A Real-Time Error Correction Method and System for Outbound Call Semantics Based on Collaborative Reasoning
By leveraging the collaborative reasoning of outbound call domain knowledge graphs and large language models, intelligent contradiction detection and automatic correction of outbound call dialogue texts are achieved, solving the problems of insufficient accuracy and consistency of dialogue content in existing technologies and improving communication efficiency.
Patent Information
- Application Number
- CN202611231827.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2026-08-14
- Publication Date
- 2026-11-10
AI Technical Summary
Existing intelligent outbound calling systems lack accuracy and consistency in detecting and correcting inconsistencies in user statements, leading to decreased communication efficiency. Current technologies lack automatic correction mechanisms driven by deep matching, large language model recognition, and semantic comparison.
By acquiring the speech recognition text of the outbound call purpose and communication with the user, using the outbound call domain knowledge graph to match knowledge data, and combining it with a large language model to identify contradictory text and generate corrected text based on semantic comparison, intelligent contradiction detection and automatic correction are achieved.
It improves the accuracy and consistency of outbound call content, reduces the risk of decreased communication efficiency due to contradictory or incorrect user statements, and improves communication quality.
Smart Images

Figure CN123072379A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of data processing technology, and in particular to a method and system for real-time error correction of outbound call semantics based on collaborative reasoning. Background Technology
[0002] With the rapid adoption of intelligent outbound calling systems in marketing, customer service, and user outreach, outbound platform operators are increasingly focusing on improving communication accuracy and efficiency through intelligent processing of dialogue text. A key technical challenge is how to detect and correct inconsistencies in user statements to prevent decreased communication efficiency. Existing technologies typically acquire speech recognition text during outbound calls, using fixed rule matching or simple keyword filtering to check text content, and then correcting it based on manual review or basic templates to support outbound dialogue management. However, existing solutions lack deep matching using knowledge graphs in the outbound calling domain, intelligent recognition of inconsistencies using large language models, and automatic correction mechanisms driven by semantic comparison. This makes it difficult to achieve real-time inconsistency detection and accurate correction of dialogue text. Commonly used static rules or manual intervention methods cannot effectively address the complexity and ambiguity of user statements, leading to insufficient accuracy and consistency in dialogue content. Inconsistencies or errors in user statements can easily cause decreased communication efficiency, misunderstandings, or subsequent process problems, limiting the overall communication quality and service effectiveness of intelligent outbound calling systems. Therefore, existing technologies have shortcomings that urgently need to be addressed. Summary of the Invention
[0003] The technical problem to be solved by the present invention is to provide a method and system for real-time semantic error correction of outbound calls based on collaborative reasoning, which can realize intelligent contradiction detection and automatic correction of outbound dialogue text, improve the accuracy and consistency of dialogue content, and reduce the risk of decreased outbound communication efficiency due to contradictions or errors in user expression.
[0004] To address the aforementioned technical problems, the first aspect of this invention discloses a real-time error correction method for outbound call semantics based on collaborative reasoning, the method comprising: Obtain the speech-recognized text of conversations with outbound users based on the purpose of the outbound call; Based on a preset outbound call domain knowledge graph, match multiple outbound call knowledge data corresponding to the outbound call purpose; Based on a large language model, contradictory text in the speech recognition text is identified according to the outbound call knowledge data for each call. Based on the semantic comparison algorithm, the corrected text corresponding to the speech recognition text is generated according to all contradictory texts in the speech recognition text.
[0005] As an optional implementation, in the first aspect of the present invention, the step of matching multiple outbound call knowledge data corresponding to the outbound call destination based on a preset outbound call domain knowledge graph includes: Obtain the outbound call script text corresponding to the stated outbound call destination; Generate the text vector corresponding to the outbound call script text; Based on the text vector, multiple outbound call knowledge data are obtained by matching in a preset outbound call domain knowledge graph library.
[0006] As an optional implementation, in the first aspect of the present invention, generating the text vector corresponding to the outbound call script text includes: Each of the outbound call script texts is input into the trained knowledge prediction model to obtain the predicted probability that each script text belongs to the knowledge category. Filter out the texts whose predicted probability is greater than a preset probability threshold to obtain multiple texts of textual knowledge. Each of the aforementioned verbal knowledge texts is input into a vectorization network to obtain the corresponding output text vector data; The text vector data of all the aforementioned script knowledge texts and the corresponding prediction probabilities are matrixed to obtain the text vectors corresponding to the outbound call script texts.
[0007] As an optional implementation, in the first aspect of the present invention, the outbound call domain knowledge graph library is pre-established through the following steps: Obtain outbound calling knowledge documents from multiple outbound calling fields; The outbound call knowledge document is standardized to obtain standardized knowledge text; Keyword identification and extraction are performed on each of the aforementioned normative knowledge texts to obtain text keywords; For each of the aforementioned standard knowledge texts, the text keywords and outbound calling domains corresponding to the standard knowledge texts are vectorized to obtain a reference vector corresponding to the standard knowledge texts. The reference vectors are used to perform similarity matching with the input text vectors, and when the matching is successful, the corresponding standard knowledge texts are identified as the outbound calling knowledge data.
