Intelligent vertical business corpus construction and real-time intention recognition system

Abstract

The invention relates to the field of intelligent corpora, in particular to an intelligent vertical business corpus construction and real-time intention recognition system which comprises equipment carrying an intention recognition system, and the intention recognition system comprises a voice input module, a semantic database and a voice output module. The semantic database is respectively connected with the voice input module and the voice output module, and the voice input module and the voice output module are respectively used for voice receiving and voice output. Through comprehensive corpus construction, fine word segmentation processing and reasonable weight presetting, various complex and diversified language expressions can be dealt with, and the intention of the user can be more accurately captured in various fields needing language interaction with the user and understanding the intention of the user, such as customer service consultation, intelligent navigation and the like.

Description

Technical Field

[0001] The present invention relates to the field of intelligent corpora, and in particular to an intelligent vertical business corpus construction and real-time intent recognition system. Background Art

[0002] Existing intelligent outbound robot technology has shown significant advantages in automatic voice response (IVR) systems. Such systems can instantly convert user voice into text by integrating automatic speech recognition (ASR) technology, and then analyze the text content to capture user intent and retrieve corresponding responses from the system. Subsequently, these responses are conveyed to users through speech synthesis technology (TTS) or preset recordings. This process constitutes a complete speech flow, which consists of multiple branch nodes, each of which carries specific trigger words and preset robot replies to map and respond to diverse user intents;

[0003] The existing technical solutions closely related to the present invention mainly rely on preset trigger words to match and infer user intentions. This method presets the keywords or sentences that the user may express at a specific dialogue node, and determines the direction of the dialogue process by comparing the speech recognition results. However, although this method has its application basis, it has obvious limitations in actual application, mainly including the following parts:

[0004] 1. Low user intent hit rate;

[0005] Due to the diversity of language habits and expressions among users, the text content obtained through voice recognition is ever-changing and difficult to be fully covered by limited preset trigger words. This directly leads to a low success rate in recognizing user intentions, and the real needs of many users are therefore ignored or misunderstood.

[0006] 2. Insufficient accuracy in identifying user intent;

[0007] In complex conversation scenarios, a single user statement often contains multiple intentions, and the existing matching method based on preset trigger words lacks dynamic analysis and deep understanding capabilities, making it difficult to accurately capture and distinguish these subtle differences in intentions. This not only limits the accuracy of the robot's answers, but also affects the user experience satisfaction.

[0008] 3. Slow response to user intent;

[0009] Because offline IoT devices have poor network signals or low computing efficiency, the artificial intelligence responses they generate are too slow, which can reduce the fluency of conversations with users, and users need to spend a lot of time waiting for feedback from the device.

[0010] In summary, existing intelligent outbound call robots have significant shortcomings in user intent recognition, and a more intelligent and flexible solution is urgently needed to overcome the above defects and improve the overall performance of the system and user satisfaction. Summary of the invention

[0011] The technical problem to be solved by the present invention is to overcome the defects of the prior art and provide an intelligent vertical business corpus construction and real-time intent recognition system to solve the problems that the corpus of current voice robots is insufficient for intent recognition and the response efficiency is too low.

[0012] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0013] In a first aspect of the present invention, an intelligent vertical business corpus construction and real-time intent recognition system is provided, and a device equipped with the intent recognition system is provided, wherein the intent recognition system comprises a speech input module, a semantic database and a speech output module, wherein the semantic database is connected to the speech input module and the speech output module respectively, and the speech input module and the speech output module are used for receiving speech and outputting speech respectively;

[0014] The semantic database also includes an intention recognition module, a semantic preservation unit, an intention association module and a semantic denoising unit, among which:

[0015] The intent recognition module is used to build the main program of the system and define intent categories, recognize the type of voice input by the user and define the intent in the voice;

[0016] The semantic storage unit is used to convert the intention in the user's voice into text and candidate word matching, quickly find the semantics of the word segmentation in the voice, and save the unmatched phrases in the storage unit;

[0017] The semantic denoising unit is used to complete the phased word deletion of the sentence under the speech by the auxiliary semantic preservation unit, split the various types of intentions in the sentence and classify them as needed;

[0018] The intention association module is used to construct a word frequency weight order for the unclear intentions in different voices, and calculate the word frequency based on the frequency of occurrence of phrases to complete the task order adjustment and recording.

