Address extraction method and device based on multi-round dialogues, equipment and storage medium

Abstract

The invention relates to the field of data processing, and provides an address extraction method and device based on multiple rounds of dialogues, computer equipment and a computer readable storage medium, and the method comprises the steps: firstly, carrying out the preprocessing of a voice dialogue based on a preprocessing module in an address extraction model, and determining the text information of the voice dialogue and a user intention; based on an entity recognition module in an address extraction model, address information in the text information is extracted in combination with a user intention, and initial address information is obtained; and further verifying the initial address information based on an address verification module in the address extraction model so as to further guide customer service personnel to perform multi-round inquiry on the current user under the condition that the initial address information is not clear or incomplete, thereby extracting accurate and complete address information through multi-round dialogues. The address extraction efficiency is improved, and the address extraction accuracy is also improved.

Description

Technical Field

[0001] The present invention relates to the fields of data processing and financial technology, and in particular to a method, device, computer equipment and computer-readable storage medium for extracting addresses based on multi-round conversations. Background Art

[0002] At present, extracting user address information from telephone communication is an important part of online financial product sales. There are two ways to extract addresses: manual extraction and system extraction. Manual extraction requires customer service staff to manually check and record customer address information, which is not only inefficient but also prone to errors, thus reducing the accuracy of the extracted addresses. The system extraction method is easily affected by the customer's ability to express himself, and cannot process unclear or incomplete addresses, thus reducing the accuracy of the extracted addresses. Summary of the invention

[0003] The main purpose of the present invention is to provide an address extraction method, device, equipment and computer-readable storage medium based on multi-round dialogues, aiming to improve the technical problem of address extraction efficiency and accuracy of extracted addresses.

[0004] To achieve the above object, the present invention provides an address extraction method based on multi-round dialogue, and the frame rate improvement method comprises the following steps:

[0005] When receiving the address extraction instruction, preprocessing the voice dialogue corresponding to the current user is performed based on the preprocessing module in the address extraction model to obtain text information corresponding to the voice dialogue and the user intention;

[0006] Based on the entity recognition module in the address extraction model, entity recognition is performed on the text information and the user intention to obtain the initial address information of the current user;

[0007] Based on the address verification module in the address extraction model, the initial address information is verified to obtain a verification result of the initial address information;

[0008] When the verification result is abnormal, the abnormal part of the address is obtained, and dialogue guidance information is generated based on the abnormal part of the address, so that the customer service staff can conduct the next round of inquiries to the current user based on the dialogue guidance information until the address information with a normal verification result is obtained as the target address information.

[0009] In addition, to achieve the above-mentioned purpose, the present invention further provides an address extraction device based on multi-round dialogue, the address extraction device based on multi-round dialogue comprising:

[0010] A dialogue preprocessing module, which is used to preprocess the voice dialogue corresponding to the current user based on the preprocessing module in the address extraction model when receiving the address extraction instruction, and obtain the text information corresponding to the voice dialogue and the user intention;

[0011] An address entity extraction module, used to perform entity recognition on the text information and user intention based on the entity recognition module in the address extraction model to obtain the initial address information of the current user;

[0012] An address information verification module, used to verify the initial address information based on the address verification module in the address extraction model, and obtain a verification result of the initial address information;

[0013] The dialogue guidance inquiry module is used to obtain the abnormal part of the address when the verification result is abnormal, and generate dialogue guidance information based on the abnormal part of the address, so that the customer service personnel can conduct the next round of inquiries to the current user based on the dialogue guidance information until the address information with a normal verification result is obtained as the target address information.

[0014] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer device, which includes a processor, a memory, and an address extraction program based on multi-round conversations stored in the memory and executable by the processor, wherein when the address extraction program based on multi-round conversations is executed by the processor, the steps of the address extraction method based on multi-round conversations as described above are implemented.

[0015] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer-readable storage medium, on which an address extraction program based on multi-round conversations is stored, wherein when the address extraction program based on multi-round conversations is executed by a processor, the steps of the address extraction method based on multi-round conversations as described above are implemented.

