Business resource scheduling method based on privacy dialogue text

Abstract

The embodiment of the invention discloses a business resource scheduling method based on a private dialogue text. A specific embodiment of the method comprises the steps of performing privacy filtering processing on a received dialogue text to obtain a privacy dialogue text; performing service identification processing on the privacy dialogue text to obtain dialogue text intermediate data and a service scene identification result; according to the dialogue text intermediate data, generating a decision basis and a confidence score of a business scene recognition result; eliminating dialogue text intermediate data; according to the service scene identification result and a preset service rule base, generating a service resource scheduling instruction sequence, and sequentially executing service resource scheduling instructions in the service resource scheduling instruction sequence; and outputting the business scene recognition result to a terminal display of the business system, and storing the decision basis, the privacy dialogue text and the business scene recognition result to a preset log register. According to the embodiment, user information leakage is reduced, and response time is shortened.

Description

Technical Field

[0001] The embodiments disclosed herein relate to the field of computer technology, and more specifically to a method for scheduling business resources based on privacy-preserving dialogue text. Background Technology

[0002] With the development of intelligent enterprise services, in the field of smart customer service, business resource scheduling through dialogue text has become a means for enterprises to improve operational efficiency. Currently, the main method for business resource scheduling is to upload dialogue data to a cloud server, then directly parse the dialogue data based on prompt words, and generate resource scheduling instructions through a predefined rule base.

[0003] However, when using the above method to parse dialogue text and generate resource scheduling instructions, the following technical problems often arise: The dialogue text contains users' private information (ID number, mobile phone number, etc.), which is easily transmitted and processed directly in the Internet environment, posing a risk of information leakage. In addition, the use of complex and lengthy prompt words to guide large models to output specific classification results results in poor model stability and excessively long response times when facing business scenarios with ambiguous semantic boundaries (a dialogue or text content may involve multiple business topics at the same time).

[0004] The information disclosed in this background section is only intended to enhance the understanding of the background of the inventive concept, and therefore may contain information that does not constitute prior art known to those skilled in the art. Summary of the Invention

[0005] The summary portion of this disclosure is intended to provide a brief overview of the concepts, which will be described in detail in the detailed description portion that follows. This summary portion is not intended to identify key or essential features of the claimed technical solutions, nor is it intended to limit the scope of the claimed technical solutions.

[0006] Some embodiments of this disclosure propose a business resource scheduling method, apparatus, electronic device, and computer-readable medium based on privacy-preserving dialogue text to address one or more of the technical problems mentioned in the background section above.

[0007] In a first aspect, some embodiments of this disclosure provide a business resource scheduling method based on privacy-preserving dialogue text. The method includes: in response to receiving dialogue text sent by a client, performing privacy filtering on the dialogue text according to a privacy entity dictionary to obtain privacy-preserving dialogue text; performing business recognition processing on the privacy-preserving dialogue text based on a business scenario recognition model to obtain intermediate dialogue text data and a business scenario recognition result; generating a decision basis and confidence score for the business scenario recognition result based on the intermediate dialogue text data; clearing the intermediate dialogue text data in response to determining that the confidence score of the business scenario recognition result exceeds a preset standard value; generating a business resource scheduling instruction sequence based on the business scenario recognition result and a preset business rule base, and sequentially executing the business resource scheduling instructions in the business resource scheduling instruction sequence; outputting the business scenario recognition result to a terminal display of the business system, and storing the decision basis, the privacy-preserving dialogue text, and the business scenario recognition result in a preset log register according to a preset format.

[0008] Secondly, some embodiments of this disclosure provide a business resource scheduling apparatus based on privacy-preserving dialogue text. The apparatus includes: a processing unit configured to, in response to receiving dialogue text sent by a client, perform privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy-preserving dialogue text; a business identification unit configured to, based on a business scenario identification model, perform business identification processing on the privacy-preserving dialogue text to obtain intermediate dialogue text data and a business scenario identification result; a first generation unit configured to, based on the intermediate dialogue text data, generate a decision basis and a confidence score for the business scenario identification result; a clearing unit configured to, in response to determining that the confidence score of the business scenario identification result exceeds a preset standard value, clear the intermediate dialogue text data; a second generation unit configured to, based on the business scenario identification result and a preset business rule base, generate a business resource scheduling instruction sequence and execute the business resource scheduling instructions in the business resource scheduling instruction sequence sequentially; and an output unit configured to output the business scenario identification result to a terminal display of a business system and store the decision basis, the privacy-preserving dialogue text, and the business scenario identification result in a preset log register according to a preset format.

[0009] Thirdly, some embodiments of this disclosure provide an electronic device, including: one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first aspect above.

[0010] Fourthly, some embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon, wherein the program, when executed by a processor, implements the method described in any of the implementations of the first aspect above.

[0011] The above embodiments of this disclosure have the following beneficial effects: The business resource scheduling method based on privacy-preserving dialogue text in some embodiments of this disclosure can reduce user information leakage and shorten response time. Specifically, the reasons for user information leakage and excessively long response time are: dialogue text contains users' privacy information (ID number, mobile phone number, etc.), which is easily transmitted and processed directly in the Internet environment, posing a risk of information leakage. Furthermore, using complex and lengthy prompt word engineering to guide large models to output specific classification results results in poor model stability when facing business scenarios with ambiguous semantic boundaries (a dialogue or text content may involve multiple business topics simultaneously), leading to excessively long response times. Based on this, the business resource scheduling method based on privacy-preserving dialogue text in some embodiments of this disclosure can first, in response to receiving dialogue text sent by the client, perform privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy-preserving dialogue text. This allows for direct filtering of privacy information in the dialogue text. Secondly, based on a business scenario recognition model, the privacy-preserving dialogue text undergoes business recognition processing to obtain intermediate dialogue text data and business scenario recognition results. This reduces the stability problems caused by complex prompt word engineering, significantly improves recognition speed and accuracy, and reduces user waiting time. Next, based on the intermediate data of the aforementioned dialogue text, the decision basis and confidence score for the business scenario identification result are generated. This reduces the time spent on repetitive operations caused by misjudgment of business scenarios. Then, in response to the determination that the confidence score of the aforementioned business scenario identification result exceeds a preset standard value, the intermediate data of the aforementioned dialogue text is cleared. This reduces the time that private data remains in the system, further protecting user privacy information. Then, based on the aforementioned business scenario identification result and a preset business rule base, a sequence of business resource scheduling instructions is generated, and the business resource scheduling instructions in the sequence are executed sequentially. This directly converts the identification result into an executable instruction sequence, improving response efficiency. Finally, the aforementioned business scenario identification result is output to the terminal display of the business system, and the aforementioned decision basis, the aforementioned private dialogue text, and the aforementioned business scenario identification result are stored in a preset log register according to a preset format. This ensures the traceability of the processing process. Ultimately, this reduces user information leakage and shortens response time. Attached Figure Description

[0012] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and elements are not necessarily drawn to scale.

