Sentence processing method and device, electronic equipment and readable storage medium

Abstract

The application provides a sentence processing method and device, electronic equipment and a readable storage medium. When a server for sentence recognition set by an operator acquires voice data to be processed, a first machine learning model or a second machine learning model with different calculation amounts is selected according to the intelligibility of the voice data to be processed, and the sentences in the voice data to be processed are processed by word segmentation and part-of-speech tagging, so as to ensure the accuracy of processing the sentences while improving the speed and efficiency of processing the sentences.

Description

Technical Field

[0001] This application relates to the field of Natural Language Processing (NLP) technology, and in particular to a sentence processing method, apparatus, electronic device, and readable storage medium. Background Technology

[0002] With the continuous development of technology, the customer service provided by operators to users is becoming more intelligent. For example, users can call the operator's customer service hotline and state the service they want to subscribe to via voice. The operator can then collect the user's voice data, recognize the sentences within the voice data, and execute the instructions corresponding to the user's statements. Therefore, for operators, after obtaining the user's spoken sentences, the first priority is to ensure accurate processing of the sentences and recognition of the user's intent in order to ensure accurate subsequent execution.

[0003] In existing technologies, to recognize user statements, the servers set up by operators typically first use a word segmentation model to segment the statement into multiple words, and then use a part-of-speech tagging model to tag each word. Finally, based on the tagged parts of speech, the intent corresponding to the user's instructions in the entire statement is further determined.

[0004] However, using existing technologies, to achieve more accurate word segmentation and part-of-speech tagging of sentences, operators typically set up segmentation and part-of-speech tagging models with large data volumes and computational demands to improve recognition accuracy. But when the model's data volume and computational demands are large, it reduces the speed and efficiency of sentence processing. Therefore, how to improve the speed and efficiency of sentence processing while ensuring accuracy is a pressing technical problem that needs to be solved in this field. Summary of the Invention

[0005] This application provides a statement processing method, apparatus, electronic device, and readable storage medium to solve the technical problem in the prior art that it is impossible to improve speed and efficiency while ensuring accuracy when processing statements.

[0006] The first aspect of this application provides a statement processing method, including:

[0007] Acquire the voice data to be processed;

[0008] Identify the statements included in the voice data;

[0009] Determine the clarity information of the sentences in the speech data;

[0010] When the clarity information meets the preset conditions, the first machine learning model is used to perform word segmentation and part-of-speech tagging on the statement.

[0011] When the clarity information does not meet the preset conditions, the second machine learning model is used to perform word segmentation and part-of-speech tagging on the statement; wherein, the amount of data of the first machine learning model is less than the amount of data of the second machine learning model.

[0012] In some embodiments, the clarity information includes the intensity of the speech data, and the preset condition includes the speech data intensity being greater than a first threshold; and / or,

[0013] The clarity information includes the intensity of noise in the speech data, and the preset condition includes that the noise intensity is less than a second threshold; and / or,

[0014] The clarity information includes the number of voiceprint features from different users included in the voice data, and the preset condition includes the number being 1.

[0015] In some embodiments, the second machine learning model includes: a first bidirectional long short-term memory (LSTM) network model for word segmentation and a second bidirectional LSTM model for part-of-speech tagging;

[0016] The step of using a second machine learning model to perform word segmentation and part-of-speech tagging on the statement includes:

[0017] The statement is input into the first bidirectional LSTM model, which then performs word segmentation on the statement to obtain at least one word in the statement.

[0018] At least one word from the statement is input into the second bidirectional LSTM model, so that the second bidirectional LSTM model performs part-of-speech tagging on at least one word from the statement.

[0019] In some embodiments, before inputting the statement into the first bidirectional LSTM model, the method further includes:

[0020] If it is determined that there is a target text in the statement that is the same as a preset text stored in the storage space, the word segmentation result of the preset text is determined as the word segmentation result corresponding to the target text; wherein, the storage space stores multiple preset texts and the word segmentation result corresponding to each preset text;

[0021] The step of inputting the statement into the first bidirectional LSTM model includes:

[0022] The statement and the annotation information used to indicate the word segmentation results of the target text are input into the first bidirectional LSTM model.

