Cross-domain multi-round multi-intention identification method, electronic equipment and storage medium

Abstract

The embodiment of the invention provides a cross-domain multi-round multi-intention identification method and system, electronic equipment and a storage medium. The method comprises the following steps: in multiple rounds of dialogues, performing activity voice detection in real time, and when detecting that a dialogue statement is input, performing intention analysis on the dialogue statement input in the current dialogue round through an intention recognition module; judging whether the intention of the dialogue statement of the current dialogue round is in an intention expectation space of a completed dialogue before the current dialogue round, and if so, performing cross-domain semantic analysis on the dialogue statement; in semantic analysis, whether the dialogue statement is complete or not is judged, if not, the dialogue statement is complemented, and a semantic complete statement is generated; and generating a response statement for the semantic complete statement, and broadcasting the response statement. According to the embodiment of the invention, an intention expectation space method is introduced to judge whether identification and broadcast interruption are needed, the limitation of a voice rejection model judged from a voice level is broken, and mistaken interruption caused by noise is avoided.

Description

Technical Field

[0001] The present invention relates to the field of intelligent speech, and in particular to a cross-domain multi-round multi-intention recognition method, system, electronic device and storage medium. Background Art

[0002] With the development of intelligent voice technology, users can have multiple rounds of conversations with voice intelligent assistants and complete more command functions. During the broadcast of the intelligent assistant, there will be cases where the user interrupts the broadcast and starts a new conversation. At the same time, in multiple rounds of conversations, users will enter omitted or referential inquiry statements, resulting in incomplete information, and the intelligent assistant cannot effectively respond to the user's inquiry statements. These require that in multiple rounds of conversations, the system needs to monitor whether there is a new intention that needs to interrupt the current broadcast while broadcasting, and rewrite the incomplete sentence into a complete sentence with complete semantic information and can be understood independently from the context, to ensure the coherence and effectiveness of the conversation.

[0003] In existing multi-round dialogue products, VAD (Voice Activity Detection) technology is usually used to identify interruptions by judging whether the user is speaking or has finished speaking based on the strength of the sound signal or the pause time. For rewriting sentences with incomplete information, existing technologies usually save the context information of the skill, and use the context to complete the user's input when the same skill is hit in the next round of dialogue; or the historical single-round dialogue and the current incomplete sentence are used as dialogue data and encoded into a sequence for training to generate sequence labels.

[0004] The existing technology can be specifically manifested as: the current multi-round dialogue intention recognition, the voice assistant uses VAD technology to determine whether there is sound at the voice level during the broadcast process, and if there is sound, it determines whether the user is talking to itself, and recognizes the user's intention. After the intention is recognized, the user's sentence is parsed and a response is made. When the user's sentence is incomplete, the sentence entered by the user is completed according to the tags in the historical single-round dialogue or the context information of the saved skills.

[0005] In the process of implementing the present invention, the inventors found that there are at least the following problems in the related art: Through VAD recognition, it is easy for the background sound or noise to be mistakenly identified as a user interruption, which may cause the smart assistant to stop broadcasting. In addition, since it is voice activity detection, it is also easy for a short pause in the user's speech to cause the machine to think that the user has finished speaking.

[0006] For multiple rounds of conversations within a single skill, only the context information of the single skill is saved. The context can only be used when the same skill is hit in the next round of conversation. Therefore, if the next round of conversation hits other skills, the data structure defined by each skill may not be the same, resulting in a mismatch in data types, and the context cannot be used to complete the user's input.

[0007] Based on the historical single-round dialogue labels, the dialogue data needs to be encoded into a sequence starting from the first sentence, and a label corresponding to the sequence needs to be generated. Subsequently, the sequence that needs to be rewritten needs to be found based on the label, which can easily lead to a long query process and slow response speed. Summary of the invention

[0008] In order to at least solve the problem of multi-round and multi-intention recognition across fields in the existing technology.

[0009] In a first aspect, an embodiment of the present invention provides a cross-domain multi-round multi-intention recognition method, including: In multiple rounds of conversation, active speech detection is performed in real time. When a conversation sentence is detected, the intention analysis of the conversation sentence input in the current conversation round is performed through the intention recognition module; Determine 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. If so, perform cross-domain semantic analysis on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue and multiple second-level intentions preset and associated with the first-level intention, and is used to determine whether the dialogue sentence of the current dialogue round needs to be recognized; In the semantic analysis, it is determined whether the dialogue sentence is complete. If it is incomplete, the dialogue sentence is supplemented to generate a semantically complete sentence. A response sentence is generated for the semantically complete sentence and is broadcasted.