[0016] The present application provides an address extraction method based on multi-round dialogues. When receiving an address extraction instruction, the method preprocesses the voice dialogue corresponding to the current user based on the preprocessing module in the address extraction model to obtain text information and user intention corresponding to the voice dialogue; performs entity recognition on the text information and user intention based on the entity recognition module in the address extraction model to obtain the initial address information of the current user; verifies the initial address information based on the address verification module in the address extraction model to obtain the verification result of the initial address information; when the verification result is abnormal, obtains the abnormal part of the address, and generates dialogue guidance information based on the abnormal part of the address, so that the customer service personnel can conduct the next round of inquiries to the current user based on the dialogue guidance information, until the address information with a normal verification result is obtained as the target address information. Through the above method, the present application first pre-processes the voice conversation based on the pre-processing module in the address extraction model to determine the text information of the voice conversation and the user intention; then, based on the entity recognition module in the address extraction model, combined with the user intention, the address information in the text information is extracted to obtain the initial address information; and then based on the address verification module in the address extraction model, the initial address information is further verified, so as to further guide the customer service personnel to conduct multiple rounds of inquiries to the current user when the initial address information is unclear or incomplete, so as to extract accurate and complete address information through multiple rounds of conversations, thereby improving the efficiency of address extraction while also improving the accuracy of the extracted address. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 It is a flowchart of a first embodiment of the address extraction method based on multi-round dialogues of the present invention;

[0018] Figure 2 It is a flow chart of a second embodiment of the address extraction method based on multi-round dialogues of the present invention;

[0019] Figure 3 It is a flowchart diagram of a third embodiment of the address extraction method based on multi-round dialogues of the present invention;

[0020] Figure 4 A schematic diagram of an address extraction device based on multi-round dialogues according to the present invention;

[0021] Figure 5 The figure is a schematic diagram of the hardware structure of a computer device involved in an embodiment of the present invention.

[0022] The realization of the purpose, functional features and advantages of the present invention will be further explained in conjunction with embodiments and with reference to the accompanying drawings. DETAILED DESCRIPTION

[0023] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0024] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0025] It should be understood that the terms used in this application specification are only for the purpose of describing specific embodiments and are not intended to limit the application. As used in this application specification and the appended claims, unless the context clearly indicates otherwise, the singular forms "a", "an" and "the" are intended to include plural forms.

[0026] It should be further understood that the term “and / or” used in the specification and appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0027] The multi-round dialogue-based address extraction method involved in the embodiment of the present invention is mainly applied to computer equipment. The multi-round dialogue-based address extraction and generation device can be a PC, a portable computer, a mobile terminal, or other device with display and processing functions.

[0028] In conjunction with the accompanying drawings, some embodiments of the present application are described in detail below. In the absence of conflict, the following embodiments and features in the embodiments can be combined with each other.

[0029] Reference Figure 1 , Figure 1 It is a flowchart diagram of the first embodiment of the address extraction method based on multi-round dialogues of the present invention.

[0030] like Figure 1 As shown, an embodiment of the present invention provides an address extraction method based on multi-round dialogues, and the address extraction method based on multi-round dialogues includes steps S10 to S40.

[0031] In this embodiment, the address extraction method based on multi-round dialogues includes the following steps:

[0032] Step S10, when receiving the address extraction instruction, preprocessing the voice dialogue corresponding to the current user based on the preprocessing module in the address extraction model to obtain text information corresponding to the voice dialogue and the user intention;

[0033] In the current financial product sales scenario, address extraction is an important link. Currently, address extraction methods include manual extraction and system extraction. However, both methods have the following shortcomings:

[0034] (1) The manual extraction method relies on manual operation and requires manual review and recording of customer address information, which is not only inefficient but also prone to errors.

[0035] (2) The system extraction method cannot automatically identify and extract the address information given by the customer on the phone. It cannot extract accurate address information from the ambiguous information provided by the customer on the phone, nor can it handle vague and incomplete address information, resulting in low accuracy of the extracted address.

