Inference type dialogue response method and system based on large model and storage medium

Abstract

The invention discloses an inference type dialogue response method and system based on a large model, and a storage medium, relates to the technical field of service consultation dialogue response, and solves the problems that the existing service consultation dialogue response is generally relatively simple, lacks an active inquiry strategy, only can be passively responded and cannot simulate a questioning mechanism of service personnel, and the service consultation dialogue response cannot be simulated. The technical problems of insufficient suggestion judgment accuracy and high misjudgment rate caused by difficulty in comprehensively and accurately collecting the conditions of the demanders in the prior art are solved. The method comprises the steps of obtaining voice data of a demander for service consultation; the method comprises the steps of performing feature extraction on voice input data to obtain condition feature data; inquiring the demander through a pre-trained condition inquiry model based on the condition characteristic data to obtain judgment data; matching from a historical consultation database based on the condition feature data and the judgment data to obtain a reasoning target; and feeding back a judgment suggestion to the demander based on the reasoning target, thereby solving the technical problem.

Description

Technical Field

[0001] The present invention belongs to the field of service consultation dialogue response, and relates to artificial intelligence model technology, specifically to an inference-based dialogue response method, system and storage medium based on a large model. Background Art

[0002] Currently, mainstream conversational systems are mostly based on rule-based engines or search-based models, and their response logic often relies on preset templates or keyword matching. These systems often fail to fully understand user intent when faced with ambiguous statements, multiple rounds of questioning, or complex semantic scenarios, resulting in inaccurate responses.

[0003] In recent years, breakthroughs in pre-trained large models have demonstrated significant advantages in generative response. However, existing dialogue systems based on large models still face the following bottleneck: most dialogue models can only passively respond to user input and lack the ability to guide the conversation through dynamic questioning and supplement key information. As a result, current dialogue methods lack proactive inquiry strategies, can only respond passively, and cannot simulate the follow-up questioning mechanism of service personnel. It also makes it difficult to comprehensively and accurately collect situational information, resulting in insufficient accuracy in judgment and suggestions and a high rate of misjudgment.

[0004] Therefore, the present invention proposes a large-model-based reasoning dialogue response method, system and storage medium to solve the above problems. Summary of the Invention

[0005] The present invention aims to solve at least one of the technical problems existing in the prior art; to this end, the present invention proposes a large-model-based reasoning dialogue response method, system and storage medium to solve the problem that the existing service consultation dialogue response is usually relatively simple, and is mostly matched according to pre-set rules, such as directly searching the database or using simple keyword matching to give fixed suggestions. This type of dialogue lacks an active inquiry strategy and can only respond passively without simulating the follow-up mechanism of service personnel. It is also difficult to comprehensively and accurately collect the situation of the demander, resulting in insufficient accuracy in judging the suggestions and a high error rate.

[0006] To achieve the above objectives, the first aspect of the present invention provides a large-model-based reasoning dialogue response method, comprising:

[0007] Step 1: Obtain the voice data used by the customer for service consultation;

[0008] Step 2: Extracting features from the voice input data to obtain situation feature data;

[0009] Step 3: Based on the situation feature data, the pre-trained situation inquiry model is used to inquire the demander and obtain judgment data;

[0010] Step 4: Obtain inference targets based on the situation feature data and judgment data from the historical consultation database;

[0011] Step 5: Provide judgment suggestions to the demander based on the reasoning goal.

[0012] Preferably, the step of obtaining the voice data of the demander for service consultation includes:

[0013] The voice data of the demander for service consultation is collected through voice collection equipment; among which, the voice data refers to the voice of the demander stating his personal needs to the medical consultation robot.

[0014] Preferably, the feature extraction of the speech input data includes:

[0015] Converting speech data into text using speech recognition technology to obtain speech text; speech recognition technology includes Google Speech-to-Text API or IBM Watson Speech to Text;

[0016] Situational features in speech text are extracted through natural language processing technology to obtain situational feature data; wherein, situational feature data refers to a collection of situational features in the demand put forward by the demander in the speech text.

[0017] It should be noted that the situation features refer to specific information related to the service consultation status extracted from the voice text provided by the demander; for example, the demander says: "I need to replace my driver's license. Which department and service window should I go to?", then the situation features extracted through natural language processing technology include "replacement", "driver's license", "department" and "window".