[0010] In a second aspect, an embodiment of the present invention provides a cross-domain multi-round multi-intention recognition system, including: The intention analysis module is used to perform real-time active voice detection in multiple rounds of dialogue. When a dialogue sentence input is detected, the intention analysis is performed on the dialogue sentence input in the current dialogue round through the intention recognition module; A semantic parsing module, used to determine 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, perform cross-domain semantic parsing on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue, and multiple second-level intentions preset and associated with the first-level intention, and is used to determine whether the dialogue sentence of the current dialogue round needs to be recognized; A supplementation module is used to determine whether the dialogue sentence is complete during the semantic analysis, and if incomplete, supplement the dialogue sentence to generate a semantically complete sentence; The broadcast module is used to generate a response sentence for the semantically complete sentence and broadcast it.

[0011] According to a third aspect, an electronic device is provided, comprising: 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 so that the at least one processor can perform the steps of the cross-domain multi-round multi-intention recognition method of any embodiment of the present invention.

[0012] In a fourth aspect, an embodiment of the present invention provides a storage medium having a computer program stored thereon, characterized in that when the program is executed by a processor, the steps of the cross-domain multi-round and multi-intention recognition method of any embodiment of the present invention are implemented.

[0013] In a fifth aspect, an embodiment of the present invention provides a computer program product, including a computer program / instructions, characterized in that when the computer program / instructions are executed by a processor, the steps of the cross-domain multi-round and multi-intention recognition method of any embodiment of the present invention are implemented.

[0014] The beneficial effects of the embodiments of the present invention are as follows: the present application takes into account the problems of broadcast interruption and sentence incompleteness in user dialogue interaction, and introduces the method of intention expectation space to determine whether recognition and broadcast interruption are required, breaking the limitations of the "speech rejection" model that judges from the voice level, and avoiding false interruptions caused by noise. And through the pre-trained rewriting model, the intention expectation space is used to supplement the semantically incomplete sentences, generate semantically complete sentences, and improve the accuracy of the response. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0016] Figure 1 is a flow chart of a cross-domain multi-round multi-intention recognition method provided by an embodiment of the present invention; Figure 2 It is a specific flow chart of a cross-domain multi-round multi-intention recognition method provided by an embodiment of the present invention; Figure 3It is a schematic diagram of the structure of a cross-domain multi-round multi-intention recognition system provided by an embodiment of the present invention; Figure 4 A schematic structural diagram of an embodiment of an electronic device for cross-domain multi-round multi-intention recognition provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0017] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0018] like Figure 1 FIG. 1 is a flow chart of a cross-domain multi-round multi-intention recognition method provided by an embodiment of the present invention, comprising the following steps: S11: in multiple rounds of dialogue, active voice detection is performed in real time, and when a dialogue sentence input is detected, the intention analysis is performed on the dialogue sentence input in the current dialogue round through the intention recognition module; S12: Determine 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. If so, perform cross-domain semantic analysis on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue and multiple second-level intentions preset and associated with the first-level intention, so as to determine whether the dialogue sentence of the current dialogue round needs to be recognized; S13: In the semantic analysis, determining whether the dialogue sentence is complete, and if incomplete, supplementing the dialogue sentence to generate a semantically complete sentence; S14: Generate a response sentence for the semantically complete sentence and broadcast it.

[0019] In conventional multi-round conversations, problems such as interruptions and incomplete sentences in the conversation interaction are taken into consideration. For example, in half-duplex and full-duplex conversations, when the intelligent voice device is giving feedback to the user, it will collect the user's input during the period of the intelligent device's feedback, and usually interrupt itself to continue the next round of conversation.

[0020] For example: User: Navigate to McDonald’s; Smart voice device: We have found 3 McDonald's nearby for you, namely: XXX Street Store, College Store, Train Store...; User (interrupting): The second one; Intelligent voice device: start guiding you...; User (the user spoke to the passenger in the front passenger seat, not to the smart voice device): Which one do you think he is linking this time? Intelligent voice device (misidentified and interrupted its own conversation): Sorry, I didn’t understand what you said.