[0036] In order to solve the above problem, this embodiment proposes a multi-round dialogue-based address extraction method, which is applied to an address extraction model. The address extraction model includes:

[0037] The voice preprocessing module is used to perform voice recognition and intent classification on the voice conversation between the customer service representative and the customer, and obtain the text information and user intent corresponding to the voice conversation;

[0038] The entity recognition module is used to perform entity recognition on text information based on user intent to obtain address information;

[0039] The address verification module is used to verify the address information extracted by the entity recognition module, and when the address information fails the verification, a guided dialogue is generated for the customer service staff to conduct multiple rounds of dialogue and inquiry with the customer until accurate and complete address information is extracted from multiple rounds of voice dialogue.

[0040] Based on the above address extraction model, the voice conversation is first preprocessed based on the preprocessing module to determine the text information of the voice conversation and the user intention; then, based on the entity recognition module, combined with the user intention, the address information in the text information is extracted to obtain the initial address information; then, based on the address verification module, the initial address information is further verified, so that when the initial address information is unclear or incomplete, the customer service staff can be further guided to conduct multiple rounds of inquiries to the current user, so as to extract accurate and complete address information through multiple rounds of conversations. Therefore, while improving the efficiency of address extraction, the accuracy of the extracted address is also improved.

[0041] Specifically, when an address extraction instruction is detected, the voice conversation of the current user is obtained, and based on the preprocessing module, speech recognition is performed on the voice conversation to obtain text information, and the text information is classified according to intent to obtain the user intent.

[0042] Furthermore, when the address extraction instruction is received, the voice dialogue corresponding to the current user is preprocessed based on the preprocessing module in the address extraction model to obtain the text information corresponding to the voice dialogue and the user intention, and further includes:

[0043] When an extraction operation by a customer service staff is detected, or when address-related voice or text is detected, the address extraction instruction is generated to trigger the address extraction process.

[0044] In this embodiment, the customer service staff can trigger the address extraction operation by pressing the address extraction button when address extraction is needed. The address extraction instruction can also be automatically generated when address-related voice or text (such as XX city, XX street, I live in XX, etc.) is detected in the conversation between the customer service and the customer.

[0045] Furthermore, the preprocessing module in the address extraction model preprocesses the voice dialogue corresponding to the current user, specifically including:

[0046] Based on the speech recognition module in the address extraction model, converting the speech dialogue into text information;

[0047] Based on the natural language processing module in the address extraction model, the text information is classified into intent to obtain the user intent.

[0048] In this embodiment, the voice dialogue between each user and the customer service is converted in real time through the voice recognition module, or the voice dialogue is converted when a conversion instruction is received.

[0049] Among them, the speech recognition module realizes the text conversion of voice dialogue through automatic speech recognition technology (Automatic Speech Recognition, ASR). The speech recognition module can call the installed speech recognition engine, such as Baidu ASR, Tencent ASR, etc., to realize the text conversion of voice dialogue.

[0050] The natural language processing module classifies the text information converted by the speech recognition module to determine the customer's telephone needs and intentions. The natural language processing module uses natural language processing (NLP) technology to achieve the intent classification of text information. A trained deep learning model (such as the BERT model) can be used as the natural language processing module.

[0051] Specifically, text information with intent labels (such as conversation text information with intent labels) can be used as a training data set, and the deep learning model can be trained through supervised learning to obtain a natural speech processing module.

[0052] Step S20, based on the entity recognition module in the address extraction model, perform entity recognition on the text information and the user intention to obtain the initial address information of the current user;

[0053] In this embodiment, based on the entity recognition module, entity recognition is performed on the text information obtained by the preprocessing module, and the customer address information is extracted from the result of the entity recognition as the initial address information.

[0054] Among them, the entity recognition module can be trained using a deep learning model (such as BERT, BiLSTM+CRF).

[0055] Specifically, text information carrying entity tags (such as conversation text information carrying entity tags) can be used as a training data set, and a deep learning model can be trained through supervised learning to obtain the entity recognition module.

[0056] Furthermore, when the customer address information extracted from the entity recognition result is incomplete, the address matching algorithm can be used to further complete the structured information of the address, such as when the city name, street name or area name in the city is missing, the address matching algorithm can be used to supplement and complete it. When a discontinued address is detected, the discontinued address can be updated to the latest address information.

[0057] Step S30, based on the address verification module in the address extraction model, verify the initial address information to obtain the verification result of the initial address information;

[0058] In this embodiment, based on the address verification module, the accuracy and completeness of the address are verified for the initial address information.

