Cross-domain multi-round multi-intention recognition method, electronic device and storage medium

Abstract

Embodiments of the present application provide a cross-domain multi-round multi-intention recognition method, system, electronic device and storage medium. The method comprises: in a multi-round dialogue, real-time activity voice detection is performed, when dialogue sentence input is detected, an intention recognition module is used to perform intention analysis on the dialogue sentence input in the current dialogue round; it is judged whether the intention of the dialogue sentence in the current dialogue round is in the intention expectation space of the completed dialogue before the current dialogue round, if yes, cross-domain semantic analysis is performed on the dialogue sentence; in the semantic analysis, it is judged whether the dialogue sentence is complete, if not, the dialogue sentence is supplemented to generate a semantic complete sentence; a response sentence is generated for the semantic complete sentence, and is broadcast. Embodiments of the present application introduce the method of intention expectation space to judge whether recognition is needed and whether the interruption needs to be broadcast, break the limitation of the voice recognition model judged from the voice level, and avoid the false interruption caused by noise.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of intelligent voice, in particular to a cross-domain multi-round multi-intent recognition method and system, electronic device and storage medium. BACKGROUND

[0002] With the development of intelligent voice technology, users can have multi-round conversations with voice intelligent assistants to complete more instruction functions. In the process of intelligent assistant broadcasting, the user may interrupt the broadcast and start a new conversation. Meanwhile, in multi-round conversations, the user may input abbreviated or referential inquiry statements, resulting in incomplete information and the intelligent assistant being unable to effectively respond to the user's inquiry statement. These make the system need to listen for new intents to interrupt the current broadcast while broadcasting in multi-round conversations, and rewrite incomplete statements into a complete statement with complete semantic information and independent understanding of the context, to ensure the coherence and effectiveness of the conversation.

[0003] Multi-round conversation products in the prior art usually use VAD (Voice Activity Detection) technology to determine whether the user is speaking or has finished speaking by judging the strength or pause time of the sound signal to perform interruption recognition. For incomplete information statement rewriting, the prior art usually saves the context information of the skill, uses the context to complete the user's input when the next round of conversation hits the same skill, or encodes the historical single-round conversation and the incomplete statement as a sequence to generate a sequence label.

[0004] The prior art can be specifically manifested as follows: current multi-round conversation intent recognition, the voice assistant determines whether there is sound from the voice level during the broadcast process through VAD technology, and if there is sound, it determines whether the user is speaking to it, and identifies the user's intent. After intent recognition, the user's statement is analyzed and a response is made, and when the user's statement is incomplete, the user's input statement is completed based on the label in the historical single-round conversation or the context information saved by the skill.

[0005] In the process of implementing the present application, the inventors have found that the related art has at least the following problems:

[0006] Through VAD recognition, it is easy to incorrectly recognize the user's interruption due to background noise or noise, which in turn causes the intelligent assistant to stop broadcasting. In addition, since it is voice activity detection, it is also easy for the machine to think that the user has finished speaking due to a short pause during the user's speech.

[0007] The multi-round dialogue within a single skill only saves the context information of the single skill, and the context information can be used only when the next round of dialogue also hits the same skill, so if the next round of dialogue hits other skills, the context information cannot be used to complete the user input speech due to the different data structures defined by each skill.

[0008] Based on the historical single-round dialogue label, the dialogue data needs to be encoded into a sequence from the first sentence, and a label corresponding to the sequence needs to be generated, and then the sequence that needs to be rewritten is searched according to the label, which is easy to cause a long query process and slow response speed. SUMMARY

[0009] In order to at least solve the problem of multi-round multi-intention recognition across domains in the prior art.

[0010] In a first aspect, an embodiment of the present application provides a multi-round multi-intention recognition method across domains, comprising:

[0011] In the multi-round dialogue, real-time activity voice detection is performed, and when it is detected that there is a dialogue sentence input, an intention analysis module is used to analyze the intention of the dialogue sentence input in the current dialogue round;

[0012] It is judged whether the intention of the dialogue sentence of the current dialogue round is in the intention expectation space of the completed dialogue before the current dialogue round, and if so, the dialogue sentence is subjected to semantic analysis across domains, wherein the intention expectation space is constructed by a first-level intention of the completed dialogue in the multi-round dialogue and a plurality of second-level intentions associated with the first-level intention, and is used to judge whether the dialogue sentence of the current dialogue round needs to be recognized;

[0013] In the semantic analysis, it is judged whether the dialogue sentence is complete, and if not, the dialogue sentence is supplemented to generate a semantically complete sentence;

[0014] A response sentence is generated for the semantically complete sentence, and is broadcasted.