[0013] Figure 1 This is a flowchart of some embodiments of the business resource scheduling method based on privacy-preserving dialogue text according to the present disclosure; Figure 2 This is a schematic diagram of the structure of some embodiments of the business resource scheduling device based on privacy dialogue text according to the present disclosure; Figure 3 This is a schematic diagram of the structure of an electronic device suitable for implementing some embodiments of the present disclosure. Detailed Implementation

[0014] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0015] It should also be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings. Unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other.

[0016] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0017] It should be noted that the terms "a" and "a plurality of" used in this disclosure are illustrative rather than restrictive, and those skilled in the art should understand that, unless otherwise expressly indicated in the context, they should be understood as "one or more".

[0018] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0019] This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0020] Figure 1A flow 100 of some embodiments of a business resource scheduling method based on privacy-preserving dialogue text according to the present disclosure is shown. This business resource scheduling method based on privacy-preserving dialogue text includes the following steps: Step 101: In response to receiving the dialogue text sent by the client, perform privacy filtering on the dialogue text according to the privacy entity dictionary to obtain the privacy dialogue text.

[0021] In some embodiments, the execution entity (e.g., a server) of the business resource scheduling method based on privacy dialogue text may, in response to receiving dialogue text sent by a client, perform privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy dialogue text.

[0022] In addressing the aforementioned technical challenges by employing technical solutions, the following technical issues often arise when processing multi-source, dynamically changing privacy text in the intended application scenario: a fixed privacy entity dictionary struggles to cover newly added privacy entities during actual conversations, leading to identification vulnerabilities in the privacy filtering process and potential leakage of user privacy information. Considering the specific requirements of this application scenario—real-time processing and high confidentiality—we have decided to adopt the following solution: In some optional implementations of certain embodiments, the aforementioned executing entity may obtain the privacy dialogue text through the following steps: Step one: In response to receiving multi-source privacy entity data, construct a privacy semantic set. In practice, firstly, the aforementioned executing entity can preprocess the received multi-source privacy entity data to remove invalid and duplicate data, obtaining processed privacy entity data. Then, the executing entity can use a clustering algorithm (such as K-means clustering) to classify semantically similar privacy entity data into a unified category, obtaining a privacy semantic set. The multi-source privacy entity data can be user privacy information, including but not limited to: name, ID number, and phone number. Invalid data can include, but is not limited to, blank data (data fields with null values), data with missing values, and data containing garbled characters.

[0023] Step two involves sampling and labeling the aforementioned privacy semantic set to generate an initial privacy entity dictionary. In practice, the executing entity can use clustering algorithms (such as K-Means clustering) to determine all semantic categories in the privacy semantic set, obtaining a semantic category set. Then, the executing entity can use keyword extraction techniques (such as TF-IDF algorithm) to determine the most frequently occurring privacy entities corresponding to each semantic category in the semantic category set, and associate the semantic categories with the corresponding privacy entities to obtain the initial privacy entity dictionary. This initial privacy entity dictionary may include entity types (semantic categories) and corresponding standard terms (privacy entities).

[0024] Step 3: Based on the business knowledge graph, expand the initial privacy entity dictionary to obtain a privacy entity dictionary. In practice, the executing entity can expand the initial privacy entity dictionary with synonyms based on the business knowledge graph to obtain a privacy entity dictionary. As an example, the executing entity can expand the entity "telephone" to entities such as "mobile phone," "telephone number," and "phone number." The business knowledge graph can be a knowledge graph built for a specific business domain (such as customer service, healthcare, finance, etc.), and can include privacy entity types (such as ID card number, bank card number, password), related business concepts (such as account opening, transfer, and card cancellation), and corresponding relationships (such as synonymous relationships and association relationships (such as "account opening" requiring "ID card")).

[0025] Step four: Based on the aforementioned privacy entity dictionary, identify the privacy entities in the dialogue text to generate a privacy entity location list. In practice, the executing entity can identify privacy entities in the dialogue text that match the privacy entity dictionary through string matching, and determine the entity type (e.g., name, ID number), original text content, and starting index position of the matching privacy entity in the dialogue text to obtain the privacy entity location list. The privacy entity location list includes: entity type, privacy entity, and starting index.

[0026] Step 5: Based on the salted hash algorithm and differential privacy mechanism, reversibly encode each privacy entity in the privacy entity location list to obtain a token mapping table set. In practice, the executing entity can generate a random salt value for each privacy entity in the privacy entity location list and determine the hash value of the privacy entity according to a salted hash algorithm (such as SHA-256 salting). Then, the executing entity can assign perturbation parameters to the hash values ​​of each privacy entity according to a differential privacy mechanism (such as Laplace noise injection) to enhance the privacy protection strength. Finally, the executing entity can store the salt value, hash value, and perturbation parameters of each privacy entity as a mapping entry to generate a token mapping table set.

[0027] Step six: Based on the aforementioned token mapping table set, perform consistent replacements on each privacy entity in the aforementioned privacy entity location list to obtain the privacy dialogue text. In practice, the executing entity can first traverse the aforementioned token mapping table set to determine the tokens corresponding to the entity types of each privacy entity, obtain a token set, and then replace each privacy entity according to the aforementioned token set to obtain the privacy dialogue text.

[0028] Step seven involves randomly sampling the aforementioned privacy dialogue text to obtain a privacy dialogue text sample set. In practice, the executing entity can sample the aforementioned privacy dialogue text according to a preset sampling ratio (e.g., 5%) to generate at least one privacy dialogue text sample, thus obtaining a privacy dialogue text sample set.

