Intention recognition method, electronic device, and program product

Abstract

The present disclosure provides an intention recognition method, an electronic device and a program product, and relates to the technical field of natural language processing, and is used to solve the problem of low accuracy of a terminal in determining a user's screen intention. The method comprises the following steps: obtaining an intention text of a to-be-recognized intention, and matching the intention text with a preset semantic rule, the preset semantic rule being determined based on a content object pronoun and / or a question pronoun in a screen; and in a case where it is determined that the intention text matches the preset semantic rule, determining that the to-be-recognized intention has a screen intention. In this way, the accuracy of the terminal in determining the user's screen intention can be improved.

Description

Technical Field

[0001] This disclosure relates to the field of natural language processing technology, and in particular to an intent recognition method, electronic device, and program product. Background Technology

[0002] "Screen questioning" refers to a user asking questions related to the content displayed on the terminal screen via voice. The terminal determines that the user's voice has initiated a "screen questioning" operation, analyzes the question, obtains the answer, and provides feedback to the user.

[0003] However, the voice questions users ask to "ask the screen" are rather arbitrary and generalized. Based solely on the user's voice input, the terminal cannot accurately determine whether the user's intention is to "ask the screen" or to interact with the terminal's voice assistant (e.g., chatting with the voice assistant, unrelated to the content displayed on the screen). Therefore, the accuracy of current terminals in recognizing user intent based on voice is relatively low. Summary of the Invention

[0004] This disclosure provides an intent recognition method, electronic device, and program product to at least address the problem of low accuracy in determining a user's intent when asking questions.

[0005] On the one hand, an intent recognition method is provided, which includes: acquiring the intent text of the intent to be recognized; matching the intent text with a preset semantic rule, wherein the preset semantic rule is determined based on content object referents and / or question referents in the screen; and determining that the intent to be recognized has a questioning intent when the intent text matches the preset semantic rule.

[0006] On the other hand, an intent recognition device is provided, which includes an acquisition module and a processing module.

[0007] The acquisition module is used to acquire the intent text of the intent to be identified;

[0008] The processing module is used to match the intent text with preset semantic rules, which are determined based on content object referents and / or question referents in the screen;

[0009] The processing module is also used to determine that the intent to be identified has a screen-questioning intent when the intent text matches a preset semantic rule.

[0010] In another aspect, an electronic device is provided, comprising: a memory and a processor; the memory and the processor are coupled; the memory is used to store instructions executable by the processor; and the processor executes the intent recognition method of any of the above embodiments when executing the instructions.

[0011] In another aspect, a computer-readable storage medium is provided, on which computer instructions are stored, which, when executed on a computer, cause the computer to perform the intent recognition method of any of the above embodiments.

[0012] In another aspect, a computer program product is provided that, when the computer program product is run on a computer, causes the computer to execute the intent recognition method of any of the above embodiments.

[0013] In this embodiment of the disclosure, the intent text of the intent to be identified is obtained and matched with preset semantic rules. If the intent text matches the preset semantic rules, the intent to be identified is determined to have a screen-questioning intent. Since the preset semantic rules are determined based on content object pronouns and / or question pronouns on the screen, the semantics of intent text with a screen-questioning intent follows specific rules. Therefore, by setting preset semantic rules that can determine whether intent text has a screen-questioning intent, the accuracy of determining whether the obtained intent text has a screen-questioning intent can be accurately determined, thereby improving the accuracy of determining the user's screen-questioning intent. Attached Figure Description

[0014] To more clearly illustrate the technical solutions in this disclosure, the accompanying drawings used in some embodiments of this disclosure will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this disclosure. For those skilled in the art, other drawings can be obtained based on these drawings.

[0015] Figure 1 This is a schematic diagram of an intent recognition scenario provided by an embodiment of the present disclosure;

[0016] Figure 2 A flowchart illustrating an intent recognition method provided in an embodiment of this disclosure;

[0017] Figure 3 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0018] Figure 4 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0019] Figure 5 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0020] Figure 6 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0021] Figure 7 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0022] Figure 8 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0023] Figure 9 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0024] Figure 10 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0025] Figure 11 A flowchart illustrating another intent recognition method provided in this disclosure embodiment;

[0026] Figure 12 This is a schematic diagram of an application environment for intent recognition provided in an embodiment of the present disclosure;

[0027] Figure 13 This is a schematic diagram of the structure of an intent recognition device provided in an embodiment of the present disclosure;

[0028] Figure 14 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this disclosure. Detailed Implementation

[0029] The technical solutions of this disclosure will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0030] It should be noted that, in this disclosure, the terms "exemplary" or "for example" are used to indicate examples, illustrations, or descriptions. Any embodiment or design described as "exemplary" or "for example" in this disclosure should not be construed as being more preferred or advantageous than other embodiments or designs. Specifically, the use of terms such as "exemplary" or "for example" is intended to present the relevant concepts in a concrete manner.

[0031] Hereinafter, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature.

[0032] In the description of this disclosure, unless otherwise stated, " / " means "or," for example, A / B can mean A or B. "And / or" in this document is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A alone, A and B simultaneously, and B alone. Furthermore, "at least one" means one or more, and "more than one" means two or more.

[0033] Before providing a detailed description of the intent recognition method provided in the embodiments of this disclosure, the implementation environment and application scenarios of the embodiments of this disclosure will be introduced first.

[0034] First, the application scenarios of the embodiments of this disclosure will be introduced.

[0035] With the continuous advancement of artificial intelligence (AI) technology, especially in applications that enable complex interactions across various scenarios, AI technologies have developed rapidly. For example, screen-based interactive technology integrates AI, image recognition, and voice interaction technologies within smart terminal systems, enabling users to conveniently access screen-related information and services, thus improving the interaction efficiency and user experience of smart terminals. Recognizing the user's intent to ask a screen question through voice is a key technology in screen-based interactive technology. Distinguishing between a user asking a screen question and engaging in a simple chat with a voice assistant, particularly based solely on the user's voice-to-text input, remains a significant challenge in this field.

[0036] For example, when a user has questions about the content displayed on an electronic device's screen, they can ask the device questions related to that content via "screen inquiry." For instance, if the image on the screen includes a plant, the user can directly ask, "What kind of plant is this?"; or, if the image includes a person, the user can directly ask, "Please describe who this person is," and so on. When the electronic device's intent recognition system determines that the user is "inquiring about the screen," it will assign the question to a multimodal large-scale model to perform image understanding tasks and then feed the analysis results back to the user.