[0015] In a second aspect, an embodiment of the present application provides a multi-round multi-intention recognition system across domains, comprising:

[0016] An intention analysis module is used to perform real-time activity voice detection in a multi-round dialogue, and when it is detected that there is a dialogue sentence input, an intention analysis module is used to analyze the intention of the dialogue sentence input in the current dialogue round;

[0017] The semantic analysis module is configured to determine whether an intent of a dialogue sentence of the current dialogue turn is in an intent expectation space of completed dialogues before the current dialogue turn, and if so, perform cross-domain semantic analysis on the dialogue sentence, wherein the intent expectation space is constructed by a first-level intent of completed dialogues in multi-turn dialogues and a plurality of second-level intents associated with the first-level intent by default, and is configured to determine whether the dialogue sentence of the current dialogue turn needs to be recognized.

[0018] The supplement module is configured to determine whether the dialogue sentence is complete in the semantic analysis, and if not, supplement the dialogue sentence to generate a semantically complete sentence.

[0019] The broadcast module is configured to generate a response sentence for the semantically complete sentence and broadcast the response sentence.

[0020] In a third aspect, an electronic device is provided, which includes at least one processor and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the cross-domain multi-turn multi-intent recognition method of any of the embodiments of the present application.

[0021] In a fourth aspect, an embodiment of the present application provides a storage medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the steps of the cross-domain multi-turn multi-intent recognition method of any of the embodiments of the present application.

[0022] In a fifth aspect, an embodiment of the present application provides a computer program product including computer programs / instructions, wherein the computer programs / instructions, when executed by a processor, implement the steps of the cross-domain multi-turn multi-intent recognition method of any of the embodiments of the present application.

[0023] The embodiments of the present application have the following beneficial effects: The present application considers the problems of broadcast interruption and incomplete sentences in dialogue interaction, introduces the method of intent expectation space to determine whether recognition is needed and whether broadcast interruption is needed, breaks the limitation of the "speech recognition" model determined from the speech level, and avoids the noise-induced false interruption. The pre-trained rewriting model uses the intent expectation space to supplement the incomplete semantic sentences to generate semantically complete sentences, and improves the accuracy of the response. BRIEF DESCRIPTION OF DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings needed in the embodiments or prior art description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can also be obtained by those skilled in the art without creative effort.

[0025] Figure 1 is a flow chart of a cross-domain multi-round multi-intention recognition method provided by an embodiment of the present application;

[0026] Figure 2 is a specific flow chart of a cross-domain multi-round multi-intention recognition method provided by an embodiment of the present application;

[0027] Figure 3 is a structural schematic diagram of a cross-domain multi-round multi-intention recognition system provided by an embodiment of the present application;

[0028] Figure 4 is a structural schematic diagram of an electronic device for cross-domain multi-round multi-intention recognition provided by an embodiment of the present application. DETAILED DESCRIPTION

[0029] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present application.

[0030] As shown in Figure 1 is a flow chart of a cross-domain multi-round multi-intention recognition method provided by an embodiment of the present application, including the following steps:

[0031] S11: In a multi-round dialogue, real-time activity voice detection is performed, when it is detected that there is a dialogue sentence input, an intention recognition module is used to perform intention analysis on the dialogue sentence input in the current dialogue round;

[0032] S12: It is judged whether the intention of the dialogue sentence in the current dialogue round is in the intention expectation space of the completed dialogue before the current dialogue round, if yes, cross-domain semantic analysis is performed on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in the multi-round dialogue and a plurality of second-level intentions preset to be associated with the first-level intention, and is used to judge whether the dialogue sentence in the current dialogue round needs to be recognized;

[0033] S13: In the semantic analysis, it is judged whether the dialogue sentence is complete, if not, the dialogue sentence is supplemented to generate a semantic complete sentence;

[0034] S14: A response sentence is generated for the semantic complete sentence, and is broadcast.