[0008] As an optional implementation, in the first aspect of the invention, the step of identifying contradictory text in the speech recognition text based on a large language model and according to each outbound call knowledge data includes: The speech recognition text is combined with preset knowledge mention judgment prompts to obtain knowledge mention prompt text; The knowledge mention prompt text is input into the trained LLM model to obtain multiple knowledge mention texts in the output speech recognition text; Based on the outbound call knowledge data and the LLM model, predict the degree of contradiction corresponding to each knowledge mention text; Texts with a degree of contradiction greater than a preset threshold are selected to obtain contradictory texts.
[0009] As an optional implementation, in the first aspect of the present invention, predicting the degree of contradiction corresponding to each knowledge mention text based on the outbound call knowledge data and the LLM model includes: For each knowledge reference text, the knowledge reference text is filled into the text position in the preset prompt word template; Each outbound call knowledge data is filled into the knowledge position in the prompt word template to obtain the corresponding contradiction judgment prompt word; the prompt word template is used to drive the LLM model to make contradiction judgments on the text in the text position based on the knowledge in the knowledge position; Input the contradiction judgment prompt word corresponding to each of the outbound knowledge data into the LLM model to obtain the contradiction degree parameter of the knowledge mention text relative to each of the outbound knowledge data; The degree of contradiction corresponding to the knowledge mention text is obtained by calculating the average value of the contradiction parameter of the knowledge mention text relative to all the outbound knowledge data.
[0010] As an optional implementation, in the first aspect of the present invention, the step of generating corrected text corresponding to the speech recognition text based on the semantic comparison algorithm according to all contradictory text in the speech recognition text includes: For each contradictory text, the corrected character corresponding to the contradictory text is matched in the preset misspelling library to obtain the matching result; When the matching result contains a correction character, the correction character is determined as the corrected text corresponding to the contradictory text; When the matching result does not contain a correction word, the outbound knowledge data with the highest contradiction degree parameter corresponding to the contradictory text is determined, and the corresponding knowledge data to be compared is obtained. Based on the knowledge data to be compared, determine the corrected text corresponding to the contradictory text; Based on all the corrected texts, generate the corrected text corresponding to the speech recognition text.
[0011] As an optional implementation, in the first aspect of the present invention, determining the corrected text corresponding to the contradictory text based on the knowledge data to be compared includes: Calculate the semantic similarity between the contradictory text and each piece of knowledge data in the knowledge database to be compared; The knowledge data with the highest semantic similarity is identified as the corrected knowledge data; The contradictory text and the corrected knowledge data are combined into a text rewriting prompt; the text rewriting prompt is used to instruct the LLM model to rewrite the corrected instruction data based on the text style of the contradictory text; The text rewriting prompt is input into the LLM model to obtain the corrected text corresponding to the contradictory text.
[0012] A second aspect of this invention discloses a real-time outbound call semantic error correction system based on collaborative reasoning, the system comprising: The acquisition module is used to acquire the speech recognition text of the communication with the outbound user based on the purpose of the outbound call; The matching module is used to match multiple outbound call knowledge data corresponding to the outbound call destination based on a preset outbound call domain knowledge graph; The recognition module is used to identify contradictory text in the speech recognition text based on a large language model and according to the outbound call knowledge data for each call. The correction module is used to generate corrected text corresponding to the speech recognition text based on the semantic comparison algorithm and all contradictory text in the speech recognition text.
[0013] As an optional implementation, in a second aspect of the invention, the specific method by which the matching module matches multiple outbound call knowledge data corresponding to the outbound call destination based on a preset outbound call domain knowledge graph includes: Obtain the outbound call script text corresponding to the stated outbound call destination; Generate the text vector corresponding to the outbound call script text; Based on the text vector, multiple outbound call knowledge data are obtained by matching in a preset outbound call domain knowledge graph library.
[0014] As an optional implementation, in the second aspect of the present invention, the specific method by which the matching module generates the text vector corresponding to the outbound call script text includes: Each of the outbound call script texts is input into the trained knowledge prediction model to obtain the predicted probability that each script text belongs to the knowledge category. Filter out the texts whose predicted probability is greater than a preset probability threshold to obtain multiple texts of textual knowledge. Each of the aforementioned verbal knowledge texts is input into a vectorization network to obtain the corresponding output text vector data; The text vector data of all the aforementioned script knowledge texts and the corresponding prediction probabilities are matrixed to obtain the text vectors corresponding to the outbound call script texts.
[0015] As an optional implementation, in a second aspect of the invention, the outbound call domain knowledge graph library is pre-established through the following steps: Obtain outbound calling knowledge documents from multiple outbound calling fields; The outbound call knowledge document is standardized to obtain standardized knowledge text; Keyword identification and extraction are performed on each of the aforementioned normative knowledge texts to obtain text keywords; For each of the aforementioned standard knowledge texts, the text keywords and outbound calling domains corresponding to the standard knowledge texts are vectorized to obtain a reference vector corresponding to the standard knowledge texts. The reference vectors are used to perform similarity matching with the input text vectors, and when the matching is successful, the corresponding standard knowledge texts are identified as the outbound calling knowledge data.
[0016] As an optional implementation, in a second aspect of the invention, the recognition module, based on a large language model, identifies the specific method of contradictory text in the speech recognition text according to each outbound call knowledge data, including: The speech recognition text is combined with preset knowledge mention judgment prompts to obtain knowledge mention prompt text; The knowledge mention prompt text is input into the trained LLM model to obtain multiple knowledge mention texts in the output speech recognition text; Based on the outbound call knowledge data and the LLM model, predict the degree of contradiction corresponding to each knowledge mention text; Texts with a degree of contradiction greater than a preset threshold are selected to obtain contradictory texts.