[0019] As a preferred technical solution of the present invention, the system main program of the intention recognition module also includes a startup voice word for waking up the semantic database, and the startup voice word completes the wake-up through the voice input module or the voice output module; the defined intention category is used to establish a special phrase, and link the semantic phrase to the corresponding trigger program of the semantic database to complete the task issuance process directly completed by the special phrase.

[0020] As a preferred technical solution of the present invention, the semantic preservation unit includes a speech recognition algorithm, which records the speech through a speech input module and converts it into text content based on the speech recognition algorithm, and then matches the candidate words with special phrases under the intention recognition module, and records the unmatched phrases.

[0021] As a preferred technical solution of the present invention, the semantic denoising unit deletes the text before and after the phrase in a sentence-descending manner, completes word segmentation matching under a set matching cycle, and forms a segmentation operation for the intent of the sentence, which is used to extract multiple different intents under a single sentence.

[0022] As a preferred technical solution of the present invention, the intention association module includes a word frequency recording unit, wherein the weight of the word frequency recording unit is calculated by the following formula:

[0023] The frequency weight of a term = (frequency dictionary) / (total number of words);

[0024] Term: the basic unit of text after word segmentation, such as words and phrases;

[0025] Frequency dictionary: used to store each term and its corresponding frequency, usually a set of key-value pairs, where the key is the term and the value is the frequency;

[0026] Total number of words: an integer value indicating the total number of terms in the document;

[0027] Frequency weights of terms: a list or array containing the frequency weight of each term, which is usually a decimal between 0 and 1;

[0028] After the weight list is completed, it can be used for subsequent manual adjustment of the semantic database and the sorting of task flows formed by the task distribution units. The specific steps are as follows:

[0029] (1) After word segmentation is completed, the word segmentation results are obtained. For the word segmentation results, each word item is traversed and a hash table is used to record the number of times each word item appears;

[0030] (2) While traversing the word segmentation results, accumulate the number of all terms to obtain the total number of words;

[0031] (3) Obtain the frequency weight of each word as the actual weight basis for subsequent modification of the system weight preset.

[0032] In a second aspect of the present invention, a method for constructing an intelligent vertical business corpus and using a real-time intent recognition system is provided, characterized in that the specific steps are as follows:

[0033] S1. The voice input module or the voice output module outputs a specific start word, wakes up the semantic database to communicate with the user, records the user's voice and converts it into text;

[0034] S2. The semantic database under the intent recognition system is constructed by the intent recognition module to directly match special phrases, and the semantic denoising unit is used to assist in parsing phrases, save unmatched phrases, and record and mark frequently occurring phrases;

[0035] S3. The intention association module completes the frequency calculation of the phrases, classifies and sorts the frequency categories of unrecorded phrases and recognized phrases according to the actual frequencies, and finally confirms with the user again through the voice output module.

[0036] In a third aspect of the present invention, a computer device is provided, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of the system of the first aspect of the present invention or the steps of the method of the second aspect of the present invention when executing the computer program.

[0037] In a fourth aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, the functions of the system of the first aspect of the present invention or the steps of the method of the second aspect of the present invention are implemented.

[0038] Compared with the prior art, the present invention has the following beneficial effects:

[0039] 1. Improve the breadth and accuracy of intent recognition;

[0040] Through comprehensive corpus construction, sophisticated word segmentation processing and reasonable weight preset, it is able to cope with various complex and diverse language expressions, and can more accurately capture user intentions in various fields that require language interaction with users and understanding their intentions, such as customer service consultation, intelligent navigation, etc.

