Man-machine conversation method and system based on semantic recognition

Abstract

The invention relates to the technical field of natural language understanding, and discloses a man-machine conversation method and system based on semantic recognition. The specification degree of the current voice instruction is predicted through the dialogue historical data of the current user to obtain the specification prediction level, so that when the planning prediction level is higher, the expansion recognition strategies applied by the planning prediction level are fewer, semantic recognition is performed on the current voice instruction according to the target strategy corresponding to the planning prediction level, and the user experience is improved. The method comprises the steps of obtaining a semantic recognition result, generating instruction feedback information according to the semantic recognition result, performing normalization prediction on a voice instruction according to historical data of a user, and determining a target strategy according to an instruction specification degree, thereby realizing dynamic adjustment between the current voice instruction and a semantic recognition strategy. And the number of the expansion identification strategies is reduced under the condition that the planning prediction level is high, so that the computing power required for executing the expansion identification strategies is reduced, and the computing power waste of semantic identification is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of natural language understanding, and in particular to a human-computer dialogue method and system based on semantic recognition. Background Art

[0002] As an important branch of artificial intelligence, human-computer dialogue technology covers key technologies such as speech recognition (ASR, Automatic Speech Recognition), natural language understanding (NLU, Natural Language Understanding) and speech synthesis (TTS, Text To Speech). Among them, natural language understanding (NLU), as one of the core tasks of human-computer dialogue technology, plays a vital role in realizing the computer system's in-depth understanding of human language and intelligent response. In the process of natural language understanding, semantic recognition technology plays a vital role. This technology enables computer systems to decipher, interpret and respond to the meaning of human language by capturing patterns, semantic relationships and contextual information in language. Among them, in the classification of semantic recognition technology, it is usually divided into accurate semantic recognition and fuzzy semantic recognition. Accurate semantic recognition requires that the recognition object is accurate, clear and conforms to language specifications. It captures and parses all information in it through comprehensive analysis of the recognition object, including vocabulary, grammar, syntactic structure and deeper semantic and pragmatic meanings. Fuzzy semantic recognition expands the semantics of the recognition object to overcome semantic ambiguity, incorrect vocabulary, semantic omissions and other situations in language input, assists semantic understanding, and thus improves the robustness and flexibility of semantic recognition to a certain extent.

[0003] However, on the one hand, due to the need to semantically expand the recognition object, the computing power required for fuzzy semantic recognition is relatively large. On the other hand, traditional semantic recognition systems often use the same set of multiple semantic recognition strategies for different recognition objects. The singleness of this strategy leads to unnecessary waste of computing power when performing semantic recognition on standardized and clear recognition objects. Therefore, these two factors jointly lead to the problem of computing power waste in the semantic recognition process. Summary of the invention

[0004] In order to provide a basic understanding of some aspects of the disclosed embodiments, a brief summary is given below. The summary is not an extensive review, nor is it intended to identify key / critical components or delineate the scope of protection of these embodiments, but rather serves as a prelude to the detailed description that follows.

[0005] In view of the above-mentioned shortcomings of the prior art, the present application provides a human-computer dialogue method and system based on semantic recognition to reduce the waste of computing power in semantic recognition.

[0006] The present application provides a human-computer dialogue method based on semantic recognition, comprising: pre-classifying a plurality of preset semantic recognition strategies according to an extension type to obtain a standard recognition strategy and an extended recognition strategy, wherein the extension type is used to characterize whether the semantic recognition strategy extends the recognition object; in response to a current user inputting a current voice command, acquiring dialogue history data corresponding to the current user; predicting the instruction standardization degree of the current voice command according to the dialogue history data to obtain a standard prediction level, and determining a target strategy from the standard recognition strategy and the extended recognition strategy according to the standard prediction level, wherein the higher the standard prediction level, the fewer extended recognition strategies corresponding to the standard prediction level; taking the current voice command as a recognition object, performing semantic recognition on the recognition object according to the target strategy to obtain a semantic recognition result, and generating instruction feedback information corresponding to the current voice command according to the semantic recognition result.

[0007] In one embodiment of the present application, conversation history data corresponding to the current user is obtained, including at least one of the following: counting the number of human-computer conversations of the current user; counting the duration of the human-computer conversations of the current user; obtaining historical voice instructions input by the current user and instruction text information corresponding to the historical voice instructions, and determining the text accuracy corresponding to the instruction text information based on the historical voice instructions, wherein the instruction text information is obtained by performing text conversion on the historical voice instructions; obtaining historical voice instructions input by the current user and historical feedback information corresponding to the historical voice instructions, and determining the feedback accuracy corresponding to the historical feedback information based on the historical voice instructions.

[0008] In one embodiment of the present application, the degree of standardization of the current voice command is predicted based on the conversation history data to obtain a standard prediction level, including: matching the conversation history data according to a preset score matching relationship to obtain a prediction sub-score corresponding to the conversation history data; obtaining the current user score by calculating the prediction sub-scores corresponding to each of the conversation history data, and determining the standard prediction level based on a comparison result between the current user score and a preset score threshold, wherein the standard prediction level is positively correlated with the number of human-computer conversations, the duration of the human-computer conversations, the text accuracy, and the feedback accuracy.

[0009] In one embodiment of the present application, semantic recognition is performed on the identified object according to the target strategy to obtain a semantic recognition result, including: if the standard prediction level is less than the preset level threshold, the standard recognition strategy and the extended recognition strategy are used as the target strategy corresponding to the standard prediction level; semantic recognition is performed on the identified object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; if the semantic recognition confidence is greater than or equal to the first preset threshold, a semantic recognition result is generated according to the first intent label; if the semantic recognition confidence is less than the first preset threshold, and the semantic recognition confidence is greater than or equal to the second preset threshold, semantic recognition is performed on the identified object according to the extended recognition strategy to obtain a second intent label, and a semantic recognition result is generated according to the second intent label, wherein the first preset threshold is greater than the second preset threshold; if the semantic recognition confidence is less than the second preset threshold, the semantic recognition result is determined as a semantic recognition failure.