[0059] The address verification module can use address verification services or geocoding services to verify the extracted addresses to ensure the accuracy and completeness of the addresses.

[0060] Specifically, the address verification module can use the API services provided by Baidu Maps, Amap, etc. to complete the integrity and accuracy verification of the address.

[0061] Step S40, when the verification result is abnormal, obtain the abnormal part of the address, and generate dialogue guidance information based on the abnormal part of the address, so that the customer service staff can conduct the next round of inquiries to the current user based on the dialogue guidance information, until the address information with a normal verification result is obtained as the target address information.

[0062] In this embodiment, when the verification result is abnormal, the abnormal part of the address is obtained, and the dialogue guidance information is generated based on the abnormal part of the address. Specifically, the TTS technology can be used to interact with the customer through the dialogue guidance information. That is, when the information provided by the customer is incomplete or unclear, the dialogue guidance information is further generated for the customer to actively guide the dialogue with the customer based on the dialogue guidance information, so that through multiple rounds of questioning and inference, the accurate and complete address information can be gradually extracted and spliced.

[0063] Furthermore, when the verification result is normal, the address information that passes the verification can be output.

[0064] The present embodiment provides an address extraction method based on multi-round dialogues. When receiving an address extraction instruction, the method preprocesses the voice dialogue corresponding to the current user based on the preprocessing module in the address extraction model to obtain text information and user intention corresponding to the voice dialogue; performs entity recognition on the text information and user intention based on the entity recognition module in the address extraction model to obtain initial address information of the current user; verifies the initial address information based on the address verification module in the address extraction model to obtain a verification result of the initial address information; when the verification result is abnormal, obtains the abnormal part of the address, and generates dialogue guidance information based on the abnormal part of the address, so that the customer service personnel can conduct the next round of inquiries to the current user based on the dialogue guidance information, until the address information with a normal verification result is obtained as the target address information. Through the above method, the present application first pre-processes the voice conversation based on the pre-processing module in the address extraction model to determine the text information of the voice conversation and the user intention; then, based on the entity recognition module in the address extraction model, combined with the user intention, the address information in the text information is extracted to obtain the initial address information; and then based on the address verification module in the address extraction model, the initial address information is further verified, so as to further guide the customer service personnel to conduct multiple rounds of inquiries to the current user when the initial address information is unclear or incomplete, so as to extract accurate and complete address information through multiple rounds of conversations, thereby improving the efficiency of address extraction while also improving the accuracy of the extracted address.

[0065] Reference Figure 2 , Figure 2 It is a flowchart diagram of the second embodiment of the address extraction method based on multi-round dialogues of the present invention.

[0066] Based on the above Figure 1 In the embodiment shown, in this embodiment, the verification result includes an accuracy verification result and an integrity verification result, and the step S30 specifically includes:

[0067] Step S31, based on the address verification module, calling the address verification service and / or the address encoding service, performing accuracy verification and integrity verification on the initial address information, and obtaining the accuracy verification result and the integrity verification result.

[0068] In this embodiment, based on the address verification module, the address verification service or geocoding service can be used to verify the extracted address, that is, the API service provided by Baidu Maps, Amaps, etc. is used to search and verify the initial address information. If the location corresponding to the initial address information can be searched based on Baidu Maps and Amaps, the accuracy verification result of the initial address information is normal; if the location corresponding to the initial address information cannot be searched based on Baidu Maps and Amaps, the accuracy verification result of the initial address information is abnormal. And when the initial address information is the same as the specific address information corresponding to the initial address information searched by Baidu Maps and Amaps, the integrity verification result of the initial address information is normal; when the initial address information lacks some key information compared to the specific address information corresponding to the initial address information searched by Baidu Maps and Amaps, the integrity verification result of the initial address information is abnormal, such as Baidu Maps and Amaps searched for at least two identical location names corresponding to the initial address information (such as chain hotels or large supermarkets in the same city, etc.), lacking the specific area name to which it belongs (such as the government district or high-tech zone, etc.).

[0069] Specifically, when both the accuracy check result and the integrity check result of the initial address information are normal, the check result of the initial address information is determined to be normal. If either the accuracy check result or the integrity check result of the initial address information is abnormal, the check result of the initial address information is determined to be abnormal.