[0037] However, the current challenge of screen-questioning intent recognition technology lies in the fact that user questions are often arbitrary and generalized. Simply performing semantic analysis on the user's input makes it difficult to distinguish whether the user is "questioning the screen" or engaging in a normal chat with the electronic device's voice assistant. Especially when a voice question containing "screen-questioning" is inserted during a multi-turn conversation between the user and the voice assistant, the electronic device is even more likely to confuse the user's intention with whether the voice is intended to "question the screen" or engage in a chat with the voice assistant.

[0038] Typically, when a user initiates a "screen inquiry" voice query, the electronic device can take a screenshot of the content displayed on the screen for analysis. This screenshot serves as supplementary information for intent recognition, increasing the accuracy of screen inquiry intent identification. In other words, based on the screenshot background and the semantics of the user's voice, it determines whether the user is asking a question related to the screen content—that is, a screen inquiry. However, because screenshot analysis is required, it consumes a significant amount of power in general voice assistant scenarios and affects the response speed of the general voice assistant. Alternatively, a "circle search" method can be used for screen inquiries. The user draws a circle on the screen of interest using a gesture and then asks by voice what the circle contains. Based on this specific marker, the user's screen inquiry intent can be naturally recognized. However, this method increases the user's interaction, making the screen inquiry process more complex.

[0039] To at least address the aforementioned problems, this disclosure provides an intent recognition method that can be applied to electronic devices or any suitable device. This disclosure can accurately identify screen-asking intents through rule matching, and can further combine it with large model technology to more effectively solve the above problems. Based on the characteristics of the user's input speech, rule matching is used to quickly determine whether the user's speech belongs to the candidate of screen-asking intents. Then, large model-specific prompt words are used to determine whether the user's speech has screen-asking characteristics. For multi-turn dialogue scenarios, correlation discrimination between multi-turn dialogues is added based on the large model to remove interference from multi-turn dialogues. This not only effectively solves the problem of difficult screen-asking intent recognition, but also does not affect the response speed of recognizing user speech in general scenarios when interacting with voice-assisted dialogue, and does not increase power consumption or user interaction excessively.

[0040] In this embodiment of the disclosure, the intent text of the intent to be identified is obtained and matched with preset semantic rules. If the intent text matches the preset semantic rules, the intent to be identified is determined to have a screen-questioning intent. Since the preset semantic rules are determined based on content object pronouns and / or question pronouns on the screen, the semantics of intent text with a screen-questioning intent follows specific rules. Therefore, by setting preset semantic rules that can determine whether intent text has a screen-questioning intent, the accuracy of determining whether the obtained intent text has a screen-questioning intent can be accurately determined, thereby improving the accuracy of determining the user's screen-questioning intent.

[0041] The embodiments disclosed herein are mainly applied to scenarios where users use voice assistants on electronic devices to ask for various information related to the content displayed on the screen or the content displayed in the camera preview interface.

[0042] The implementation environment of the embodiments of this disclosure is described below.

[0043] like Figure 1The diagram illustrates an intent recognition system provided in this embodiment of the present disclosure, applied to an electronic device. The intent recognition system may include an input device 101, a screen 102, and a processor 103. The input device 101 may include a voice input device, a text input device, and a gesture input device. The voice input device receives voice input from the user, the text input device receives text input from the user, and the gesture input device receives gesture operations from the user. The screen 102 may be a display screen built into the electronic device or an external display screen connected to the electronic device.

[0044] When the electronic device displays content on screen 102, the user can "ask the screen" about the content displayed on screen 102 by inputting voice (or text, gesture operation, etc.) into the electronic device through input device 101 to ask questions related to the content displayed on screen 102, so that the electronic device responds to the user's voice and answers the user's questions.

[0045] That is, the input device 101 can acquire the intent text of the user's input intent to be recognized.

[0046] The processor 103 can match the intent text with preset semantic rules, so that if the intent text matches the preset semantic rules, it can determine that the intent to be identified has a screen-asking intent.

[0047] The electronic device can be a mobile phone, tablet computer, laptop computer, ultra-mobile personal computer (UMPC), handheld computer, netbook, personal digital assistant (PDA), wearable electronic device, vehicle device, or any other device including a screen. This application does not impose any restrictions on this.

[0048] For example, electronic devices may also include processors, screens, cameras, etc. The processor may include one or more processing units, such as application processors, modem processors, graphics processors, image signal processors, controllers, video codecs, digital signal processors, baseband processors, and / or neural network processors. Different processing units may be independent devices or integrated into one or more processors.

[0049] Electronic devices can display interfaces or images on their screens, allowing users to "question" the content displayed on the screen. Users can input voice commands to the electronic device to ask questions related to the content displayed on the screen, and the device will respond to the user's voice and answer their questions.

[0050] Of course, the electronic device provided in this application embodiment may also include one or more of the following components: battery, antenna, mobile communication module, wireless communication module, audio module, speaker, receiver, microphone, headphone jack, buttons, motor, indicator, and SIM card interface. This application embodiment does not impose any limitations on these components.

[0051] After introducing the application scenarios and implementation environments of the embodiments of this disclosure, the intent recognition method provided by the embodiments of this disclosure will be described in detail below in conjunction with the above implementation environment.

[0052] This disclosure provides an intent recognition method applied to electronic devices, such as... Figure 2 As shown, the intent recognition method may include: S201-S203.

[0053] S201. Obtain the intent text of the intent to be identified.

[0054] In this embodiment of the disclosure, when the electronic device displays an interface or image on its screen, the user can ask questions about the content displayed on the screen after viewing it. The user can ask the electronic device questions related to the content displayed on the screen via voice, thereby enabling the electronic device to respond to the user's voice and answer their questions.

[0055] In one possible implementation, users can "ask the screen" by inputting voice or text on the screen.

[0056] In one possible implementation, after acquiring the user's voice input (i.e., the intent to be recognized), the electronic device can perform speech recognition to obtain the corresponding intent text. Alternatively, the electronic device can directly acquire the text entered by the user on the screen.

[0057] In one possible implementation, when the electronic device is running a voice recognition or query function, the electronic device determines the intended intent to be recognized based on the received user input (voice, text, or operation), and provides feedback to the user input (i.e., the intended intent to be recognized) after recognizing the content displayed on the screen using a camera or image recognition function.

[0058] S202. Match the intent text with the preset semantic rules.

[0059] The preset semantic rules are determined based on the content object referents and / or question referents on the screen.

[0060] In one possible implementation, the electronic device can use a rule matching module to match the acquired intent text with preset semantic rules. That is, the rule matching module stores preset semantic rules in advance.