[0035] In a conventional multi-round dialogue, the user's broadcast interruption and incomplete sentence in the dialogue interaction are considered. For example, in half-duplex and full-duplex dialogue, when the intelligent voice device feeds back to the user, the user inputs sound during the feedback of the intelligent device, which usually interrupts the next round of dialogue.

[0036] For example:

[0037] User: Navigate to McDonald's;

[0038] Intelligent voice device: I have found 3 nearby McDonald's, which are: Xx Street Store, College Store, Train…

[0039] User (interrupted): The second one;

[0040] Intelligent voice device: Start guiding you…

[0041] User (the user said to the passenger in the co-pilot, not to the intelligent voice device): Which one did he say this time?

[0042] Intelligent voice device (misrecognized and interrupted its own dialogue): Sorry, I didn't understand what you said.

[0043] That is, the prior art has the following problems: in order to make the communication between the intelligent voice device and the user more intelligent, the user's voice is often recognized in real time, and the user's voice is quickly recognized to ensure the real-time nature of the user interaction. However, this also has new problems. Ideally, the user and the intelligent device have a 1:1 dialogue, however, in real scenarios, for example, multiple people watch an intelligent TV together, and multiple people are present in a vehicle, it is possible that the user interacts with others during the user's interaction with the intelligent voice recognition. Therefore, in multi-round dialogue interaction, the intelligent voice device also (the user did not say to the intelligent voice recognition) recognizes, interrupts and responds to the dialogue between the user and others.

[0044] Therefore, in order to solve the above problems, the main improvement of the present application is: 1. How to ensure that the intelligent voice device is not interrupted by noise or the dialogue between the user and others in multi-round dialogue interaction; 2. If it is interrupted, how to determine the interruption logic.

[0045] For step S11, the dialogue system is started, the user performs voice interaction with the intelligent voice device, and in the multi-round voice interaction process, the intelligent voice device performs real-time VAD (Voice activity detection) detection, and in the multi-path dialogue, it is detected whether there is a new dialogue sentence input. When it is detected that there is a dialogue sentence input, the intention analysis module is used to analyze the intention of the dialogue sentence input in the current dialogue round;

[0046] For step S12, the method pre-constructs an intention expectation space. First, the intention determined by the completed dialogue round is set as a first-level intention, and the associated intention is preset according to the first-level intention and is set as a second-level intention. The first-level intention and the second-level intention jointly constitute the intention expectation space.

[0047] For example, the first-level intention is navigation, and the associated second-level intention can include weather, group purchase, ticket booking, etc. The above is an example, which indicates that the first-level intention can be associated with multiple second-level intentions, which is not limited herein. For example, in the first dialogue round, the user's intention is navigation. The second dialogue round can be group purchase. It should be noted that the application sets all the intentions determined by the completed dialogue round as the first-level intention, rather than the intention of the last completed dialogue round. Therefore, in the third dialogue round, the first-level intention includes "navigation, group purchase". The second-level intention includes the intention associated with navigation and group purchase. With the continuous dialogue of the user, the first-level intention can be continuously expanded.

[0048] The design concept is that in the multi-round dialogue of the user, there will indeed be cross-domain situations, but the cross-domain still has a certain logical range. For example, the user asks "Introduce some delicious food in Suzhou", and the intelligent voice device replies "Suzhou fresh meat moon cake, green dumplings, Ozo noodles, steamed buns, lotus root with glutinous rice, and steamed ribs…". At this time, for example, this part of interaction is in the food field. At this time, the user inputs "How long does it take to go from Wuhan" (at this time, from the food field to the navigation field), and the intelligent voice device replies "It takes 10 hours to drive there". The user continues to ask "What's tomorrow like over there" (from the navigation field to the weather field). That is, although the user's multi-round dialogue will continuously cross the field, it is still following the logic of the dialogue. Even if it crosses the field, the field it crosses is also related. Therefore, the intention expectation space of the application is to construct the related cross-domain.