[0029] Step eight: Based on the aforementioned token mapping table set, perform reverse decoding verification operations on each privacy dialogue text sample in the aforementioned privacy dialogue text sample set to generate verification results, thus obtaining a verification result set. In practice, the aforementioned executing entity can perform reverse decoding verification operations on each privacy dialogue text sample in the aforementioned privacy dialogue text sample set based on the aforementioned token mapping table set to generate verification results (such as "successful match", "decoding failed", or "entity mismatch"), thus obtaining a verification result set.

[0030] Step nine: Based on the aforementioned verification result set, generate an error report set and store it in the aforementioned preset log database. In practice, the executing entity can obtain an abnormal verification result set by filtering out verification results marked as failed or abnormal from the aforementioned verification result set. Then, the executing entity generates an error report set based on each abnormal verification result in the aforementioned abnormal verification result set and stores it in the aforementioned preset log database. The error report may include the identifier, location, expected original entity, decoded entity, and error type of the erroneous text. The aforementioned preset log database can be a pre-defined database for storing the aforementioned error report set, and can be a relational database (such as MySQL).

[0031] Step 10: Perform frequency statistics on each token mapping table in the aforementioned token mapping table set to generate a high-frequency token set. In practice, the executing entity can determine the aforementioned token mapping table as a high-frequency token set in response to the determination that the number of times the aforementioned token mapping table appears in historical privacy dialogue text exceeds a preset threshold (which could be 50 times).

[0032] Step 11: Based on the aforementioned high-frequency token set, perform an incremental update operation on the aforementioned privacy entity dictionary to generate an updated privacy entity dictionary. In practice, the executing entity can determine whether the privacy entity corresponding to each high-frequency token in the aforementioned high-frequency token set is a new privacy entity type or a synonym of an existing privacy entity type. In response to the privacy entity type in the aforementioned privacy entity dictionary not including the privacy entity type corresponding to the aforementioned high-frequency token, the privacy entity type corresponding to the aforementioned high-frequency token is added as a new entity type, and the aforementioned privacy entity dictionary is updated to generate an updated privacy entity dictionary.

[0033] Steps one through eleven above, and related content, constitute an inventive point of this disclosure, solving the technical problem that "it is difficult to cover the privacy entities in actual dialogues using a fixed, single privacy entity dictionary, resulting in identification vulnerabilities during privacy filtering and leading to the leakage of user privacy information." Factors leading to the leakage of user privacy information often include: the inability to cover the privacy entities in actual dialogues using a fixed, single privacy entity dictionary, resulting in identification vulnerabilities during privacy filtering and leading to the leakage of user privacy information. Solving these factors can resolve the problem of user privacy information leakage. To achieve this effect, the first step is to construct a privacy semantic set in response to receiving multi-source privacy entity data. This allows for the clustering of dispersed privacy entities, reducing identification vulnerabilities caused by unknown types. The second step is to sample and label the aforementioned privacy semantic set to generate an initial privacy entity dictionary. This improves the coverage of the privacy entity dictionary. The third step is to expand the initial privacy entity dictionary based on a business knowledge graph to obtain a new privacy entity dictionary. This reduces the problem of being unable to perform dynamic identification due to semantic rigidity. The fourth step is to identify the privacy entities in the aforementioned dialogue text based on the privacy entity dictionary to generate a privacy entity location list. This allows for more accurate entity localization, solving the problem of missed or false detections caused by inaccurate identification in traditional methods. Fifth, based on the salted hash algorithm and differential privacy mechanism, each privacy entity in the aforementioned privacy entity localization list is reversibly encoded to obtain a token mapping table set. This solves the problem of text leakage caused by directly processing text. Sixth, based on the aforementioned token mapping table set, each privacy entity in the aforementioned privacy entity localization list undergoes consistent replacement to obtain the privacy dialogue text. This ensures the semantic coherence of the text. Seventh, the aforementioned privacy dialogue text undergoes random sampling to obtain a privacy dialogue text sample set. This reduces computational overhead. Eighth, based on the aforementioned token mapping table set, each privacy dialogue text sample in the aforementioned privacy dialogue text sample set undergoes reverse decoding verification to generate verification results, obtaining a verification result set. This verifies the accuracy of the privacy filtering process. Ninth, based on the aforementioned verification result set, an error report set is generated and stored in the aforementioned preset log database. This allows for the aggregation of error types, ensuring rapid identification of the root cause of the problem. Step 10: Perform frequency statistics on each token mapping table in the aforementioned token mapping table set to generate a high-frequency token set. This solves the problem of delayed dictionary updates. Step 11: Based on the high-frequency token set, perform incremental updates on the aforementioned privacy entity dictionary to generate an updated privacy entity dictionary. This enables real-time dynamic updates to the privacy dictionary. Ultimately, new privacy entities in the dialogue text can be identified based on the updated privacy entity dictionary, thereby solving the problem of user privacy information leakage.

[0034] Step 102: Based on the business scenario recognition model, perform business recognition processing on the privacy dialogue text to obtain intermediate dialogue text data and business scenario recognition results.

[0035] In some embodiments, the aforementioned execution entity may perform business recognition processing on the aforementioned privacy dialogue text based on a business scenario recognition model to obtain intermediate dialogue text data and business scenario recognition results. The aforementioned intermediate dialogue text data includes a dialogue feature vector set and an abstract semantic representation set.

[0036] In practice, the aforementioned implementing entities can construct a business scenario identification model through the following steps: Step 1: Obtain the dialogue text set and the corresponding business tag set. In practice, the aforementioned execution entity can collect dialogue text and corresponding business tags from various business systems (such as intelligent customer service platforms and voice call record databases) through predefined API interfaces (such as RESTful API interfaces), thus obtaining the dialogue text set and the business tag set. The business tags can be manually labeled and represent the dialogue text scenario category, and may include, but are not limited to, "card cancellation and reactivation," "account registration," and "service activation."

[0037] Step two: Based on the preset dialogue model and dialogue rewriting prompts, the above dialogue text set is rewritten for privacy, resulting in a de-identified text set. In practice, the executing entity can import the Large Language Model (LLM) into the local area network via encrypted storage media (such as an external hard drive) to ensure data security. Then, the executing entity can input the dialogue rewriting prompts into the preset dialogue model to obtain the de-identified text. This can be achieved using the following formula: .