[0061] It should be noted that the preset semantic rules are semantic rules determined based on the analysis of multiple screen-questioning voices, and the preset semantic rules conform to the semantics of the voices with screen-questioning intent. That is, when the screen-questioning voice or intent text matches the preset semantic rules, it can be determined that the screen-questioning voice or intent text has a screen-questioning intent.

[0062] In one possible implementation, the electronic device can analyze and process the intent text based on preset semantic rules to determine whether the intent text conforms to a specific semantic pattern (specific text structure) indicated by the preset semantic rules.

[0063] Optionally, various semantic rules for voice assistants to recognize voice intent (user intent) can be pre-stored in the semantic rule base of the electronic device to support the electronic device in recognizing user voice intent in various scenarios.

[0064] S203. If the intent text matches the preset semantic rules, determine that the intent to be identified has the intent to ask the screen.

[0065] It should be noted that since the preset semantic rules are determined based on the speech or text with the intent to ask the screen, when the intent text matches the preset semantic rules, it can be determined that the intent to be recognized corresponding to the intent text has the intent to ask the screen.

[0066] In this embodiment, the intent text of the intent to be identified is obtained and matched with preset semantic rules. If the intent text matches the preset semantic rules, it is determined that the intent to be identified has a screen-asking intent. Since the semantics of intent text with a screen-asking intent follow specific rules, by setting preset semantic rules that can determine whether intent text has a screen-asking intent, the accuracy of determining whether the obtained intent text has a screen-asking intent can be improved. This enhances the accuracy of user intent recognition based on user speech.

[0067] In some embodiments, such as Figure 3 As shown, the intent recognition method may specifically include: S301-S303, and the above-mentioned S203 can be implemented through S303.

[0068] S301. Obtain the intent text of the intent to be identified.

[0069] S302. Match the intent text with the preset semantic rules.

[0070] For a detailed description of S301 and S302, please refer to the steps above; they will not be repeated here.

[0071] S303. If the text structure of the intent text matches the preset text structure indicated by the preset semantic rule, it is determined that the intent to be identified has a screen-questioning intent.

[0072] The preset semantic rules include: first-precision semantic rules and second-precision semantic rules, which indicate different preset text structures.

[0073] In one possible implementation, when matching the intent text with preset semantic rules, text structure analysis can be performed on the intent text to determine its text structure. Then, the text structure of the intent text is compared with the preset text structure indicated by the preset semantic rules to determine whether the text structure of the intent text matches the preset text structure indicated by the preset semantic rules.

[0074] It can be understood that preset semantic rules (first-precision semantic rules and second-precision semantic rules) are used to indicate preset text structures, and the intent of the intent text constituted by these preset text structures satisfies the screen-questioning intent. That is, the text structure of intent text with screen-questioning intent is the preset text structure.

[0075] In one possible implementation, the matching of the text structure of the intent text with the preset text structure indicated by the preset semantic rule can be understood as the similarity between the text structure of the intent text and the preset text structure indicated by the preset semantic rule satisfying the preset similarity.

[0076] It should be noted that the embodiments disclosed herein use the example of preset semantic rules including: a first-precision semantic rule (i.e., a high-precision semantic rule) and a second-precision semantic rule (i.e., a low-precision semantic rule) for illustrative purposes. In specific application scenarios, the preset semantic rules can be set to multiple precision semantic rules based on requirements. For example, three precision semantic rules can be set, including: a high-precision semantic rule, a medium-precision semantic rule, and a low-precision semantic rule; or five precision semantic rules can be set, including: a first-level precision semantic rule, a second-level precision semantic rule, a third-level precision semantic rule, a fourth-level precision semantic rule, and a fifth-level precision semantic rule.

[0077] In this embodiment of the disclosure, it is determined whether the text structure of the intent text matches the preset text structure indicated by the preset semantic rules, in order to determine whether the intent to be identified has a screen-asking intent. Thus, since the structure of text with a screen-asking intent usually has a specific text structure, when the text structure of the intent text conforms to the specific text structure (i.e., the preset text structure), it can be determined that the intent text corresponding to the intent text has a screen-asking intent.

[0078] In some embodiments, the preset semantic rules may include: a first-precision semantic rule; such as Figure 4As shown, the intent recognition method may specifically include: S401-S403, and the above-mentioned S203 can be implemented through S403.

[0079] S401. Obtain the intent text of the intent to be identified.

[0080] S402. Match the intent text with the preset semantic rules.

[0081] For a detailed description of S401 and S402, please refer to the steps above; they will not be repeated here.

[0082] S403. If the intent text matches the first-precision semantic rule, determine that the intent to be identified has the intent to ask for a screen.

[0083] In some embodiments, the first precision semantic rule is used to constrain the intent text to consist of content object referents and question referents on the screen, and that there are no other intent texts preceding the intent text.

[0084] It is understandable that the first-precision semantic rule, by restricting the intent text to include content object referents on the screen and question referents, can determine with greater accuracy whether the intent text has a questioning intent.

[0085] In one possible implementation, the first-precision semantic rule further constrains that there are no other intent texts between the user and the electronic device before the intent text is obtained (i.e., the user has not entered any other speech or text within a preset time before entering the intent to be recognized).

[0086] For example, the first precision semantic rule refers to the text structure of the schematic diagram text as follows: <begin>+[Pronoun referring to the content on the screen]+[Pronoun referring to the question]+ <end>Among them, [content object referents on the screen] are referents used to indicate objects displayed on the screen, such as: this, this one, he, it, she, in the picture, etc. [question referents] are interrogative words corresponding to the question screen, such as: what is it, what, what is something, who is it, what are they, how are they, what does it do, etc. <begin>This indicates that no other intent text was obtained within the preset time period prior to obtaining this intent text; <end>This indicates that no other intent texts were obtained after this intent text was obtained.

[0087] For example, if the intent text is "Who is he?" or "What is this?", it can be determined that these intent texts match the first-precision semantic rules and have a questioning intent.

[0088] That is, after matching the intent text with the preset semantic rules, if the intent text matches the first-precision semantic rule, it can be directly determined that the intent text corresponds to the intent text and has the intent to ask the screen.

[0089] In this embodiment of the disclosure, multiple semantic rules of different precision are set, and each semantic rule of different precision is used to constrain the intent text to have a different text structure. Thus, by using semantic rules of different precision, the semantic rules that the intent text matches are determined. If the intent text matches the first precision semantic rule (high precision semantic rule), it can be directly determined that the intent to be identified has a screen-asking intent.