[0018] Preferably, the step of querying the demander through a pre-trained situation query model based on the situation feature data includes:

[0019] Input the situation feature number into the pre-trained situation inquiry model to obtain the inquiry question;

[0020] Return the inquiry to the requester and collect the requester's answer voice in real time;

[0021] The answering voice is converted into text form by using speech recognition technology to obtain the answering text;

[0022] The situation features in the answer text are extracted through natural language processing technology to obtain judgment data; wherein, the judgment data refers to the collection of situation features in the answer text.

[0023] Preferably, the training method of the pre-trained situation inquiry model includes:

[0024] Obtain questions asked by service personnel regarding certain situations based on historical consultation service records;

[0025] Integrate several situations into standard input data, and integrate the inquiry questions corresponding to several situations into standard output data;

[0026] An artificial intelligence model is trained based on standard input data and standard output data to obtain a situation inquiry model; wherein the artificial intelligence model includes: a convolutional neural network or a deep belief network.

[0027] It should be noted that the inquiry questions refer to questions that the service personnel or equipment ask further based on the situation described by the demander to obtain more detailed information about the situation; for example, in the case of replacing a driver's license, the inquiry questions are: whether you have participated in a physical examination in the hospital, whether the physical examination included color blindness and color weakness detection, etc.

[0028] Preferably, the inference target is obtained by matching the situation feature data and the judgment data from the historical consultation database, including:

[0029] The situation feature data and judgment data are converted into corresponding word vectors through a word vector model to obtain situation word vector data; wherein the word vector model includes: Word2Vec or BERT;

[0030] The situations corresponding to several situations in the historical consultation database are converted into word vectors through the word vector model to obtain the situation word vector data corresponding to the several situations;

[0031] Calculating similarities between the situation word vector data and the situation word vector data corresponding to a plurality of situations by a similarity algorithm to obtain a plurality of similarities; wherein the similarity algorithm includes: cosine similarity or Euclidean distance;

[0032] Based on several similarities, they are sorted in descending order, and the disease corresponding to the first-ranked similarity is selected as the inference target.

[0033] It should be noted that the historical consultation database is a data set on the Internet that stores known situations and corresponding situations and precautions, etc., and the historical consultation database is manually entered.

[0034] Preferably, the feedback of judgment suggestions to the demander based on the reasoning goal includes:

[0035] Obtain the precautions and processing methods corresponding to the reasoning target based on the historical consultation database;

[0036] Get the processing department corresponding to the reasoning target;

[0037] Feedback the precautions and processing methods corresponding to the reasoning target and the corresponding processing department to the demander.

[0038] A second aspect of the present invention provides a large model-based inference-based dialogue response system, comprising: a data acquisition module, a data analysis module, and a dialogue response module;

[0039] Data collection module: obtains voice data used by users for service consultation;

[0040] Data analysis module: extracts features from voice input data to obtain situation feature data; based on the situation feature data, the pre-trained situation inquiry model is used to inquire the user and obtain judgment data;

[0041] Dialogue response module: Based on the situation feature data and judgment data, the inference target is matched from the historical consultation database; based on the inference target, judgment suggestions are fed back to the demander.

[0042] Preferably, the data analysis module is in communication and / or electrically connected with the data acquisition module and the dialogue response module respectively.

[0043] A third aspect of the present invention provides a storage medium for large-model based reasoning dialogue response, wherein the storage medium stores computer program instructions, which implement the above method steps when executed by a processor.

[0044] Compared with the prior art, the present invention has the following beneficial effects:

[0045] The existing technology lacks an active inquiry strategy and can only respond passively without simulating the follow-up questioning mechanism of service personnel. It is also difficult to comprehensively and accurately collect situation information, resulting in insufficient accuracy of judgment suggestions and a high misjudgment rate. The present invention obtains the voice data of the demander and performs feature extraction to accurately identify the situation characteristics, laying the foundation for subsequent judgments. Based on the pre-trained situation inquiry model, the system actively asks questions to the demander, simulates the follow-up questioning mechanism of the service personnel, and deeply explores more details of the symptoms, making up for the deficiency of the existing technology that can only respond passively. By using the word vector model and similarity algorithm, the system accurately matches the inference target from the historical consultation database, effectively improving the accuracy of the judgment suggestion and reducing the misjudgment rate. According to the inference target, detailed judgment suggestions are fed back, including precautions, processing methods and processing departments, etc., to provide the demander with comprehensive and personalized service consultation, which significantly improves the service quality and reliability of the service consultation dialogue response system. BRIEF DESCRIPTION OF THE DRAWINGS