[0021] That is to say, the existing technology exists: 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 quickly recognized to ensure the real-time nature of the user interaction. However, this also brings new problems. Ideally, the user has a one-to-one conversation with the intelligent device. However, in real scenarios, for example, when multiple people watch a smart TV together or there are multiple people in a car, there is a possibility that the user will interact with other people while interacting with the intelligent voice recognition. Therefore, it will appear that in multiple rounds of dialogue interaction, the intelligent voice device also recognizes, interrupts and responds to the conversation between the user and other people (the user is not speaking to the intelligent voice recognition).

[0022] Therefore, in order to solve the above problems, the main improvements of this application are: 1. How to ensure that the intelligent voice device will not be accidentally interrupted by noise or conversations between the user and other people during multi-round dialogue interactions; 2. If interruption is required, how to determine the interruption logic.

[0023] In step S11, the dialogue system is turned on, and the user interacts with the intelligent voice device by voice. During multiple rounds of voice interaction, the intelligent voice device performs VAD (Voice activity detection) detection in real time, and detects whether there is a new dialogue sentence input in the multi-way dialogue. When a dialogue sentence input is detected, the intention recognition module performs an intention analysis on the dialogue sentence input in the current dialogue round; For step S12, the method pre-constructs an intention expectation space. First, the intention determined by the completed dialogue round is set as the first-level intention, and the associated intention is preset according to the first-level intention and set as the second-level intention. The first-level intention and the second-level intention jointly constitute the intention expectation space.

[0024] For example, the first-level intent is navigation, and the second-level intents associated with it may include: weather, group buying, ticket booking, etc. The above are examples, indicating that the first-level intent can be associated with multiple second-level intents, which are not limited here. For example, in the first round of conversation, the user's intention is navigation. The second round of conversation can be group buying. It should be noted that this application sets the intentions determined by all completed conversation rounds as first-level intentions, rather than "the intentions of the previous completed conversation round." Therefore, in the third round of conversation, the first-level intent at this time includes "navigation, group buying." The second-level intent includes intentions that are preset and associated with navigation and group buying. As the user continues to talk, the first-level intent can continue to expand.

[0025] The idea of ​​this design is that in the multi-round dialogue of users, cross-domain situations will indeed occur, but there is still a certain logical range across domains. For example, the user asks "What delicacies are there in Suzhou?", and the intelligent voice device replies "Su-style fresh meat mooncakes, green dumplings, Aozao noodles, pan-fried, osmanthus cooked lotus root, steamed pork ribs..." At this time, for example, this part of the interaction is all in the field of food. At this time, the user inputs "How long does it take to get there from Wuhan" (at this time, it crosses from the field of food to the field of navigation), and the intelligent voice device replies "It takes 10 hours to drive there." The user continues to ask "How will it be tomorrow over there" (and crosses from the field of navigation to the field of weather). In other words, although the user's multi-round dialogue will continue to cross domains, they all follow the logic of the dialogue. Even if it crosses domains, the fields crossed are also related. Therefore, the intended expectation space of this application is to construct related cross-domains.

[0026] For example, the intention of the current dialogue sentence of "How long does it take to get there from Wuhan" is "navigation", and the previous completed dialogue is "Introduce what delicacies are there in Suzhou". Assume that the intention expectation space of the completed dialogue is "First-level intention: food; second-level intention: navigation, group buying, takeaway". The intention of the dialogue sentence of the current dialogue round exists in the second-level intention of the intention expectation space of the completed dialogue. If it exists, it means that the sentence said by the user is related, rather than an unfounded conversation (for example, the user turns his head and says something to others, or noise). In this way, it is determined whether the dialogue sentence of the current dialogue round needs to be recognized. If it exists, it means that it needs to be recognized in multiple rounds of dialogue. If it does not exist, it will not be recognized in multiple rounds of dialogue. In other words, when a human voice is detected, it is determined whether it has an intention. If the new dialogue sentence has an intention, the user's sentence intention is determined whether it belongs to the intention expectation space. If it belongs to the intention expectation space, the input sentence parsing module parses the semantics of the sentence to be answered input by the user in the current round.