[0041] 2. Enhance the system's adaptability and self-optimization capabilities;

[0042] The processing mechanism for unrecognized intent and the word frequency recording function enable the system to continuously learn and improve. This feature is of great value in many scenarios that rely on user input and feedback to optimize services. For example, online education platforms can understand students' knowledge weaknesses based on the word frequency of students' questions and optimize teaching content.

[0043] 3. Optimize resource allocation and process efficiency;

[0044] By setting the intention trigger frequency and accurately identifying the user's intention, resources can be allocated reasonably, avoiding excessive consumption of system resources in processing repeated or invalid intentions. This resource optimization idea is of great significance in various systems that rely on user interaction. For example, in the field of logistics, transportation routes and vehicles can be reasonably arranged according to the customer's requirements for the delivery time of goods.

[0045] 4. Improve user experience and satisfaction;

[0046] Accurately understanding user intentions and providing satisfactory service responses are crucial for any scenario involving user services. Whether it is after-sales consultation for e-commerce shopping or demand processing for financial services, users can experience better service quality. For example, in e-commerce after-sales service, accurately understand the user's description of product quality issues and quickly provide solutions.

[0047] 5. Quickly respond to user questions;

[0048] In the process of users repeatedly mentioning and quickly conversing, the corpus stored locally can complete the interaction process more quickly, avoiding the user's continuous waiting after a conversation. The corresponding reply can be completed within 1-2 seconds without long waiting. BRIEF DESCRIPTION OF THE DRAWINGS

[0049] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0050] Figure 1 It is a schematic diagram of the system structure of the present invention;

[0051] Figure 2 is a schematic diagram of the corpus reaction time of the present invention;

[0052] Figure 3 It is a schematic diagram of the construction of the main process of the present invention;

[0053] Figure 4 It is a schematic diagram of the word segmentation process of the present invention;

[0054] Figure 5 It is a flow chart of the method of use of the present invention. DETAILED DESCRIPTION

[0055] The preferred embodiments of the present invention are described below in conjunction with the accompanying drawings. It should be understood that the preferred embodiments described herein are only used to illustrate and explain the present invention, and are not used to limit the present invention.

[0056] like Figure 1-2As shown, the present invention provides an intelligent vertical business corpus construction and real-time intention recognition system, which has a device equipped with an intention recognition system, characterized in that the intention recognition system includes a speech input module, a semantic database and a speech output module, the semantic database is connected to the speech input module and the speech output module respectively, and the speech input module and the speech output module are used for speech reception and speech output respectively;

[0057] The semantic database also includes an intention recognition module, a semantic preservation unit, an intention association module and a semantic denoising unit, among which:

[0058] The intent recognition module is used to build the main program of the system and define intent categories, recognize the type of voice input by the user and define the intent in the voice;

[0059] The system main program of the intention recognition module also includes a start-up voice word for waking up the semantic database, and the start-up voice word completes the wake-up through the voice input module or the voice output module;

[0060] Construction of the main process of the system

[0061] When the robot calls a customer or is called by a customer, after answering the call, the robot will execute the inquiry according to the main process. At this stage, the robot can only recognize the customer's yes or no intention.

[0062] 1. Define the opening statement

[0063] Design an opening statement to introduce the topic.

[0064] 2. Define the business inquiry

[0065] Design main line business inquiry.

[0066] 3. Definition of the closing remarks

[0067] Based on the customer's answer, give appropriate closing remarks and hang up the phone.

[0068] like Figure 3 As shown, the set process is only used as a regular wake-up process, that is, the semantic database is in a pre-start state at this time, but not directly started. If the main process is able to process the task, the next step will not be performed.

[0069] Define intent categories to create special phrases, and link semantic phrases to corresponding triggers in the semantic database to complete the task dispatching process directly completed by special phrases;

[0070] That is, when a customer makes an inquiry, in order to ensure that the robot can handle user inquiries outside the main process, the system builds a comprehensive corpus. First, it is necessary to define the intent category, such as:

[0071] Purpose: To clarify intent categories to cover various user inquiries and requests that may be encountered in food delivery scenarios.