[0046] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0047] Figure 1 Schematic diagram of the method steps of an embodiment of the present invention;

[0048] Figure 2 Schematic diagram of system modules according to an embodiment of the present invention. DETAILED DESCRIPTION

[0049] The technical solutions of the present invention will be clearly and completely described below in conjunction with the embodiments. Obviously, the embodiments described are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0050] See also Figure 1 The first embodiment of the present invention provides a large model-based reasoning dialogue response method, including:

[0051] Step 1: Obtain the voice data used by the customer for service consultation;

[0052] Step 2: Extracting features from the voice input data to obtain situation feature data;

[0053] Step 3: Based on the situation feature data, the pre-trained situation inquiry model is used to inquire the demander and obtain judgment data;

[0054] Step 4: Obtain inference targets based on the situation feature data and judgment data from the historical consultation database;

[0055] Step 5: Provide judgment suggestions to the demander based on the reasoning goal

[0056] In this embodiment, the voice data of the demander for service consultation is obtained, including:

[0057] The voice data of the demander for service consultation is collected through voice collection equipment; among which, the voice data refers to the voice of the demander stating his personal needs to the medical consultation robot.

[0058] Furthermore, feature extraction is performed on the voice input data, including:

[0059] Converting speech data into text using speech recognition technology to obtain speech text; speech recognition technology includes Google Speech-to-Text API or IBM Watson Speech to Text;

[0060] Situational features in speech text are extracted through natural language processing technology to obtain situational feature data; wherein, situational feature data refers to a collection of situational features in the demand put forward by the demander in the speech text.

[0061] Based on the situation feature data, the pre-trained situation inquiry model is used to inquire the demander, including:

[0062] Input the situation feature number into the pre-trained situation inquiry model to obtain the inquiry question;

[0063] Return the inquiry to the requester and collect the requester's answer voice in real time;

[0064] The answering voice is converted into text form by using speech recognition technology to obtain the answering text;

[0065] The situation features in the answer text are extracted through natural language processing technology to obtain judgment data; wherein, the judgment data refers to the collection of situation features in the answer text.

[0066] In the case of pre-training, the model training method is asked, including:

[0067] Obtain questions asked by service personnel regarding certain situations based on historical consultation service records;

[0068] Integrate several situations into standard input data, and integrate the inquiry questions corresponding to several situations into standard output data;

[0069] An artificial intelligence model is trained based on standard input data and standard output data to obtain a situation inquiry model; wherein the artificial intelligence model includes: a convolutional neural network or a deep belief network.

[0070] The reasoning targets are obtained by matching the situation feature data and judgment data from the historical consultation database, including:

[0071] The situation feature data and judgment data are converted into corresponding word vectors through a word vector model to obtain situation word vector data; wherein the word vector model includes: Word2Vec or BERT;

[0072] The situations corresponding to several situations in the historical consultation database are converted into word vectors through the word vector model to obtain the situation word vector data corresponding to the several situations;

[0073] Calculating similarities between the situation word vector data and the situation word vector data corresponding to a plurality of situations by a similarity algorithm to obtain a plurality of similarities; wherein the similarity algorithm includes: cosine similarity or Euclidean distance;

[0074] Based on several similarities, they are sorted in descending order, and the disease corresponding to the first-ranked similarity is selected as the inference target.

[0075] Feedback to the user based on the reasoning goal includes:

[0076] Obtain the precautions and processing methods corresponding to the reasoning target based on the historical consultation database;

[0077] Get the processing department corresponding to the reasoning target;

[0078] Feedback the precautions and processing methods corresponding to the reasoning target and the corresponding processing department to the demander.

[0079] For example, a customer inquires about his or her situation, as follows:

[0080] 1. Step 1: Obtain voice data;

[0081] Customer input: The customer says to the service consulting robot via voice: "I need to renew my driver's license. Which department and service window should I go to?"