[0049] For example, the intent of the dialogue sentence of the current dialogue turn of "How long does it take to go from Wuhan" is "navigation", and the completed dialogue before is "Introduce some food in Suzhou", assuming that the intent expectation space of the completed dialogue is "first level intent: food; second level intent: navigation, group purchase, takeout". The intent of the dialogue sentence of the current dialogue turn exists in the second level intent of the intent expectation space of the completed dialogue. If it exists, it means that the user's sentence is related, rather than irrelevant conversation (for example, the user turns his head and talks to others, or noise). By this way, it is determined whether the dialogue sentence of the current dialogue turn needs to be recognized. If it exists, it means that it needs to be recognized in the multi-turn dialogue. If it does not exist, it does not need to be recognized in the multi-turn dialogue. That is, when human voice is detected, it is determined whether it has an intent. If the new dialogue sentence has an intent, it is determined whether the intent of the user's sentence belongs to the intent expectation space range. If it belongs to the intent expectation space range, the semantic analysis module is input to analyze the semantic of the current turn input by the user.

[0050] For step S13, in the multi-turn dialogue, the user often omits part of the vocabulary in the historical dialogue sentence in the current turn to be answered, or replaces the part of the vocabulary with a representative vocabulary, so that the sentence is incomplete. In order to solve this problem, the present application needs to supplement the dialogue sentence to generate a semantic complete sentence. Specifically, the supplement is processed by a rewriting model pre-trained by a historical dialogue sentence, including:

[0051] Before cross-domain semantic analysis, the content in the intent expectation space is used as the rewriting direction of the rewriting model, and the rewriting model is used for supplement.

[0052] In the embodiment, for example:

[0053] User: "How far is Suzhou from Wuhan"

[0054] Intelligent voice device: "About 743 kilometers"

[0055] The current turn sentence to be answered is:

[0056] User: "What are the foods there" - using the rewriting model to rewrite: "What are the foods in Suzhou".

[0057] Further, the rewriting model has a large divergence space. In order to make the supplemented content more consistent with the user's own intent, the content in the intent expectation space is used as the rewriting direction of the rewriting model, so as to avoid generating content that still cannot explain the specific intent or the wrong intent, which is difficult to truly understand the user's demand and give wrong feedback.

[0058] For step S14, a response sentence is generated for the semantic complete sentence and is broadcast. In the broadcasting, there is also a problem of interruption. Specifically, the completed dialogue includes a neighboring dialogue within a preset dialogue turn and a historical dialogue outside the preset dialogue turn.

[0059] The broadcasting of the response sentence generated for the semantic complete sentence includes:

[0060] After the response sentence is generated, it is determined whether the broadcasting of the completed dialogue is currently being performed. If the broadcasting is already being performed, it is determined whether the response sentence can interrupt the broadcasting of the completed dialogue based on the intent expectation space:

[0061] If, in the intent expectation space, the intent of the dialogue sentence of the current dialogue turn as a second-level intent is associated with the intent of the neighboring dialogue as a first-level intent, the broadcasting of the completed dialogue is interrupted, and the response sentence is played,

[0062] If, in the intent expectation space, the intent of the dialogue sentence of the current dialogue turn as a second-level intent is associated with the intent of the historical dialogue as a first-level intent, the response sentence is played after the broadcasting of the completed dialogue is completed.

[0063] If the response sentence is a preset forced broadcasting, the broadcasting of the completed dialogue is interrupted, and the response sentence is played.

[0064] In this embodiment, considering the multi-turn dialogue, if the response sentence of the previous turn of interaction is still being broadcast, the broadcasting needs to be interrupted if it is still being broadcast. In a simple manner, after the complete semantic sentence is obtained, it is determined whether the current broadcasting is being performed and is forced broadcasting. If so, the broadcasting is waited to be completed, and the to-be-responded sentence is generated. Otherwise, a break instruction is directly initiated.

[0065] This manner needs to set the "forced" setting in advance for various broadcastings. If too many settings are set, any sentence will be interrupted. If the settings are not accurate, the effect of the user experience of the interruption is not necessarily good.