[0027] Regarding step S13, in multiple rounds of dialogue, users often omit some words that appear in the historical dialogue sentences in the current round of waiting for response, or replace these words with referential words, so the sentences will be incomplete. In order to solve this problem, the present application needs to supplement the dialogue sentences to generate semantically complete sentences. Specifically, the supplementation is processed by the rewriting model pre-trained for the historical dialogue sentences, including: Before cross-domain semantic analysis, the content in the intention expectation space is used as the rewriting direction of the rewriting model, and the rewriting model is used for supplementation.

[0028] In this embodiment, for example: User: "How far is Suzhou from Wuhan?" Smart voice device: "About 743 kilometers" The statements to be answered in the current round are: User: “What kind of delicious food is there?” – Use the rewriting model to rewrite: “What kind of delicious food is there in Suzhou?”

[0029] Furthermore, since the rewriting model has a large divergent space, in order to make the supplemented content more in line with the user's original intention, the content in the intention expectation space is used as the rewriting direction of the rewriting model to avoid the generated content still not being able to explain the specific intention or the wrong intention, making it difficult to truly understand user needs and giving wrong feedback.

[0030] For step S14, a response sentence is generated for the semantically complete sentence and broadcasted. In the broadcasting, there will also be interruptions. Specifically, the completed conversation includes: adjacent conversations within the preset conversation rounds, and historical conversations outside the preset conversation rounds; Generating a response sentence for the semantically complete sentence and broadcasting it includes: After generating a response statement, determine whether the completed conversation is currently being broadcasted. If it is already being broadcasted, determine whether the response statement can interrupt the broadcast of the completed conversation based on the intended expectation space: If, in the intention expectation space, the intention of the dialogue sentence of the current dialogue turn as the second-level intention is associated with the intention of the adjacent dialogue as the first-level intention, the broadcast of the completed dialogue is interrupted and the response sentence is played. If in the intention expectation space, the intention of the dialogue sentence of the current dialogue round as the second-level intention is associated with the intention of the historical dialogue as the first-level intention, then wait until the broadcast of the completed dialogue is finished before playing the response sentence.

[0031] If the response statement is a preset mandatory broadcast, the broadcast of the completed dialogue is interrupted and the response statement is played.

[0032] In this embodiment, taking into account multiple rounds of conversations, if the response statement of the previous round of interaction is still being broadcast, it needs to be interrupted if it is still being broadcast. A simple way is to wait for the complete semantic sentence to be broadcast and determine whether it is currently being broadcast and it is a forced broadcast. If so, wait for the broadcast to be completed and generate a statement to be answered, otherwise directly initiate an interrupt command.

[0033] This method requires setting "forced" settings for various broadcasts in advance. If too many settings are set, all statements will be interrupted. If the settings are inaccurate, the interruption effect and user experience may not be good.

[0034] Therefore, the interruption of this application is related to the intention expectation space. Therefore, the completed conversation is divided into: adjacent conversations within the preset conversation rounds, and historical conversations outside the preset conversation rounds. For example, there are 7 rounds of multi-round conversations, and the conversations within 3 rounds are adjacent conversations, and the conversations outside are historical conversations. It is also considered that if the intention of the conversation statement in the current round may appear in both the adjacent conversation and the historical conversation, it will be processed as an adjacent conversation. The priority of the adjacent conversation is greater than the historical conversation. The adjacent conversation within the preset conversation round indicates that the user and the intelligent voice device are communicating about the hot topic content. At this time, they can interrupt each other, which can be closer to the full-duplex conversation, making the voice interaction more flexible and fast. The historical conversation outside the preset conversation round indicates that the intended topic may be inserted by the user temporarily, and the immediacy of the conversation is relatively weak. Therefore, it is more suitable for the half-duplex interactive conversation and improves the user interaction experience. In other words, if the user's sentence information is incomplete, the sentence is rewritten by the rewriting module. The system responds to the statement. If there is a broadcast currently and the broadcast content is mandatory, the system will respond after the broadcast is completed. Otherwise, the system will directly interrupt the broadcast to respond. Therefore, the intended expectation space of this application is used to determine whether the dialogue statement of the current dialogue round needs to be recognized, to determine whether the response statement can interrupt the broadcast of the completed dialogue, and to supplement the rewriting model. The overall process is as follows Figure 2 shown.