[0023] In some embodiments, before sequentially inputting the at least one word into the second bidirectional LSTM model, the method further includes:

[0024] If it is determined that there is a target word in the at least one word that is the same as a preset word stored in the storage space, the part-of-speech tagging result of the target word is determined as the part-of-speech tagging result corresponding to the target word; wherein, the storage space stores multiple preset words and the part-of-speech tagging result corresponding to each preset word;

[0025] The step of sequentially inputting the at least one word into the second bidirectional LSTM model includes:

[0026] The target vocabulary from the at least one vocabulary and the annotation information used to indicate the part-of-speech tagging results of the target vocabulary are input into the second bidirectional LSTM model.

[0027] In some embodiments, the method further includes:

[0028] Retrieve multiple preset statements, along with word segmentation and part-of-speech tagging information for each statement;

[0029] The multiple preset sentences and the word segmentation information are input into the first bidirectional LSTM model, so that the first bidirectional LSTM model trains the model parameters for word segmentation from the forward and backward directions of the preset sentences respectively.

[0030] The multiple preset statements and the part-of-speech tagging information are input into the second bidirectional LSTM model, so that the second bidirectional LSTM model trains the model parameters for part-of-speech tagging from the forward and backward directions of the preset statements respectively.

[0031] In some embodiments, after recognizing the statements included in the speech data, the process includes:

[0032] If the statement is determined to be the same as a preset statement stored in the storage space, the word segmentation result and part-of-speech tagging result of the preset statement are used as the word segmentation result and part-of-speech tagging result of the statement; wherein, the storage space stores multiple preset statements, as well as the word segmentation result and part-of-speech tagging result corresponding to each preset statement.

[0033] A second aspect of this application provides a statement processing apparatus, comprising:

[0034] The acquisition module is used to acquire the voice data to be processed;

[0035] A recognition module is used to recognize the statements included in the voice data;

[0036] A determination module is used to determine the clarity information of sentences in the speech data;

[0037] The first processing module is used to perform word segmentation and part-of-speech tagging on the statement using a first machine learning model when the clarity information meets the preset conditions.

[0038] The second processing module is used to perform word segmentation and part-of-speech tagging on the statement using a second machine learning model when the clarity information does not meet the preset conditions; wherein the amount of data in the first machine learning model is less than the amount of data in the second machine learning model.

[0039] A third aspect of this application provides an electronic device, comprising: a memory and a processor; the memory storing computer-executable instructions; the processor executing the computer-executable instructions stored in the memory, causing the processor to perform the statement processing method provided in the first aspect of this application.

[0040] A fourth aspect of this application provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, are used to implement the statement processing method provided in the first aspect of this application.

[0041] The statement processing method, apparatus, electronic device, and readable storage medium provided in this application, when a server set up by an operator for statement recognition acquires speech data to be processed, including statements, selects either a first machine learning model or a second machine learning model with different computational requirements based on the clarity of the speech data to be processed. These models perform word segmentation and part-of-speech tagging on the statements in the speech data. When the statements in the speech data are relatively clear and a certain level of recognition accuracy can be guaranteed, the first machine learning model with lower computational requirements is used for word segmentation and part-of-speech tagging to reduce unnecessary computation. Conversely, when the statements in the speech data are unclear, the second machine learning model with higher computational requirements is used for word segmentation and part-of-speech recognition to improve the accuracy of word segmentation and part-of-speech tagging. Ultimately, while ensuring the accuracy of statement processing, the speed and efficiency of statement processing are improved, thereby increasing the overall processing speed and efficiency of the intelligent customer service system where the server is located, and enhancing the user experience of the intelligent customer service provided by the operator. Attached Figure Description

[0042] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a schematic diagram illustrating the application scenario of this application;

[0044] Figure 2 A flowchart illustrating an embodiment of the statement processing method provided in this application;

[0045] Figure 3 This is a schematic diagram of the structure of an embodiment of the second machine learning model provided in this application;

[0046] Figure 4 A schematic diagram of the training process for the first bidirectional LSTM model provided in this application;

[0047] Figure 5 A schematic diagram of the training process for the second bidirectional LSTM model provided in this application;

[0048] Figure 6 A flowchart illustrating another embodiment of the statement processing method provided in this application;

[0049] Figure 7 A schematic diagram of the structure of an embodiment of the statement processing apparatus provided in this application;

[0050] Figure 8 A schematic diagram of the structure of an embodiment of the electronic device provided in this application. Detailed Implementation

[0051] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0052] The terms "first" and "second" in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented, for example, in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0053] Before formally introducing the embodiments of this application, the application scenario and the problems existing in the scenario will be explained in conjunction with the accompanying drawings.