[0010] In one embodiment of the present application, the extended recognition strategy expands the recognition object in at least one of the following ways: obtaining a correction word library, wherein the correction word library includes error-prone speech and common texts corresponding to the error-prone speech; obtaining the words to be corrected in the recognition object by performing word segmentation processing on the recognition object, and matching the words from the correction word library according to the speech corresponding to the words to be corrected to obtain replacement texts corresponding to the words to be corrected; obtaining multiple replacement objects by splicing the replacement texts, and determining similar objects from each of the replacement objects according to the pronunciation similarity between the replacement objects and the recognition object, so as to replace the recognition object according to the similar objects; obtaining the historical voice instructions of the current user; and taking the historical voice instructions as new recognition objects to perform semantic recognition on the historical voice instructions.

[0011] In one embodiment of the present application, the method also includes: if the semantic recognition result is a semantic recognition failure, obtaining a guidance instruction database, wherein the guidance instruction database includes standard feature words, standard voice instructions, and the association relationship between the standard feature words and the standard voice instructions; obtaining the text field corresponding to the recognition object by performing domain classification on the recognition object, and determining the domain feature words corresponding to the text field from the recognition object; querying the guidance instruction database according to the matching result between the domain feature words and the standard feature words, and obtaining the associated voice instructions corresponding to the domain feature words; determining similar voice instructions from the associated voice instructions according to the text similarity between the associated voice instructions and the recognition object; generating instruction guidance information according to the similar voice instructions, wherein the instruction guidance information is used to recommend the similar voice instructions to the user.

[0012] In one embodiment of the present application, a first preset threshold and a second preset threshold are obtained in the following manner: a first accuracy rate and a second accuracy rate are obtained, wherein the first accuracy rate is greater than the second accuracy rate; semantic recognition is performed on a preset test sample set through the standard recognition strategy to obtain test results corresponding to a plurality of test samples and result confidences corresponding to each of the test results, wherein the test sample set is composed of the test samples; a first preset threshold is determined according to the first accuracy rate, so that the recognition accuracy rate corresponding to the first sample is greater than or equal to the first accuracy rate, wherein the first sample includes a test sample whose result confidence is greater than the first preset threshold, and the recognition accuracy rate corresponding to the first sample is obtained by comparing the test result corresponding to the first sample with the first sample; a second preset threshold is determined according to the second accuracy rate, so that the recognition accuracy rate corresponding to the second sample is greater than or equal to the second accuracy rate, wherein the second sample includes a test sample whose result confidence is greater than the second preset threshold, and the recognition accuracy rate corresponding to the second sample is obtained by comparing the test result corresponding to the second sample with the second sample.

[0013] In one embodiment of the present application, semantic recognition is performed on the identified object according to the target strategy to obtain a semantic recognition result, including: if the standard prediction level is greater than or equal to the preset level threshold, the standard recognition strategy is used as the target strategy corresponding to the standard prediction level; semantic recognition is performed on the identified object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; if the semantic recognition confidence is greater than or equal to the first preset threshold, a semantic recognition result is generated according to the first intent label; if the semantic recognition confidence is less than the first preset threshold, the semantic recognition result is determined as a semantic recognition failure.

[0014] In one embodiment of the present application, command feedback information corresponding to the current voice command is generated according to the semantic recognition result, including: obtaining an intention label instruction template, wherein the intention label instruction template is composed of multiple intention label fields; extracting the intention labels corresponding to each of the intention label fields from the semantic recognition result, and filling the intention labels into the intention label instruction template to obtain the current intention instruction; matching according to the current intention instruction to obtain a reply template corresponding to the current intention instruction, wherein the reply template is composed of multiple template fields; selecting a target recommended text from the preset recommended texts corresponding to each of the template fields, and filling the target recommended text into the reply template to obtain command feedback information.

[0015] The present application provides a human-computer dialogue system based on semantic recognition, including: a classification module, used to classify a plurality of preset semantic recognition strategies according to an extension type, and obtain a standard recognition strategy and an extended recognition strategy, wherein the extension type is used to characterize whether the semantic recognition strategy extends the recognition object; an acquisition module, used to respond to a current user inputting a current voice command, and obtain dialogue history data corresponding to the current user; a prediction module, used to predict the instruction standardization degree of the current voice command according to the dialogue history data, and obtain a standard prediction level, and determine a target strategy from the standard recognition strategy and the extended recognition strategy according to the standard prediction level, wherein the higher the standard prediction level, the fewer extended recognition strategies corresponding to the standard prediction level; a feedback module, used to take the current voice command as a recognition object, perform semantic recognition on the recognition object according to the target strategy, and obtain a semantic recognition result, and generate instruction feedback information corresponding to the current voice command according to the semantic recognition result.

[0016] Beneficial effects of this application:

[0017] The standardization degree of the current voice command is predicted through the current user's conversation history data to obtain the standardization prediction level. Therefore, when the planning prediction level is higher, the fewer extended recognition strategies are applied to the planning prediction level, so as to perform semantic recognition on the current voice command according to the target strategy corresponding to the planning prediction level, obtain the semantic recognition result, and generate command feedback information according to the semantic recognition result. In this way, the voice command is normatively predicted according to the user's historical data to determine the target strategy according to the standardization degree of the command, realize the dynamic adjustment between the current voice command and the semantic recognition strategy, and reduce the number of extended recognition strategies when the planning prediction level is high, thereby reducing the computing power required to execute the extended recognition strategy and reducing the computing power waste of semantic recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0018] Figure 1 It is a flowchart of a human-computer dialogue method based on semantic recognition in an embodiment of the present application;

[0019] Figure 2 It is a flowchart of a method for obtaining a voice command in an embodiment of the present application;

[0020] Figure 3 It is a flowchart of a method for converting a voice command into text in an embodiment of the present application;

[0021] Figure 4 It is a flowchart of a method for generating instruction feedback information in an embodiment of the present application;

[0022] Figure 5It is a flowchart of a voice playback method in an embodiment of the present application;

[0023] Figure 6 It is a schematic diagram of the structure of a guidance instruction database in an embodiment of the present application;

[0024] Figure 7 is a flowchart of another human-computer dialogue method based on semantic recognition in an embodiment of the present application;

[0025] Figure 8 It is a structural diagram of a human-computer dialogue system based on semantic recognition in an embodiment of the present application. DETAILED DESCRIPTION

[0026] The following describes the embodiments of the present invention through specific examples, and those skilled in the art can easily understand other advantages and effects of the present invention from the contents disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and the details in this specification can also be modified or changed in various ways based on different viewpoints and applications without departing from the spirit of the present invention. It should be noted that the following embodiments and sub-samples in the embodiments can be combined with each other without conflict.