[0017] As an optional implementation, in a second aspect of the invention, the specific method by which the identification module predicts the degree of contradiction corresponding to each knowledge mention text based on the outbound call knowledge data and the LLM model includes: For each knowledge reference text, the knowledge reference text is filled into the text position in the preset prompt word template; Each outbound call knowledge data is filled into the knowledge position in the prompt word template to obtain the corresponding contradiction judgment prompt word; the prompt word template is used to drive the LLM model to make contradiction judgments on the text in the text position based on the knowledge in the knowledge position; Input the contradiction judgment prompt word corresponding to each of the outbound knowledge data into the LLM model to obtain the contradiction degree parameter of the knowledge mention text relative to each of the outbound knowledge data; The degree of contradiction corresponding to the knowledge mention text is obtained by calculating the average value of the contradiction parameter of the knowledge mention text relative to all the outbound knowledge data.
[0018] As an optional implementation, in a second aspect of the invention, the specific method by which the correction module generates the corrected text corresponding to the speech recognition text based on a semantic comparison algorithm and according to all contradictory texts in the speech recognition text includes: For each contradictory text, the corrected character corresponding to the contradictory text is matched in the preset misspelling library to obtain the matching result; When the matching result contains a correction character, the correction character is determined as the corrected text corresponding to the contradictory text; When the matching result does not contain a correction word, the outbound knowledge data with the highest contradiction degree parameter corresponding to the contradictory text is determined, and the corresponding knowledge data to be compared is obtained. Based on the knowledge data to be compared, determine the corrected text corresponding to the contradictory text; Based on all the corrected texts, generate the corrected text corresponding to the speech recognition text.
[0019] As an optional implementation, in a second aspect of the invention, the specific method by which the correction module determines the corrected text corresponding to the contradictory text based on the knowledge data to be compared includes: Calculate the semantic similarity between the contradictory text and each piece of knowledge data in the knowledge database to be compared; The knowledge data with the highest semantic similarity is identified as the corrected knowledge data; The contradictory text and the corrected knowledge data are combined into a text rewriting prompt; the text rewriting prompt is used to instruct the LLM model to rewrite the corrected instruction data based on the text style of the contradictory text; The text rewriting prompt is input into the LLM model to obtain the corrected text corresponding to the contradictory text.
[0020] A third aspect of this invention discloses another real-time outbound call semantic error correction system based on collaborative reasoning, the system comprising: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute some or all of the steps in the outbound call semantic real-time error correction method based on collaborative reasoning disclosed in the first aspect of the present invention.
[0021] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the outbound call semantic real-time error correction method based on collaborative reasoning disclosed in the first aspect of the present invention.
[0022] Compared with the prior art, the embodiments of the present invention have the following beneficial effects: This invention acquires speech recognition text from communication with outbound users based on the purpose of the outbound call, matches outbound knowledge data based on an outbound domain knowledge graph, uses a large language model to identify contradictory text, and generates corrected text based on semantic comparison. This enables intelligent contradiction detection and automatic correction of outbound dialogue text, improves the accuracy and consistency of dialogue content, and reduces the risk of decreased outbound communication efficiency due to contradictory or erroneous user statements. Attached Figure Description
[0023] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.
[0024] Figure 1 This is a flowchart illustrating a real-time error correction method for outbound call semantics based on collaborative reasoning disclosed in an embodiment of the present invention.
[0025] Figure 2 This is a schematic diagram of the structure of a real-time error correction system for outbound call semantics based on collaborative reasoning disclosed in an embodiment of the present invention.
[0026] Figure 3 This is a schematic diagram of another outbound call semantic real-time error correction system based on collaborative reasoning disclosed in an embodiment of the present invention. Detailed Implementation
[0027] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.
[0028] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.
[0029] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.
[0030] This invention discloses a real-time semantic error correction method and system for outbound calls based on collaborative reasoning. By acquiring the speech recognition text of the communication between the caller and the user based on the call purpose, matching outbound knowledge data based on an outbound domain knowledge graph, using a large language model to identify contradictory text, and generating corrected text based on semantic comparison, this method enables intelligent contradiction detection and automatic correction of outbound dialogue text, improving the accuracy and consistency of dialogue content and reducing the risk of decreased communication efficiency due to contradictory or erroneous user statements. Detailed explanations follow.
[0031] Example 1 Please see Figure 1 , Figure 1 This is a flowchart illustrating a real-time semantic error correction method for outbound calls based on collaborative reasoning, as disclosed in an embodiment of the present invention. Figure 1 The described real-time semantic error correction method for outbound calls based on collaborative reasoning can be applied to data processing systems / data processing devices / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 1 As shown, the real-time error correction method for outbound call semantics based on collaborative reasoning can include the following operations: 101. Obtain the speech recognition text of the caller when communicating with the caller based on the purpose of the call.
[0032] Optionally, the purpose of the outbound call can be to remind customers of overdue loan products, promote anti-telecom fraud awareness, conduct after-sales satisfaction follow-up, retain customers for value-added services, or confirm the intention to renew course enrollment. This invention does not limit this purpose.
[0033] Optionally, the speech-recognized text can be a JSON text document transcribed in real time by a streaming speech recognition (ASR) engine, a short text segment stream stored in a memory cache queue, or speech text with timestamps collected and transcribed by call center hardware. This invention does not impose any limitations.