[0082] Voice collection: Voice collection devices (such as smart speakers) record the voice of the demander in real time and generate voice data.

[0083] 2. Step 2: Feature extraction;

[0084] Speech recognition: Use the Google Speech-to-Text API to convert speech into text: "I need to renew my driver's license. Which department or service window should I go to?"

[0085] Situation feature extraction:

[0086] Natural Language Processing (NLP): Extracting situational features through entity recognition technology

[0087] Situation characteristic data: replacement, driver's license, department, window.

[0088] 3. Step 3: Active questioning and judgment data collection;

[0089] The situation asks the model to generate questions:

[0090] Input situation features: Model input "replacement, driver's license, department, window".

[0091] Generate follow-up questions (based on the agent question logic in the training data):

[0092] 1. “Have you ever participated in a hospital experience?”

[0093] 2. "Would you like to have a physical examination with us?"

[0094] 3. "Does the physical examination include testing for color blindness and color deficiency?"

[0095] Demander's answer:

[0096] Voice answer: "I had a medical examination at the hospital last weekend to renew my driver's license."

[0097] Judgment data extraction:

[0098] Judgment data: Last weekend, the driver’s license physical examination was done.

[0099] 4. Step 4: Disease matching;

[0100] Word vector conversion:

[0101] Symptom word vector data: The condition features (last weekend, already done, driver's license, physical examination) are converted into high-dimensional vectors through the BERT model.

[0102] Disease word vector data: Extract situations such as "already done", "physical examination for driver's license", and "last weekend" from the historical consultation database (such as "replace, driver's license, already, completed, physical examination" for replacing a driver's license) and convert them into vectors.

[0103] Similarity calculation:

[0104] Cosine similarity calculation:

[0105] Similarity between the hypothetical situation vector and the driver's license replacement: 0.85;

[0106] Similarity between the hypothetical situation vector and the inquiry system: 0.72;

[0107] Reasoning target: Select the situation corresponding to the maximum similarity as the reasoning target, that is, changing the driver's license (0.85).

[0108] 5. Step 5: Determine the feedback;

[0109] Get situation information:

[0110] Renewing your driver's license:

[0111] Notes: You need to queue up to get a number to avoid getting the wrong number and provide a medical examination certificate.

[0112] How to deal with it: Go to service window No. 5 and ask the service staff to handle it.

[0113] Processing department: Service Hall.

[0114] After the system integrates the information, it will be fed back to the requester in text form:

[0115] Existing technologies may directly match situations based solely on “replacement, driver’s license”, leading to misjudgments (such as ignoring key information such as “experience”).

[0116] The present invention actively asks questions through a pre-trained model, collects details such as "already", "driver's license", and "experience", and significantly improves the accuracy of judgment.

[0117] See Figure 2 , a second aspect of the present invention provides an inference-based dialogue response system based on a large model, comprising: a data acquisition module, a data analysis module and a dialogue response module;

[0118] Data collection module: obtains voice data used by users for service consultation;

[0119] Data analysis module: extracts features from voice input data to obtain situation feature data; based on the situation feature data, the pre-trained situation inquiry model is used to inquire the user and obtain judgment data;

[0120] Dialogue response module: Based on the situation feature data and judgment data, the inference target is matched from the historical consultation database; based on the inference target, judgment suggestions are fed back to the demander.

[0121] The third aspect of the present invention provides a storage medium for large-model based reasoning dialogue response, wherein the storage medium stores computer program instructions, which implement the above method steps when executed by a processor.

[0122] Working principle of the present invention:

[0123] The present invention obtains voice data used by the demander for service consultation; obtains situation feature data by performing feature extraction on the voice input data; obtains judgment data by questioning the demander through a pre-trained situation inquiry model based on the situation feature data; obtains an inference target from a historical consultation database based on matching the situation feature data and the judgment data; and feeds back judgment suggestions to the demander based on the inference target. This solves the technical problems that the existing technology lacks an active inquiry strategy, can only respond passively but cannot simulate the follow-up questioning mechanism of service personnel, and is difficult to fully and accurately collect situation information, resulting in insufficient accuracy of judgment suggestions and a high misjudgment rate.

[0124] The above embodiments are only used to illustrate the technical method of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical method of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical method of the present invention.