[0066] Therefore, the interrupting of the present application is related to the intention expectation space. Therefore, the completed dialogue is divided into: adjacent dialogue within a preset dialogue turn, and historical dialogue outside the preset dialogue turn. For example, a multi-turn dialogue has 7 turns, and the dialogue within 3 turns is adjacent dialogue, and the outside is historical dialogue. It is also considered that if the intention of the dialogue sentence of the current turn can appear in both adjacent dialogue and historical dialogue, it is processed as adjacent dialogue. The priority of adjacent dialogue is higher than that of historical dialogue. The adjacent dialogue within the preset dialogue turn indicates that the user and the intelligent voice device are communicating on the hot topic content, at which time they can interrupt each other, can be closer to the full-duplex dialogue, and the voice interaction is more flexible and fast. The historical dialogue outside the preset dialogue turn indicates that the intention topic can be temporarily inserted by the user, and the immediacy of the dialogue is relatively weak, so it is more suitable for half-duplex interactive dialogue, and the user interaction experience is improved. That is, if the user sentence information is incomplete, the sentence is rewritten through the rewriting module. The system implements response to the sentence, and if the current has a broadcast and the broadcast content is mandatory broadcast content, the response is entered after the broadcast is completed, otherwise the broadcast is directly interrupted to respond. Therefore, the intention expectation space of the present application is used to judge whether the dialogue sentence of the current dialogue turn needs to be recognized, whether the response sentence can interrupt the broadcast of the completed dialogue, and also used in the rewriting model for supplement. The overall process is shown in Figure 2

[0067] As can be seen from the embodiment, the present application considers the problems of broadcast interruption and incomplete sentences in dialogue interaction, introduces the method of intention expectation space to judge whether recognition is needed and whether broadcast interruption is needed, breaks the limitation of the "speech recognition" model judged from the speech level, and avoids the noise from causing false interruption. And through the pre-trained rewriting model, the intention expectation space is used to supplement the incomplete semantic sentences, generate complete semantic sentences, and improve the accuracy of response.

[0068] Simply put, when the user is in a multi-turn dialogue with an intelligent dialogue system, sometimes the user will choose to interrupt the current dialogue and start a new dialogue when the system is broadcasting. And some of the words in the historical dialogue sentence are omitted in the current turn to be responded, or the part of the words is replaced by a referential word, for example:

[0069] User: "Introduce some delicious food in Suzhou"

[0070] Intelligent voice system: "Suzhou fresh meat moon cake, green dumplings, Ozo noodles, steamed buns, lotus root with glutinous rice, steamed pork ribs……"

[0071] User (interrupting): "How long does it take to go from Wuhan"

interrupted recognition, skill switching, incomplete sentence--omission

[0072] ​Smart voice system: "It will take 10 hours to drive there"

[0073] User: "How about tomorrow over there" [Incomplete sentence -- reference]

[0074] Step 1: The dialogue system dialogue starts, and VAD detection is performed to detect whether there is a new dialogue sentence.

[0075] Step 2: The detected new dialogue is analyzed by the intent recognition module to determine whether it has an intent. If the new dialogue sentence has an intent, it is determined whether the intent is within the expected intent space; after determining the expected intent space, the to-be-answered dialogue sentence is input into the semantic analysis module.

[0076] Step 3: If the to-be-answered dialogue sentence is incomplete, the semantic rewriting module is used to rewrite the semantics, the pre-trained rewriting model is used to determine the rewriting type (omission, reference), and the complete semantic information is obtained by rewriting the incomplete semantics according to the rewriting type in combination with at least one historical to-be-answered sentence and the corresponding system response sentence. After obtaining the complete semantic sentence, it is determined whether there is a broadcast at present and whether it is a mandatory broadcast, if so, wait for the broadcast to be completed and generate the to-be-answered sentence, otherwise directly initiate a break instruction.

[0077] Step 4: A break instruction is initiated to output a response sentence. And return to step 1.

[0078] The intent expectation space includes primary intent and secondary intent, which is used to determine whether the detected new dialogue sentence is noise, thereby avoiding the false interruption of human-computer dialogue by noise. The intent expectation space is constructed based on the primary intent (the dialogue intent of the user determined by the completed dialogue round) and the secondary intent (the associated intent preset based on the primary intent). If the new dialogue sentence intent conforms to the primary intent, it indicates that the new dialogue sentence is related to the completed dialogue, so it can be determined that the detected new dialogue sentence is not noise, and the new dialogue sentence is responded to, thereby avoiding the false interruption of human-computer dialogue by noise. If the new dialogue intent conforms to the secondary intent, the new dialogue sentence can also be determined as non-noise, thereby avoiding the misrecognition of the dialogue sentence of the target user as noise, thereby avoiding the user's interruption demand being ignored when the new dialogue intent corresponding to the dialogue sentence of the target user does not completely match the current dialogue intent.