[0034] 102. Based on the preset outbound call domain knowledge graph, match multiple outbound call knowledge data corresponding to the outbound call purpose.
[0035] Optionally, the outbound call domain knowledge graph can be a financial compliance entity relationship network stored in the Neo4j graph database, a telecommunications script strategy topology graph stored in a knowledge base system, or an automotive technical parameter knowledge graph built based on the RDF triple architecture. This invention does not limit the scope of the knowledge graph.
[0036] 103. Based on a large language model, identify contradictory text in speech recognition text according to the knowledge data of each outbound call.
[0037] 104. Based on the semantic comparison algorithm, generate the corrected text corresponding to the speech recognition text according to all contradictory texts in the speech recognition text.
[0038] As can be seen, the above-mentioned embodiments of the invention acquire speech recognition text when communicating with outbound users based on the purpose of outbound calls, match outbound knowledge data based on outbound domain knowledge graphs, use large language models to identify contradictory texts, and generate corrected texts based on semantic comparisons. This enables intelligent contradiction detection and automatic correction of outbound dialogue texts, improves the accuracy and consistency of dialogue content, and reduces the risk of decreased outbound communication efficiency due to contradictory or incorrect user statements.
[0039] As an optional embodiment, the step above, matching multiple outbound call knowledge data corresponding to the outbound call purpose based on a preset outbound call domain knowledge graph, includes: Obtain the outbound call script text corresponding to the outbound call destination; Generate the text vector corresponding to the outbound call script; Based on text vectors, multiple outbound call knowledge data are obtained by matching in a pre-set outbound call domain knowledge graph library.
[0040] Optionally, the outbound call knowledge data can be specific policy and regulatory texts, official standard product tariff rate details, collection business red line script rules, or a list of entity attributes matched from a graph database. This invention does not limit this.
[0041] As can be seen, through the above optional embodiments, by obtaining the outbound call script text to generate text vectors, and matching multiple outbound call knowledge data in the preset outbound call domain knowledge graph library based on the vectors, the script text can be accurately semantically mapped to knowledge data, thereby improving the efficiency and relevance of knowledge matching and reducing the risk of knowledge retrieval deviation caused by inaccurate text vector representation.
[0042] As an optional embodiment, the step of generating the text vector corresponding to the outbound call script text in the above steps includes: Each outbound call script is input into a trained knowledge prediction model to obtain the predicted probability that each script belongs to the knowledge category. Select texts with predicted probabilities greater than a preset probability threshold to obtain multiple texts of textual knowledge. Each dialogue knowledge text is input into a vectorization network to obtain the corresponding text vector data of the output; The text vector data of all the call script knowledge texts and their corresponding prediction probabilities are matrixed to obtain the text vectors corresponding to the outbound call script texts.
[0043] Optionally, the knowledge prediction model can be a text binary classification network based on BERT fine-tuning, a deep bidirectional long short-term memory network (BiLSTM) classifier, or a conditional random field (CRF) sequence labeling model; this invention does not limit the model.
[0044] Optionally, the text vector data can be a 1024-dimensional dense vector output by the Embedding layer, a floating-point feature tensor generated based on a vectorized network, or a low-dimensional dense encoding in the latent semantic space; this invention does not impose any limitations.
[0045] As can be seen, through the above optional embodiments, by inputting each outbound call script text into a trained knowledge prediction model to obtain the predicted probability of belonging to knowledge information, filtering high-probability script texts to obtain script knowledge texts, inputting them into a vectorization network to obtain text vector data, and finally performing matrix processing to obtain the text vector corresponding to the outbound call script text, the knowledge information of the script text can be accurately extracted and vectorized, improving the semantic richness and knowledge relevance of the text vectors, and assisting in the accurate matching of subsequent outbound call knowledge data.
[0046] As an optional embodiment, the outbound call domain knowledge graph library in the above steps is pre-established through the following steps: Obtain outbound calling knowledge documents from multiple outbound calling fields; Standardize the knowledge documents for outbound calls to obtain standardized knowledge text; Keyword identification and extraction are performed on each normative knowledge text to obtain text keywords; For each standard knowledge text, the corresponding text keywords and outbound calling domain are vectorized to obtain the reference vector corresponding to the standard knowledge text.
[0047] Optionally, the reference vector is used to perform similarity matching with the input text vector, and when the match is successful, the corresponding canonical knowledge text is identified as outbound call knowledge data.
[0048] Optionally, the outbound call knowledge document can be a PDF business white paper, an internal Word training script manual, or a TXT compliance quality inspection case set stored in a non-relational database. This invention does not impose any limitations on this.
[0049] Optionally, the normalization process can be achieved by using a text cleaning script to remove meaningless garbled characters and special control characters, performing syntactic-level sentence segmentation and paragraph reconstruction on long documents, or unifying the text encoding to the UTF-8 standard format. This invention does not limit the specific methods.
[0050] As can be seen, through the above optional embodiments, reference vectors are obtained by standardizing, extracting keywords, and vectorizing outbound call knowledge documents in the outbound call field. These reference vectors are then used for subsequent text vector matching to determine outbound call knowledge data. This enables the structured construction and efficient retrieval of the outbound call knowledge graph library, improves the standardization and usability of knowledge data, and reduces the risk of knowledge matching difficulties caused by unstructured documents.