[0054] Figure 1 This diagram illustrates the application scenario of this application. In it, an operator can provide communication services to electronic devices 11, such as mobile phones, used by user 10. With continuous technological advancements, the customer service provided by operators is becoming increasingly intelligent. For example, user 10 can use electronic device 11 to call the operator's customer service hotline via communication network 20 and state the desired service via voice. The base station 21 set up by the operator in communication network 20 receives the user's voice data and sends it to server 22 for sentence recognition. Server 22 then executes the instructions corresponding to the user's statements. For instance, when a user states they want to check their phone bill, server 22, recognizing the intent in the statement, can directly send a text message to the electronic device 11 used by user 10 to inform them of the bill information. Throughout this process, the operator no longer needs to set up human customer service representatives 30, reducing labor costs and the office costs associated with hiring human customer service representatives 30 and the associated equipment 31, thus achieving a more intelligent customer service approach.

[0055] Then in the above Figure 1In the scenario of intelligent customer service shown, for the server 22 set up by the operator, after obtaining the statement spoken by user 10, the first step is to ensure accurate processing of the statement and then identify the intent of the user's instructions in the statement in order to ensure accurate subsequent execution. In some technologies, in order to identify the obtained user's statement, the server usually first uses a word segmentation model to segment the statement into multiple words, and then uses a part-of-speech tagging model to tag each word. The word segmentation model and part-of-speech tagging model used can be a one-way long short-term memory network (LSTM), etc. However, due to the deficiency of these models in not considering the context information of the statement, they generally only perform word segmentation and part-of-speech tagging according to the order from the beginning to the end of the sentence, resulting in insufficient accuracy. Therefore, in some other technologies, a bidirectional LSTM-based word segmentation model and part-of-speech tagging model can be set up in the server. This allows for two word segmentation and part-of-speech tagging operations based on the sentence from beginning to end and from end to beginning. The accuracy of the results can be improved through two recognitions. However, this method is equivalent to performing two LSTM calculations on the sentence, which is twice as computationally intensive as a single-item LSTM. This results in a large computational load and a large number of models to be stored, which reduces the speed and efficiency of the server in processing sentences.

[0056] Therefore, how to improve the speed and efficiency of statement processing while ensuring accuracy is a pressing technical problem in this field. This application provides a statement processing method and apparatus, applicable to applications such as... Figure 1 In the scenario shown, when acquiring speech data including sentences, machine learning models with varying computational requirements are selected based on the speech data's clarity for word segmentation and part-of-speech tagging. When the sentences in the speech data are relatively clear and a certain level of recognition accuracy can be maintained, a less computationally intensive machine learning model is used for word segmentation and part-of-speech tagging to reduce unnecessary computation. Conversely, when the sentences in the speech data are unclear, a more computationally intensive machine learning model is used for word segmentation and part-of-speech tagging to improve the accuracy of these processes. Ultimately, this approach ensures both accuracy and speed in sentence processing, thereby improving overall performance. Figure 1 The intelligent customer service system shown improves the speed and efficiency of sentence processing, enhancing the user experience of the services provided by the operator.

[0057] The technical solutions of this application will be described in detail below with specific embodiments. The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments.

[0058] Figure 2A flowchart illustrating an embodiment of the statement processing method provided in this application is shown below. Figure 2 The statement processing method shown can be applied to, for example... Figure 1 In the intelligent customer service application scenario provided by the operator shown, it is implemented by server 22 set up by the operator within network 20. Specifically, the statement processing method provided in this embodiment includes:

[0059] S101: Obtain the voice data to be processed. This involves recording the voice data obtained from the user by the server as the voice data to be processed, and then performing word segmentation and part-of-speech tagging on the voice data in subsequent steps.

[0060] S102: Recognize the statements included in the above-mentioned voice data. This can be achieved by using a voice recognition interface or other means to convert the voice data into text. This embodiment of the application does not limit the specific form of converting the voice data into text.

[0061] S103: Determine the speech clarity information in the above speech data.

[0062] S104: Determine whether the resolution information meets the preset conditions. If yes, proceed to S105; otherwise, proceed to S106.