[0027] It should be noted that the illustrations provided in the following embodiments are only schematic illustrations of the basic concept of the present invention, and thus the drawings only show components related to the present invention rather than being drawn according to the number, shape and size of components in actual implementation. In actual implementation, the type, quantity and proportion of each component may be changed arbitrarily, and the component layout may also be more complicated.

[0028] In the following description, numerous details are discussed to provide a more thorough explanation of the embodiments of the present invention. However, it is obvious to those skilled in the art that the embodiments of the present invention can be implemented without these specific details. In other embodiments, well-known structures and devices are shown in the form of block diagrams rather than in detail to avoid making the embodiments of the present invention difficult to understand.

[0029] The terms "first", "second", etc. in the specification and claims of the embodiments of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the terms used in this way can be interchanged where appropriate, so that the embodiments of the embodiments of the present disclosure described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions.

[0030] Unless otherwise stated, the term "plurality" means two or more.

[0031] In the embodiment of the present disclosure, the character " / " indicates that the preceding and following objects are in an "or" relationship. For example, A / B indicates: A or B.

[0032] The term "and / or" is a description of the association relationship between objects, indicating that three relationships can exist. For example, A and / or B means: A or B, or, A and B.

[0033] Combination Figure 1 As shown, the embodiment of the present disclosure provides a human-computer dialogue method based on semantic recognition, including:

[0034] Step S101, pre-classifying a plurality of preset semantic recognition strategies according to the extension type to obtain a standard recognition strategy and an extended recognition strategy;

[0035] Among them, the extension type is used to indicate whether the semantic recognition strategy extends the recognition object;

[0036] Step S102, in response to the current user inputting the current voice command, obtaining the conversation history data corresponding to the current user;

[0037] Step S103, predicting the standardization degree of the current voice command according to the dialogue history data, obtaining a standardization prediction level, and determining a target strategy from the standard recognition strategy and the extended recognition strategy according to the standardization prediction level;

[0038] Among them, the higher the normative prediction level, the fewer the extended identification strategies corresponding to the normative prediction level;

[0039] Step S104, taking the current voice command as a recognition object, performing semantic recognition on the recognition object according to the target strategy, obtaining a semantic recognition result, and generating command feedback information corresponding to the current voice command according to the semantic recognition result.

[0040] The human-computer dialogue method based on semantic recognition provided by the embodiment of the present disclosure is adopted to predict the standardization degree of the current voice command through the dialogue history data of the current user to obtain the standardization prediction level, so that when the planning prediction level is higher, the fewer the extended recognition strategies applied to the planning prediction level are, so as to perform semantic recognition on the current voice command according to the target strategy corresponding to the planning prediction level, obtain the semantic recognition result, and generate instruction feedback information according to the semantic recognition result. In this way, the voice command is normatively predicted according to the user's historical data to determine the target strategy according to the standardization degree of the command, so as to realize the dynamic adjustment between the current voice command and the semantic recognition strategy, and reduce the number of extended recognition strategies when the planning prediction level is high, thereby reducing the computing power required to execute the extended recognition strategy and reducing the computing power waste of semantic recognition.

[0041] Combination Figure 2, the embodiment of the present disclosure provides a method for obtaining a voice command, including: setting a voice input module, wherein the voice input module is provided with a microphone; user 1, user 2, and user 3 all input voices through the microphone, so that the voice input module performs preprocessing according to the input voice, for example, noise reduction, voice enhancement, frequency adjustment, etc.; saving the preprocessed voice in a preset format to obtain a voice command, wherein the voice command is a WAV (Waveform Audio File) format file with a sampling rate of 16k, and the voice command includes 1.WAV file, 2.WAV file, and 3.WAV file; performing voiceprint recognition on the voice command, and determining the current user corresponding to the voice command according to the voiceprint recognition result.

[0042] Combination Figure 3 As shown, an embodiment of the present disclosure provides a method for text conversion of voice commands, including: performing voice recognition on each voice command by calling a voice recognition interface provided by a supplier to convert the voice command into a text command, wherein 1. the text command corresponding to the WAV file is "adjust the air conditioner to 25 degrees Celsius", 2. the text command corresponding to the WAV file is "please turn on the lathe", and 3. the text command corresponding to the WAV file is "a little higher".

[0043] Optionally, obtaining the conversation history data corresponding to the current user includes: counting the number of human-computer conversations of the current user, and / or counting the duration of the human-computer conversations of the current user.

[0044] In some embodiments, the number of human-computer dialogues or the duration of human-computer dialogues can represent the current user's experience in using human-computer dialogues. If the number of human-computer dialogues is large or the duration of human-computer dialogues is long, it means that the current user has experience in using human-computer dialogues and is more likely to input voice commands in a planned manner.

[0045] In some embodiments, if the number of human-computer dialogues of the current user is greater than or equal to 500 times, the predicted sub-score corresponding to the number of human-computer dialogues is 90 points; if the number of human-computer dialogues of the current user is less than 500 times, and the number of human-computer dialogues of the current user is greater than or equal to 100 times, the predicted sub-score corresponding to the number of human-computer dialogues is 60 points; if the number of human-computer dialogues of the current user is less than 100 times, the predicted sub-score corresponding to the number of human-computer dialogues is 30 points.

[0046] Optionally, the conversation history data corresponding to the current user is obtained, including: obtaining historical voice commands input by the current user, instruction text information corresponding to the historical voice commands, and determining the text accuracy corresponding to the instruction text information based on the historical voice commands, wherein the instruction text information is obtained by converting the historical voice commands into text.

[0047] In some embodiments, if in the historical data, the current user has input 10 historical voice commands; based on natural language processing technology, context analysis, text conversion confidence, user feedback and other methods, it is determined that 8 historical voice commands have been accurately converted into correct text, then the text accuracy of the current user is calculated to be 80%.

[0048] In some embodiments, a preset text benchmark score is calculated based on the text accuracy to obtain a prediction sub-score corresponding to the text accuracy. For example, the current user has input 1,000 historical voice commands, of which 900 historical voice commands were accurately converted into correct text. The text accuracy of the current user is 90%, and the prediction sub-score corresponding to the text accuracy is 90 points.

[0049] Optionally, obtaining conversation history data corresponding to the current user includes: obtaining historical voice commands input by the current user, historical feedback information corresponding to the historical voice commands, and determining a feedback accuracy rate corresponding to the historical feedback information based on the historical voice commands.