[0072] Process: Based on the characteristics of the food delivery business and user needs, a series of intent categories are pre-defined, including but not limited to "ask about home delivery time", "confirm pick-up point", "request to change delivery address", "ask about order status", etc.

[0073] Then the user triggers the words and phrases, and the robot’s reply is collected;

[0074] Purpose: To build a comprehensive and accurate corpus to support the robot to understand users' inquiries, requests, and responses about meal collection methods.

[0075] Process: Collect relevant words, phrases and sentences through multiple channels such as user feedback from food delivery platforms, customer service records, social media discussions, and remove duplicate content;

[0076] Finally, the word and sentence intentions are classified and entered;

[0077] Purpose: Enter the system based on the intent category and collected words and phrases.

[0078] Process: Enter the collected data into the system. For example, classify the user's question phrase "put it in the takeaway cabinet" or "put it downstairs for me" as the "user pick up" intent, and configure the corresponding robot reply for the intent, such as "OK, we will put it according to your request. Is there anything else you need help with?"

[0079] After completing the above steps, a corpus entry with semantic intent related to takeout can be formed, which can form quick responses and quick matches for special phrases with higher accuracy.

[0080] The semantic storage unit is used to convert the intention in the user's voice into text and candidate word matching, quickly find the semantics of the word segmentation in the voice, and save the unmatched phrases in the storage unit;

[0081] The semantic preservation unit includes a speech recognition algorithm, which records the speech through a speech input module and converts it into text content based on the speech recognition algorithm. It then matches the candidate words with special phrases under the intention recognition module and records the unmatched phrases.

[0082] In actual use, the semantic preservation unit includes a speech recognition algorithm, which records the speech through a speech input module and converts it into text content based on the speech recognition algorithm, and then matches the candidate words with special phrases under the intention recognition module, and records the unmatched phrases.

[0083] After the above process, the main process will also be associated;

[0084] Purpose: To ensure that after processing inquiries outside the main process, the robot can smoothly return to the main process and continue to execute the remaining steps.

[0085] Process: Establish a bridging intent mechanism. When a customer triggers a bridging intent in the corpus, such as "putting takeout in the cabinet" is identified as the "user picks up the item and ends the call" intent, the robot will automatically associate it with the ending node of the main process, execute the corresponding ending and end the call.

[0086] For content that does not exist in the corpus, it is judged as unrecognized intent;

[0087] Purpose: To handle situations that are not within the preset trigger phrases or the user does not clearly express the intention, so as to prevent the robot from being unresponsive or providing irrelevant information.

[0088] Process: When user input does not match any preset intent, the robot can first ask the user if they need help, or ask what they want to do specifically, such as: "I don't quite understand what you mean, what information do you want to know?", and establish a library of unmatched preset intents to collect user input for subsequent manual expansion of the corpus.

[0089] The semantic denoising unit is used to complete the phased word deletion of the sentence under the speech by the auxiliary semantic preservation unit, split the various types of intentions in the sentence and classify them as needed;

[0090] The semantic denoising unit deletes the text before and after the phrase in a sentence-decrementing manner, completes word segmentation matching under the set matching cycle, and forms a segmentation operation for the sentence intent to extract multiple different intents under a single sentence.

[0091] That is, the intention trigger frequency setting is set;

[0092] Purpose: To ensure stable operation of the system and avoid infinite loops, the system allows you to set a maximum number of triggers for a certain intent. This measure is intended to prevent excessive consumption of system resources due to repeated inquiries or misoperations by users, while ensuring a smooth call process.

[0093] Process: Based on historical data analysis, set a reasonable maximum triggering threshold for each user intent. This threshold should take into account normal user inquiry needs and potential abnormal situations to ensure that user needs are met while preventing abuse. When the triggering number of an intent reaches the set maximum threshold, the system should automatically enter the closing statement in the main process and end the call.