[0035] It can be seen from this implementation that this application takes into account the problems of interruptions and incomplete sentences in user dialogue interactions, and introduces the method of intention expectation space to determine whether recognition and interruption are required, breaking the limitations of the "speech rejection" model that judges from the voice level and avoiding false interruptions caused by noise. And through the pre-trained rewriting model, the intention expectation space is used to supplement the semantically incomplete sentences, generate semantically complete sentences, and improve the accuracy of the response.

[0036] Simply put, when a user has multiple rounds of conversations with an intelligent dialogue system, he or she may choose to interrupt the current conversation and start a new one while the system is broadcasting. In addition, some words that appeared in the historical conversation sentences are omitted in the current round of waiting answers, or these words are replaced with referential words, for example: User: “Please introduce some delicious food in Suzhou” Intelligent voice system: "Su-style fresh meat mooncakes, green dumplings, Aozao noodles, fried dumplings, osmanthus cooked lotus root, steamed spareribs..." User (interruption): "How long does it take to get there from Wuhan?" [interruption recognition, skill switching, incomplete sentence - omitted] Intelligent voice system: "It takes 10 hours to drive there" User: "How's it going tomorrow over there?" [Incomplete sentence - reference] Step 1: The dialogue system dialogue is started, and VAD detection is performed to detect whether there are new dialogue sentences.

[0037] Step 2: The intention recognition module analyzes the detected new dialogue to determine whether it has an intention. If the new dialogue sentence has an intention, determine whether the intention is within the expected intention space; after determining that it is within the expected intention space, input the dialogue sentence to be answered into the semantic parsing module.

[0038] Step 3: If the dialogue sentence to be answered is incomplete, the semantics is rewritten through the rewriting module, and the rewriting type (omitted, reference) is determined through the pre-trained rewriting model. According to the rewriting type, the incomplete semantics is rewritten in combination with at least one historical sentence to be answered and the corresponding system response sentence to obtain complete semantic information. After obtaining the complete semantic sentence, determine whether there is a broadcast at present and it is a mandatory broadcast. If there is, wait for the broadcast to be completed and generate the sentence to be answered, otherwise directly initiate an interruption command.

[0039] Step 4: Initiate an interrupt command to output a response statement and return to step 1.

[0040] The intention expectation space includes the first-level intention and the second-level intention, which are used to determine whether the new dialogue intention has the conditions to interrupt the current dialogue. This is used to determine whether the detected new dialogue sentence is noise, thereby avoiding the accidental interruption of the human-computer dialogue by noise. The intention expectation space is constructed based on the first-level intention (the user's dialogue intention determined by the completed dialogue round) and the second-level intention (the associated intention preset based on the first-level intention). If the new dialogue sentence intention meets the first-level intention, it indicates that the new dialogue sentence is related to the completed dialogue, so that it can be determined that the detected new dialogue sentence is not noise, and respond to the new dialogue sentence, thereby avoiding the accidental interruption of the human-computer dialogue by noise. If the new dialogue intention meets the second-level intention, the new dialogue sentence can also be determined as non-noise, thereby avoiding the target user's dialogue sentence being mistakenly identified as noise, thereby avoiding the user's interruption demand being ignored when the new dialogue intention corresponding to the target user's dialogue sentence does not fully match the current dialogue intention.

[0041] The rewriting model is trained based on the historical dialogue sentences of the dialogue system. The model rewrites the omissions, references, and other incomplete sentences based on the rewriting type and at least one round of historical dialogue and the sentences to be answered. The omitted or referenced inquiry sentences input by the user are rewritten to make the sentence semantics complete, so that the intelligent dialogue system can make effective responses to the user's inquiry sentences.

[0042] like Figure 3 The figure shows a schematic diagram of the structure of a cross-domain multi-round multi-intention recognition system provided by an embodiment of the present invention. The system can execute the cross-domain multi-round multi-intention recognition method described in any of the above embodiments and be configured in a terminal.

[0043] The present embodiment provides a cross-domain multi-round multi-intention recognition system 10 including: an intention analysis module 11, a semantic parsing module 12, a supplementary writing module 13 and a reporting module 14.