[0050] In some embodiments, if the current user inputs the voice command "adjust the air conditioner temperature to 25 degrees" and the feedback information generated is "OK, the air conditioner has been adjusted to 25 degrees", it means that the feedback information corresponding to the voice command is correct information; if in the historical data, the current user has input 10 historical voice commands; according to natural language processing technology, context analysis, text conversion confidence, user feedback and other methods, it is determined that the feedback information corresponding to the 9 historical voice commands are all correct information, then according to the calculation, the feedback accuracy of the current user is 90%.

[0051] In some embodiments, a preset feedback benchmark score is calculated based on the feedback accuracy to obtain a predicted sub-score corresponding to the feedback accuracy. For example, the current user has input 1,000 historical voice commands, of which 500 of the feedback information corresponding to the historical voice commands are correct information. Then, the text accuracy of the current user is 50%, and the predicted sub-score corresponding to the text accuracy is 50 points.

[0052] Optionally, the degree of standardization of the current voice command is predicted based on the conversation history data to obtain a standard prediction level, including: matching the conversation history data according to a preset score matching relationship to obtain a prediction sub-score corresponding to the conversation history data; obtaining the current user score by calculating the prediction sub-scores corresponding to each conversation history data, and determining the standard prediction level based on a comparison result between the current user score and a preset score threshold, wherein the standard prediction level is positively correlated with the number of human-computer conversations, the duration of the human-computer conversations, the text accuracy, and the feedback accuracy, respectively.

[0053] In some embodiments, the score weights corresponding to the prediction sub-scores are obtained. For example, the weight of the number of human-computer dialogues is 0.2, the score weight corresponding to the text accuracy is 0.4, and the score weight corresponding to the feedback accuracy is 0.4; each prediction sub-score is weighted according to the score weight, and the current user score is obtained as (90×0.2)+(90×0.4)+(50×0.4)=79 points.

[0054] In some embodiments, the preset score threshold is 80 points, wherein, if the current user score is greater than or equal to 80 points, the standard prediction level is a high standard level, and if the current user score is less than 80 points, the planned prediction level is a low standard level.

[0055] In some embodiments, the preset score thresholds are 80 points and 50 points, where if the current user score is greater than or equal to 80 points, the standard prediction level is level 3, and the number of extended identification strategies corresponding to the standard prediction level is 0; if the current user score is less than 80 points and the current user score is greater than or equal to 50 points, the planning prediction level is level 2, and the number of extended identification strategies corresponding to the standard prediction level is 1; if the current user score is less than 50 points, the standard prediction level is level 1, and the number of extended identification strategies corresponding to the standard prediction level is 2.

[0056] Optionally, command feedback information corresponding to the current voice command is generated based on the semantic recognition result, including: obtaining an intent label command template, wherein the intent label command template consists of multiple intent label fields; extracting the intent labels corresponding to each intent label field from the semantic recognition result, and filling the intent labels into the intent label command template to obtain the current intent command; matching according to the current intent command to obtain a reply template corresponding to the current intent command, wherein the reply template consists of multiple template fields; selecting a target recommended text from the preset recommended texts corresponding to each template field, and filling the target recommended text into the reply template to obtain command feedback information.

[0057] In some embodiments, the intent tag instruction template is designed as "name-function-action-value", where name is the device intent field, function is the function intent field, action is the operation intent field, value is the numerical intent field, and - is a separator.

[0058] In some embodiments, the intent tag field also includes a position intent field position.

[0059] In some embodiments, if the user inputs a voice command of "turn on the air conditioner", the current intent command corresponding to the voice command is airConditioning-o-open-o, where the device intent field name=airConditioning represents air conditioning, the function intent field function=o represents an empty field value, the operation intent field action=open represents turning on, and the numerical intent field value=o represents an empty field value.

[0060] In some embodiments, if the user inputs a voice command of "adjust the air conditioning temperature to 25 degrees", the current intent command corresponding to the voice command is airConditioning-temperature-set-temperature_value, where the device intent field name=airConditioning represents air conditioning, the function intent field function=temperature represents temperature, the operation intent field action=set represents setting, and the numerical intent field value=temperature_value represents the temperature value.

[0061] Combination Figure 4 As shown, the embodiment of the present disclosure provides a method for generating instruction feedback information, including:

[0062] Step S401, performing semantic recognition on the recognition object to obtain a semantic recognition result;

[0063] Step S402, extracting the intent labels corresponding to the intent label fields from the semantic recognition results, and filling the intent labels into the intent label instruction template to obtain the current intent instruction;

[0064] Step S403, matching according to the current intention instruction to obtain a reply template corresponding to the current intention instruction;

[0065] Among them, the reply template consists of multiple template fields;

[0066] Step S404, selecting a target recommended text from the preset recommended texts corresponding to each template field, and filling the target recommended text into the reply template to obtain instruction feedback information.

[0067] In some embodiments, if the identification object is "adjust the height of the driver's window to 25%", the current intention instruction includes name = window, function = height, action = adjust to, position = driver's seat, value = 25%; according to the current intention instruction, a match is performed from a preset template database to obtain a reply template corresponding to the current intention instruction "OK, I'll adjust the height of the driver's window to 25% for you right away", wherein the template fields in the reply template are divided into auxiliary feedback words and intention feedback words, the auxiliary feedback words include "OK", "right away for you", "Yeah", and the intention feedback words include "window", "height", "adjust to", "driver's seat", "25%"; obtain the preset recommended texts corresponding to each template field, for example, the preset recommended texts corresponding to "OK" include ["OK", "Here I am", "Great", "I'm here too"], and the preset recommended texts corresponding to "right away for you" include ["Immediately for you", "right here for you", "immediately for you", "immediately for you", "right away for you"], the preset recommended texts corresponding to "Yeah" include ["ha", "wow", "la"], the preset recommended texts corresponding to "car window" include ["window", "window", "window glass", "car window"], the preset recommended texts corresponding to "height" include ["height", "high"], the preset recommended texts corresponding to "adjust to" include ["adjust to", "adjust to", "adjust to", "adjust"], and the preset recommended texts corresponding to "main driver" include ["driving seat", "main driver", "cab", "driver"], etc.; random selection is performed according to the preset recommended texts to form 3 sentences to be selected, including "OK, I'll adjust the car window height for you right away", "Okay, I'll adjust the window height a little higher for you right away", "Here, I'll adjust the window height for you a little bit right away, wow"; a text to be selected is randomly selected as the command feedback information.