[0051] As an optional embodiment, the step above, identifying contradictory text in the speech recognition text based on a large language model and according to the knowledge data of each outbound call, includes: The speech recognition text is combined with preset knowledge mention judgment prompts to obtain knowledge mention prompt text; The knowledge mention cue text is input into the trained LLM model to obtain multiple knowledge mention texts in the output speech recognition text; Based on outbound call knowledge data and an LLM model, predict the degree of contradiction corresponding to each knowledge mention text; Texts with a degree of contradiction greater than a preset threshold are selected to obtain contradictory texts.
[0052] Optionally, the LLM model can be a privately deployed open-source large language model, a large-scale cloud-based commercial model called through an API gateway, or a lightweight text generation model customized and fine-tuned for specific businesses. This invention does not limit the scope of the model.
[0053] Optionally, the knowledge reference text may be a text fragment containing key business entities, such as a specific time point, amount, interest rate, or package name expressed by the outbound caller or customer service robot in the voice text. This invention does not limit this.
[0054] As can be seen, through the above optional embodiments, by combining the speech recognition text with the knowledge mention judgment prompt words to form a prompt word text, inputting it into the LLM model to obtain the knowledge mention text, and filtering contradictory texts based on the degree of contradiction prediction based on outbound call knowledge data, intelligent contradiction recognition based on prompt engineering and knowledge comparison is realized, improving the accuracy and automation of contradictory text detection, and reducing the risk of contradiction omission due to insufficient knowledge coverage.
[0055] As an optional embodiment, the step above, predicting the degree of contradiction corresponding to each knowledge mention text based on outbound call knowledge data and an LLM model, includes: For each knowledge mention text, fill the text mention text into the text position of the preset prompt word template; Each outbound call knowledge data is filled into the knowledge position in the prompt word template to obtain the corresponding contradiction judgment prompt word; optionally, the prompt word template is used to drive the LLM model to make contradiction judgments on the text in the text position based on the knowledge in the knowledge position; Input the contradiction judgment prompt words corresponding to each outbound knowledge data into the LLM model to obtain the contradiction degree parameter of the knowledge mention text relative to each outbound knowledge data. Calculate the average value of the contradiction parameter of the knowledge mention text relative to all outbound knowledge data to obtain the degree of contradiction corresponding to the knowledge mention text.
[0056] Optionally, the prompt template can be a pre-set XML-formatted structured text in the system configuration file, a system prompt containing role constraints and Few-Shot examples, or a dynamically concatenated Markdown template string; this invention does not impose any limitations.
[0057] As can be seen, through the above optional embodiments, by filling each knowledge mention text into the text position in the prompt word template and filling the outbound knowledge data into the knowledge position to obtain the contradiction judgment prompt word, inputting it into the LLM model to obtain the contradiction degree parameter and calculating the average value to obtain the contradiction degree, a precise contradiction quantification assessment based on prompt engineering and knowledge comparison is achieved, improving the intelligence and accuracy of contradiction identification and reducing the risk of contradiction detection omission due to insufficient knowledge matching.
[0058] As an optional embodiment, the step described above, generating corrected text corresponding to the speech recognition text based on a semantic comparison algorithm and all contradictory text in the speech recognition text, includes: For each contradictory text, the corrected character corresponding to the contradictory text is matched in the preset misspelling database to obtain the matching result; If a correction word exists in the matching result, the correction word will be identified as the corrected text corresponding to the contradictory text. When no correction word is found in the matching result, the outbound knowledge data with the highest contradiction degree parameter corresponding to the contradictory text is determined to obtain the corresponding knowledge data to be compared. Based on the knowledge data to be compared, determine the corrected text corresponding to the contradictory text; Based on all the corrected text, generate the corrected text corresponding to the speech recognition text.
[0059] Optionally, the misspelling database can be an industry high-frequency easily confused word mapping table stored in Redis cache, a homophone misspelling correction dictionary built based on sound similarity, or a one-dimensional enumeration error correction table embedded in local memory. This invention does not limit the scope of the database.
[0060] Optionally, the knowledge data to be compared can be the normative knowledge record with the highest confidence obtained from the graph library, or an authoritative text entity read from a distributed knowledge base; this invention does not impose any limitations.
[0061] As can be seen, through the above optional embodiments, by matching typos for each contradictory text or determining the corrected text based on the knowledge data of the highest degree of contradiction, and generating the corrected text, intelligent correction and text optimization of contradictory text are achieved, thereby improving the final quality and readability of speech recognition text and reducing the risk of misunderstanding of outbound dialogue due to uncorrected contradictions.
[0062] As an optional embodiment, the step above, determining the corrected text corresponding to the contradictory text based on the knowledge data to be compared, includes: Calculate the semantic similarity between the contradictory text and each piece of knowledge data in the knowledge database to be compared; The knowledge data with the highest semantic similarity is identified as the corrected knowledge data; The contradictory text and the corrected knowledge data are combined into text rewriting prompts; optionally, the text rewriting prompts are used to instruct the LLM model to rewrite the corrected instruction data based on the text style of the contradictory text. Input the text rewriting prompts into the LLM model to obtain the corrected text corresponding to the contradictory text.
[0063] Optionally, the text rewriting prompt can be a rewriting control command that instructs the large model to maintain the user's original tone and manner and correct erroneous facts into corrected knowledge data content; this invention does not limit this.