[0044] Among them, the intention analysis module 11 is used to perform real-time active voice detection in multiple rounds of dialogue. When a dialogue sentence input is detected, the intention recognition module performs intention analysis on the dialogue sentence input in the current dialogue round; the semantic parsing module 12 is used to determine 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 so, the dialogue sentence is subjected to cross-domain semantic parsing, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue, and a plurality of second-level intentions preset and associated with the first-level intention, and is used to determine whether the dialogue sentence in the current dialogue round needs to be recognized; the supplementation module 13 is used to determine whether the dialogue sentence is complete in the semantic parsing. If not, the dialogue sentence is supplemented to generate a semantically complete sentence; the broadcasting module 14 is used to generate a response sentence for the semantically complete sentence and broadcast it.

[0045] The embodiment of the present invention further provides a non-volatile computer storage medium, the computer storage medium stores computer executable instructions, and the computer executable instructions can execute the cross-domain multi-round multi-intention recognition method in any of the above method embodiments; As an implementation mode, the non-volatile computer storage medium of the present invention stores computer executable instructions, and the computer executable instructions are configured as follows: In multiple rounds of conversation, active speech detection is performed in real time. When a conversation sentence is detected, the intention analysis of the conversation sentence input in the current conversation round is performed through the intention recognition module; Determine 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. If so, perform cross-domain semantic analysis on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue and multiple second-level intentions preset and associated with the first-level intention, and is used to determine whether the dialogue sentence of the current dialogue round needs to be recognized; In the semantic analysis, it is determined whether the dialogue sentence is complete. If it is incomplete, the dialogue sentence is supplemented to generate a semantically complete sentence. A response sentence is generated for the semantically complete sentence and is broadcasted.

[0046] 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 present invention. 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 of the above method embodiments is executed.

[0047] Figure 4 is a schematic diagram of the hardware structure of an electronic device of a cross-domain multi-round multi-intention recognition method provided by another embodiment of the present application, such as Figure 4 As shown, the device includes: One or more processors 410 and memory 420, Figure 4 A processor 410 is taken as an example. The device of the cross-domain multi-round multi-intention recognition method may also include: an input device 430 and an output device 440.

[0048] The processor 410, the memory 420, the input device 430 and the output device 440 may be connected via a bus or other means. Figure 4 The example of connecting through bus is taken in the following.

[0049] The memory 420, as a non-volatile computer-readable storage medium, 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 present 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 of the above method embodiment is realized.

[0050] The memory 420 may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data, etc. In addition, the memory 420 may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the memory 420 may optionally include a memory remotely arranged relative to the processor 410, and these remote memories may be connected to the mobile device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0051] The input device 430 can receive input digital or character information. The output device 440 can include a display device such as a display screen.

[0052] 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-round multi-intention recognition method in any of the above method embodiments.

[0053] The above-mentioned product can execute the method provided in the embodiment of the present application, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not fully described in this embodiment, please refer to the method provided in the embodiment of the present application.

[0054] The non-volatile computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system, an application required for at least one function; the data storage area may store data created according to the use of the device, etc. In addition, the non-volatile computer-readable storage medium may include a high-speed random access memory, and may also include a non-volatile memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the non-volatile computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the device via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0055] An embodiment of the present invention also provides an electronic device, comprising: 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-round and multi-intention recognition method of any embodiment of the present invention.

[0056] The electronic device of the embodiment of the present application exists in various forms, including but not limited to: (1) Mobile communication equipment: This type of equipment is characterized by having mobile communication functions and its main purpose is to provide voice and data communications. This type of terminal includes: smart phones, multimedia phones, functional phones, and low-end phones.

[0057] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDAs, MIDs, and UMPC devices, such as tablet computers.

[0058] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players, handheld game consoles, e-books, smart toys, and portable car navigation devices.

[0059] (4) Other electronic devices with data processing functions.

[0060] In this article, relational terms such as first and second, etc. are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include" and "comprise" include not only those elements, but also other elements not explicitly listed, or also include elements inherent to such processes, methods, articles or equipment. In the absence of further restrictions, the elements defined by the statement "include..." do not exclude the existence of other identical elements in the process, method, article or equipment that includes the elements.