[0038] .

[0039] in, This indicates a pre-defined dialogue model. This indicates a local area network (LAN). Represents a large language model. This indicates desensitized text. This indicates the dialogue text. The dialogue rewriting prompt can be used to specify the rules and output requirements that the aforementioned preset dialogue model should follow when processing dialogue text. As an example, the dialogue rewriting prompt could be: "Based on the following customer service call record, remove redundant content from the call (such as interjections and greetings like 'um,' 'ah,' 'this,' 'that,' etc.), retain private information, preserve the original text, and output a more fluent call text according to the original call format."

[0040] Step 3: Based on the aforementioned preset dialogue model and preset tag augmentation prompts, semantically augment the aforementioned business tag set to obtain a descriptive text set. In practice, the aforementioned executing entity can input the aforementioned preset tag augmentation prompts and the aforementioned business tags into the aforementioned isolation model to obtain the descriptive text. The descriptive text can be obtained using the following formula: .

[0041] .

[0042] in, This indicates preset tag augmentation suggestions. This refers to template functions used to generate tag augmentation suggestions (such as descriptions of "..."). "Business Scenario Categories" This indicates a pre-defined dialogue model. This indicates the dialogue text. This represents descriptive text. This represents a business tag. The template function described above can be a predefined string generation rule. As an example, the template function above can generate a string from the input value " Fill into the preset text frame.

[0043] Step four: Based on the aforementioned dialogue text set, the aforementioned anonymized text set, and the aforementioned description text set, generate at least one dialogue sample to obtain a dialogue sample set. In practice, the aforementioned executing entity can combine the aforementioned dialogue text set, the aforementioned anonymized text set, and the aforementioned description text set into a single dialogue sample set.

[0044] Step 5: Based on the aforementioned dialogue sample set, adjust the instructions of the aforementioned preset dialogue model to obtain the adjusted preset dialogue model, which serves as the business scenario recognition model. In practice, the aforementioned executing entity can adjust the instructions of the aforementioned preset dialogue model based on the aforementioned dialogue sample set to obtain the adjusted preset dialogue model, which serves as the business scenario recognition model, through the following steps: Sub-step one involves masking irrelevant words in each dialogue sample in the aforementioned dialogue sample set, resulting in a masked dialogue sample set. In practice, the aforementioned executing entity can use masking techniques to conceal irrelevant words in the dialogue samples that are not relevant to the business scenario recognition task. These irrelevant words can include stop words (such as interjections and greetings) and domain-irrelevant words (such as advertising content).

[0045] Sub-step two involves encoding each dialogue sample in the masked dialogue sample set to obtain a dialogue feature vector set. In practice, the execution entity can freeze the parameters of the text mapping part of the Recurrent Neural Network (RNN) to encode each dialogue sample in the masked dialogue sample set, generating the masked dialogue feature vector set. The masked dialogue feature vector set can be obtained using the following formula: .

[0046] .

[0047] in, This represents the merged text. This indicates the dialogue text. This indicates desensitized text. This represents a recurrent neural network. This represents the dialogue feature vector. This represents the merge operator.

[0048] Sub-step three involves adding a low-rank matrix to the output layer of the aforementioned pre-trained dialogue model and updating the original weights of the model. In practice, the execution entity can be fine-tuned using the LoRA method for the pre-trained weight matrix. The incremental parameter matrix of full parameter fine-tuning Represented as a matrix with two smaller parameters and The low-rank similarity matrix can be obtained using the following formula: .

[0049] in, This represents the weight matrix. This represents the incremental parameter matrix. Let represent a dimension-reduced matrix with dimensions d×r and rank d less than r. This represents an upgraded matrix with dimensions r×d.

[0050] Sub-step four: Based on the above-mentioned preset dialogue model, determine the projection vector of each dialogue feature vector in the above-mentioned dialogue feature vector set to obtain the projection vector set.

[0051] .

[0052] in, This represents the weight matrix. This represents a dimension reduction matrix. This represents an upgraded matrix. This represents the bias term, with a dimension of llm. This represents the dialogue feature vector. This represents the learnable weight matrix, with dimensions llm×d. This represents the projection vector. llm represents the spatial vector dimension of the large model.

[0053] Sub-step five involves determining the similarity between the aforementioned projection vector set and the aforementioned business tag set, thus obtaining a similarity set. This similarity can be cosine similarity.

[0054] Sub-step six: Based on the aforementioned similarity set, determine the loss value of the preset dialogue model, and adjust the preset dialogue model according to the loss value to obtain the business scenario recognition model. In practice, the executing entity can determine the loss values ​​of the low-rank matrix and projection vector based on the aforementioned similarity set, and adjust the preset dialogue model according to the loss value to obtain the business scenario recognition model. The loss value can be obtained using the following formula: .

[0055] in, This indicates the business tag index. This indicates a text index. This represents the index variable, used to iterate through all samples. This indicates the total number of text samples. Indicates the first Predictive business tags. Indicates the first The truth value of the label for each text. This represents the loss value. This represents the similarity function, which can be a cosine similarity function. Indicates the first Candidate business tags. Represents the natural constant.

[0056] In some optional implementations of certain embodiments, the aforementioned execution entity can obtain intermediate dialogue text data and business scenario recognition results through the following steps: Step one involves standardizing the aforementioned privacy dialogue text to obtain standardized privacy dialogue text. In practice, the executing entity can use natural language processing tools to remove redundant information (such as interjections like "um," "ah," and other polite greetings), complete any missing words, and segment and reorganize the text according to predetermined formatting specifications to obtain standardized privacy dialogue text. These predetermined formatting specifications may include punctuation and character encoding standards (e.g., uniform use of full-width characters) and segmentation and line break standards (e.g., one paragraph per speaker's privacy dialogue text). The natural language processing tools can be HanLP (Han Language Processing), NLTK (Natural Language Toolkit), or Spacy (Natural Language Processing Library).