[0070] Furthermore, when the address extraction instruction is received, the voice dialogue corresponding to the current user is preprocessed based on the preprocessing module in the address extraction model to obtain the text information corresponding to the voice dialogue and the user intention, and further includes:

[0071] Fine-tuning the text information and the user intention through a fine-tuning module in the address extraction model to reduce speech translation errors; or

[0072] After performing entity recognition on the text information and the user intention based on the entity recognition module in the address extraction model to obtain the initial address information of the current user, the method further includes:

[0073] The initial address information is fine-tuned through the fine-tuning module to reduce entity extraction errors.

[0074] Furthermore, before the base fine-tunes the text information and the user intention through the fine-tuning module in the address extraction model to reduce speech translation errors, it also includes:

[0075] Collecting historical conversation data as a sample data set, wherein the historical conversation data includes historical voice conversations, historical text information corresponding to the historical voice conversations, and historical intention information;

[0076] Marking the conversation samples containing address information in the sample data set as positive samples, and marking the conversation samples not containing address information as negative samples;

[0077] Based on the sample data set, the positive samples and the negative samples, the initial model is trained to obtain an address extraction model including the entity recognition module.

[0078] Due to the limitations of the address extraction model, various errors may occur during the interaction process, such as ASR translation errors or NER extraction errors. In this embodiment, in order to reduce the above problems, an address extraction model (such as a seq2seq model) is used to fine-tune the instructions. Specifically:

[0079] 1. Data Collection:

[0080] Collect conversation data sets between customers and customer service (including manual and robot customer service), including user voice, text, and customer service responses;

[0081] According to the situations in which customers mention addresses in the conversation, the conversation dataset is annotated, and the conversations containing address information are marked as positive samples, and the conversations that do not contain address information are marked as negative samples.

[0082] 2. Build a seq2seq model:

[0083] Use language models, such as GPT-3, T5, etc., as basic models;

[0084] Fine-tune the pre-annotated conversation dataset to make the seq2seq model focus on extracting address-related content;

[0085] When fine-tuning, the conversation dataset is taken as input and the output is sequence labeled to mark address-related entities.

[0086] 3. Address extraction:

[0087] Preprocess the text of the dialogue dataset in advance, such as word segmentation and removal of stop words.

[0088] The preprocessed text is input into the fine-tuned seq2seq model to generate entity annotation results for the text by the seq2seq model.

[0089] Extract the part marked as address entity in the seq2seq model output as the final extraction result.

[0090] Through the above method, this embodiment uses a large-scale language model to deeply understand the conversation history and customer input in the telephone communication, so as to realize the automatic extraction of address information. When the customer mentions the address information, the address extraction model can automatically identify and extract the key information. If the information provided by the customer is incomplete or unclear, the address extraction model can actively guide the conversation and gradually piece together the accurate address information through multiple rounds of questioning and inference. In addition, the address extraction model also has powerful natural language processing capabilities. No matter how the customer expresses the address information, such as spoken, written or dialect, the address extraction model can accurately understand and parse it, thereby greatly improving the accuracy and adaptability of address extraction.

[0091] Reference Figure 3 , Figure 3 It is a flowchart diagram of a third embodiment of the address extraction method based on multi-round dialogues of the present invention.

[0092] Based on the above Figure 1 In the embodiment shown, in this embodiment, after step S20, the following steps are further included:

[0093] Step S050, based on the address matching rules, the initial address information is improved, and based on the user characteristics of the current user, the improved initial address is corrected, wherein the user characteristics include dialect, accent and / or landmark targets contained in the voice dialogue.

[0094] In this embodiment, when the customer address information extracted from the entity recognition result is incomplete, the address matching algorithm can be used to further complete the structured information of the address. For example, when the city name, street name, or area name in the city is missing, the address matching algorithm can be used to supplement and complete it. When a deactivated address is detected, the deactivated address can be updated to the latest address information. Specifically, the specific address of the current user can be reasonably inferred through the user's identity information (such as place of origin or residential address), the user's accent and dialect during the call, and the landmark targets mentioned by the user during the process, and the inference result can be matched with the initial address information for completion and correction.