[0063] In some embodiments, clarity information refers to information that can be used to measure whether the corresponding part of a sentence in the speech data to be processed is clear. For example, clarity information may include the intensity of the speech data. When the intensity of the speech data to be processed is greater than a first threshold, it indicates that the user's voice is loud, and the recognized text is likely to be more accurate, thus allowing for more accurate subsequent word segmentation and part-of-speech tagging. Conversely, when the intensity of the speech data to be processed is less than the first threshold, it indicates that the user's voice is soft, the recognized text is less accurate, and the accuracy of subsequent word segmentation and part-of-speech tagging is also lower. In this case, the preset condition could be that the intensity of the speech data is greater than the first threshold. Alternatively, clarity information could also be the intensity of noise in the speech data. When the noise intensity in the speech data to be processed is less than a second threshold, it indicates that the sentences in the speech data are relatively clear. When the noise intensity in the speech data to be processed is not less than the second threshold, it indicates that the sentences in the speech data are unclear. In this case, the preset condition could be that the noise intensity in the speech data is less than the second threshold. For example, clarity information can also be the number of voiceprint features from different users included in the speech data. For instance, after separating the voices from different users in the speech data based on voiceprint features, if it is determined that there are multiple users' voices, it indicates that the environment in which the user speaks is relatively complex and the clarity of the sentence is not high; if it is determined that there is only one user's voiceprint information in the speech data to be processed, then the clarity of the sentence is high.

[0064] In some embodiments, the clarity information may be one or more of the intensity, noise intensity, and number of voiceprint features of the speech data to be processed. The preset conditions may also be set accordingly based on the clarity information. The specific implementation method and principle are the same, only the quantity is changed, and will not be described in detail here.

[0065] S105. Use the first machine learning model to perform word segmentation and part-of-speech tagging on the above statement.

[0066] S106. Use the second machine learning model to perform word segmentation and part-of-speech tagging on the above statement.

[0067] In some embodiments, when the aforementioned clarity information meets the preset conditions, the computational load can be reduced by using a first machine learning model with a smaller data volume and computational load to perform word segmentation and part-of-speech tagging on the sentences in the speech data to be processed; when the aforementioned clarity information does not meet the preset conditions, the word segmentation and part-of-speech tagging on the sentences in the speech data to be processed can be performed by using a second machine learning model with a larger data volume and computational load.

[0068] In some embodiments, the first machine learning model may be a unidirectional LSTM model, then S105 specifically includes: inputting the sentence into the first unidirectional LSTM model, causing the first unidirectional LSTM model to perform word segmentation on the sentence; and inputting at least one word from the sentence into the second unidirectional LSTM model, causing the second unidirectional LSTM model to perform part-of-speech tagging on at least one word from the sentence. Alternatively, the first machine learning model may also be a model such as an RNN used for unidirectional word segmentation and part-of-speech tagging of sentences. This application does not limit the specific implementation of the first machine learning model, but the computational cost and the amount of data involved in the computation of the first machine learning model are less than those of the second machine learning model.

[0069] In some embodiments, Figure 3 A schematic diagram of the structure of an embodiment of the second machine learning model provided in this application is shown below. Figure 3 This illustrates one possible structure of the second machine learning model used in S106 provided in this application, such as... Figure 3 The second machine learning model shown includes: a first bidirectional LSTM model and a second bidirectional LSTM model. The first bidirectional LSTM model is used to perform word segmentation on the sentence to obtain the word segmentation result, and the second bidirectional LSTM model is used to perform part-of-speech tagging on the sentence to obtain the part-of-speech tagging result.

[0070] In some embodiments, after obtaining the word segmentation and part-of-speech tagging results of a statement, the server can further determine the intent of the statement based on these results, and then execute the corresponding instructions according to the intent in the statement, such as the server returning the information required by the user or the server executing the instructions issued by the user.