[0057] Step two involves inputting the processed privacy-preserving dialogue text into the business scenario recognition model to obtain a dialogue feature vector set. In practice, the executing entity can input the processed privacy-preserving dialogue text into the business scenario recognition model, and through the encoding layer of the business scenario recognition model, convert the processed privacy-preserving dialogue text into a high-dimensional numerical vector to generate a dialogue feature vector.

[0058] Step 3: Based on the aforementioned dialogue feature vector set, generate an abstract semantic representation set. In practice, the aforementioned execution entity can use the deep neural network of the aforementioned business scenario recognition model to convert the aforementioned dialogue feature vector set into an abstract semantic representation set. As an example, the aforementioned execution entity can convert "I want to cancel my credit card" and "How to cancel the card" into "User intent: Terminate card service".

[0059] Step four: Based on preset business scenario labels, determine the business scenario probability of each dialogue feature vector in the above dialogue feature vector set, thus obtaining a business scenario probability set. In practice, the executing entity can input the above dialogue feature vector set into the classification layer of the above business scenario recognition model based on preset business scenario labels to determine the business scenario probability of each dialogue feature vector in the above dialogue feature vector set, thereby obtaining a business scenario probability set. The preset business scenario labels can be predefined business category labels, which may include, but are not limited to, card cancellation and reopening, new card application, and business inquiries (such as points inquiry and application progress inquiry).

[0060] Step 5: Based on the aforementioned business scenario probability set, determine the business scenario identification result. In practice, the executing entity can determine the business scenario with the highest probability in the aforementioned business scenario probability set as the business scenario identification result.

[0061] In addressing the technical challenges mentioned above, the application scenario of extracting text from audio data for business scenario recognition often presents the following problems: the original audio environment contains noise pollution, semantic ambiguity, and invalid segments. Directly converting the original audio data into text leads to a high error rate and long recognition time in speech recognition. Considering the following requirements for this application scenario—high noise robustness and real-time processing—we have decided to adopt the following solution: In practice, the aforementioned implementing entity can obtain the dialogue text through the following steps: The first step is to obtain the dialogue audio data based on the business system. In practice, the aforementioned execution entity can obtain the dialogue audio data containing the user's call content with customer service by calling the predefined API interface in the business system.

[0062] The second step is to perform noise reduction processing on the aforementioned dialogue audio data to obtain noise-reduced audio data. In practice, the executing entity can use an audio noise reduction algorithm to determine the frequency domain characteristics representing environmental background noise and sudden interference impulse noise in the aforementioned dialogue audio data, and filter these frequency domain characteristics to obtain noise-reduced audio data. The aforementioned environmental background noise can be steady-state noise, including but not limited to fan noise and the sound of an air conditioner's outdoor unit rotating. The aforementioned sudden interference impulse noise can be transient noise, including but not limited to coughing sounds, the sound of objects falling, and the sound of metal knocking. The aforementioned audio noise reduction algorithm can be a real-time noise reduction algorithm based on a statistical model.

[0063] The third step involves endpoint segmentation of the aforementioned noise-reduced audio data to obtain short-time audio data. In practice, the executing entity can use an End-point Detection (EPD) algorithm to determine the short-time energy and zero-crossing rate of the noise-reduced audio data, thereby identifying the start and end positions of each valid speech segment in the audio data. Based on these start and end positions, the audio data is then segmented into multiple audio segments, which serve as short-time audio data. This short-time audio data can include multiple audio segments of valid speech. These valid speech segments can represent the content of the speaker's speech.

[0064] The fourth step involves performing voiceprint recognition on the aforementioned short-time audio data to obtain identifiable audio data. In practice, the executing entity can determine the Mel-frequency cepstral coefficients of each audio segment in the short-time audio data and the cosine similarity between these Mel-frequency cepstral coefficients and the user's voiceprint in a pre-set voiceprint library to generate the user tag most relevant to the audio segment, thus obtaining identifiable audio data. The pre-set voiceprint library can be a voiceprint model of a registered user. The user tag can be a tag representing the user's identity, such as "Customer Service A" or "User A".

[0065] The fifth step is to convert the aforementioned audio data into text data to obtain the original dialogue text. In practice, the executing entity can use Automatic Speech Recognition (ASR) technology to convert the aforementioned audio data into text data to obtain the original dialogue text. The original dialogue text includes user identity information and corresponding dialogue information.

[0066] The sixth step is to remove redundant information from the original dialogue text to obtain the cleaned text. This redundant information can include interjections, repeated words, and polite greetings (such as "hello" and "thank you").

[0067] The seventh step is to complete the cleaned text by adding missing words, resulting in the completed text. In practice, the aforementioned execution entity can use the natural language processing tools and contextual information to complete text containing missing or incorrect words due to unclear speech or noise interference.

[0068] Step 8: Sentence segmentation and reorganization of the completed text to obtain the reorganized text. In practice, the executing entity can segment and reorganize lengthy sentences in the completed text based on punctuation and semantic boundaries to obtain the reorganized text. For example, "Hello customer, I am customer service representative Zhang San. How can I help you?" can be reorganized into "Hello customer," "I am customer service representative Zhang San," and "How can I help you?"

[0069] The ninth step involves standardizing the format of the reconstructed text to obtain the standardized original dialogue text, which is then sent to the client. In practice, the executing entity can replace full-width characters in the reconstructed text with half-width characters and standardize the format of each sentence in the reconstructed text according to the preset template of "Speaker: Dialogue Content" to obtain the dialogue text.