[0095] See also Figure 4 , Figure 4 It is a functional module diagram of an address extraction device based on multi-round dialogue provided in an embodiment of the present application.

[0096] like Figure 4As shown, the multi-round dialogue-based address extraction device 400 includes:

[0097] The dialogue preprocessing module 10 is used to preprocess the voice dialogue corresponding to the current user based on the preprocessing module in the address extraction model when receiving the address extraction instruction, and obtain the text information corresponding to the voice dialogue and the user intention;

[0098] An address entity extraction module 20, configured to perform entity recognition on the text information and user intention based on the entity recognition module in the address extraction model, and obtain the initial address information of the current user;

[0099] An address information verification module 30, configured to verify the initial address information based on the address verification module in the address extraction model, and obtain a verification result of the initial address information;

[0100] The dialogue guidance inquiry module 40 is used to obtain the abnormal part of the address when the verification result is abnormal, and generate dialogue guidance information based on the abnormal part of the address, so that the customer service staff can conduct the next round of inquiries to the current user based on the dialogue guidance information until the address information with a normal verification result is obtained as the target address information.

[0101] Furthermore, the dialogue preprocessing module 10 specifically includes:

[0102] A speech recognition unit, configured to convert the speech dialogue into text information based on the speech recognition module in the address extraction model;

[0103] The intention recognition unit is used to classify the text information according to its intention based on the natural language processing module in the address extraction model to obtain the user intention.

[0104] Furthermore, the multi-round dialogue-based address extraction device 400 further includes an extraction instruction generation module, which is used to:

[0105] When an extraction operation by a customer service staff is detected, or when address-related voice or text is detected, the address extraction instruction is generated to trigger the address extraction process.

[0106] Furthermore, the address information verification module 30 is specifically used to:

[0107] Based on the address verification module, an address verification service and / or an address encoding service is called to perform accuracy verification and integrity verification on the initial address information to obtain the accuracy verification result and the integrity verification result.

[0108] Furthermore, the multi-round dialogue-based address extraction device 400 further includes an intention fine-tuning module, which is used to:

[0109] Fine-tuning the text information and the user intention through a fine-tuning module in the address extraction model to reduce speech translation errors;

[0110] Furthermore, the multi-round dialogue-based address extraction device 400 further includes an entity fine-tuning module, which is used to:

[0111] The initial address information is fine-tuned through the fine-tuning module to reduce entity extraction errors.

[0112] Furthermore, the multi-round dialogue-based address extraction device 400 further includes an extraction model training module, which is used to:

[0113] Collecting historical conversation data as a sample data set, wherein the historical conversation data includes historical voice conversations, historical text information corresponding to the historical voice conversations, and historical intention information;

[0114] Marking the conversation samples containing address information in the sample data set as positive samples, and marking the conversation samples not containing address information as negative samples;

[0115] Based on the sample data set, the positive samples and the negative samples, the initial model is trained to obtain an address extraction model including the entity recognition module.

[0116] Furthermore, the multi-round dialogue-based address extraction device 400 further includes an address matching improvement module, which is used to:

[0117] Based on the address matching rules, the initial address information is improved, and based on the user characteristics of the current user, the improved initial address is corrected, wherein the user characteristics include dialect, accent and / or landmark targets included in the voice conversation.

[0118] It should be noted that those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working process of the above-described device and each module can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0119] The above-mentioned device can be implemented in the form of a computer program. Figure 5 Runs on the computer device shown.

[0120] See also Figure 5 , Figure 51 is a schematic block diagram of a computer device provided in an embodiment of the present application. The computer device may be a server.

[0121] See also Figure 5 The computer device includes a processor, a memory and a network interface connected through a system bus, wherein the memory may include a non-volatile storage medium and an internal memory.

[0122] The non-volatile storage medium can store an operating system and a computer program. The computer program includes program instructions, and when the program instructions are executed, the processor can execute any address extraction method based on multiple rounds of dialogue.

[0123] The processor is used to provide computing and control capabilities and support the operation of the entire computer equipment.

[0124] The internal memory provides an environment for the operation of the computer program in the non-volatile storage medium. When the computer program is executed by the processor, the processor can execute any address extraction method based on multi-round dialogues.