[0071] In summary, in the above... Figure 2 In the sentence processing method shown, when the server set up by the operator for sentence recognition acquires the voice data to be processed, including sentences, it selects either a first machine learning model or a second machine learning model with different computational requirements based on the clarity of the voice data. The models perform word segmentation and part-of-speech tagging on the sentences in the voice data. When the sentences in the voice data are relatively clear and a certain level of recognition accuracy can be guaranteed, the first machine learning model with lower computational requirements is used for word segmentation and part-of-speech tagging to reduce unnecessary computation. Conversely, when the sentences in the voice data are unclear, the second machine learning model with higher computational requirements is used for word segmentation and part-of-speech recognition to improve the accuracy of word segmentation and part-of-speech tagging. Ultimately, while ensuring the accuracy of sentence processing, the speed and efficiency of sentence processing are improved, thereby increasing the overall processing speed and efficiency of the intelligent customer service system where the server is located, and enhancing the user experience of the intelligent customer service provided by the operator.

[0072] In some embodiments, in order to achieve such Figure 2 The statement processing method shown in this application also provides a training method for the first bidirectional LSTM model and the second bidirectional LSTM model in a second machine learning model. Among them, Figure 4 This is a schematic diagram of the training process for the first bidirectional LSTM model provided in this application. Figure 5 This is a schematic diagram illustrating the training process of the second bidirectional LSTM model provided in this application. Figure 4 As shown, to train the first bidirectional LSTM model, multiple preset sentences are first obtained, and the word segmentation and part-of-speech tagging results of the preset sentences are labeled. The word segmentation and part-of-speech tagging results are used as known quantities. Subsequently, the word segmentation information representing the word segmentation results and the multiple preset sentences are input as follows: Figure 4 The first bidirectional LSTM model shown performs word segmentation twice: once in the forward direction from the beginning to the end of the sentence, and again in the reverse direction from the end to the beginning. If either result is incorrect, the model is reset to adjust the parameters used for word segmentation. After processing multiple preset sentences, the trained first bidirectional LSTM model is finally obtained, which can be used for, for example... Figure 2 The statement processing shown. Figure 5 The training process of the second bidirectional LSTM model shown is the same as... Figure 4Similarly, after inputting the part-of-speech tagging information and preset sentences to represent the part-of-speech tagging results into the second bidirectional LSTM model, the second bidirectional LSTM model performs part-of-speech tagging twice: once in the forward direction from the beginning to the end of the sentence, and again in the reverse direction from the end to the beginning. If either of the two results is incorrect, the model is returned to adjust the model parameters used for part-of-speech tagging in the second bidirectional LSTM model. After processing multiple preset sentences, the finally trained second bidirectional LSTM model is obtained, which can be used for, for example... Figure 2 The statement processing shown.

[0073] In some embodiments, since the statements spoken by users often share commonalities, such as saying "I want to check my phone bill" or similar phrases, the server can pre-store some preset statements and their word segmentation and part-of-speech tagging results. This allows the server to determine whether the received speech data matches the preset statements in the storage space after acquiring the speech data and identifying the statements within it. If they match, there's no need to determine clarity information or process the statements using a first or second machine learning model. Instead, the word segmentation and part-of-speech tagging results of the preset statements can be directly retrieved from the storage space and used as the result for the statements in the current speech data for subsequent calculations. Therefore, this embodiment can further improve the processing speed and efficiency in the speech processing process.

[0074] In some embodiments, when such Figure 2 The second machine learning model in the statement processing method shown employs, as follows: Figure 3In the structure shown, following the same logic, the server's storage space can pre-store multiple different preset texts and their corresponding word segmentation results. Before the server obtains a sentence and inputs it into the first bidirectional LSTM model, it can first identify target texts in the sentence that are identical to the preset texts stored in the storage space, and use the word segmentation results of the preset texts in the storage space as the word segmentation results of the target texts in the sentence. Subsequently, when inputting the sentence into the first bidirectional LSTM model, the sentence and the annotation information used to indicate the word segmentation results of the target texts can be input into the first bidirectional LSTM model. This allows the first bidirectional LSTM model to perform word segmentation only on the other texts in the sentence, without performing word segmentation on the target texts, and the word segmentation results of the target texts can also assist the first bidirectional LSTM model in performing word segmentation on the other texts. For example, if the storage space contains preset words such as "I," "want," "query," and "phone bill" along with their word segmentation results, then when the sentence in the voice data to be processed is "I want to query ringback tones," before inputting the sentence into the first bidirectional LSTM model, the target words "I," "want," and "query" can be determined first, and their word segmentation results can be marked. Then, the sentence can be input into the first bidirectional LSTM model, allowing it to segment "ringback tone" given the known word segmentation results for "I," "want," and "query." This reduces the computational load of the first bidirectional LSTM model and improves the processing speed and efficiency. Correspondingly, when calculating the first bidirectional LSTM model, in addition to following... Figure 4 The training can be performed in the manner shown. Multiple training sentences and the word segmentation results labeled in the training sentences can also be input into the model. For example, in the sentence "I want to check my phone bill", the model can be trained by inputting the labeled information of the three word segmentation results "I", "want", and "check", as well as the word segmentation result of "phone bill". This allows the first bidirectional LSTM model to learn the word segmentation result of "phone bill" given the three word segmentation results.