[0064] As can be seen, through the above optional embodiments, by calculating the semantic similarity between the contradictory text and each knowledge data in the knowledge database to be compared, the knowledge data with the highest similarity is selected as the corrected knowledge data. The contradictory text and the corrected knowledge data are combined into text rewriting prompt words and input into the LLM model to obtain the corrected text. This realizes intelligent contradictory text correction based on semantic similarity and prompt rewriting, improves the quality and consistency of text correction, and reduces the risk of text rewriting deviation caused by improper selection of corrected knowledge.
[0065] Example 2 Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a real-time outbound call semantic error correction system based on collaborative reasoning, as disclosed in an embodiment of the present invention. Figure 2The described real-time error correction system for outbound call semantics based on collaborative reasoning can be applied to data processing systems / data processing devices / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 2 As shown, the outbound call semantic real-time error correction system based on collaborative reasoning may include: The acquisition module 201 is used to acquire the speech recognition text when communicating with the outbound user based on the purpose of the outbound call.
[0066] The matching module 202 is used to match multiple outbound call knowledge data corresponding to the outbound call purpose based on a preset outbound call domain knowledge graph.
[0067] The recognition module 203 is used to identify contradictory text in speech recognition text based on a large language model and knowledge data of each outbound call.
[0068] The correction module 204 is used to generate the corrected text corresponding to the speech recognition text based on the semantic comparison algorithm and all contradictory texts in the speech recognition text.
[0069] As can be seen, the above-mentioned embodiments of the invention acquire speech recognition text when communicating with outbound users based on the purpose of outbound calls, match outbound knowledge data based on outbound domain knowledge graphs, use large language models to identify contradictory texts, and generate corrected texts based on semantic comparisons. This enables intelligent contradiction detection and automatic correction of outbound dialogue texts, improves the accuracy and consistency of dialogue content, and reduces the risk of decreased outbound communication efficiency due to contradictory or incorrect user statements.
[0070] As an optional embodiment, the matching module matches multiple outbound call knowledge data corresponding to an outbound call purpose based on a preset outbound call domain knowledge graph, including the following specific methods: Obtain the outbound call script text corresponding to the outbound call destination; Generate the text vector corresponding to the outbound call script; Based on text vectors, multiple outbound call knowledge data are obtained by matching in a pre-set outbound call domain knowledge graph library.
[0071] As can be seen, through the above optional embodiments, by obtaining the outbound call script text to generate text vectors, and matching multiple outbound call knowledge data in the preset outbound call domain knowledge graph library based on the vectors, the script text can be accurately semantically mapped to knowledge data, thereby improving the efficiency and relevance of knowledge matching and reducing the risk of knowledge retrieval deviation caused by inaccurate text vector representation.
[0072] As an optional embodiment, the specific method by which the matching module generates the text vector corresponding to the outbound call script text includes: Each outbound call script is input into a trained knowledge prediction model to obtain the predicted probability that each script belongs to the knowledge category. Select texts with predicted probabilities greater than a preset probability threshold to obtain multiple texts of textual knowledge. Each dialogue knowledge text is input into a vectorization network to obtain the corresponding text vector data of the output; The text vector data of all the call script knowledge texts and their corresponding prediction probabilities are matrixed to obtain the text vectors corresponding to the outbound call script texts.
[0073] As can be seen, through the above optional embodiments, by inputting each outbound call script text into a trained knowledge prediction model to obtain the predicted probability of belonging to knowledge information, filtering high-probability script texts to obtain script knowledge texts, inputting them into a vectorization network to obtain text vector data, and finally performing matrix processing to obtain the text vector corresponding to the outbound call script text, the knowledge information of the script text can be accurately extracted and vectorized, improving the semantic richness and knowledge relevance of the text vectors, and assisting in the accurate matching of subsequent outbound call knowledge data.
[0074] As an optional embodiment, the outbound call domain knowledge graph library is pre-built through the following steps: Obtain outbound calling knowledge documents from multiple outbound calling fields; Standardize the knowledge documents for outbound calls to obtain standardized knowledge text; Keyword identification and extraction are performed on each normative knowledge text to obtain text keywords; For each standard knowledge text, the corresponding text keywords and outbound calling domain are vectorized to obtain the reference vector corresponding to the standard knowledge text. The reference vector is used to perform similarity matching with the input text vector, and when the matching is successful, the corresponding standard knowledge text is identified as outbound calling knowledge data.
[0075] As can be seen, through the above optional embodiments, reference vectors are obtained by standardizing, extracting keywords, and vectorizing outbound call knowledge documents in the outbound call field. These reference vectors are then used for subsequent text vector matching to determine outbound call knowledge data. This enables the structured construction and efficient retrieval of the outbound call knowledge graph library, improves the standardization and usability of knowledge data, and reduces the risk of knowledge matching difficulties caused by unstructured documents.
[0076] As an optional embodiment, the recognition module, based on a large language model, identifies the specific methods of contradictory text in the speech recognition text according to the knowledge data of each outbound call, including: The speech recognition text is combined with preset knowledge mention judgment prompts to obtain knowledge mention prompt text; The knowledge mention cue text is input into the trained LLM model to obtain multiple knowledge mention texts in the output speech recognition text; Based on outbound call knowledge data and an LLM model, predict the degree of contradiction corresponding to each knowledge mention text; Texts with a degree of contradiction greater than a preset threshold are selected to obtain contradictory texts.