[0125] The network interface is used for network communication, such as sending assigned tasks, etc. Those skilled in the art will appreciate that Figure 5 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0126] It should be understood that the processor may be a central processing unit (CPU), and the processor may also be other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor, etc.

[0127] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0128] When receiving the address extraction instruction, preprocessing the voice dialogue corresponding to the current user is performed based on the preprocessing module in the address extraction model to obtain text information corresponding to the voice dialogue and the user intention;

[0129] Based on the entity recognition module in the address extraction model, entity recognition is performed on the text information and the user intention to obtain the initial address information of the current user;

[0130] Based on the address verification module in the address extraction model, the initial address information is verified to obtain a verification result of the initial address information;

[0131] When the verification result is abnormal, the abnormal part of the address is obtained, and dialogue guidance information is generated based on the abnormal part of the address, so that the customer service staff can conduct the next round of inquiries to the current user based on the dialogue guidance information until the address information with a normal verification result is obtained as the target address information.

[0132] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0133] Based on the speech recognition module in the address extraction model, converting the speech dialogue into text information;

[0134] Based on the natural language processing module in the address extraction model, the text information is classified into intent to obtain the user intent.

[0135] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0136] When an extraction operation by a customer service staff is detected, or when address-related voice or text is detected, the address extraction instruction is generated to trigger the address extraction process.

[0137] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0138] Based on the address verification module, an address verification service and / or an address encoding service is called to perform accuracy verification and integrity verification on the initial address information to obtain the accuracy verification result and the integrity verification result.

[0139] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0140] Fine-tuning the text information and the user intention through a fine-tuning module in the address extraction model to reduce speech translation errors;

[0141] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0142] The initial address information is fine-tuned through the fine-tuning module to reduce entity extraction errors.

[0143] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0144] Collecting historical conversation data as a sample data set, wherein the historical conversation data includes historical voice conversations, historical text information corresponding to the historical voice conversations, and historical intention information;

[0145] Marking the conversation samples containing address information in the sample data set as positive samples, and marking the conversation samples not containing address information as negative samples;

[0146] Based on the sample data set, the positive samples and the negative samples, the initial model is trained to obtain an address extraction model including the entity recognition module.

[0147] In one embodiment, the processor is used to run a computer program stored in the memory to implement the following steps:

[0148] Based on the address matching rules, the initial address information is improved, and based on the user characteristics of the current user, the improved initial address is corrected, wherein the user characteristics include dialect, accent and / or landmark targets included in the voice conversation.

[0149] A computer-readable storage medium is also provided in an embodiment of the present application, wherein the computer-readable storage medium stores a computer program, wherein the computer program includes program instructions, and the processor executes the program instructions to implement any of the multi-round dialogue-based address extraction methods provided in the embodiments of the present application.

[0150] The computer-readable storage medium may be an internal storage unit of the computer device described in the foregoing embodiment, such as a hard disk or memory of the computer device. The computer-readable storage medium may also be an external storage device of the computer device, such as a plug-in hard disk, a smart memory card (Smart Medi a Card, SMC), a secure digital (Secure Digital, SD) card, a flash memory card (Flash Card), etc., equipped on the computer device.

[0151] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed in the present application, and these modifications or replacements should be included in the protection scope of the present application. Therefore, the protection scope of the present application shall be based on the protection scope of the claims.

Claims

1. A method for address extraction based on multi-round dialogue, characterized in that: The address extraction method comprises the following steps: When receiving the address extraction instruction, preprocessing the voice dialogue corresponding to the current user is performed based on the preprocessing module in the address extraction model to obtain text information corresponding to the voice dialogue and the user intention; Based on the entity recognition module in the address extraction model, entity recognition is performed on the text information and the user intention to obtain the initial address information of the current user; Based on the address verification module in the address extraction model, the initial address information is verified to obtain a verification result of the initial address information; When the verification result is abnormal, the abnormal part of the address is obtained, and dialogue guidance information is generated based on the abnormal part of the address, so that the customer service staff can conduct the next round of inquiries to the current user based on the dialogue guidance information until the address information with a normal verification result is obtained as the target address information.