[0075] In some embodiments, the server's storage space can pre-store multiple different preset words and the part-of-speech tagging results corresponding to each preset word. Before the server obtains and inputs at least one word of the sentence into the second bidirectional LSTM model, it can first determine the target words in the sentence that are the same as the preset words stored in the storage space, and use the part-of-speech tagging results of the preset words in the storage space as the part-of-speech tagging results of the target words in the sentence. Subsequently, when inputting the sentence into the second bidirectional LSTM model, the sentence and the tagging information used to indicate the part-of-speech tagging results of the target words can be input into the second bidirectional LSTM model. This allows the second bidirectional LSTM model to perform part-of-speech tagging on the target words without performing part-of-speech tagging on the target words, but only on the other words in the sentence. Furthermore, the part-of-speech tagging results of the target words can also assist the second bidirectional LSTM model in performing part-of-speech tagging on the other words. This reduces the computational load of the first bidirectional LSTM model and improves the processing speed and efficiency of the sentence. For example, if the storage space contains preset words such as "I," "want," "query," and "phone bill," and their part-of-speech tagging results are noun, verb, verb, and noun respectively, then when the sentence in the voice data to be processed is "I want to query ringback tones," the part-of-speech tagging results for "I," "want," and "query" can be tagged as noun, verb, and verb respectively, and then input into the second bidirectional LSTM model. This will cause the second bidirectional LSTM model to output the part-of-speech tagging result for "phone bill" in the sentence as a noun. Accordingly, when calculating the second bidirectional LSTM model, in addition to following the... Figure 5 The training method shown can also be used to input multiple training sentences and their part-of-speech tagging results into the model. For example, training can be performed on the sentence "I want to check my phone bill". For instance, the model can be trained by inputting the part-of-speech tagging results for "I", "want", and "check", as well as the part-of-speech tagging result for "phone bill" as a noun. This allows the second bidirectional LSTM model to learn that the part-of-speech tagging result for "phone bill" is a noun, following a noun, verb, and then a verb, given the part-of-speech tagging results for these three words. This enables the second bidirectional LSTM model to directly determine that the part-of-speech tagging result for "ring tone" is a noun when it receives the sentence "I want to check my ringback tone" and the part-of-speech tagging results for "I", "want", and "check".

[0076] In some embodiments, Figure 6 A flowchart illustrating another embodiment of the statement processing method provided in this application, wherein, as Figure 6After the process shown begins, the server first collects the question raised by the client (user) and determines whether the question is text information. If it is text, it proceeds directly to subsequent recognition and processing. If it is voice data, it is converted into a text statement through a voice recognition interface. Then, it checks if the same statement exists in the database. If so, the intent corresponding to the stored statement is processed directly. If not, a first bidirectional LSTM model is used to segment the statement, and a second bidirectional LSTM model is used to perform part-of-speech tagging. Then, semantic analysis is performed on the tagged text to determine the intent, and the corresponding instruction is processed, such as returning the information the user needs or further resolving the question, finally ending the entire process. Figure 6 The first and second bidirectional LSTM models shown are similar to those described above. Figure 3 The same applies as shown, so I will not repeat it again.

[0077] In the foregoing embodiments, the statement processing method provided by the embodiments of this application has been described. To implement the functions of the statement processing method provided by the embodiments of this application, the server, as the execution entity, may include hardware structures and / or software modules, implementing the above functions in the form of hardware structures, software modules, or a combination of hardware structures and software modules. Whether a particular function is executed in the form of hardware structures, software modules, or a combination of hardware structures and software modules depends on the specific application and design constraints of the technical solution.