[0068] Combination Figure 5 As shown, the embodiment of the present disclosure provides a voice playback method, including:

[0069] Step S501, obtaining instruction feedback information;

[0070] Step S502, preprocessing the instruction feedback information;

[0071] Among them, information preprocessing includes judging grammatical correctness and writing-to-reading method;

[0072] Step S503, converting the pinyin corresponding to the instruction feedback information into phonemes;

[0073] Step S504, synthesizing the phonemes according to the speech synthesis model to obtain synthesized speech corresponding to the instruction feedback information, and playing the synthesized speech.

[0074] Optionally, the first preset threshold and the second preset threshold are obtained in the following manner: a first accuracy rate and a second accuracy rate are obtained, wherein the first accuracy rate is greater than the second accuracy rate; semantic recognition is performed on a preset test sample set through a standard recognition strategy to obtain test results corresponding to multiple test samples and result confidences corresponding to each test result, wherein the test sample set is composed of test samples; a first preset threshold is determined according to the first accuracy rate, so that the recognition accuracy rate corresponding to the first sample is greater than or equal to the first accuracy rate, wherein the first sample includes a test sample whose result confidence is greater than the first preset threshold, and the recognition accuracy rate corresponding to the first sample is obtained by comparing the test results corresponding to the first sample with the first sample; a second preset threshold is determined according to the second accuracy rate, so that the recognition accuracy rate corresponding to the second sample is greater than or equal to the second accuracy rate, wherein the second sample includes a test sample whose result confidence is greater than the second preset threshold, and the recognition accuracy rate corresponding to the second sample is obtained by comparing the test results corresponding to the second sample with the second sample.

[0075] In some embodiments, 100 test samples are obtained, and the first accuracy rate is set to 100% and the second accuracy rate is set to 80%; semantic recognition is performed on each test sample through a standard recognition strategy to obtain test results and result confidence levels, wherein when the result confidence level is greater than 0.9, the recognition accuracy rate of the standard recognition strategy reaches 100%, therefore, 0.9 is used as the first preset threshold; when the result confidence level is greater than 0.6, the recognition accuracy rate of the standard recognition strategy reaches 80%, therefore, 0.6 is used as the second preset threshold.

[0076] Optionally, semantic recognition is performed on the identified object according to the target strategy to obtain a semantic recognition result, including: if the normative prediction level is greater than or equal to a preset level threshold, the standard recognition strategy is used as the target strategy corresponding to the normative prediction level; semantic recognition is performed on the identified object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; if the semantic recognition confidence is greater than or equal to the first preset threshold, a semantic recognition result is generated according to the first intent label; if the semantic recognition confidence is less than the first preset threshold, the semantic recognition result is determined as a semantic recognition failure.

[0077] In some embodiments, the standard recognition strategy is trained based on the Bert text classification model. The Bert text classification model is a pre-trained model built on the encoder part of the Transformer architecture. It has mastered rich language representation capabilities through deep training on large-scale unlabeled text data. It is fine-tuned through intent classification samples to obtain an intent classification model, thereby achieving accurate semantic recognition of the recognition object.

[0078] In some embodiments, if the specification prediction level is a high specification level, semantic recognition is performed on the recognition object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; if the semantic recognition confidence is greater than or equal to 0.9, the intent label instruction template is directly filled in according to the intent label obtained by semantic recognition to obtain the current intent instruction; if the semantic recognition confidence is less than 0.9, the semantic recognition result is determined as a semantic recognition failure, and the user is prompted with feedback information "I don't quite understand, please re-enter the instruction."

[0079] In some embodiments, if the user inputs a voice command of "adjust the air conditioner to 25 degrees", the voice command is semantically recognized and the current intention command corresponding to the voice command is obtained as "airConditioning-temperature-set-temperature_value", and the semantic recognition confidence is 0.98, then feedback information is generated according to the current intention command.

[0080] Optionally, semantic recognition is performed on the identified object according to the target strategy to obtain a semantic recognition result, including: if the standard prediction level is less than a preset level threshold, the standard recognition strategy and the extended recognition strategy are used as the target strategy corresponding to the standard prediction level; semantic recognition is performed on the identified object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; if the semantic recognition confidence is greater than or equal to the first preset threshold, a semantic recognition result is generated according to the first intent label; if the semantic recognition confidence is less than the first preset threshold, and the semantic recognition confidence is greater than or equal to the second preset threshold, semantic recognition is performed on the identified object according to the extended recognition strategy to obtain a second intent label, and a semantic recognition result is generated according to the second intent label, wherein the first preset threshold is greater than the second preset threshold; if the semantic recognition confidence is less than the second preset threshold, the semantic recognition result is determined as a semantic recognition failure.

[0081] In some embodiments, if the standard prediction level is a low standard level, semantic recognition is performed on the recognition object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; if the semantic recognition confidence is greater than or equal to 0.9, the intent label instruction template is directly filled in according to the intent label obtained by semantic recognition to obtain the current intent instruction; if the semantic recognition confidence is less than 0.9, and the semantic recognition confidence is greater than or equal to 0.6, semantic recognition is performed on the recognition object according to the extended recognition strategy to obtain a second intent label, and a semantic recognition result is generated according to the second intent label, wherein the extended semantic recognition includes a text error correction strategy and a context extension strategy; if the semantic recognition confidence is less than 0.6, similar voice instructions are recommended to the user.

[0082] Optionally, the extended recognition strategy expands the recognition object in the following ways: obtaining a correction word library, wherein the correction word library includes easily-mistaken speech and commonly used texts corresponding to the easily-mistaken speech; obtaining the words to be corrected in the recognition object by performing word segmentation processing on the recognition object, and matching them from the correction word library according to the word speech corresponding to the words to be corrected, to obtain replacement texts corresponding to the words to be corrected; obtaining multiple replacement objects by splicing the replacement texts, and determining similar objects from each replacement object according to the pronunciation similarity between the replacement object and the recognition object, so as to replace the recognition object according to the similar objects.

