息对应的各指定回复信息，确定该历史输出信息对应的至少一个关键词；根据各历史输出信息对应的关键词，确定历史输出信息与关键词之间的对应关系。
- 如权利要求9所述的装置，所述对应关系确定模块，用于针对每个历史回复信息，将该历史回复信息输入到预训练的分类模型，得到所述分类模型输出的该历史回复信息对应于各历史输出信息的概率；确定该历史回复信息对应于各历史输出信息的概率之间的差异，将差异高于预设差异阈值的各历史回复信息作为各候选回复信息；针对每个历史输出信息，根据所述各候选回复信息分别对应于该历史输出信息的概率，从所述各历史回复信息中筛选出该历史输出信息对应的各指定回复信息。
- 如权利要求10所述的装置，所述装置还包括：训练模块，用于预先针对每轮历史对话记录，以该轮历史对话记录中的所述参考用户输入的历史回复信息为第二训练样本，以该轮历史对话记录中所述历史回复信息回应的历史输出信息为所述第二训练样本的标签；将所述第二训练样本输入到待训练的分类模型中，得到所述分类模型输出的所述第二训练样本的预测信息；以所述第二训练样本的预测信息与所述第二训练样本的标签之间的差异最小化为训练目标，训练所述分类模型。
- 如权利要求9所述的装置，所述对应关系确定模块，用于对该历史输出信息对应的各指定回复信息分别进行分词，得到各候选词；将所述各候选词分别输入预训练的意图识别模型，根据所述意图识别模型的输出从所述各候选词中筛选出该历史输出信息对应的至少一个关键词。
- 如权利要求8所述的装置，所述待识别信息回应的所述智能对话系统的输出信息为：用于询问所述目标用户是否存在异常行为的问题；所述查找模块在所述查找模块查找与所述目标关键词对应的参考输出信息，以及所 述目标用户针对所述参考输出信息输入的参考回复信息之前，还用于确定所述待识别信息包含否定词。
- 如权利要求8所述的装置，所述装置还包括：调整模块，用于获取所述参考回复信息对应的所述目标用户的意图；根据所述目标意图，调整所述参考回复信息对应的所述目标用户的意图，以根据调整后的意图判断所述目标用户是否存在异常行为。
- 一种计算机可读存储介质，所述存储介质存储有计算机程序，所述计算机程序被处理器执行时实现权利要求1～7任一项所述的方法。
- 一种电子设备，包括存储器、处理器及存储在存储器上并可在处理器上运行的计算机程序，所述处理器执行所述程序时实现权利要求1～7任一项所述的方法。
Applications Claiming Priority (2)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211370268.5A CN115759122A (zh) | 2022-11-03 | 2022-11-03 | 一种意图识别方法、装置、设备及可读存储介质 |
| CN202211370268.5 | 2022-11-03 |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| WO2024093403A1 true WO2024093403A1 (zh) | 2024-05-10 |
Family
ID=85357603
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| PCT/CN2023/110632 WO2024093403A1 (zh) | 2022-11-03 | 2023-08-01 | 一种意图识别方法、装置、设备及可读存储介质 |
Country Status (2)
| Country | Link |
|---|---|
| CN (1) | CN115759122A (zh) |
| WO (1) | WO2024093403A1 (zh) |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118312244A (zh) * | 2024-06-07 | 2024-07-09 | 安徽高灯微行科技有限公司 | 业务数据处理方法、装置、计算机设备和存储介质 |
| CN118468045A (zh) * | 2024-07-12 | 2024-08-09 | 之江实验室 | 一种模型训练加速方法、装置及存储介质 |
| CN118503395A (zh) * | 2024-07-17 | 2024-08-16 | 浪潮通用软件有限公司 | 一种多轮业务数据查询方法、设备及介质 |
| CN118568244A (zh) * | 2024-08-01 | 2024-08-30 | 之江实验室 | 一种基于智能问答的业务执行方法、装置以及电子设备 |
Families Citing this family (3)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN115759122A (zh) * | 2022-11-03 | 2023-03-07 | 支付宝(杭州)信息技术有限公司 | 一种意图识别方法、装置、设备及可读存储介质 |
| CN116503879B (zh) * | 2023-05-22 | 2024-01-19 | 广东骏思信息科技有限公司 | 应用于电商平台的威胁行为识别方法及装置 |