Claims

1. A large-model-based reasoning dialogue response method, characterized by: include: Step 1: Obtain the voice data used by the customer for service consultation; Step 2: Extracting features from the voice input data to obtain situation feature data; Step 3: Based on the situation feature data, the pre-trained situation inquiry model is used to inquire the demander and obtain judgment data; Step 4: Obtain inference targets based on the situation feature data and judgment data from the historical consultation database; Step 5: Provide judgment suggestions to the demander based on the reasoning goal.

2. The large model-based reasoning dialogue response method according to claim 1, characterized in that: The acquisition of the voice data of the demander for service consultation includes: The voice data of the demander for service consultation is collected through voice collection equipment; among which, the voice data refers to the voice of the demander stating his personal needs to the medical consultation robot.

3. The large model-based reasoning dialogue response method according to claim 1, characterized in that: The feature extraction of the speech input data includes: Converting speech data into text using speech recognition technology to obtain speech text; speech recognition technology includes Google Speech-to-Text API or IBM Watson Speech to Text; Situational features in speech text are extracted through natural language processing technology to obtain situational feature data; wherein, situational feature data refers to a collection of situational features in the demand put forward by the demander in the speech text.

4. The large model-based reasoning dialogue response method according to claim 1, characterized in that: The method of querying the demander based on the situation feature data through a pre-trained situation query model includes: Input the situation feature number into the pre-trained situation inquiry model to obtain the inquiry question; Return the inquiry to the requester and collect the requester's answer voice in real time; The answering voice is converted into text form by using speech recognition technology to obtain the answering text; The situation features in the answer text are extracted through natural language processing technology to obtain judgment data; wherein, the judgment data refers to the collection of situation features in the answer text.

5. The method for inference-based dialogue response based on a large model according to claim 4, characterized in that: The pre-training query model training method includes: Obtain questions asked by service personnel regarding certain situations based on historical consultation service records; Integrate several situations into standard input data, and integrate the inquiry questions corresponding to several situations into standard output data; An artificial intelligence model is trained based on standard input data and standard output data to obtain a situation inquiry model; wherein the artificial intelligence model includes: a convolutional neural network or a deep belief network.

6. The large model-based reasoning dialogue response method according to claim 1, characterized in that: The inference target is obtained by matching the situation feature data and the judgment data from the historical consultation database, including: The situation feature data and judgment data are converted into corresponding word vectors through a word vector model to obtain situation word vector data; wherein the word vector model includes: Word2Vec or BERT; The situations corresponding to several situations in the historical consultation database are converted into word vectors through the word vector model to obtain the situation word vector data corresponding to the several situations; Calculating similarities between the situation word vector data and the situation word vector data corresponding to a plurality of situations by a similarity algorithm to obtain a plurality of similarities; wherein the similarity algorithm includes: cosine similarity or Euclidean distance; Based on several similarities, they are sorted in descending order, and the disease corresponding to the first-ranked similarity is selected as the inference target.

7. The method for inference-based dialogue response based on a large model according to claim 1, characterized in that: The judgment suggestions fed back to the demander based on the reasoning goal include: Obtain the precautions and processing methods corresponding to the reasoning target based on the historical consultation database; Get the processing department corresponding to the reasoning target; Feedback the precautions and processing methods corresponding to the reasoning target and the corresponding processing department to the demander.

8. A large-model-based reasoning dialogue response system, configured to execute the large-model-based reasoning dialogue response method according to any one of claims 1 to 7, characterized in that: include: Data collection module, data analysis module and dialogue response module; Data collection module: obtains voice data used by users for service consultation; Data analysis module: extracts features from voice input data to obtain situation feature data; based on the situation feature data, the pre-trained situation inquiry model is used to inquire the user and obtain judgment data; Dialogue response module: This module obtains inference targets based on situation feature data and judgment data matched from the historical consultation database; Provide judgment suggestions to the demander based on the reasoning goal.

9. The large model-based reasoning dialogue response system according to claim 8, characterized in that: The data analysis module is in communication and / or electrically connected with the data acquisition module and the dialogue response module respectively.

10. A large-scale model-based inference dialogue response storage medium, characterized in that: The storage medium stores computer program instructions, which, when executed by a processor, implement the method steps described in any one of claims 1 to 7.

Images