[0094] The intention association module is used to construct a word frequency weight order for the unclear intentions in different voices, and calculate the word frequency based on the frequency of occurrence of phrases to complete the task order adjustment and recording.

[0095] The intention association module contains a word frequency recording unit, where the weight of the word frequency recording unit is calculated by the following formula:

[0096] The frequency weight of a term = (frequency dictionary) / (total number of words);

[0097] Term: the basic unit of text after word segmentation, such as words and phrases;

[0098] Frequency dictionary: used to store each term and its corresponding frequency, usually a set of key-value pairs, where the key is the term and the value is the frequency;

[0099] Total number of words: an integer value indicating the total number of terms in the document;

[0100] Frequency weights of terms: a list or array containing the frequency weight of each term, which is usually a decimal between 0 and 1;

[0101] After the weight list is completed, it can be used for subsequent manual adjustment of the semantic database and the sorting of task flows formed by the task distribution units. The specific steps are as follows:

[0102] (1) After word segmentation is completed, the word segmentation results are obtained. For the word segmentation results, each word item is traversed and a hash table is used to record the number of times each word item appears;

[0103] (2) While traversing the word segmentation results, accumulate the number of all terms to obtain the total number of words;

[0104] (3) Obtain the frequency weight of each word as the actual weight basis for subsequent modification of the system weight preset.

[0105] The weight preset of the intent name can make the task process smoother, and is the upper-level judgment method of the word frequency weight of the word item;

[0106] Purpose: To preset higher weights for common or important meal collection method intentions to optimize system response.

[0107] Process: Based on business data and business experience, estimate the usage frequency of different intents. According to the estimated frequency, assign an initial weight to each intent. The default weight is 0. The larger the value, the higher the weight. Intents with high weights are matched first during recognition.

[0108] Taking the takeaway system as an example, when presetting the word segmentation dictionary of the corpus, all user-triggered phrases collected in the intent recognition module are added to the word segmentation dictionary and sorted by the number of characters and weights, so that the user input content can be accurately identified and segmented in the word segmentation processing stage, such as the following table:

[0109]

[0110] Word segmentation dictionary table: meaning Figure 1

[0111]

[0112] Word segmentation dictionary table: meaning Figure 2

[0113]

[0114] Word segmentation dictionary table: meaning Figure 3

[0115]

[0116] Word segmentation dictionary table: Unrecognized meaning Figure 1

[0117]

[0118] Participle dictionary table: Bridge meaning Figure 1

[0119]

[0120]

[0121] Word segmentation dictionary table: Intent name weight table

[0122] Specifically, the overall system mainly establishes a corpus directly, that is, a semantic database, establishes a main process based on the intention recognition unit, sets up an additional communication process for business inquiries in the main process, and establishes a word segmentation dictionary based on the local semantic database. The weight values ​​constructed by the word segmentation dictionary correspond to each other and are sorted. After obtaining the user's intention, the main program process is finally completed to complete the user's consultation and the user's task requirements; the word segmentation process will also record frequently occurring phrases, and the frequently occurring phrases will be sorted during the next system update, prompting the user to re-enter the corresponding business process and reply process.

[0123] In another embodiment, the present invention provides a method for constructing an intelligent vertical business corpus and using a real-time intent recognition system. Still taking the takeaway system as an example, in a question about the customer's meal delivery method, assuming that the user says "Uh, I'm busy now, can you deliver it later or put it in the takeaway cabinet!", the specific steps are as follows:

[0124] S1. The voice input module or the voice output module outputs a specific start word to wake up the semantic database to communicate with the user;

[0125] 1. Speech-to-text conversion and preprocessing

[0126] 1.1: Voice reception and conversion

[0127] (1) Receive user's voice input.

[0128] (2) Use ASR technology to convert speech into text.

[0129] 1.2: Text Preprocessing

[0130] (1) Remove noise data from the text, including terminal punctuation (such as commas, periods, exclamation marks, etc.) and special symbols (such as modal particles such as "er" and "ba").

[0131] (2) After processing, a text message is obtained: "I am not free now, please deliver it later or put it in the takeaway cabinet."