| CN117708337B (zh) * | 2024-02-05 | 2024-04-26 | 杭州杰竞科技有限公司 | 一种面向复杂定域的人机交互方法和系统 |
Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN111428483A (zh) * | 2020-03-31 | 2020-07-17 | 华为技术有限公司 | 语音交互方法、装置和终端设备 |
| CN112613308A (zh) * | 2020-12-17 | 2021-04-06 | 中国平安人寿保险股份有限公司 | 用户意图识别方法、装置、终端设备及存储介质 |
| US20210150385A1 (en) * | 2019-11-20 | 2021-05-20 | International Business Machines Corporation | Automatically generated conversation output |
| CN113158062A (zh) * | 2021-05-08 | 2021-07-23 | 清华大学深圳国际研究生院 | 一种基于异构图神经网络的用户意图识别方法及装置 |
| CN114706945A (zh) * | 2022-03-24 | 2022-07-05 | 马上消费金融股份有限公司 | 意图识别方法、装置、电子设备及存储介质 |
| CN114880472A (zh) * | 2022-04-28 | 2022-08-09 | 支付宝(杭州)信息技术有限公司 | 数据处理方法、装置及设备 |
| CN114970560A (zh) * | 2022-05-19 | 2022-08-30 | 深圳市优必选科技股份有限公司 | 对话意图识别方法、装置、存储介质和智能设备 |
| CN115033676A (zh) * | 2022-06-22 | 2022-09-09 | 支付宝(杭州)信息技术有限公司 | 一种意图识别模型训练及用户意图识别方法及装置 |
| CN115186071A (zh) * | 2021-04-06 | 2022-10-14 | 中移(上海)信息通信科技有限公司 | 意图识别方法、装置、电子设备及可读存储介质 |
| CN115203394A (zh) * | 2022-07-20 | 2022-10-18 | 北京三快在线科技有限公司 | 一种模型训练的方法、业务执行的方法及装置 |
| CN115759122A (zh) * | 2022-11-03 | 2023-03-07 | 支付宝(杭州)信息技术有限公司 | 一种意图识别方法、装置、设备及可读存储介质 |
-
2022
- 2022-11-03 CN CN202211370268.5A patent/CN115759122A/zh active Pending
-
2023
- 2023-08-01 WO PCT/CN2023/110632 patent/WO2024093403A1/zh unknown
Patent Citations (11)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| US20210150385A1 (en) * | 2019-11-20 | 2021-05-20 | International Business Machines Corporation | Automatically generated conversation output |
| CN111428483A (zh) * | 2020-03-31 | 2020-07-17 | 华为技术有限公司 | 语音交互方法、装置和终端设备 |
| CN112613308A (zh) * | 2020-12-17 | 2021-04-06 | 中国平安人寿保险股份有限公司 | 用户意图识别方法、装置、终端设备及存储介质 |
| CN115186071A (zh) * | 2021-04-06 | 2022-10-14 | 中移(上海)信息通信科技有限公司 | 意图识别方法、装置、电子设备及可读存储介质 |
| CN113158062A (zh) * | 2021-05-08 | 2021-07-23 | 清华大学深圳国际研究生院 | 一种基于异构图神经网络的用户意图识别方法及装置 |
| CN114706945A (zh) * | 2022-03-24 | 2022-07-05 | 马上消费金融股份有限公司 | 意图识别方法、装置、电子设备及存储介质 |
| CN114880472A (zh) * | 2022-04-28 | 2022-08-09 | 支付宝(杭州)信息技术有限公司 | 数据处理方法、装置及设备 |
| CN114970560A (zh) * | 2022-05-19 | 2022-08-30 | 深圳市优必选科技股份有限公司 | 对话意图识别方法、装置、存储介质和智能设备 |
| CN115033676A (zh) * | 2022-06-22 | 2022-09-09 | 支付宝(杭州)信息技术有限公司 | 一种意图识别模型训练及用户意图识别方法及装置 |