[0070] The first to ninth steps and related content described above constitute an inventive point of this disclosure, solving the technical problem that "the original audio environment contains noise pollution, semantic ambiguity, and invalid segments, leading to high speech recognition error rates and long recognition times when directly converting the original audio data to text." The factors contributing to high speech recognition error rates and long recognition times are often as follows: the original audio environment contains noise pollution, semantic ambiguity, and invalid segments, leading to high speech recognition error rates and long recognition times when directly converting the original audio data to text. Solving these factors can resolve the problems of high speech recognition error rates and long recognition times. To achieve this effect, the first step involves acquiring dialogue audio data based on the business system. This yields the audio data to be recognized. The second step involves denoising the dialogue audio data to obtain denoised audio data. This reduces the noise in the audio data. The third step involves segmenting the denoised audio data into short-time audio data. This eliminates invalid audio segments and reduces the amount of data processing. The fourth step involves performing voiceprint recognition on the short-time audio data to obtain identifier audio data. This reduces recognition errors caused by multi-person dialogues. The fifth step is to convert the aforementioned audio data into text data to obtain the original dialogue text, which includes user identity information and dialogue information. This converts audio data into text data. The sixth step removes redundant information from the original dialogue text to obtain cleaned text. This eliminates interference from irrelevant words. The seventh step performs word completion on the cleaned text to obtain completed text. This improves the semantic integrity of the text and reduces semantic ambiguity. The eighth step restructures the completed text into sentences to obtain restructured text. This makes the text more conform to natural language norms, reduces processing delays caused by sentence structure confusion, and improves recognition efficiency. The ninth step standardizes the format of the restructured text to obtain standardized original dialogue text, which is then sent to the client. This unifies the text format. Ultimately, this reduces the speech recognition error rate and shortens the recognition time.

[0071] Step 103: Based on the intermediate data of the dialogue text, generate the decision basis and confidence score for the business scenario recognition result.

[0072] In some embodiments, the aforementioned executing entity may generate the decision basis and confidence score of the aforementioned business scenario identification result based on the aforementioned intermediate dialogue text data.

[0073] In practice, the aforementioned implementing entities can obtain decision-making basis and confidence scores through the following steps: Step 1: Extract semantic elements from the intermediate data of the above dialogue text to obtain a semantic element set. These semantic elements can include user intent, core operations, involved business entities (such as product or service names), and key modifiers (such as words indicating degree).

[0074] Step two involves performing a decision transformation on the aforementioned semantic element set to obtain the decision basis. In practice, the executing entity can select the corresponding pre-set decision template from the pre-set decision template library based on the business scenario identification results, and fill the corresponding position of the aforementioned semantic element set into the pre-set decision template library to obtain the decision basis. The aforementioned pre-set decision template can be a predefined decision template for various business scenarios. As an example, the aforementioned pre-set decision template could be: "Identification basis: The user expressed a [intent] need, involving [entity] business, requiring the execution of [business resource scheduling] operations."

[0075] Step 3: Determine the highest probability value set and the second highest probability value set of the above business scenario probability set. In practice, the executing entity can traverse the above business scenario probability set to generate probability values ​​belonging to the above business scenario identification results, thus obtaining the highest probability value set, and the highest probability value not belonging to the above business scenario identification results, thus obtaining the second highest probability value set.

[0076] Step four: Based on the aforementioned set of highest probability values ​​and the set of second-highest probability values, determine the set of absolute differences. In practice, the executing entity can obtain the set of absolute differences by iterating through the second-highest probability values ​​in the set of second-highest probability values ​​corresponding to each highest probability value in the set of highest probability values.

[0077] Step 5: Generate a confidence score set based on the highest probability value set and the absolute difference set. The confidence score in the confidence score set is calculated as: highest probability value in the highest probability value set × (1 + α × absolute difference in the absolute difference set). α can be a preset adjustment coefficient for different business scenarios, with a value range of [0, 1]. For example, in the "card cancellation and reopening" scenario, α can be 0.85. In the "new card application" scenario, α can be 0.75.

[0078] Step six: Encapsulate the aforementioned confidence score set and decision-making criteria to obtain an evidence package. In practice, the executing entity can encapsulate the aforementioned confidence score set and decision-making criteria using a preset format to obtain the evidence package. The preset format can be JSON or XML.

[0079] Step 104: In response to the determination that the confidence score of the business scenario recognition result exceeds the preset standard value, clear the intermediate data of the dialogue text.

[0080] In some embodiments, the aforementioned execution entity may clear the aforementioned intermediate dialogue text data in response to determining that the confidence score of the aforementioned business scenario identification result exceeds a preset standard value.

[0081] In practice, the aforementioned executing entity generates a clearing instruction in response to determining that the confidence score of the business scenario identification result exceeds a preset standard value. Then, the executing entity can clear the intermediate data of the dialogue text according to the clearing instruction to eliminate the risk of information leakage. Finally, the executing entity can generate a clearing log record and write the clearing log record to the preset log database. The preset standard value can be 0.75. The clearing instruction can be an instruction to delete the physical address of the intermediate data of the dialogue text. The clearing log record can include the clearing time, business scenario identification result identifier, and confidence score.

[0082] Step 105: Based on the business scenario identification results and the preset business rule base, generate a sequence of business resource scheduling instructions, and execute the business resource scheduling instructions in the sequence in sequence.

[0083] In some embodiments, the aforementioned execution entity may generate a sequence of business resource scheduling instructions based on the business scenario identification results and a preset business rule base, and execute the business resource scheduling instructions in the sequence of business resource scheduling instructions in sequence.

[0084] In practice, the aforementioned executing entity can generate a sequence of business resource scheduling instructions based on the business scenario identification results and the preset business rule base through the following steps, and then execute the business resource scheduling instructions in the sequence in sequence: The first step is to generate an initial instruction set based on the business scenario identification results. In practice, the executing entity can determine the initial instruction set corresponding to the business scenario identification results based on a preset business rule base. This preset business rule base can be a database that includes business scenarios, business resource scheduling instructions, and the mapping relationship between business scenarios and business resource scheduling instructions.

[0085] The second step involves updating the initial instruction set to obtain a business resource scheduling instruction sequence. In practice, the executing entity can populate the initial instruction set with business operation objects (such as users, user IDs, and products) and sort the initial instruction set according to the order of business logic to obtain the business resource scheduling instruction sequence. The business resource scheduling instructions in this sequence may include, but are not limited to, database update instructions, message notification instructions, machine call instructions, and resource allocation instructions. Database update instructions can be used to modify or update user account information, business status, or transaction records in the database. Message notification instructions can be used to send business processing notifications or business confirmation information to users via communication channels (such as SMS or email). Machine call instructions can be used to control associated business robots to perform relevant business operations (such as the business identification result representing "card application," and the business robot selecting and packaging bank cards from the warehouse). Resource allocation instructions are used to reserve computing resources and network bandwidth for high-priority businesses to avoid resource conflicts.