[0083] In some embodiments, error-prone speech and common text corresponding to the error-prone speech are stored in a Redis database to form an error correction word library, for example, error-prone pronunciation key = "chechuang", common text value = ["car window", ...]; useless words in the recognition object are deleted, and the recognition object after the useless words are deleted is subjected to binary word segmentation to obtain the words to be corrected, and the word speech corresponding to the words to be corrected is obtained, for example, the recognition object is "please open the lathe", the useless word "please" in the recognition object is deleted, and binary word segmentation is performed on "open the lathe" to obtain the words to be corrected ["open", "drive", "car window"], and they are converted into word speech ["dakai", "kaiche", "chechuang"]; according to each word speech, a query is performed in the error correction word library to obtain a replacement text, for example, "d The replacement text corresponding to "akai" is "open", the replacement text corresponding to "kaiche" is "drive", and the replacement text corresponding to "chechuang" is "car window"; the replacement text is used to obtain a new replacement object, for example, the replacement objects include 3, namely "open and drive", "drive car window", and "open car window"; the voice similarity between the replacement object and the recognition object is calculated, that is, the pinyin similarity between the two is calculated, for example, the voice similarity between "open and drive" and "open lathe" is 0.5, the voice similarity between "drive car window" and "open lathe" is 0, and the voice similarity between "open car window" and "open lathe" is 0.75; according to the voice similarity, "open car window" is determined as a similar object, and "open car window" replaces the recognition object, so as to obtain the corresponding semantic recognition result according to "open car window".

[0084] Optionally, the extended recognition strategy expands the recognition object in the following manner: acquiring historical voice commands of the current user; and using the historical voice commands as new recognition objects to perform semantic recognition on the historical voice commands.

[0085] In some embodiments, if the current user inputs a voice command of "a little higher", the historical voice command input by the current user last time is obtained, wherein the historical voice command includes "adjust the air conditioning temperature to 25 degrees Celsius"; semantic recognition is performed based on the historical voice command to obtain an intent label, wherein the intent includes "name" = "airConditioning", "function" = "temperature"; the intent label of the historical semantic instruction is combined with the current voice command, and the current intent instruction corresponding to the current voice command is "adjust the air conditioning temperature a little higher".

[0086] Optionally, the method also includes: if the semantic recognition result is a semantic recognition failure, obtaining a guidance instruction database, wherein the guidance instruction database includes standard feature words, standard voice instructions, and the association relationship between standard feature words and standard voice instructions; obtaining the text field corresponding to the recognition object by performing domain classification on the recognition object, and determining the domain feature words corresponding to the text field from the recognition object; querying the guidance instruction database according to the matching results between the domain feature words and the standard feature words, and obtaining the associated voice instructions corresponding to the domain feature words; determining similar voice instructions from the associated voice instructions according to the text similarity between the associated voice instructions and the recognition object; generating instruction guidance information according to the similar voice instructions, wherein the instruction guidance information is used to recommend similar voice instructions to users.

[0087] In some embodiments, a standard instruction library is established according to different text fields, wherein a part of the standard instruction library is shown in Table 1; the standard instruction library is stored in the knowledge graph to form a guide instruction database, wherein the structure of the guide instruction database is as shown in Figure 6 As shown; at the same time, the human-computer dialogue log is analyzed periodically, and the knowledge graph is updated according to the standard voice commands with higher dialogue frequency.

[0088] Table 1

[0089]

[0090]

[0091] In some embodiments, if the recognized object is "air conditioning a little bit", the recognized object is classified into a field and the text field is obtained as "airConditioning", and the field feature word corresponding to the text field, namely "air conditioning" is determined from the recognized object; the guidance instruction database is queried according to the field feature word "air conditioning" to obtain the associated voice instructions corresponding to the field feature word, and the associated voice instructions include "adjust the air conditioning temperature a little bit", "adjust the air conditioning temperature by 2 degrees Celsius", "turn on the air conditioner", "increase the air volume of the air conditioner", etc.; similar voice instructions are determined from the associated voice instructions according to the text similarity between the associated voice instructions and the recognized object, among which the similar voice instruction corresponding to "air conditioning a little bit" is "adjust the air conditioning temperature a little bit"; instruction guidance information is generated according to the similar voice instructions, for example, a guidance prefix "I don't seem to understand, you can say", and then the similar voice instruction "adjust the air conditioning temperature a little bit" is spliced ​​to obtain the instruction guidance information as "I don't seem to understand, you can say: adjust the air conditioning temperature a little bit"; the instruction guidance information is pushed to the user.

[0092] Combination Figure 7 As shown, the embodiment of the present disclosure provides a human-computer dialogue method based on semantic recognition, including:

[0093] Step S701, obtaining the current voice command of the current user, and converting the current voice command into the current text command;

[0094] Step S702, obtaining the historical conversation data of the current user;

[0095] Step S703, predicting the standardization degree of the current voice command according to the dialogue history data to obtain a standardization prediction level;

[0096] Step S704, determine whether the standard prediction level is greater than the preset level threshold, if so, jump to step S705, if not, jump to step S709;

[0097] Step S705, performing semantic recognition on the recognition object according to the standard recognition strategy to obtain a first intent label and semantic recognition confidence;

[0098] Step S706, determining whether the semantic recognition confidence is greater than or equal to a first preset threshold, if so, jump to step S707, if not, jump to step S708;

[0099] Step S707, generating a semantic analysis result according to the first intention tag, and generating instruction feedback information corresponding to the current voice instruction according to the semantic recognition result.

[0100] Step S708: determining the semantic recognition result as semantic recognition failure.

[0101] Step S709, performing semantic recognition on the recognition object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence;

[0102] Step S710, determining whether the semantic recognition confidence is greater than or equal to a first preset threshold, if so, jump to step S707, if not, jump to step S711;

[0103] Step S711, determining whether the semantic recognition confidence is greater than or equal to a second preset threshold, if so, jump to step S712, if not, jump to step S714;

[0104] Wherein, the first preset threshold is greater than the second preset threshold;

[0105] Step S712, performing semantic recognition on the recognition object according to the extended recognition strategy to obtain a second intent label;

[0106] Step S713, generating a semantic recognition result according to the second intention tag, so as to generate instruction feedback information corresponding to the current voice instruction according to the semantic recognition result;

[0107] Step S714: recommend similar voice commands to the user.

[0108] The human-computer dialogue method based on semantic recognition provided by the embodiment of the present disclosure is adopted to predict the standardization degree of the current voice command through the dialogue history data of the current user to obtain the standardization prediction level, so that when the planning prediction level is higher, the fewer the extended recognition strategies applied to the planning prediction level are, so as to perform semantic recognition on the current voice command according to the target strategy corresponding to the planning prediction level, obtain the semantic recognition result, and generate instruction feedback information according to the semantic recognition result. In this way, the voice command is normatively predicted according to the user's historical data to determine the target strategy according to the standardization degree of the command, so as to realize the dynamic adjustment between the current voice command and the semantic recognition strategy, and reduce the number of extended recognition strategies when the planning prediction level is high, thereby reducing the computing power required to execute the extended recognition strategy and reducing the computing power waste of semantic recognition.