[0078] For example, Figure 7 A schematic diagram of the structure of an embodiment of the statement processing apparatus provided in this application is shown below. Figure 7 This application also provides a sentence processing device 700, comprising: an acquisition module 701, an identification module 702, a determination module 703, a first processing module 704, and a second processing module 705; wherein, the acquisition module 701 is used to acquire speech data to be processed; the identification module 702 is used to identify sentences included in the speech data; the determination module 703 is used to determine the clarity information of the sentences in the speech data to be processed; the first processing module 704 is used to perform word segmentation and part-of-speech tagging processing on the sentences using a first machine learning model when the clarity information meets preset conditions; the second processing module 705 is used to perform word segmentation and part-of-speech tagging processing on the sentences using a second machine learning model when the clarity information does not meet the preset conditions; wherein, the data volume of the first machine learning model is less than the data volume of the second machine learning model.

[0079] Specifically, the specific principles and implementation methods of the above steps executed by each module in the statement processing device can be referred to the statement processing method in the foregoing embodiments of this application, and will not be repeated here.

[0080] It should be noted that the division of the various modules in the above device is merely a logical functional division. In actual implementation, they can be fully or partially integrated into a single physical entity, or they can be physically separated. These modules can be implemented entirely in software via processing element calls; they can be fully implemented in hardware; or some modules can be implemented by processing element calls to software, while others are implemented in hardware. They can be separate processing elements, integrated into a chip within the device, or stored as program code in the device's memory, invoked and executed by a processing element. The implementation of other modules is similar. Furthermore, these modules can be fully or partially integrated together, or implemented independently. The processing element described here can be an integrated circuit with signal processing capabilities. In the implementation process, each step of the above method or each of the above modules can be completed through integrated logic circuits in the hardware of the processor element or through software instructions.

[0081] For example, these modules can be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs). As another example, when a module is implemented using processing element scheduler code, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of calling program code. Furthermore, these modules can be integrated together to implement a system-on-a-chip (SOC).

[0082] In the above embodiments, implementation can be achieved entirely or partially through software, hardware, firmware, or any combination thereof. When implemented using software, it can be implemented entirely or partially in the form of a computer program product. The computer program product includes one or more computer instructions. When the computer program instructions are loaded and executed on a computer, all or part of the processes or functions described in the embodiments of this application are generated. The computer can be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions can be stored in a computer-readable storage medium or transmitted from one computer-readable storage medium to another. For example, the computer instructions can be transmitted from one website, computer, server, or data center to another website, computer, server, or data center via wired (e.g., coaxial cable, fiber optic, digital subscriber line (DSL)) or wireless (e.g., infrared, wireless, microwave, etc.) means. The computer-readable storage medium can be any available medium that a computer can access or a data storage device such as a server or data center that integrates one or more available media. The available medium can be a magnetic medium (e.g., floppy disk, hard disk, magnetic tape), an optical medium (e.g., DVD), or a semiconductor medium (e.g., solid-state disk (SSD)).

[0083] This application also provides an electronic device, which may be a server or a computer, for example. Figure 8 A schematic diagram of the structure of an embodiment of the electronic device provided in this application is shown below. Figure 8 The illustrated electronic device 800 includes a processor 8020 and a memory 803; wherein the memory 803 stores a computer program, and when the processor 802 executes the computer program, the processor can be used to execute any of the statement processing methods in the foregoing embodiments of this application. For example, the processor 802 can also acquire voice data to be processed through a communication interface 801.

[0084] This application also provides a computer-readable storage medium storing a computer program, which, when executed, can be used to perform any of the statement processing methods described in the foregoing embodiments of this application.

[0085] This application also provides a chip for executing instructions, the chip being used to execute statement processing methods executed by a server as described in any of the foregoing embodiments of this application.

[0086] Those skilled in the art will understand that all or part of the steps of the above-described method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When executed, the program performs the steps of the above-described method embodiments; and the aforementioned storage medium includes various media capable of storing program code, such as ROM, RAM, magnetic disks, or optical disks.

[0087] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of this application.

Claims

1. A statement processing method, characterized in that, include: Acquire the voice data to be processed; Identify the statements included in the voice data; Determine the clarity information of the sentences in the speech data; The clarity information of the statement is used to characterize the recognizability and reliability of the statement in the speech data. The clarity information includes the intensity of the speech data, the intensity of noise in the speech data, and the number of voiceprint features from different users. When the clarity information meets the preset conditions, the first machine learning model is used to perform word segmentation and part-of-speech tagging on the statement. When the clarity information does not meet the preset conditions, the second machine learning model is used to perform word segmentation and part-of-speech tagging on the statement; wherein, the amount of data of the first machine learning model is less than the amount of data of the second machine learning model.