[0086] The third step involves executing the business resource scheduling instructions in the aforementioned sequence sequentially via a pre-defined instruction execution interface. In practice, the executing entity can use this pre-defined instruction execution interface to distribute the business resource scheduling instructions in the sequence to the corresponding target modules for execution. This pre-defined instruction execution interface can be a RESTful API.

[0087] Step 106: Output the business scenario recognition result to the terminal display of the business system, and store the privacy dialogue text and the business scenario recognition result in the preset log register according to the preset format.

[0088] In some embodiments, the aforementioned execution entity may output the aforementioned business scenario identification results to the terminal display of the business system, and store the aforementioned privacy dialogue text and the aforementioned business scenario identification results in a preset log register according to a preset format.

[0089] In practice, the aforementioned executing entity can output the business scenario identification results to the terminal display of the business system via a network communication interface. Then, according to the aforementioned preset format, the aforementioned privacy dialogue text and the aforementioned business scenario identification results are stored in a preset log register. The aforementioned network communication interface can be a RESTful API interface. The aforementioned preset log register can be a file system log. The aforementioned preset format can be a predefined, standardized, machine-readable data structure and encoding rule format, and can be JSON format.

[0090] The above embodiments of this disclosure have the following beneficial effects: The business resource scheduling method based on privacy-preserving dialogue text in some embodiments of this disclosure can reduce user information leakage and shorten response time. Specifically, the reasons for user information leakage and excessively long response time are: dialogue text contains users' privacy information (ID number, mobile phone number, etc.), which is easily transmitted and processed directly in the Internet environment, posing a risk of information leakage. Furthermore, using complex and lengthy prompt word engineering to guide large models to output specific classification results results in poor model stability when facing business scenarios with ambiguous semantic boundaries (a dialogue or text content may involve multiple business topics simultaneously), leading to excessively long response times. Based on this, the business resource scheduling method based on privacy-preserving dialogue text in some embodiments of this disclosure can first, in response to receiving dialogue text sent by the client, perform privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy-preserving dialogue text. This allows for direct filtering of privacy information in the dialogue text. Secondly, based on a business scenario recognition model, the privacy-preserving dialogue text undergoes business recognition processing to obtain intermediate dialogue text data and business scenario recognition results. This reduces the stability problems caused by complex prompt word engineering, significantly improves recognition speed and accuracy, and reduces user waiting time. Next, based on the intermediate data of the aforementioned dialogue text, the decision basis and confidence score for the business scenario identification result are generated. This reduces the time spent on repetitive operations caused by misjudgment of business scenarios. Then, in response to the determination that the confidence score of the aforementioned business scenario identification result exceeds a preset standard value, the intermediate data of the aforementioned dialogue text is cleared. This reduces the time that private data remains in the system, further protecting user privacy information. Then, based on the aforementioned business scenario identification result and a preset business rule base, a sequence of business resource scheduling instructions is generated, and the business resource scheduling instructions in the sequence are executed sequentially. This directly converts the identification result into an executable instruction sequence, improving response efficiency. Finally, the aforementioned business scenario identification result is output to the terminal display of the business system, and the aforementioned decision basis, the aforementioned private dialogue text, and the aforementioned business scenario identification result are stored in a preset log register according to a preset format. This ensures the traceability of the processing process. Ultimately, this reduces user information leakage and shortens response time.

[0091] Further reference Figure 2 As an implementation of the methods shown in the above figures, this disclosure provides some embodiments of a business resource scheduling method apparatus based on privacy-preserving dialogue text. These apparatus embodiments are similar to... Figure 1 Corresponding to the method embodiments shown, this business resource scheduling device based on privacy dialogue text can be specifically applied to various electronic devices.

[0092] like Figure 2As shown, a business resource scheduling device 200 based on privacy dialogue text includes: a processing unit 201, a business identification unit 202, a first generation unit 203, a clearing unit 204, a second generation unit 205, and an output unit 206. The system includes the following components: a processing unit 201, configured to perform privacy filtering on the dialogue text sent by the client in response to receiving the dialogue text, based on a privacy entity dictionary, to obtain privacy dialogue text; a business identification unit 202, configured to perform business identification processing on the privacy dialogue text based on a business scenario identification model, to obtain intermediate dialogue text data and business scenario identification results; a first generation unit 203, configured to generate decision criteria and confidence scores for the business scenario identification results based on the intermediate dialogue text data; a clearing unit 204, configured to clear the intermediate dialogue text data in response to determining that the confidence score of the business scenario identification results exceeds a preset standard value; a second generation unit 205, configured to generate a business resource scheduling instruction sequence based on the business scenario identification results and a preset business rule base, and to execute the business resource scheduling instructions in the sequence sequentially; and an output unit 206, configured to output the business scenario identification results to the terminal display of the business system, and to store the decision criteria, the privacy dialogue text, and the business scenario identification results in a preset log register according to a preset format.

[0093] It is understandable that the units and references recorded in the business resource scheduling device 200 based on privacy dialogue text... Figure 1 The steps in the described method correspond accordingly. Therefore, the operations, features, and beneficial effects described above for the method also apply to the business resource scheduling device 200 based on privacy-preserving dialogue text and the units contained therein, and will not be repeated here.

[0094] The following is for reference. Figure 3 It illustrates electronic devices suitable for implementing some embodiments of the present disclosure (such as...). Figure 1 The diagram shows the structure of the computing device 101) 300. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments of this disclosure.

[0095] like Figure 3As shown, the electronic device 300 may include a processing unit (e.g., a central processing unit, a graphics processing unit, etc.) 301, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 302 or a program loaded from a storage device 308 into a random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing unit 301, ROM 302, and RAM 303 are interconnected via a bus 304. An input / output (I / O) interface 305 is also connected to the bus 304.

[0096] Typically, the following devices can be connected to I / O interface 305: input devices 306 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 307 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 308 including, for example, magnetic tapes, hard disks, etc.; and communication devices 309. Communication device 309 allows electronic device 300 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 3 An electronic device 300 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively. Figure 3 Each box shown can represent a device or multiple devices as needed.

[0097] In particular, according to some embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, some embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 309, or installed from a storage device 308, or installed from a ROM 302. When the computer program is executed by the processing device 301, it performs the functions defined in the methods of some embodiments of this disclosure.