[0090] In some embodiments, the preset semantic rules further include: a second precision semantic rule, wherein the precision of the first precision semantic rule indicating that the text has a screen-questioning intent is higher than the precision of the second precision semantic rule indicating that the text has a screen-questioning intent.

[0091] In some embodiments, such as Figure 5 As shown, the intent recognition method may specifically include: S501-S503.

[0092] S501. Obtain the intent text of the intent to be identified.

[0093] S502. Match the intent text with the preset semantic rules.

[0094] For a detailed description of S501 and S502, please refer to the steps above; they will not be repeated here.

[0095] S503. If it is determined that the intent text does not match the first-precision semantic rule, the screen-questioning intent to be identified is determined based on the matching result of the intent text and the second-precision semantic rule.

[0096] In one possible implementation, when it is determined that the intent text does not match the higher-precision semantic rule (i.e., the first-precision semantic rule), it can be further determined whether the intent text matches the lower-precision semantic rule (i.e., the second-precision semantic rule) to determine whether the intent to be identified has a screen-asking intent.

[0097] In some embodiments, the second precision semantic rule is used to constrain the intent text to include at least content object referents and question referents on the screen.

[0098] It is understandable that the second-precision semantic rule restricts the intent text to include at least the content object referents and question referents on the screen, but does not restrict whether the intent text also includes other referential content. Thus, it is possible to determine whether the intent text has a questioning intent on a broader scale.

[0099] Furthermore, the second-precision semantic rules do not constrain whether there are other intent texts between the user and the electronic device before the intent text is obtained.

[0100] For example, the second-precision semantic rule refers to the text structure of the schematic diagram text as: [content object referents on the screen] + <whatever>+[pronoun referring to the problem]. Among them... <whatever>The intent text can include any other content between [content object referent on screen] and [question referent].

[0101] For example, if the intent text is "What brand of car is this?" or "What holidays are there this month?", it can be determined that these intent texts match the second-precision semantic rules.

[0102] It should be noted that when the intent text does not match the first-precision semantic rule but matches the second-precision semantic rule, it cannot be directly determined that the intent to be identified has a screen-questioning intent. In this case, further analysis of the intent text is required (e.g., as described in the following embodiments) to determine whether the intent to be identified has a screen-questioning intent.

[0103] In this embodiment of the disclosure, if the intent text does not match the first precision semantic rule, it can be further determined whether the intent text matches the second precision semantic rule. Based on the matching result of the intent text and the second precision semantic rule, it can be determined whether the intent to be identified has a screen-asking intent. In this way, the intent text can be further analyzed to determine whether the intent to be identified has a screen-asking intent through multiple methods, thereby improving the accuracy of determining whether the intent to be identified has a screen-asking intent.

[0104] In some embodiments, such as Figure 6 As shown, the intent recognition method may specifically include: S601-S603, and the above-mentioned S503 can be implemented through S603.

[0105] S601. Obtain the intent text of the intent to be identified.

[0106] S602. Match the intent text with the preset semantic rules.

[0107] For a detailed description of S601 and S602, please refer to the steps above; they will not be repeated here.

[0108] S603. If it is determined that the intent text does not match the first precision semantic rule, the intent text matches the second precision semantic rule, and a preset operation associated with the screen-asking intent is received, then the intent to be identified is determined to have a screen-asking intent.

[0109] In one possible implementation, if the intent text does not match the first-precision semantic rule, and it is further determined that the intent text matches the second-precision semantic rule, then it can be determined that the intent to be identified may have a screen-questioning intent.

[0110] In this scenario, it can be further determined whether a preset operation associated with the screen-questioning intent was received within a preset time period before or after the intent to be identified was received, thereby confirming whether the intent to be identified has a screen-questioning intent. If a preset operation associated with the screen-questioning intent is received, the intent to be identified can be directly confirmed to have a screen-questioning intent based on the result of matching the intent text with the second-precision semantic rules. However, if no preset operation associated with the screen-questioning intent is received, it is not possible to directly confirm the intent to be identified to have a screen-questioning intent based on the result of matching the intent text with the second-precision semantic rules. Further consideration of other content (such as prompt words and history) is needed to determine whether the intent to be identified has a screen-questioning intent.

[0111] In one possible implementation, the preset operation associated with the screen-asking intent can be an auxiliary operation performed by the user on the screen, or the user's voice may include content related to "screen-asking" (such as "on the screen", "I want to ask the screen", etc.).

[0112] For example, the user's auxiliary operations on the screen may include: the user clicking on an object or person displayed on the screen, the user selecting a certain area on the screen, and the user operating on preset controls.

[0113] Understandably, in this situation, if the electronic device receives a preset operation associated with the screen-asking intent and determines that the intent text matches the second-precision semantic rule, it can directly determine that the intent to be identified has a screen-asking intent.

[0114] In this embodiment of the disclosure, when the intent text does not match the first precision semantic rule but matches the second precision semantic rule, the user's auxiliary operation (i.e., the preset operation associated with the screen-asking intent) can be combined to determine whether the intent to be identified has a screen-asking intent. Thus, it is possible to determine whether the intent to be identified has a screen-asking intent through multiple methods, thereby improving the accuracy of determining whether the intent to be identified has a screen-asking intent.

[0115] In some embodiments, such as Figure 7 As shown, the intent recognition method may specifically include: S701-S703, and the above-mentioned S503 can be implemented through S703.

[0116] S701. Obtain the intent text of the intent to be identified.

[0117] S702. Match the intent text with the preset semantic rules.

[0118] For a detailed description of S701 and S702, please refer to the steps above; they will not be repeated here.

[0119] S703. If the intent text does not match the first-precision semantic rule and the intent text does not match the second-precision semantic rule, determine that the intent to be identified does not have the intent to ask the screen.

[0120] It is understandable that if the intent text does not match either the first-precision semantic rule or the second-precision semantic rule, it can be directly determined that the intent to be identified does not have a screen-questioning intent.

[0121] In some embodiments, such as Figure 8 As shown, the intent recognition method may specifically include: S801-S804, and the above-mentioned S503 can be implemented through S803 and S804.

[0122] S801. Obtain the intent text of the intent to be identified.

[0123] S802. Match the intent text with the preset semantic rules.

[0124] For a detailed description of S801 and S802, please refer to the steps above; they will not be repeated here.

[0125] S803. When it is determined that the intent text does not match the first-precision semantic rule, but the intent text matches the second-precision semantic rule, identify the type of the intent text.

[0126] S804. Based on the type of intent text, determine the screen-questioning intent to be identified.