2. The method according to claim 1, characterized in that, The clarity information includes the intensity of the speech data, and the preset condition includes the speech data intensity being greater than a first threshold; and / or, The clarity information includes the intensity of noise in the speech data, and the preset condition includes that the noise intensity is less than a second threshold; and / or, The clarity information includes the number of voiceprint features from different users included in the voice data, and the preset condition includes the number being 1.

3. The method according to claim 1 or 2, characterized in that, The second machine learning model includes: a first bidirectional LSTM model for word segmentation and a second bidirectional LSTM model for part-of-speech tagging; The step of using a second machine learning model to perform word segmentation and part-of-speech tagging on the statement includes: The statement is input into the first bidirectional LSTM model, which then performs word segmentation on the statement to obtain at least one word in the statement. At least one word from the statement is input into the second bidirectional LSTM model, so that the second bidirectional LSTM model performs part-of-speech tagging on at least one word from the statement.

4. The method according to claim 3, characterized in that, Before inputting the statement into the first bidirectional LSTM model, the method further includes: If it is determined that there is a target text in the statement that is the same as a preset text stored in the storage space, the word segmentation result of the preset text is determined as the word segmentation result corresponding to the target text; wherein, the storage space stores multiple preset texts and the word segmentation result corresponding to each preset text; The step of inputting the statement into the first bidirectional LSTM model includes: The statement and the annotation information used to indicate the word segmentation results of the target text are input into the first bidirectional LSTM model.

5. The method according to claim 3, characterized in that, Before sequentially inputting the at least one word into the second bidirectional LSTM model, the method further includes: If it is determined that there is a target word in the at least one word that is the same as a preset word stored in the storage space, the part-of-speech tagging result of the target word is determined as the part-of-speech tagging result corresponding to the target word; wherein, the storage space stores multiple preset words and the part-of-speech tagging result corresponding to each preset word; The step of sequentially inputting the at least one word into the second bidirectional LSTM model includes: The target vocabulary from the at least one vocabulary and the annotation information used to indicate the part-of-speech tagging results of the target vocabulary are input into the second bidirectional LSTM model.

6. The method according to claim 4 or 5, characterized in that, The method further includes: Retrieve multiple preset statements, along with word segmentation and part-of-speech tagging information for each statement; The multiple preset sentences and the word segmentation information are input into the first bidirectional LSTM model, so that the first bidirectional LSTM model trains the model parameters for word segmentation from the forward and backward directions of the preset sentences respectively. The multiple preset statements and the part-of-speech tagging information are input into the second bidirectional LSTM model, so that the second bidirectional LSTM model trains the model parameters for part-of-speech tagging from the forward and backward directions of the preset statements respectively.

7. The method according to claim 1, characterized in that, After recognizing the statements included in the speech data, the process includes: If the statement is determined to be the same as a preset statement stored in the storage space, the word segmentation result and part-of-speech tagging result of the preset statement are used as the word segmentation result and part-of-speech tagging result of the statement; wherein, the storage space stores multiple preset statements, as well as the word segmentation result and part-of-speech tagging result corresponding to each preset statement.

8. A statement processing device, characterized in that, include: The acquisition module is used to acquire the voice data to be processed; A recognition module is used to recognize the statements included in the voice data; A determination module is used to determine the clarity information of sentences in the speech data; The clarity information of the statement is used to characterize the recognizability and reliability of the statement in the speech data. The clarity information includes the intensity of the speech data, the intensity of noise in the speech data, and the number of voiceprint features from different users. The first processing module is used to perform word segmentation and part-of-speech tagging on the statement using a first machine learning model when the clarity information meets the preset conditions. The second processing module is used to perform word segmentation and part-of-speech tagging on the statement using a second machine learning model when the clarity information does not meet the preset conditions; wherein the amount of data in the first machine learning model is less than the amount of data in the second machine learning model.

9. An electronic device, characterized in that, include: Memory and processor; The memory stores computer-executed instructions; The processor executes computer execution instructions stored in the memory, causing the processor to perform the statement processing method as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, are used to implement the statement processing method as described in any one of claims 1 to 7.

Images