[0132] S2. The semantic database under the intent recognition system is constructed by the intent recognition module to directly match special phrases, and the semantic denoising unit is used to assist in parsing phrases, save unmatched phrases, and record and mark frequently occurring phrases;

[0133] Word segmentation

[0134] initialization

[0135] (1) Get the maximum length of all entries in the word segmentation dictionary and set it to length. In this case, length = 8.

[0136] (2) Set an empty list to store the word segmentation results.

[0137] Scan and Match

[0138] (1) Starting from the beginning of the text, each time a string of length length is taken as a candidate word.

[0139] (2) Match the candidate words with the entries in the dictionary, comparing them one by one from the beginning to the end of the dictionary.

[0140] Recursive matching and segmentation

[0141] (1) If a match is found, then:

[0142] (1.1) Add the word to the word segmentation result list.

[0143] (1.2) Update the text, remove the words that have been matched, and if the remaining text begins with punctuation, remove it as well.

[0144] (1.3) Continue scanning from the beginning of the remaining text.

[0145] (2) If no match is found, then:

[0146] (2.1) Reduce the length of the candidate word by 1 (i.e. remove the rightmost word).

[0147] (2.2) If the length is greater than 1, regenerate the candidate word and try to match it.

[0148] (3) If no match is found after the length is reduced to 1, in this scenario, a single character is usually not directly used for intent recognition, so the second round of matching ends.

[0149] Repeat steps

[0150] Repeat steps S2 and S3 until the entire text has been processed.

[0151] Output

[0152] (1) Output the word segmentation result list, which contains all the words after the text is segmented.

[0153] (2) After the above steps, the obtained word segmentation and its corresponding intention weight results are "I am not free now|1", "Later|2", "Put it in the takeaway cabinet|3"

[0154] The word segmentation process is Figure 4 shown.

[0155] S3. The intention association module completes the frequency calculation of the phrases, classifies and sorts the frequency categories of the unrecorded phrases and the recognized phrases by the actual frequency, and finally confirms with the user again through the voice output module;

[0156] Word segmentation result processing:

[0157] (1) Select the word or phrase with the highest weight value as the best result;

[0158] (2) Therefore, “Put it in the takeaway cabinet” is selected as the best matching result;

[0159] (3) Reverse search the intent of “put in takeout cabinet” to obtain the robot’s reply statement;

[0160] (4) Play the speech to the user through speech synthesis, completing an IVR interaction;

[0161] Word frequency record:

[0162] The frequency weight of a term = (frequency of the term) / (total number of words).

[0163] Term: The basic unit of text after word segmentation, such as words, phrases, etc.

[0164] Frequency dictionary: used to store each term and its corresponding frequency. Usually a set of key-value pairs, where the key is the term and the value is the frequency.

[0165] Total words: An integer value representing the total number of terms in the document.

[0166] Frequency weight list: A list or array containing the frequency weight of each term. The weight is usually a decimal between 0 and 1.

[0167] step:

[0168] (1) After word segmentation, if the word segmentation results are obtained, for the word segmentation results, traverse each word item and use a hash table to record the number of times each word item appears.

[0169] (2) While traversing the word segmentation results, accumulate the number of all terms to obtain the total number of words.

[0170] (3) Obtain the frequency weight of each word as the actual weight basis for subsequent modification of the system weight preset.

[0171] In another embodiment, the present invention provides a computer device, including a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of the method embodiment or other corresponding system embodiments when executing the computer program, which will not be described in detail herein.

[0172] In another embodiment, the present invention provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the functions of the method embodiment or other corresponding system embodiments described above are implemented, which will not be described in detail herein.