[0079] The rewriting model is trained based on the historical dialogue sentences of the dialogue system. The model rewrites the omission or reference of the user's input inquiry sentence based on the rewriting type in combination with at least one round of historical dialogue and the to-be-answered sentence. The rewritten sentence has complete semantics, and the intelligent dialogue system can effectively respond to the user's inquiry sentence.

[0080] As Figure 3 Fig. 1 shows a structural schematic diagram of a cross-domain multi-turn multi-intention recognition system according to an embodiment of the present application, which can implement the cross-domain multi-turn multi-intention recognition method according to any of the above embodiments and is configured in a terminal.

[0081] The cross-domain multi-turn multi-intention recognition system 10 according to the embodiment includes an intention analysis module 11, a semantic analysis module 12, a supplement module 13 and a broadcast module 14.

[0082] The intention analysis module 11 is configured to, in a multi-turn conversation, perform active speech detection in real time, and when detecting that a conversation sentence is input, perform intention analysis on the conversation sentence input in the current conversation turn by an intention recognition module; the semantic analysis module 12 is configured to determine whether the intention of the conversation sentence in the current conversation turn is in an intention expectation space of a completed conversation before the current conversation turn, and if so, perform cross-domain semantic analysis on the conversation sentence, wherein the intention expectation space is constructed by a first-level intention of the completed conversation in the multi-turn conversation and a plurality of second-level intentions associated with the first-level intention by default, and is used to determine whether the conversation sentence in the current conversation turn needs to be recognized; the supplement module 13 is configured to, in the semantic analysis, determine whether the conversation sentence is complete, and if not, supplement the conversation sentence to generate a semantic complete sentence; and the broadcast module 14 is configured to generate a response sentence for the semantic complete sentence and broadcast the response sentence.

[0083] The embodiment of the present application further provides a non-volatile computer storage medium, which stores computer executable instructions, and the computer executable instructions can implement the cross-domain multi-turn multi-intention recognition method in any of the above method embodiments.

[0084] As an implementation form, the non-volatile computer storage medium of the present application stores computer executable instructions, and the computer executable instructions are configured to:

[0085] In a multi-turn conversation, active speech detection is performed in real time, and when detecting that a conversation sentence is input, intention analysis is performed on the conversation sentence input in the current conversation turn by an intention recognition module;

[0086] Determine whether the intention of the conversation sentence in the current conversation turn is in an intention expectation space of a completed conversation before the current conversation turn, and if so, perform cross-domain semantic analysis on the conversation sentence, wherein the intention expectation space is constructed by a first-level intention of the completed conversation in the multi-turn conversation and a plurality of second-level intentions associated with the first-level intention by default, and is used to determine whether the conversation sentence in the current conversation turn needs to be recognized;

[0087] In the semantic analysis, it is judged whether the dialogue sentence is complete, if not, the dialogue sentence is supplemented to generate a semantic complete sentence;

[0088] A response sentence is generated for the semantic complete sentence, and is broadcasted.

[0089] As a non-volatile computer readable storage medium, it can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the method in the embodiment of the application. One or more program instructions are stored in the non-volatile computer readable storage medium, and when executed by the processor, the cross-domain multi-round multi-intention recognition method in any method embodiment described above is executed.

[0090] Figure 4 The hardware structure diagram of the electronic device for the cross-domain multi-round multi-intention recognition method provided by another embodiment of the application is as shown in Figure 4 The device comprises:

[0091] One or more processors 410 and memories 420, Figure 4 In the embodiment, the processor 410 is taken as an example. The device for the cross-domain multi-round multi-intention recognition method can further comprise an input device 430 and an output device 440.

[0092] The processor 410, the memory 420, the input device 430 and the output device 440 can be connected through a bus or other means, Figure 4 In the embodiment, the connection through the bus is taken as an example.

[0093] The memory 420 is a non-volatile computer readable storage medium, which can be used to store non-volatile software programs, non-volatile computer executable programs and modules, such as program instructions / modules corresponding to the cross-domain multi-round multi-intention recognition method in the embodiment of the application. The processor 410 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 420, that is, the cross-domain multi-round multi-intention recognition method in the above method embodiment is realized.