[0127] In one possible implementation, if the intent text does not match a first-precision semantic rule, and further determines that the intent text matches a second-precision semantic rule, then it can be determined that the intent to be identified may have a screen-asking intent. Therefore, the type of the intent text can be further identified to determine whether the intent to be identified has a screen-asking intent based on the type of the intent text.

[0128] In one possible implementation, the type of intent text can be identified through analysis using a Large Language Model (LLM).

[0129] In one possible implementation, the intent text type can include: check the weather, set an alarm, chat, question and answer, etc. It should be noted that if the intent text matches the "chat" or "question and answer" types, it can be determined that the intent text may have a screen-questioning intent; however, if the intent text matches the "check the weather" or "set an alarm" types, it can be determined that the intent text does not have a screen-questioning intent, because intent text of the "check the weather" and "set an alarm" types are user control voice commands to the electronic device, which clearly do not conform to a screen-questioning intent.

[0130] That is, if the intent text does not match the first-precision semantic rule but matches the second-precision semantic rule, the intent to be identified may have a screen-questioning intent by determining that the type of the intent text is "casual conversation" or "question and answer". Then, the intent text is further analyzed and identified through a large language model to further determine whether the intent to be identified has a screen-questioning intent.

[0131] In this embodiment of the disclosure, when the intent text does not match the first precision semantic rule but matches the second precision semantic rule, the type of the intent text can be further combined to determine whether the intent to be identified has a screen-questioning intent. Thus, it is possible to determine whether the intent to be identified has a screen-questioning intent through multiple methods, thereby improving the accuracy of determining whether the intent to be identified has a screen-questioning intent.

[0132] In some embodiments, such as Figure 9 As shown, the intent recognition method may specifically include: S901-S905, and the above-mentioned S804 can be implemented by S904 or S905.

[0133] S901. Obtain the intent text of the intent to be identified.

[0134] S902. Match the intent text with the preset semantic rules.

[0135] S903. When it is determined that the intent text does not match the first-precision semantic rule, but the intent text matches the second-precision semantic rule, identify the type of the intent text.

[0136] Specifically, the descriptions of S901, S902, and S903 can be found in the steps described above, and will not be repeated here.

[0137] S904. When the type of intent text is related to screen prompts, determine the screen prompt intent of the intent to be identified based on the screen prompt words.

[0138] It should be noted that the types related to screen chat can include "casual chat" and "question and answer".

[0139] In one possible implementation, intent text can be identified through large language model analysis, and screen prompt words can be classified to determine the screen prompt words corresponding to the intent text.

[0140] It is understandable that, after determining through the above steps that the intent text may have a screen-questioning intent, it is possible to further determine whether the intent to be identified has a screen-questioning intent by identifying the screen-questioning prompt words corresponding to the intent text, based on the content indicated by the prompt words.

[0141] In some embodiments, screen prompts are used to classify the screen prompt scenarios to which the intent text belongs. The screen prompts include screen prompt voice examples and / or screen prompt voice descriptions for each screen prompt scenario.

[0142] In other words, the screen prompt words provide a detailed description of the corresponding screen prompting scenario. By determining the screen prompt words through a large language model, it is possible to more accurately identify whether the intent to be identified has a screen prompting intent.

[0143] For example, after the user inputs the intent to be recognized, the large language model can classify the intent (or intent text) to be recognized based on the meanings of multiple pre-set screen prompt words (screen prompt voice examples and / or screen prompt voice descriptions) to determine the screen prompt words corresponding to the intent to be recognized.

[0144] In one possible implementation, by inputting the screen prompt word corresponding to the determined intent text into a large language model, the reasoning ability of the large language model can be used to infer whether the intent text has a screen prompt intent.

[0145] For example, pre-defined screen prompts may include A, B, C, D, E, and F. Here, A indicates that the user's question (intent to be identified) concerns information about a specific object on the current screen, such as its name, attributes, location, purpose, or content. Questions typically begin with a demonstrative pronoun (e.g., "this") or a personal pronoun (e.g., "he"), such as: "What color is this cup?", "What breed is the cat on the far right?", "What kind of plant is this?", "What brand is this car?", "What animals are in the picture?", etc.; or, the question instructs the user to perform an action on a specific object on the screen, such as "Calculate this problem?", "Zoom in on this pattern?", etc.; or, the user inquires whether a person on the screen has participated in or experienced a certain activity, and this person may begin with a demonstrative pronoun (e.g., "this") or a personal pronoun (e.g., "he / she"), such as "Has he / she recently appeared in a movie?", "What awards has this scientist received?", etc. B indicates that the user's question or instruction is an encyclopedic or common-sense question, such as "What kind of waste is a plastic bag?" or "What kind of animal is a frog?". C indicates that the user wants to navigate to the address displayed on the screen. D indicates that the user's question is a tool instruction completely unrelated to the content displayed on the screen. E indicates that the user's question requires extracting and saving text from an image displayed on the screen. F indicates that the user's question requires image processing, image understanding, or image recognition operations on the image or photo currently displayed on the screen.

[0146] It is understandable that the meanings corresponding to the screen prompts B and D are clearly unrelated to the content displayed on the screen. Therefore, when the screen prompt corresponding to the intent text is B or D, it can be determined that the intent to be identified does not have a screen prompt intent. However, the meanings corresponding to the screen prompts A, C, E, and F are questions or related to the content displayed on the screen. Therefore, when the screen prompt corresponding to the intent text is A, C, E, or F, it can be determined that the intent to be identified may have a screen prompt intent.

[0147] S905. If the type of the intent text is unrelated to screen asking, determine that the intent to be identified does not have a screen asking intent.

[0148] In one possible implementation, if the type of intent text is not related to screen asking (e.g., "check the weather" or "set an alarm"), then it can be directly determined that the intent to be identified does not have screen asking intent.

[0149] In this embodiment of the disclosure, if it is determined that the type of the intent text is related to screen prompting, then the screen prompt word corresponding to the intent text can be further determined, thereby determining whether the intent to be identified has screen prompting intent based on the determined screen prompt word. In this way, it is possible to determine whether the intent to be identified has screen prompting intent through multiple methods, thereby improving the accuracy of determining whether the intent to be identified has screen prompting intent.

[0150] In some embodiments, such as Figure 10 As shown, the intent recognition method may specifically include: S1001-S1005, and the above-mentioned S904 can be implemented through S1004 and S1005.

[0151] S1001. Obtain the intent text of the intent to be identified.

[0152] S1002. Match the intent text with the preset semantic rules.

[0153] S1003. When it is determined that the intent text does not match the first precision semantic rule, but the intent text matches the second precision semantic rule, identify the type of the intent text.