[0173] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. An intelligent vertical business corpus construction and real-time intent recognition system, including a device equipped with an intent recognition system, characterized in that: The intention recognition system includes a speech input module, a semantic database and a speech output module, wherein the semantic database is connected to the speech input module and the speech output module respectively, and the speech input module and the speech output module are used for speech reception and speech output respectively; The semantic database also includes an intention recognition module, a semantic preservation unit, an intention association module and a semantic denoising unit, among which: The intent recognition module is used to build the main program of the system and define intent categories, recognize the type of voice input by the user and define the intent in the voice; The semantic storage unit is used to convert the intention in the user's voice into text and candidate word matching, quickly find the semantics of the word segmentation in the voice, and save the unmatched phrases in the storage unit; The semantic denoising unit is used to complete the phased word deletion of the sentence under the speech by the auxiliary semantic preservation unit, split the various types of intentions in the sentence and classify them as needed; The intention association module is used to construct a word frequency weight order for the unclear intentions in different voices, and calculate the word frequency based on the frequency of occurrence of phrases to complete the task order adjustment and recording.

2. According to claim 1, an intelligent vertical business corpus construction and real-time intent recognition system is characterized in that: The system main program of the intention recognition module also includes a start-up voice word for waking up the semantic database, and the start-up voice word completes the wake-up through the voice input module or the voice output module; The defined intention category is used to establish a special phrase, and link the semantic phrase to the corresponding trigger program of the semantic database to complete the delivery process of the task directly completed by the special phrase.

3. According to claim 2, an intelligent vertical business corpus construction and real-time intent recognition system is characterized in that: The semantic preservation unit includes a speech recognition algorithm, which records the speech through a speech input module and converts it into text content based on the speech recognition algorithm, and then matches the candidate words with special phrases under the intention recognition module, and records the unmatched phrases.

4. According to claim 3, an intelligent vertical business corpus construction and real-time intent recognition system is characterized in that: The semantic denoising unit deletes the text before and after the phrase in a sentence-decrementing manner, completes word segmentation matching under a set matching cycle, and forms a segmentation operation for the intention of the sentence to extract multiple different intentions under a single sentence.

5. According to claim 4, an intelligent vertical business corpus construction and real-time intent recognition system is characterized in that: The intention association module includes a word frequency recording unit, wherein the weight of the word frequency recording unit is calculated by the following formula: The frequency weight of a term = (frequency dictionary) / (total number of words); Term: the basic unit of text after word segmentation, such as words and phrases; Frequency dictionary: used to store each term and its corresponding frequency, usually a set of key-value pairs, where the key is the term and the value is the frequency; Total number of words: an integer value indicating the total number of terms in the document; Frequency weights of terms: a list or array containing the frequency weight of each term, which is usually a decimal between 0 and 1; After the weight list is completed, it can be used for subsequent manual adjustment of the semantic database and the sorting of task flows formed by the task distribution units. The specific steps are as follows: (1) After word segmentation is completed, the word segmentation results are obtained. For the word segmentation results, each word item is traversed and a hash table is used to record the number of times each word item appears; (2) While traversing the word segmentation results, accumulate the number of all terms to obtain the total number of words; (3) Obtain the frequency weight of each word as the actual weight basis for subsequent modification of the system weight preset.

6. According to claim 1, a method for constructing an intelligent vertical business corpus and using a real-time intent recognition system is characterized in that: The specific steps are as follows: S1. The voice input module or the voice output module outputs a specific start word, wakes up the semantic database to communicate with the user, records the user's voice and converts it into text; S2. The semantic database under the intent recognition system is constructed by the intent recognition module to directly match special phrases, and the semantic denoising unit is used to assist in parsing phrases, save unmatched phrases, and record and mark frequently occurring phrases; S3. The intention association module completes the frequency calculation of the phrases, classifies and sorts the frequency categories of unrecorded phrases and recognized phrases according to the actual frequencies, and finally confirms with the user again through the voice output module.

7. A computer device, comprising a processor, a memory, and a computer program stored in the memory and executable on the processor, wherein the processor implements the functions of the system according to any one of claims 1 to 5 or the steps of the method according to claim 5 when executing the computer program.

8. A computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements the functions of the system according to any one of claims 1 to 5 or the steps of the method according to claim 6.

Images