2. The address extraction method based on multi-round dialogue as claimed in claim 1, characterized in that: The preprocessing module in the address extraction model preprocesses the voice dialogue corresponding to the current user, specifically including: Based on the speech recognition module in the address extraction model, converting the speech dialogue into text information; Based on the natural language processing module in the address extraction model, the text information is classified into intent to obtain the user intent.

3. The address extraction method based on multi-round dialogue as claimed in claim 1, characterized in that: When the address extraction instruction is received, the method pre-processes the voice dialogue corresponding to the current user based on the pre-processing module in the address extraction model to obtain the text information corresponding to the voice dialogue and the user intention, and further includes: When an extraction operation by a customer service staff is detected, or when address-related voice or text is detected, the address extraction instruction is generated to trigger the address extraction process.

4. The address extraction method based on multi-round dialogue as claimed in claim 1, characterized in that: The verification result includes an accuracy verification result and an integrity verification result. The address verification module in the address extraction model verifies the initial address information to obtain the verification result of the initial address information, which specifically includes: Based on the address verification module, an address verification service and / or an address encoding service is called to perform accuracy verification and integrity verification on the initial address information to obtain the accuracy verification result and the integrity verification result.

5. The address extraction method based on multi-round dialogue as claimed in claim 1, characterized in that: When the address extraction instruction is received, the voice dialogue corresponding to the current user is preprocessed based on the preprocessing module in the address extraction model to obtain text information corresponding to the voice dialogue and the user intention, and further includes: Fine-tuning the text information and the user intention through a fine-tuning module in the address extraction model to reduce speech translation errors; or After performing entity recognition on the text information and the user intention based on the entity recognition module in the address extraction model to obtain the initial address information of the current user, the method further includes: The initial address information is fine-tuned through the fine-tuning module to reduce entity extraction errors.

6. The address extraction method based on multi-round dialogue as claimed in claim 5, characterized in that: Before the base fine-tunes the text information and the user intention through the fine-tuning module in the address extraction model to reduce speech translation errors, it also includes: Collecting historical conversation data as a sample data set, wherein the historical conversation data includes historical voice conversations, historical text information corresponding to the historical voice conversations, and historical intention information; Marking the conversation samples containing address information in the sample data set as positive samples, and marking the conversation samples not containing address information as negative samples; Based on the sample data set, the positive samples and the negative samples, the initial model is trained to obtain an address extraction model including the entity recognition module.

7. The address extraction method based on multi-round dialogue according to any one of claims 1 to 6, characterized in that: After performing entity recognition on the text information and the user intention based on the entity recognition module in the address extraction model to obtain the initial address information of the current user, the method further includes: Based on the address matching rules, the initial address information is improved, and based on the user characteristics of the current user, the improved initial address is corrected, wherein the user characteristics include dialect, accent and / or landmark targets included in the voice conversation.

8. An address extraction device based on multi-round dialogue, characterized in that: The address extraction device based on multiple rounds of dialogues comprises: A dialogue preprocessing module, which is used to preprocess the voice dialogue corresponding to the current user based on the preprocessing module in the address extraction model when receiving the address extraction instruction, and obtain the text information corresponding to the voice dialogue and the user intention; An address entity extraction module, used to perform entity recognition on the text information and user intention based on the entity recognition module in the address extraction model to obtain the initial address information of the current user; An address information verification module, used to verify the initial address information based on the address verification module in the address extraction model, and obtain a verification result of the initial address information; The dialogue guidance inquiry module is used to obtain the abnormal part of the address when the verification result is abnormal, and generate dialogue guidance information based on the abnormal part of the address, so that the customer service personnel can conduct the next round of inquiries to the current user based on the dialogue guidance information until the address information with a normal verification result is obtained as the target address information.

9. A computer device, characterized in that: The computer device includes a processor, a memory, and an address extraction program based on multi-round dialogues stored in the memory and executable by the processor, wherein the address extraction program based on multi-round dialogues, when executed by the processor, implements the steps of the address extraction method based on multi-round dialogues as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an address extraction program based on multi-round dialogues, wherein when the address extraction program based on multi-round dialogues is executed by a processor, the steps of the address extraction method based on multi-round dialogues as described in any one of claims 1 to 7 are implemented.

Images