| CN115203394A (zh) * | 2022-07-20 | 2022-10-18 | 北京三快在线科技有限公司 | 一种模型训练的方法、业务执行的方法及装置 |
| CN115759122A (zh) * | 2022-11-03 | 2023-03-07 | 支付宝(杭州)信息技术有限公司 | 一种意图识别方法、装置、设备及可读存储介质 |
Cited By (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN118312244A (zh) * | 2024-06-07 | 2024-07-09 | 安徽高灯微行科技有限公司 | 业务数据处理方法、装置、计算机设备和存储介质 |
| CN118468045A (zh) * | 2024-07-12 | 2024-08-09 | 之江实验室 | 一种模型训练加速方法、装置及存储介质 |
| CN118503395A (zh) * | 2024-07-17 | 2024-08-16 | 浪潮通用软件有限公司 | 一种多轮业务数据查询方法、设备及介质 |
| CN118568244A (zh) * | 2024-08-01 | 2024-08-30 | 之江实验室 | 一种基于智能问答的业务执行方法、装置以及电子设备 |
Also Published As
| Publication number | Publication date |
|---|---|
| CN115759122A (zh) | 2023-03-07 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| WO2024093403A1 (zh) | 一种意图识别方法、装置、设备及可读存储介质 | |
| CN111984779B (zh) | 一种对话文本分析方法、装置、设备和可读介质 | |
| CN107437416A (zh) | 一种基于语音识别的咨询业务处理方法及装置 | |
| WO2023246393A1 (zh) | 意图识别模型训练及用户意图识别 | |
| US11908477B2 (en) | Automatic extraction of conversation highlights | |
| CN112735374B (zh) | 一种自动语音交互的方法及装置 | |
| US20240177510A1 (en) | Model training method and apparatus, service processing method and apparatus, storage medium, and device | |
| CN109299276B (zh) | 一种将文本转化为词嵌入、文本分类方法和装置 | |
| CN113779208A (zh) | 用于人机对话的方法和装置 | |
| CN111488737B (zh) | 文本识别方法、装置及设备 | |
| CN112908315B (zh) | 一种基于声音特征和语音识别的问答意图判断方法 | |
| CN117648437A (zh) | 银行智能业务引导方法及系统 | |
| CN117688155A (zh) | 业务问题的回复方法、装置、存储介质以及电子设备 | |
| CN112256836A (zh) | 记录数据的处理方法、装置和服务器 | |
| WO2023000954A1 (zh) | 票据处理 | |
| CN114970559B (zh) | 智能应答方法及装置 | |
| CN112163078B (zh) | 智能应答方法、装置、服务器及存储介质 | |
| CN114722821A (zh) | 一种文本匹配方法、装置、存储介质及电子设备 | |
| CN110276001B (zh) | 盘点页识别方法、装置、计算设备和介质 | |
| CN109902170B (zh) | 一种文本的分类方法、装置及电子设备 | |
| CN115081006A (zh) | 一种敏感数据的处理方法、装置及设备 | |
| CN115599891B (zh) | 一种确定异常对话数据方法、装置、设备及可读存储介质 | |
| CN108804603B (zh) | 人机书面对话方法及系统、服务器、介质 | |
| US11520839B2 (en) | User based network document modification | |
| CN117035695B (zh) | 一种信息预警的方法、装置、可读存储介质以及电子设备 |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| 121 | Ep: the epo has been informed by wipo that ep was designated in this application |
Ref document number: 23884342 Country of ref document: EP Kind code of ref document: A1 |