[0154] Specifically, the descriptions of S1001, S1002, and S1003 can be found in the steps above, and will not be repeated here.

[0155] S1004. If the type of the intent text is screen-related and there is a history record before the intent to be identified, determine the contextual relationship between the intent to be identified and the history record.

[0156] In one possible implementation, the history could be the user's past conversations with the voice assistant, or the user's past voice recordings.

[0157] It should be noted that determining whether there is a history record before the intent to be identified is to determine whether the intent input by the current user corresponds to multiple rounds of historical session records. If the intent to be identified does not correspond to multiple rounds of historical session records, it can be directly determined that the intent to be identified has a screen-asking intent.

[0158] It can be understood that whether the intent to be identified corresponds to multiple rounds of historical conversation records means whether other voice input from the user was received within a preset time period before the electronic device received the intent to be identified. If other voice input from the user was received within the preset time period before the intent to be identified was received, then it can be determined that the intent to be identified corresponds to multiple rounds of historical conversation records (i.e., a historical record exists).

[0159] In one possible implementation, a large language model can be used to verify whether there is a contextual relationship between the multi-turn historical conversation records and the intent to be identified. If there is no contextual relationship, it can be directly determined that the intent to be identified has a screen-questioning intent.

[0160] In one possible implementation, if there is a contextual relationship between the multi-turn historical session records and the intent to be identified, it is possible to further determine what the pronouns in the intent text refer to. If the content referred to by the pronouns in the intent text appears in the multi-turn historical session records, it means that the pronoun refers to the content in the multi-turn historical session records, that is, the intent text (the intent to be identified) is unrelated to screen asking and does not have a screen asking intent.

[0161] S1005. When the intent to be identified does not have a contextual relationship with the history, determine that the intent to be identified has a screen-questioning intent based on the screen prompt words.

[0162] It is understandable that if the intent to be identified does not have a contextual relationship with the history, the intent to be identified can be determined directly based on the screen prompt words.

[0163] In this embodiment of the disclosure, if a historical record exists prior to the intent to be identified, it is further determined whether there is a contextual relationship between the intent to be identified and the historical record, thereby determining whether the intent to be identified has a screen-questioning intent. If there is no contextual relationship between the intent to be identified and the historical record, it can be directly determined that the intent to be identified has a screen-questioning intent based on the screen-questioning prompt. Thus, it is possible to determine whether the intent to be identified has a screen-questioning intent through multiple methods, thereby improving the accuracy of determining whether the intent to be identified has a screen-questioning intent.

[0164] In some embodiments, such as Figure 11 As shown, the intent recognition method may specifically include: S1101-S1105.

[0165] S1101. Obtain the intent text of the intent to be identified.

[0166] S1102. Match the intent text with the preset semantic rules.

[0167] S1103. When it is determined that the intent text does not match the first precision semantic rule, but the intent text matches the second precision semantic rule, identify the type of the intent text.

[0168] S1104. If the type of the intent text is screen-related and there is a history record before the intent to be identified, determine the contextual relationship between the intent to be identified and the history record.

[0169] Specifically, the descriptions of S1101, S1102, S1103 and S1104 can be found in the steps above, and will not be repeated here.

[0170] S1105. When the intent to be identified has a contextual relationship with the history, determine that the intent to be identified does not have a screen-asking intent.

[0171] It is understandable that when the intent to be identified has a contextual relationship with the history, it can be directly determined that the intent to be identified does not have a screen-asking intent.

[0172] In this embodiment of the disclosure, when the intent to be identified is contextually related to the history, the intent to be identified input by the current user can be considered as a continuation of the historical dialogue rather than a screen-asking voice. Therefore, it can be directly determined that the intent to be identified does not have a screen-asking intent.

[0173] For example, a photo of person A is displayed on the screen. A user can ask, "Who is he?" based on the photo of person A. Analyzing "Who is he?" reveals that the intent text consists of the content object referent "he" on the screen and the question referent "who is he?", which matches the first-precision semantic rule. Therefore, "Who is he?" can be input into a multimodal large model for screen-based question processing, so that the multimodal large model can answer: He is A, and he recently starred in the TV series "B".

[0174] Furthermore, if the user continues to ask, "How is this TV series?", analysis of "How is this TV series?" reveals that the intent text consists of the on-screen content referent "this," the question referent "how is," and other referents of "TV series," matching the second-precision semantic rules. Therefore, further analysis of "How is this TV series?" using a large language model (general intent recognition) is needed to determine if the intent of "How is this TV series?" is a question-and-answer intent. Then, the type of intent text can be further determined using a large language model (specific question-and-answer intent recognition) to determine if it is a question-and-answer related type. However, "How is this TV series?" also has application history (i.e., "Who is he?" and "He is A, and he recently starred in TV series 'B'"). Therefore, it is also necessary to determine whether there is a contextual relationship between "How is this TV series?" and "Who is he?" and / or "He is A, and he recently starred in TV series 'B'." This determines that the referent "this" refers to "TV series 'B'," and "TV series 'B'" appears in the historical dialogue. Therefore, it is finally determined that "How is this TV series?" does not belong to the question-and-answer intent category.

[0175] This disclosed embodiment can be divided into two stages. The first stage directly outputs the intent text as having a screen-questioning intent, or potentially having a screen-questioning intent (candidate screen-questioning intent), through a general intent recognition process (based on semantic rule matching and general intent recognition using a large language model). The second stage, for candidate screen-questioning intents, uses a large language model to perform dedicated screen-questioning recognition and multi-turn dialogue relevance verification. In this way, in the general intent recognition scenario of voice assistants, obvious screen-questioning intents can be quickly identified without affecting the response speed of other general intent recognitions, and less obvious and uncertain screen-questioning intents and multi-turn interference problems can be better handled.

[0176] This embodiment establishes semantic rules of varying precision. Text conforming to high-precision semantic rules (or conforming to low-precision semantic rules but receiving a preset operation associated with a screen-questioning intent) is directly identified as having a screen-questioning intent, while text not matching any semantic rules is identified as not having a screen-questioning intent. Furthermore, text conforming to low-precision semantic rules undergoes further screen-questioning-specific recognition and multi-round verification. The recognition results from the general recognition process and the matching results of the low-precision semantic rules are combined to determine whether the text might have a screen-questioning intent. Additionally, screen-questioning prompts from a large language model can be used to identify whether the text has a specific screen-questioning intent. Moreover, when it is determined that the text corresponds to a historical record, the large language model is used to further identify whether the text has a contextual relationship with the historical dialogue record, thereby determining whether the text has a screen-questioning intent.