[0098] It should be noted that, in some embodiments of this disclosure, the computer-readable medium may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In some embodiments of this disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In some embodiments of this disclosure, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0099] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and peer-to-peer networks (e.g., ad hoc peer-to-peer networks), as well as any currently known or future-developed networks.

[0100] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently without being assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: upon receiving dialogue text sent by a client, perform privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy dialogue text; perform business recognition processing on the privacy dialogue text based on a business scenario recognition model to obtain intermediate dialogue text data and business scenario recognition results; generate decision criteria and confidence scores for the business scenario recognition results based on the intermediate dialogue text data; clear the intermediate dialogue text data upon determining that the confidence score of the business scenario recognition results exceeds a preset standard value; generate a sequence of business resource scheduling instructions based on the business scenario recognition results and a preset business rule base, and execute the business resource scheduling instructions in the sequence sequentially; output the business scenario recognition results to the terminal display of the business system, and store the decision criteria, the privacy dialogue text, and the business scenario recognition results in a preset log register according to a preset format.

[0101] Computer program code for performing operations of some embodiments of this disclosure can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0102] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0103] The units described in some embodiments of this disclosure can be implemented in software or hardware. The described units can also be housed in a processor; for example, a processor may be described as including a processing unit, a service identification unit, a first generation unit, a clearing unit, a second generation unit, and an output unit. The names of these units do not necessarily limit the unit itself; for example, a processing unit may also be described as "a unit that, in response to receiving dialogue text sent by a client, performs privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy dialogue text."

[0104] The functions described above in this document can be performed at least in part by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), complex programmable logic devices (CPLDs), and so on.

[0105] The above description is merely a selection of preferred embodiments of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in the embodiments of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above-described features with (but not limited to) technical features with similar functions disclosed in the embodiments of this disclosure.

Claims

1. A business resource scheduling method based on privacy-preserving dialogue text, comprising: In response to receiving the dialogue text sent by the client, the dialogue text is subjected to privacy filtering processing according to the privacy entity dictionary to obtain privacy dialogue text; Based on the business scenario recognition model, the privacy dialogue text is processed for business recognition to obtain intermediate dialogue text data and business scenario recognition results. Based on the intermediate data of the dialogue text, the decision basis and confidence score of the business scenario recognition result are generated; In response to determining that the confidence score of the business scenario identification result exceeds a preset standard value, the intermediate data of the dialogue text is cleared; Based on the business scenario identification results and the preset business rule base, a business resource scheduling instruction sequence is generated, and the business resource scheduling instructions in the business resource scheduling instruction sequence are executed sequentially. The business scenario identification result is output to the terminal display of the business system, and the decision basis, the privacy dialogue text, and the business scenario identification result are stored in a preset log register according to a preset format.

2. The method according to claim 1, wherein, The business scenario recognition model is generated through the following steps: Obtain the dialogue text set and the corresponding business tag set; Based on a preset dialogue model and dialogue rewriting prompts, the dialogue text set is rewritten for privacy to obtain a desensitized text set. Based on the preset dialogue model and preset tag augmentation prompts, the business tag set is semantically augmented to obtain a descriptive text set; Based on the dialogue text set, the anonymized text set, and the description text set, at least one dialogue sample is generated to obtain a dialogue sample set. Based on the set of dialogue samples, the preset dialogue model is adjusted according to instructions to obtain the adjusted preset dialogue model, which serves as the business scenario recognition model.

3. The method according to claim 2, wherein, The step of adjusting the preset dialogue model according to the dialogue sample set to obtain an adjusted preset dialogue model, which serves as a business scenario recognition model, includes: For each dialogue sample in the dialogue sample set, irrelevant words are masked to obtain the masked dialogue sample set; Encode each dialogue sample in the masked dialogue sample set to obtain a dialogue feature vector set; Add a low-rank matrix to the output layer of the preset dialogue model and update the original weights of the preset dialogue model; Based on the preset dialogue model, determine the projection vector of each dialogue feature vector in the dialogue feature vector set to obtain the projection vector set; Determine the similarity between the projection vector set and the business tag set to obtain a similarity set; Based on the similarity set, the loss value of the preset dialogue model is determined, and the preset dialogue model is adjusted according to the loss value to obtain a business scenario recognition model.

4. The method according to claim 1, wherein, The business scenario recognition model performs business recognition processing on the privacy dialogue text to obtain intermediate dialogue text data and business scenario recognition results, including: The privacy dialogue text is standardized to obtain standardized privacy dialogue text; The processed privacy dialogue text is input into the business scenario recognition model to obtain a dialogue feature vector set; Based on the dialogue feature vector set, an abstract semantic representation set is generated; Based on preset business scenario labels, the business scenario probability of each dialogue feature vector in the dialogue feature vector set is determined to obtain a business scenario probability set. The business scenario identification result is determined based on the probability set of the business scenario.

5. The method according to claim 1, wherein, The intermediate data of the dialogue text includes a set of dialogue feature vectors and a set of abstract semantic representations.

6. A business resource scheduling device based on privacy-preserving dialogue text, comprising: The processing unit is configured to, in response to receiving dialogue text sent by the client, perform privacy filtering processing on the dialogue text according to a privacy entity dictionary to obtain privacy dialogue text; The business identification unit performs business identification processing on the privacy dialogue text based on the business scenario identification model to obtain intermediate dialogue text data and business scenario identification results. The first generation unit generates the decision basis and confidence score of the business scenario recognition result based on the intermediate data of the dialogue text. The clearing unit, in response to determining that the confidence score of the business scenario identification result exceeds a preset standard value, clears the intermediate data of the dialogue text; The second generation unit generates a sequence of business resource scheduling instructions based on the business scenario identification results and the preset business rule base, and executes the business resource scheduling instructions in the sequence of business resource scheduling instructions in sequence. The output unit outputs the business scenario identification result to the terminal display of the business system, and stores the decision basis, the privacy dialogue text and the business scenario identification result in a preset log register according to a preset format.

7. An electronic device, comprising: One or more processors; A storage device on which one or more programs are stored; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1 to 5.

8. A computer-readable medium having a computer program stored thereon, wherein, When the program is executed by the processor, it implements the method as described in any one of claims 1 to 5.

Images