[0094] The memory 420 can include a program storage area and a data storage area. The program storage area can store an operating system, application programs, and the like needed by at least one function. The data storage area can store data, etc. In addition, the memory 420 can include a high-speed random access memory, and can further include a nonvolatile memory such as at least one disk memory device, flash memory device, or other nonvolatile solid-state memory device. In some embodiments, the memory 420 can optionally include a memory disposed remotely from the processor 410, which can be connected to the mobile device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0095] The input device 430 can receive input digital or character information. The output device 440 can include a display screen and the like display device.

[0096] The one or more modules are stored in the memory 420 and, when executed by the one or more processors 410, perform the cross-domain multi-turn multi-intent recognition method of any of the method embodiments described above.

[0097] The above product can perform the method provided by the embodiments of the present application, has the corresponding function modules and beneficial effects of performing the method. Technical details not described in detail in the embodiments can be referred to the method provided by the embodiments of the present application.

[0098] The non-volatile computer readable storage medium can include a program storage area and a data storage area. The program storage area can store an operating system, application programs, and the like needed by at least one function. The data storage area can store data created according to the use of the device, etc. In addition, the non-volatile computer readable storage medium can include a high-speed random access memory, and can further include a non-volatile memory such as at least one disk memory device, flash memory device, or other non-volatile solid-state memory device. In some embodiments, the non-volatile computer readable storage medium can optionally include a memory disposed remotely from the processor, which can be connected to the device through a network. Examples of the network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0099] The embodiments of the present application also provide an electronic device, which includes at least one processor, and a memory connected in communication with the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the cross-domain multi-turn multi-intent recognition method of any of the embodiments of the present application.

[0100] The electronic device of the embodiments of the present application exists in various forms, including but not limited to:

[0101] (1) Mobile communication device: This kind of device is characterized by having mobile communication function, and the main target is to provide voice and data communication. This kind of terminal includes smart phone, multimedia phone, functional phone, and low-end phone.

[0102] (2) Ultra-mobile personal computer device: This kind of device belongs to the category of personal computer, has computing and processing function, and generally has mobile internet feature. This kind of terminal includes PDA, MID and UMPC device, such as tablet computer.

[0103] (3) Portable entertainment device: This kind of device can display and play multimedia content. This kind of device includes audio and video player, palm game machine, electronic book, and smart toy and portable car navigation device.

[0104] (4) Other electronic devices with data processing function.

[0105] In this document, the terms "first", "second", etc. are used merely to distinguish one entity or action from another, and do not necessarily require or imply any actual such relationship or order between or among the entities or actions. Moreover, the terms "comprises", "comprising", "includes", "including", "has", "having" and the like are not limited to the listed elements only, but rather include other elements not expressly listed or inherent to the process, method, article, or apparatus. The term "consisting of" is used to refer to only the specified elements of any process, method, article, or apparatus, and does not foreclose the addition of additional unspecified elements, including elements inherent to the process, method, article, or apparatus. The term "consisting essentially of" is used to refer to the specified elements of any process, method, article, or apparatus, and does not foreclose the addition of additional unspecified elements, including elements inherent to the process, method, article, or apparatus.

[0106] The device embodiments described above are only illustrative, and the units described as separate components can or can not be physically separated, and the components displayed as units can or can not be physical units, i.e. can be located in one place or distributed to multiple network units. Part or all of the modules can be selected to achieve the purpose of the embodiment according to actual needs. Those skilled in the art can understand and implement without creative labor.

[0107] Those skilled in the art can clearly understand the technical solutions of the various embodiments from the above description of the embodiments, and the various embodiments can be implemented by means of software with the necessary general hardware platforms, and of course, can also be implemented by hardware. Based on such understanding, the above technical solutions, essentially or in other words, the part of the prior art that makes a contribution, can be embodied in the form of a software product, which can be stored in a computer readable storage medium, such as a ROM / RAM, a magnetic disk, an optical disk, and the like, and includes a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0108] Finally, it should be noted that: the above embodiments are only used to illustrate the technical solutions of the present application, rather than limit them; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that: it can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacement for some technical features therein; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present application.