[0177] In one possible implementation, such as Figure 12 The diagram shown is an application environment diagram of intent recognition provided in an embodiment of this disclosure. The application environment includes modules such as: main control module 1201, rule matching module 1202, general intent recognition module 1203, screen-based intent recognition module 1204, multi-turn dialogue verification module 1205, large language model 1206, and multimodal large model 1207.

[0178] The main control module 1201 is used to acquire the user's input intent to be recognized and to identify the corresponding intent text.

[0179] The rule matching module 1202 is used to match the intent text with preset semantic rules to determine the semantic rules (first-precision semantic rules or second-precision semantic rules) for matching the intent text.

[0180] The general intent recognition module 1203 is used to analyze the intent to be recognized (intent text) and determine the intent of the intent to be recognized when it is determined that the intent to be recognized does not have a screen-asking intent.

[0181] The screen-questioning intent recognition module 1204 is used to analyze the intent to be recognized (intent text) and determine the response content of the intent to be recognized when it is determined that the intent to be recognized has a screen-questioning intent.

[0182] The multi-turn dialogue verification module 1205 is used to determine whether other historical dialogues (historical records) have been received before the intent to be identified is received, thereby determining whether the intent to be identified has a screen-asking intent.

[0183] The large language model 1206 is used to analyze intent text, determine the type of intent text, or identify prompt words, and determine whether the intent to be identified has a prompt intent based on the prompt words.

[0184] The multimodal large model 1207 is used to analyze the intent to be identified (intent text) and determine the response content of the intent to be identified when it is determined that the intent to be identified does not have a screen-asking intent.

[0185] It should be noted that when implementing the intent recognition function (or intent recognition method), the electronic device may include some of the modules mentioned above (e.g., main control module 1201, rule matching module 1202, general intent recognition module 1203, screen inquiry intent recognition module 1204, and multi-turn dialogue verification module 1205); or, the electronic device may include all of the modules mentioned above (i.e., main control module 1201, rule matching module 1202, general intent recognition module 1203, screen inquiry intent recognition module 1204, multi-turn dialogue verification module 1205, large language model 1206, and multimodal large model 1207); or, the electronic device may include more modules than those mentioned above.

[0186] It is understood that, in order to achieve the above-mentioned functions, electronic devices include hardware structures and / or software modules corresponding to the execution of each function. Those skilled in the art should readily recognize that, based on the algorithmic steps of the examples described in conjunction with the embodiments of this disclosure, this disclosure can be implemented in hardware or a combination of hardware and computer software. Whether a function is executed in hardware or by computer software driving hardware depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this disclosure.

[0187] This disclosure embodiment can divide the data sending end and data receiving end into functional modules according to the above method embodiment. For example, each function can be divided into a separate functional module, or two or more functions can be integrated into one functional module. The integrated module can be implemented in hardware or software. It should be noted that the module division in this disclosure embodiment is illustrative and only represents one logical functional division. In actual implementation, there may be other division methods. The following description uses the example of dividing each functional module according to each function.

[0188] Figure 13 This is a schematic diagram of an intent recognition device provided in an embodiment of the present disclosure, applied to an electronic device. The intent recognition device 1300 can perform the methods described in the above embodiments. Figure 3 The intent recognition method shown. Figure 13 As shown, the intent recognition device 1300 includes an acquisition module 1301 and a processing module 1302.

[0189] The acquisition module 1301 is used to acquire the intent text of the intent to be identified;

[0190] The processing module 1302 is used to match the intent text with preset semantic rules, which are determined based on content object referents and / or question referents in the screen;

[0191] The processing module 1302 is also used to determine that the intent to be identified has a screen-asking intent when the intent text matches the preset semantic rules.

[0192] Optionally, the processing module 1302 is specifically used to determine that the intent to be identified has a screen-asking intent when the text structure of the intent text matches the preset text structure indicated by the preset semantic rules; the preset semantic rules include: a first precision semantic rule and a second precision semantic rule, the first precision semantic rule and the second precision semantic rule indicating different preset text structures.

[0193] Optionally, the preset semantic rules include: a first-precision semantic rule, which is used to constrain the intent text to consist of content object referents and question referents on the screen, and that there are no other intent texts before the intent text; the processing module 1302 is specifically used to determine that the intent to be identified has a screen question intent when the intent text matches the first-precision semantic rule.

[0194] Optionally, the preset semantic rules also include: a second-precision semantic rule, which is used to constrain the intent text to include at least the content object referent and the question referent in the screen, and the precision of the first-precision semantic rule indicating that the text has a questioning intent is higher than the precision of the second-precision semantic rule indicating that the text has a questioning intent; the processing module 1302 is also used to determine the questioning intent of the intent to be identified based on the matching result of the intent text and the second-precision semantic rule when it is determined that the intent text does not match the first-precision semantic rule.

[0195] Optionally, the processing module 1302 is specifically used to determine that the intent to be identified does not have a screen-asking intent when the intent text does not match the second-precision semantic rules;

[0196] Alternatively, processing module 1302 is specifically used to identify the type of intent text when the intent text matches the second-precision semantic rule;

[0197] The processing module 1302 is specifically used to determine the screen intent to be identified based on the type of intent text.

[0198] Optionally, the processing module 1302 is specifically used to determine the screen-questioning intent of the intent to be identified based on the screen-questioning prompt words when the type of the intent text is a screen-questioning related type.

[0199] Alternatively, processing module 1302 is specifically used to determine that the intent to be identified does not have a screen-questioning intent when the type of the intent text is unrelated to screen-questioning.

[0200] Optionally, screen prompts are used to classify the screen prompt scenarios to which the intent text belongs. Screen prompts include screen prompt voice examples and / or screen prompt voice descriptions for each screen prompt scenario.

[0201] Optionally, the processing module 1302 is specifically used to determine the contextual association between the intent to be identified and the historical records when there are historical records prior to the intent to be identified.

[0202] Processing module 1302 is specifically used to determine that the intent to be identified has a screen-questioning intent based on the screen-questioning prompt words when the intent to be identified has no contextual relationship with the history.

[0203] Alternatively, the processing module 1302 is further configured to determine that the intent to be identified does not have a screen-asking intent when the intent to be identified has a contextual relationship with the history.

[0204] Optionally, the processing module 1302 is specifically used to determine that the intent to be identified has a screen-questioning intent when the intent text matches the second-precision semantic rule and a preset operation associated with the screen-questioning intent is received.

[0205] In the case where the functions of the integrated modules described above are implemented in hardware, this disclosure provides a schematic diagram of the structure of an electronic device for the intent recognition device involved in the above embodiments. For example... Figure 14 As shown, the electronic device 1400 includes a memory 1401 and a processor 1402. Optionally, the electronic device 1400 may also include a communication interface 1403 and a bus 1404.