[0061] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0062] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0063] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A cross-domain multi-round multi-intention recognition method, comprising: In multiple rounds of conversation, active speech detection is performed in real time. When a conversation sentence is detected, the intention analysis of the conversation sentence input in the current conversation round is performed through the intention recognition module; Determine 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. If so, perform cross-domain semantic analysis on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue and multiple second-level intentions preset and associated with the first-level intention, and is used to determine whether the dialogue sentence of the current dialogue round needs to be recognized; In the semantic analysis, it is determined whether the dialogue sentence is complete. If it is incomplete, the dialogue sentence is supplemented to generate a semantically complete sentence. A response sentence is generated for the semantically complete sentence and is broadcasted.

2. The method according to claim 1, wherein: The completed conversations include: adjacent conversations within a preset conversation round, and historical conversations outside the preset conversation round; Generating a response sentence for the semantically complete sentence and broadcasting it includes: After generating a response statement, determine whether the completed conversation is currently being broadcasted. If it is already being broadcasted, determine whether the response statement can interrupt the broadcast of the completed conversation based on the intended expectation space: If, in the intention expectation space, the intention of the dialogue sentence of the current dialogue turn as the second-level intention is associated with the intention of the adjacent dialogue as the first-level intention, the broadcast of the completed dialogue is interrupted and the response sentence is played. If in the intention expectation space, the intention of the dialogue sentence of the current dialogue round as the second-level intention is associated with the intention of the historical dialogue as the first-level intention, then wait until the broadcast of the completed dialogue is finished before playing the response sentence.

3. The method according to claim 2, wherein: The determining whether the response statement can interrupt the broadcast of the completed dialogue further includes: If the response statement is a preset mandatory broadcast, the broadcast of the completed dialogue is interrupted and the response statement is played.

4. The method according to claim 1, wherein: The rewriting is processed by a rewriting model pre-trained on historical dialogue sentences, including: Before cross-domain semantic analysis, the content in the intention expectation space is used as the rewriting direction of the rewriting model, and the rewriting model is used for supplementation.

5. A cross-domain multi-round multi-intention recognition system, comprising: The intention analysis module is used to perform real-time active voice detection in multiple rounds of dialogue. When a dialogue sentence input is detected, the intention analysis is performed on the dialogue sentence input in the current dialogue round through the intention recognition module; A semantic parsing module, used to determine 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, perform cross-domain semantic parsing on the dialogue sentence, wherein the intention expectation space is constructed by the first-level intention of the completed dialogue in multiple rounds of dialogue, and multiple second-level intentions preset and associated with the first-level intention, and is used to determine whether the dialogue sentence of the current dialogue round needs to be recognized; A supplementation module is used to determine whether the dialogue sentence is complete during the semantic analysis, and if incomplete, supplement the dialogue sentence to generate a semantically complete sentence; The broadcast module is used to generate a response sentence for the semantically complete sentence and broadcast it.

6. The system according to claim 5, wherein: The completed conversations include: adjacent conversations within a preset conversation round, and historical conversations outside the preset conversation round; The broadcast module is used for: After generating the response statement, determining whether the completed conversation is currently being broadcasted, and if it is already being broadcasted, determining whether the response statement can interrupt the broadcasting of the completed conversation based on the intended expectation space; If, in the intention expectation space, the intention of the dialogue sentence of the current dialogue turn as the second-level intention is associated with the intention of the adjacent dialogue as the first-level intention, the broadcast of the completed dialogue is interrupted and the response sentence is played. If in the intention expectation space, the intention of the dialogue sentence of the current dialogue round as the second-level intention is associated with the intention of the historical dialogue as the first-level intention, then wait until the broadcast of the completed dialogue is finished before playing the response sentence.

7. The system according to claim 5, wherein: The write-in module is used for: Before cross-domain semantic analysis, the content in the intention expectation space is used as the rewriting direction of the rewriting model, and the rewriting model is used for supplementation.

8. A storage medium having a computer program product stored thereon, characterized in that: When the program is executed by a processor, the steps of the method described in any one of claims 1 to 4 are implemented.

9. A computer program product having instructions embedded on a storage medium, wherein the instructions implement the steps of the method according to any one of claims 1 to 4.

10. An electronic device, comprising: 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 method described in any one of claims 1 to 4.

Images