[0077] As can be seen, through the above optional embodiments, by combining the speech recognition text with the knowledge mention judgment prompt words to form a prompt word text, inputting it into the LLM model to obtain the knowledge mention text, and filtering contradictory texts based on the degree of contradiction prediction based on outbound call knowledge data, intelligent contradiction recognition based on prompt engineering and knowledge comparison is realized, improving the accuracy and automation of contradictory text detection, and reducing the risk of contradiction omission due to insufficient knowledge coverage.
[0078] As an optional implementation, the identification module predicts the degree of contradiction corresponding to each knowledge mention text based on outbound call knowledge data and an LLM model, including the following specific methods: For each knowledge mention text, fill the text mention text into the text position of the preset prompt word template; Each outbound call knowledge data is filled into the knowledge position in the prompt word template to obtain the corresponding contradiction judgment prompt word; optionally, the prompt word template is used to drive the LLM model to make contradiction judgments on the text in the text position based on the knowledge in the knowledge position; Input the contradiction judgment prompt words corresponding to each outbound knowledge data into the LLM model to obtain the contradiction degree parameter of the knowledge mention text relative to each outbound knowledge data. Calculate the average value of the contradiction parameter of the knowledge mention text relative to all outbound knowledge data to obtain the degree of contradiction corresponding to the knowledge mention text.
[0079] As can be seen, through the above optional embodiments, by filling each knowledge mention text into the text position in the prompt word template and filling the outbound knowledge data into the knowledge position to obtain the contradiction judgment prompt word, inputting it into the LLM model to obtain the contradiction degree parameter and calculating the average value to obtain the contradiction degree, a precise contradiction quantification assessment based on prompt engineering and knowledge comparison is achieved, improving the intelligence and accuracy of contradiction identification and reducing the risk of contradiction detection omission due to insufficient knowledge matching.
[0080] As an optional embodiment, the correction module generates the corrected text corresponding to the speech recognition text based on a semantic comparison algorithm and all contradictory text in the speech recognition text, including: For each contradictory text, the corrected character corresponding to the contradictory text is matched in the preset misspelling database to obtain the matching result; If a correction word exists in the matching result, the correction word will be identified as the corrected text corresponding to the contradictory text. When no correction word is found in the matching result, the outbound knowledge data with the highest contradiction degree parameter corresponding to the contradictory text is determined to obtain the corresponding knowledge data to be compared. Based on the knowledge data to be compared, determine the corrected text corresponding to the contradictory text; Based on all the corrected text, generate the corrected text corresponding to the speech recognition text.
[0081] As can be seen, through the above optional embodiments, by matching typos for each contradictory text or determining the corrected text based on the knowledge data of the highest degree of contradiction, and generating the corrected text, intelligent correction and text optimization of contradictory text are achieved, thereby improving the final quality and readability of speech recognition text and reducing the risk of misunderstanding of outbound dialogue due to uncorrected contradictions.
[0082] As an optional embodiment, the correction module determines the specific method by which it determines the corrected text corresponding to the contradictory text based on the knowledge data to be compared, including: Calculate the semantic similarity between the contradictory text and each piece of knowledge data in the knowledge database to be compared; The knowledge data with the highest semantic similarity is identified as the corrected knowledge data; The contradictory text and the corrected knowledge data are combined into text rewriting prompts; optionally, the text rewriting prompts are used to instruct the LLM model to rewrite the corrected instruction data based on the text style of the contradictory text. Input the text rewriting prompts into the LLM model to obtain the corrected text corresponding to the contradictory text.
[0083] As can be seen, through the above optional embodiments, by calculating the semantic similarity between the contradictory text and each knowledge data in the knowledge database to be compared, the knowledge data with the highest similarity is selected as the corrected knowledge data. The contradictory text and the corrected knowledge data are combined into text rewriting prompt words and input into the LLM model to obtain the corrected text. This realizes intelligent contradictory text correction based on semantic similarity and prompt rewriting, improves the quality and consistency of text correction, and reduces the risk of text rewriting deviation caused by improper selection of corrected knowledge.
[0084] Example 3 Please see Figure 3 , Figure 3 This is another outbound call semantic real-time error correction system based on collaborative reasoning disclosed in the embodiments of the present invention. Figure 3 The described real-time error correction system for outbound call semantics based on collaborative reasoning is applied in data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 3As shown, the outbound call semantic real-time error correction system based on collaborative reasoning may include: Memory 301 storing executable program code; Processor 302 coupled to memory 301; The processor 302 calls the executable program code stored in the memory 301 to execute the steps of the outbound call semantic real-time error correction method based on collaborative reasoning described in Embodiment 1.
[0085] Example 4 This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps of the outbound call semantic real-time error correction method based on collaborative reasoning described in Embodiment 1.
[0086] Example 5 This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps of the outbound call semantic real-time error correction method based on collaborative reasoning described in Embodiment 1.
[0087] The foregoing has described specific embodiments of this specification; other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily have to follow the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.
[0088] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.
[0089] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.
[0090] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.
[0091] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.
[0092] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.
[0093] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.
[0094] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.
[0095] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.
[0096] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.
[0097] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.
[0098] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.
[0099] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.
[0100] Finally, it should be noted that the real-time error correction method and system for outbound call semantics based on collaborative reasoning disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention, and are only used to illustrate the technical solutions of the present invention, not to limit it; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.
Claims
1. A real-time error correction method for outbound call semantics based on collaborative reasoning, characterized in that, The method includes: Obtain the speech-recognized text of conversations with outbound users based on the purpose of the outbound call; Based on a preset outbound call domain knowledge graph, match multiple outbound call knowledge data corresponding to the outbound call purpose; Based on a large language model, contradictory text in the speech recognition text is identified according to the outbound call knowledge data for each call. Based on the semantic comparison algorithm, the corrected text corresponding to the speech recognition text is generated according to all contradictory texts in the speech recognition text.
2. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 1, characterized in that, The method of matching multiple outbound call knowledge data corresponding to the outbound call destination based on a preset outbound call domain knowledge graph includes: Obtain the outbound call script text corresponding to the stated outbound call destination; Generate the text vector corresponding to the outbound call script text; Based on the text vector, multiple outbound call knowledge data are obtained by matching in a preset outbound call domain knowledge graph library.
3. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 2, characterized in that, The process of generating the text vector corresponding to the outbound call script includes: Each of the outbound call script texts is input into the trained knowledge prediction model to obtain the predicted probability that each script text belongs to the knowledge category. Filter out the texts whose predicted probability is greater than a preset probability threshold to obtain multiple texts of textual knowledge. Each of the aforementioned verbal knowledge texts is input into a vectorization network to obtain the corresponding output text vector data; The text vector data of all the aforementioned script knowledge texts and the corresponding prediction probabilities are matrixed to obtain the text vectors corresponding to the outbound call script texts.
4. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 2, characterized in that, The outbound call domain knowledge graph library is pre-established through the following steps: Obtain outbound calling knowledge documents from multiple outbound calling fields; The outbound call knowledge document is standardized to obtain standardized knowledge text; Keyword identification and extraction are performed on each of the aforementioned normative knowledge texts to obtain text keywords; For each of the aforementioned standard knowledge texts, the text keywords and outbound calling domains corresponding to the standard knowledge texts are vectorized to obtain a reference vector corresponding to the standard knowledge texts. The reference vectors are used to perform similarity matching with the input text vectors, and when the matching is successful, the corresponding standard knowledge texts are identified as the outbound calling knowledge data.
5. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 1, characterized in that, The method of identifying contradictory text in the speech recognition text based on a large language model and according to the outbound call knowledge data for each call includes: The speech recognition text is combined with preset knowledge mention judgment prompts to obtain knowledge mention prompt text; The knowledge mention prompt text is input into the trained LLM model to obtain multiple knowledge mention texts in the output speech recognition text; Based on the outbound call knowledge data and the LLM model, predict the degree of contradiction corresponding to each knowledge mention text; Texts with a degree of contradiction greater than a preset threshold are selected to obtain contradictory texts.
6. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 5, characterized in that, The prediction of the degree of contradiction corresponding to each knowledge mention text based on the outbound call knowledge data and the LLM model includes: For each knowledge mention text, the knowledge mention text is filled into the text position in the preset prompt word template; Each outbound call knowledge data is filled into the knowledge position in the prompt word template to obtain the corresponding contradiction judgment prompt word; the prompt word template is used to drive the LLM model to make contradiction judgments on the text in the text position based on the knowledge in the knowledge position; Input the contradiction judgment prompt word corresponding to each of the outbound knowledge data into the LLM model to obtain the contradiction degree parameter of the knowledge mention text relative to each of the outbound knowledge data; The degree of contradiction corresponding to the knowledge mention text is obtained by calculating the average value of the contradiction parameter of the knowledge mention text relative to all the outbound knowledge data.
7. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 1, characterized in that, The semantic comparison algorithm generates corrected text corresponding to the speech recognition text based on all contradictory text in the speech recognition text, including: For each contradictory text, the corrected character corresponding to the contradictory text is matched in the preset misspelling library to obtain the matching result; When the matching result contains a correction character, the correction character is determined as the corrected text corresponding to the contradictory text; When the matching result does not contain a correction word, the outbound knowledge data with the highest contradiction degree parameter corresponding to the contradictory text is determined, and the corresponding knowledge data to be compared is obtained. Based on the knowledge data to be compared, determine the corrected text corresponding to the contradictory text; Based on all the corrected texts, generate the corrected text corresponding to the speech recognition text.
8. The real-time error correction method for outbound call semantics based on collaborative reasoning according to claim 7, characterized in that, The step of determining the corrected text corresponding to the contradictory text based on the knowledge data to be compared includes: Calculate the semantic similarity between the contradictory text and each piece of knowledge data in the knowledge database to be compared; The knowledge data with the highest semantic similarity is identified as the corrected knowledge data; The contradictory text and the corrected knowledge data are combined into a text rewriting prompt; the text rewriting prompt is used to instruct the LLM model to rewrite the corrected instruction data based on the text style of the contradictory text; The text rewriting prompt is input into the LLM model to obtain the corrected text corresponding to the contradictory text.
9. A real-time semantic error correction system for outbound calls based on collaborative reasoning, characterized in that, The system includes: The acquisition module is used to acquire the speech recognition text of the communication with the outbound user based on the purpose of the outbound call; The matching module is used to match multiple outbound call knowledge data corresponding to the outbound call destination based on a preset outbound call domain knowledge graph; The recognition module is used to identify contradictory text in the speech recognition text based on a large language model and according to the outbound call knowledge data for each call. The correction module is used to generate corrected text corresponding to the speech recognition text based on the semantic comparison algorithm and all contradictory text in the speech recognition text.
10. A real-time semantic error correction system for outbound calls based on collaborative reasoning, characterized in that, The system includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the outbound call semantic real-time error correction method based on collaborative reasoning as described in any one of claims 1-8.