[0109] Combination Figure 8 As shown, the embodiment of the present disclosure provides a human-computer dialogue system based on semantic recognition, including a classification module 801 , an acquisition module 802 , a prediction module 803 and a feedback module 804 .

[0110] The classification module 801 is used to classify the preset multiple semantic recognition strategies according to the extension type to obtain the standard recognition strategy and the extended recognition strategy, wherein the extension type is used to indicate whether the semantic recognition strategy extends the recognition object.

[0111] The acquisition module 802 is used to acquire the conversation history data corresponding to the current user in response to the current user inputting the current voice command.

[0112] The prediction module 803 is used to predict the degree of standardization of the current voice command based on the dialogue history data, obtain the standard prediction level, and determine the target strategy from the standard recognition strategy and the extended recognition strategy based on the standard prediction level. The higher the standard prediction level, the fewer extended recognition strategies corresponding to the standard prediction level.

[0113] The feedback module 804 is used to take the current voice instruction as a recognition object, perform semantic recognition on the recognition object according to the target strategy, obtain a semantic recognition result, and generate instruction feedback information corresponding to the current voice instruction according to the semantic recognition result.

[0114] The human-computer dialogue system based on semantic recognition provided by the embodiment of the present disclosure is adopted to predict the standardization degree of the current voice command through the dialogue history data of the current user to obtain the standardization prediction level, so that when the planning prediction level is higher, the fewer the extended recognition strategies applied to the planning prediction level are, so as to perform semantic recognition on the current voice command according to the target strategy corresponding to the planning prediction level, obtain the semantic recognition result, and generate instruction feedback information according to the semantic recognition result. In this way, the voice command is predicted to be normative according to the user's historical data, so as to determine the target strategy according to the standardization degree of the command, realize the dynamic adjustment between the current voice command and the semantic recognition strategy, and reduce the number of extended recognition strategies when the planning prediction level is high, thereby reducing the computing power required to execute the extended recognition strategy and reducing the computing power waste of semantic recognition.

[0115] The above description and the accompanying drawings fully illustrate the embodiments of the present disclosure so that those skilled in the art can practice them. Other embodiments may include structural, logical, electrical, process and other changes. The embodiments represent possible changes only. Unless explicitly required, separate components and functions are optional, and the order of operation can be changed. Parts and sub-samples of some embodiments may be included in or replace parts and sub-samples of other embodiments. Moreover, the words used in this application are only used to describe the embodiments and are not used to limit the claims. As used in the description of the embodiments and the claims, unless the context clearly indicates, the singular forms of "a", "an" and "the" are intended to include plural forms as well. Similarly, the term "and / or" as used in this application refers to any and all possible combinations of listings containing one or more associated ones. In addition, when used in the present application, the term "comprise" and its variants "comprises" and / or comprising refer to the presence of a stated sub-sample, whole, step, operation, element, and / or component, but do not exclude the presence or addition of one or more other sub-samples, wholes, steps, operations, elements, components and / or groups of these. In the absence of further restrictions, the elements defined by the sentence "comprising a ..." do not exclude the presence of other identical elements in the process, method or device comprising the elements. In this article, each embodiment may focus on the differences from other embodiments, and the same and similar parts between the various embodiments may refer to each other. For the methods, products, etc. disclosed in the embodiments, if they correspond to the method part disclosed in the embodiments, then the relevant parts can refer to the description of the method part.

[0116] Those skilled in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are performed in hardware or software may depend on the specific application and design constraints of the technical solution. Technicians can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the embodiments of the present disclosure. Technicians can clearly understand that for the convenience and simplicity of description, the specific working process of the above-described system, device and unit can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0117] In the embodiments disclosed herein, the disclosed methods and products (including but not limited to devices, equipment, etc.) can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of units can be only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some sub-samples can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between each other shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms. 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 units may be selected according to actual needs to implement this embodiment. In addition, each functional unit in the embodiment of the present disclosure may be integrated in a processing unit, or each unit may exist physically alone, or two or more units may be integrated in one unit.

[0118] The flowchart and block diagram in the accompanying drawings show the possible architecture, functions and operations of the system, method and computer program product according to the embodiment of the present disclosure. In this regard, each box in the flowchart or block diagram can represent a module, a program segment or a part of the code, and a part of the module, program segment or code contains one or more executable instructions for realizing the specified logical function. In some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two consecutive boxes can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. In the description corresponding to the flowchart and block diagram in the accompanying drawings, the operations or steps corresponding to different boxes can also occur in a different order from the order disclosed in the description, and sometimes there is no specific order between different operations or steps. For example, two consecutive operations or steps can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, which can depend on the functions involved. Each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, may be implemented by a dedicated hardware-based system that performs the specified functions or actions, or may be implemented by a combination of dedicated hardware and computer instructions.

Claims

1. A human-computer dialogue method based on semantic recognition, characterized in that: include: Pre-classifying a plurality of preset semantic recognition strategies according to the extension type to obtain a standard recognition strategy and an extended recognition strategy, wherein the extension type is used to characterize whether the semantic recognition strategy extends the recognition object; In response to a current user inputting a current voice command, acquiring conversation history data corresponding to the current user; Predicting the instruction standardization degree of the current voice instruction according to the conversation history data to obtain a standard prediction level, and determining a target strategy from the standard recognition strategy and the extended recognition strategy according to the standard prediction level, wherein the higher the standard prediction level, the fewer extended recognition strategies corresponding to the standard prediction level; The current voice instruction is taken as a recognition object, semantic recognition is performed on the recognition object according to the target strategy to obtain a semantic recognition result, and instruction feedback information corresponding to the current voice instruction is generated according to the semantic recognition result.

2. The method according to claim 1, characterized in that Get the conversation history data corresponding to the current user, including at least one of the following: Counting the number of human-computer dialogues of the current user; Counting the duration of the human-computer dialogue of the current user; Acquire the historical voice instructions input by the current user and the instruction text information corresponding to the historical voice instructions, and determine the text accuracy corresponding to the instruction text information according to the historical voice instructions, wherein the instruction text information is obtained by performing text conversion on the historical voice instructions; Acquire the historical voice instructions input by the current user and the historical feedback information corresponding to the historical voice instructions, and determine the feedback accuracy corresponding to the historical feedback information according to the historical voice instructions.