[0206] The memory 1401 may be a read-only memory (ROM) or other type of static storage device capable of storing static information and instructions, random access memory (RAM) or other type of dynamic storage device capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), disk storage medium or other magnetic storage device, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but is not limited thereto.

[0207] Processor 1402 may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with embodiments of this disclosure. Processor 1402 may be a central processing unit, a general-purpose processor, a digital signal processor, an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It may implement or execute various exemplary logic blocks, modules, and circuits described in conjunction with embodiments of this disclosure. Processor 1402 may also be a combination of functions implementing computational capabilities, such as a combination of one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0208] Communication interface 1403 is used to connect to other devices via a communication network. This communication network can be Ethernet, wireless access network, wireless local area network (WLAN), etc.

[0209] The 1404 bus can be an extended industry standard architecture (EISA) bus, etc. The 1404 bus can be divided into address bus, data bus, control bus, etc. For ease of representation, Figure 14 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0210] As one possible implementation, the memory 1401 can exist independently of the processor 1402. The memory 1401 can be connected to the processor 1402 via a bus 1404 and is used to store instructions or program code. When the processor 1402 calls and executes the instructions or program code stored in the memory 1401, it can implement the intent recognition method provided in the embodiments of this disclosure.

[0211] In another possible implementation, the memory 1401 can also be integrated with the processor 1402.

[0212] Some embodiments of this disclosure provide a computer-readable storage medium (e.g., a non-transitory computer-readable storage medium) storing computer program instructions that, when executed on a computer, cause the computer to perform an intent recognition method as described in any of the above embodiments.

[0213] For example, the aforementioned computer-readable storage media may include, but are not limited to: magnetic storage devices (e.g., hard disks, floppy disks, or magnetic tapes), optical disks (e.g., compact disks (CDs), digital versatile disks (DVDs), etc.), smart cards, and flash memory devices (e.g., erasable programmable read-only memory (EPROMs), cards, sticks, or key drives, etc.). The various computer-readable storage media described in this disclosure may represent one or more devices for storing information and / or other machine-readable storage media. The term "machine-readable storage media" may include, but is not limited to, wireless channels and various other media capable of storing, containing, and / or carrying instructions and / or data.

[0214] This disclosure provides a computer program product containing instructions that, when run on a computer, cause the computer to execute the intent recognition method of any of the above embodiments.

[0215] The above are merely specific embodiments of this disclosure, but the scope of protection of this disclosure is not limited thereto. Any changes or substitutions within the technical scope disclosed in this disclosure should be included within the scope of protection of this disclosure. Therefore, the scope of protection of this disclosure should be determined by the scope of the claims.< / whatever> < / whatever> < / end> < / begin> < / end> < / begin>

Claims

1. An intent recognition method, characterized in that, The method includes: Obtain the intent text of the intent to be identified; The intent text is matched with preset semantic rules, which are determined based on content object referents and / or question referents on the screen. If the intent text matches the preset semantic rule, the intent to be identified is determined to have a screen-asking intent.

2. The method according to claim 1, characterized in that, The step of determining that the intent to be identified has a screen-asking intent when the intent text matches the preset semantic rule includes: If the text structure of the intent text matches the preset text structure indicated by the preset semantic rule, it is determined that the intent to be identified has a screen-asking intent; the preset semantic rule includes: a first precision semantic rule and a second precision semantic rule, wherein the first precision semantic rule and the second precision semantic rule indicate different preset text structures.

3. The method according to claim 1, characterized in that, The preset semantic rules include: a first precision semantic rule, which is used to constrain the intent text to consist of content object referents and question referents on the screen, and that there are no other intent texts before the intent text; The step of determining that the intent to be identified has a screen-asking intent when the intent text matches the preset semantic rule includes: If the intent text is determined to match the first precision semantic rule, the intent to be identified is determined to have a screen-questioning intent.

4. The method according to claim 3, characterized in that, The preset semantic rules also include: a second precision semantic rule, which is used to constrain the intent text to include at least content object referents and question referents on the screen, and the precision of the first precision semantic rule indicating that the text has a questioning intent is higher than the precision of the second precision semantic rule indicating that the text has a questioning intent. The method further includes: If it is determined that the intent text does not match the first precision semantic rule, the screen-questioning intent of the intent to be identified is determined based on the matching result of the intent text and the second precision semantic rule.

5. The method according to claim 4, characterized in that, The step of determining the screen-questioning intent of the intent to be identified based on the matching result between the intent text and the second precision semantic rule includes: If the intent text does not match the second precision semantic rule, it is determined that the intent to be identified does not have a screen-asking intent; or, If the intent text matches the second precision semantic rule, the type of the intent text is identified; Based on the type of the intent text, the screen-questioning intent of the intent to be identified is determined.

6. The method according to claim 5, characterized in that, Determining the screen-reading intent of the intent to be identified based on the type of the intent text includes: If the type of the intent text is related to screen prompts, the screen prompt intent of the intent to be identified is determined based on the screen prompt words; or, If the type of the intent text is unrelated to screen asking, it is determined that the intent to be identified does not have a screen asking intent.

7. The method according to claim 6, characterized in that, The screen prompts are used to classify the screen prompt scenarios to which the intent text belongs. The screen prompts include screen prompt voice examples and / or screen prompt voice descriptions for each screen prompt scenario.

8. The method according to claim 6, characterized in that, The process of determining the intent to be identified based on the screen prompt words includes: If a historical record exists prior to the intent to be identified, determine the contextual association between the intent to be identified and the historical record; If the intent to be identified has no contextual relevance to the historical records, the intent to be identified is determined to have a screen-questioning intent based on the screen prompt words; or, If the intent to be identified has a contextual relationship with the history, it is determined that the intent to be identified does not have a screen-asking intent.

9. The method according to claim 4, characterized in that, The step of determining the screen-questioning intent of the intent to be identified based on the matching result between the intent text and the second precision semantic rule includes: If the intent text matches the second precision semantic rule and a preset operation associated with the screen-asking intent is received, it is determined that the intent to be identified has a screen-asking intent.

10. An electronic device, characterized in that, include: Memory and processor; Memory and processor are coupled; The memory is used to store instructions that can be executed by the processor; When the processor executes the instructions, it performs the method as described in any one of claims 1-9.

11. A computer program product, characterized in that, When the computer program product is run on a computer, it causes the computer to perform the method as described in any one of claims 1-9.

Images