Claims

1. A cross-domain multi-turn multi-intent recognition method, comprising: in a multi-turn dialogue, real-time active speech detection is performed, when it is detected that there is a dialogue sentence input, an intent analysis module is used to analyze the intent of the dialogue sentence input in the current dialogue turn; it is judged whether the intent of the dialogue sentence in the current dialogue turn is in the intent expectation space of the completed dialogue before the current dialogue turn, if yes, semantic analysis of the dialogue sentence is performed in a cross-domain manner, wherein the intent expectation space is constructed by a first-level intent of the completed dialogue in the multi-turn dialogue and a plurality of second-level intents associated with the first-level intent in a cross-domain manner, which are used to judge whether the dialogue sentence in the current dialogue turn needs to be recognized; in the semantic analysis, it is judged whether the dialogue sentence is complete, if not, the dialogue sentence is supplemented to generate a semantic complete sentence, wherein the supplement is processed by a rewriting model pre-trained by historical dialogue sentences, including: before cross-domain semantic analysis, the content in the intent expectation space is used as the rewriting direction of the rewriting model, and the rewriting model is used for supplement; a response sentence is generated for the semantic complete sentence, and the response sentence is broadcasted.

2. The method of claim 1, wherein, The completed dialogue includes adjacent dialogue within a preset dialogue turn and historical dialogue outside the preset dialogue turn; the response sentence is generated for the semantic complete sentence, and the response sentence is broadcasted, including: after the response sentence is generated, it is judged whether the current dialogue is being broadcasted, if yes, based on the intent expectation space, it is judged whether the response sentence can interrupt the broadcast of the completed dialogue: if the intent of the dialogue sentence in the current dialogue turn as a second-level intent is associated with the intent of the adjacent dialogue as a first-level intent in the intent expectation space, the broadcast of the completed dialogue is interrupted, and the response sentence is played, if the intent of the dialogue sentence in the current dialogue turn as a second-level intent is associated with the intent of the historical dialogue as a first-level intent in the intent expectation space, the response sentence is played after the broadcast of the completed dialogue is played.

3. The method of claim 2, wherein, The judgment of whether the response sentence can interrupt the broadcast of the completed dialogue further includes: if the response sentence is a preset forced broadcast, the broadcast of the completed dialogue is interrupted, and the response sentence is played.

4. A cross-domain multi-turn multi-intent recognition system, comprising: an intent analysis module, used to, in a multi-turn dialogue, perform real-time active speech detection, when it is detected that there is a dialogue sentence input, analyze the intent of the dialogue sentence input in the current dialogue turn by an intent recognition module; The semantic analysis module is configured to determine whether an intent of a dialogue sentence of the current dialogue round is in an intent expectation space of completed dialogues before the current dialogue round, and if so, perform cross-domain semantic analysis on the dialogue sentence, wherein the intent expectation space is constructed by a first-level intent of completed dialogues in multi-round dialogues and a plurality of second-level intents associated with the first-level intent by default for cross-domain semantic analysis, and is configured to determine whether the dialogue sentence of the current dialogue round needs to be recognized. The supplement module is configured to determine whether the dialogue sentence is complete in the semantic analysis, and if not, supplement the dialogue sentence to generate a semantically complete sentence, wherein the supplement is processed by a rewriting model pre-trained by historical dialogue sentences, including: before cross-domain semantic analysis, taking content in the intent expectation space as a rewriting direction of the rewriting model, and supplementing by using the rewriting model. The broadcasting module is configured to generate a response sentence for the semantically complete sentence and broadcast the response sentence.

5. The system of claim 4, wherein, The completed dialogues include adjacent dialogues within a preset dialogue round and historical dialogues outside the preset dialogue round. The broadcasting module is configured to: After generating the response sentence, determine whether the completed dialogues are being broadcasted, and if so, determine whether the response sentence can interrupt the broadcasting of the completed dialogues based on the intent expectation space. If, in the intent expectation space, an intent of a dialogue sentence of the current dialogue round as a second-level intent is associated with an intent of an adjacent dialogue as a first-level intent, the broadcasting of the completed dialogues is interrupted, and the response sentence is played. If, in the intent expectation space, an intent of a dialogue sentence of the current dialogue round as a second-level intent is associated with an intent of a historical dialogue as a first-level intent, the response sentence is played after the broadcasting of the completed dialogues is completed.

6. A storage medium having stored thereon a computer program product, characterized in that, The program is executed by a processor to implement the steps of the method of any one of claims 1-3.

7. A computer program product embedded with instructions on a storage medium, the instructions implementing the steps of the method of any one of claims 1-3.

8. An electronic device comprising: At least one processor and a memory connected to the at least one processor in communication, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the steps of the method of any one of claims 1-3.

Images