3. The method according to claim 2, characterized in that Predicting the degree of standardization of the current voice command according to the conversation history data to obtain a standardization prediction level includes: Matching the conversation history data according to a preset score matching relationship to obtain a predicted sub-score corresponding to the conversation history data; The current user score is obtained by calculating the prediction sub-scores corresponding to each of the conversation history data, and the standard prediction level is determined based on the comparison result between the current user score and the preset score threshold, wherein the standard prediction level is positively correlated with the number of human-computer conversations, the duration of the human-computer conversations, the text accuracy, and the feedback accuracy.

4. The method according to claim 1, characterized in that Performing semantic recognition on the recognition object according to the target strategy to obtain a semantic recognition result includes: If the normative prediction level is less than the preset level threshold, the standard recognition strategy and the extended recognition strategy are used as target strategies corresponding to the normative prediction level; Performing semantic recognition on the recognition object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; If the semantic recognition confidence is greater than or equal to a first preset threshold, generating a semantic recognition result according to the first intent label; If the semantic recognition confidence is less than the first preset threshold, and the semantic recognition confidence is greater than or equal to the second preset threshold, semantic recognition is performed on the recognition object according to the extended recognition strategy to obtain a second intent label, and a semantic recognition result is generated according to the second intent label, wherein the first preset threshold is greater than the second preset threshold; If the semantic recognition confidence is less than the second preset threshold, the semantic recognition result is determined as a semantic recognition failure.

5. The method according to claim 4, characterized in that The extended recognition strategy expands the recognition object in at least one of the following ways: Acquire a correction word library, wherein the correction word library includes error-prone speech and common texts corresponding to the error-prone speech; obtain words to be corrected in the recognition object by performing word segmentation processing on the recognition object, and match the words to be corrected from the correction word library according to the speech corresponding to the words to be corrected to obtain replacement texts corresponding to the words to be corrected; obtain multiple replacement objects by splicing the replacement texts, and determine similar objects from each of the replacement objects according to the pronunciation similarity between the replacement objects and the recognition object, so as to replace the recognition object according to the similar objects; Acquire the historical voice commands of the current user; and use the historical voice commands as new recognition objects to perform semantic recognition on the historical voice commands.

6. The method according to claim 4, characterized in that The method further comprises: If the semantic recognition result is semantic recognition failure, obtaining a guidance instruction database, wherein the guidance instruction database includes standard feature words, standard voice instructions, and an association relationship between the standard feature words and the standard voice instructions; By performing field classification on the recognition object, a text field corresponding to the recognition object is obtained, and field feature words corresponding to the text field are determined from the recognition object; According to the matching result between the domain feature word and the standard feature word, the guidance instruction database is queried to obtain the associated voice instruction corresponding to the domain feature word; determining similar voice instructions from the associated voice instructions according to text similarity between the associated voice instructions and the recognition object; Instruction guidance information is generated according to the similar voice instructions, wherein the instruction guidance information is used to recommend the similar voice instructions to the user.

7. The method according to claim 4, characterized in that The first preset threshold and the second preset threshold are obtained in the following manner: Obtaining a first accuracy rate and a second accuracy rate, wherein the first accuracy rate is greater than the second accuracy rate; Performing semantic recognition on a preset test sample set by using the standard recognition strategy to obtain test results corresponding to a plurality of test samples and result confidences corresponding to each of the test results, wherein the test sample set is composed of the test samples; Determining a first preset threshold value according to the first accuracy rate, so that the recognition accuracy rate corresponding to the first sample is greater than or equal to the first accuracy rate, wherein the first sample includes a test sample whose result confidence is greater than the first preset threshold value, and the recognition accuracy rate corresponding to the first sample is obtained by comparing the first sample with the test result corresponding to the first sample; A second preset threshold is determined based on the second accuracy rate, so that the recognition accuracy rate corresponding to the second sample is greater than or equal to the second accuracy rate, wherein the second sample includes a test sample whose result confidence is greater than the second preset threshold value, and the recognition accuracy rate corresponding to the second sample is obtained by comparing the second sample with the test result corresponding to the second sample.

8. The method according to claim 1, characterized in that Performing semantic recognition on the recognition object according to the target strategy to obtain a semantic recognition result includes: If the normative prediction level is greater than or equal to a preset level threshold, the standard identification strategy is used as the target strategy corresponding to the normative prediction level; Performing semantic recognition on the recognition object according to the standard recognition strategy to obtain a first intent label and a semantic recognition confidence; If the semantic recognition confidence is greater than or equal to a first preset threshold, generating a semantic recognition result according to the first intent label; If the semantic recognition confidence is less than the first preset threshold, the semantic recognition result is determined as a semantic recognition failure.

9. The method according to any one of claims 1 to 8, characterized in that: Generating command feedback information corresponding to the current voice command according to the semantic recognition result, including: Obtaining an intent tag instruction template, wherein the intent tag instruction template is composed of a plurality of intent tag fields; Extracting the intent labels corresponding to the intent label fields from the semantic recognition result, and filling the intent labels into the intent label instruction template to obtain the current intent instruction; Matching according to the current intention instruction to obtain a reply template corresponding to the current intention instruction, wherein the reply template is composed of multiple template fields; A target recommended text is selected from the preset recommended texts corresponding to the template fields, and the target recommended text is filled into the reply template to obtain instruction feedback information.

10. A human-computer dialogue system based on semantic recognition, characterized in that: include: A classification module, used to classify the preset multiple semantic recognition strategies according to the extension type to obtain the standard recognition strategy and the extended recognition strategy, wherein the extension type is used to characterize whether the semantic recognition strategy extends the recognition object; An acquisition module, configured to acquire the conversation history data corresponding to the current user in response to the current user inputting the current voice command; A prediction module, configured to predict the degree of instruction standardization of the current voice instruction according to the conversation history data, obtain a standard prediction level, and determine a target strategy from the standard recognition strategy and the extended recognition strategy according to the standard prediction level, wherein the higher the standard prediction level, the fewer extended recognition strategies corresponding to the standard prediction level; The feedback module is used to take the current voice instruction as a recognition object, perform semantic recognition on the recognition object according to the target strategy, obtain a semantic recognition result, and generate instruction feedback information corresponding to the current voice instruction based on the semantic recognition result.

Images