Interrogation information extraction method, system and device and storage medium

Abstract

The invention discloses an interrogation information extraction method, system and device and a storage medium, and relates to the technical field of knowledge graph construction and computer artificial intelligence, and the method comprises the steps: obtaining interrogation record data, extracting text information, matching keywords in the text information through a regular expression, and storing the keywords in the text information; carrying out anaphora resolution on the substitute of the interrogated person in the text information by using a standard character matching method; using a T5 pre-training model to carry out anaphora resolution and sentence pattern simplification on pronouns called by other people, using a BiLstm-CRF model to identify substitutions of all people, and carrying out pairwise pairing on the substitutions of all people; carrying out character relationship extraction on each pair of entities based on a Bert pre-training model; establishing a knowledge graph by taking the character as a node, the extracted character relationship as a node relationship and the basic information of the character as a node attribute; therefore, the interrogation information is easier to understand and analyze, and the case reasoning efficiency is improved.

Description

Technical Field

[0001] The present invention relates to the field of knowledge graph construction and computer artificial intelligence technology, and specifically to an interrogation information extraction method, system, device and storage medium. Background Art

[0002] In the process of handling criminal cases, it is necessary to analyze the interrogation records, and have a dedicated person review the interrogation records to obtain character information and their relationship network, and make tables to store this key information. In this process, some human resources will be consumed. In addition, this method of manually analyzing interrogation records is often difficult to fully and accurately analyze and express the character information and their relationship network in the interrogation records. This will cause certain obstacles to subsequent case analysis, formulation of secondary interrogation strategies, and related investigation and evidence collection activities. In the prior art, there is a lack of tools for analyzing character information and their relationship network in interrogation records.

[0003] At present, the published patent CN110991165A proposes a method for extracting character relationships in text, but the method only extracts sentences with two or more character entities, and the remaining sentences are discarded; most of the interrogation records are question-answer sentences, and personal pronouns are used to replace a certain entity name in the sentence. This method will cause the key information in such sentences to be lost. Patent CN111428041A discloses a method for generating case summaries, including extracting entities and entity attribute information from the interrogation records of the case, and determining the relationship between the entity attribute information and the entity; this method cannot judge the relationship between two name entities, but can only judge the relationship between entity attributes and entities, and the method does not construct a knowledge graph based on the extraction results that is conducive to subsequent analysis and judgment.

[0004] In summary, the existing interrogation information extraction methods only extract sentences with two or more person entities, resulting in the loss of key information and the inability to judge the relationship between two entities. They can only judge the relationship between entity attributes and entities. Summary of the invention

[0005] In view of the deficiency of the prior art that it can only judge the relationship between entity attributes and entities, but cannot judge the relationship between two entities, the present invention proposes an interrogation information extraction method, system, device and storage medium, which solves the problem that the prior art cannot judge the relationship between two entities by extracting character information and relationship networks in interrogation records and integrating them into a knowledge graph.

[0006] A method for extracting interrogation information comprises the following steps:

[0007] Obtain interrogation record data and extract text information;

[0008] Match keywords in text information through regular expressions, and extract corresponding content after the successfully matched keywords in the text information;

[0009] Based on the extracted keywords and their corresponding contents, the standard character matching method is used to resolve the pronouns of the person being interrogated in the text information; the T5 pre-trained model is used to resolve the pronouns of other personal pronouns in the text information;

[0010] The BiLstm-CRF model is used to identify the aliases of all characters in the text information after the reference resolution, and all the aliases of characters are paired;

[0011] Extract the person relationship for each pair of entities based on the Bert pre-trained model;

[0012] A knowledge graph is constructed by using characters and extracted character relationships as node relationships, and basic character information as node attributes;

[0013] Extract the information of the person to be interrogated and his / her relationship network based on the knowledge graph.

[0014] Furthermore, the step of obtaining the interrogation record data and extracting the text information specifically includes the following steps:

[0015] Automatically determine and pre-process the data format of interrogation records;

[0016] If the data is detected to be in image format, the ultra-lightweight PP-OCRv3 model is used for text recognition to extract the text in the image;

[0017] If the data is detected as a PDF or Word file, the interrogation record text is automatically extracted from the file.

[0018] Furthermore, the use of the T5 pre-trained model to perform reference resolution on other personal pronouns in the text information specifically includes the following steps:

[0019] Establish a reference resolution and sentence simplification model based on the T5 pre-trained model; the reference resolution and sentence simplification model includes a training data generation module, a text data representation module and a reference resolution and sentence simplification module;

[0020] The coreference resolution and sentence simplification model is defined and trained in the form of summary generation by a training data generation module;

[0021] Modify the tokenizer of the Bert pre-trained model through the text data representation module to imitate the tokenizer of the T5 pre-trained model and add the word segmentation function. Input the original text data into the tokenizer to obtain the word embedding representation of the text data.

[0022] The coreference resolution and sentence simplification module is used to perform coreference resolution and sentence simplification on the input text in a generative manner, converting the single-round dialogue in the interrogation record into structured declarative sentences.

[0023] Furthermore, the BiLstm-CRF model is used to identify the aliases of all characters in the text information after the reference resolution, and all the aliases of the characters are paired, which includes the following steps:

[0024] Based on the pre-trained BERT, the original text data is input into the pre-trained BERT to obtain the word embedding representation of the text data;

[0025] Use BiLSTM to predict the probability of the given label of the input word embedding and output EmissionScore; the label is annotated in BIO format, B represents the beginning of an entity word, I represents the middle of an entity word, and O represents a non-entity word;

[0026] For a given Emission Score, a conditional random field is used to predict the predicted label sequence with the maximum probability that meets the label transfer constraints; the output is a BIO label sequence, and the entire continuous BIO label sequence is scanned, and the word between each B and the last I after it is a target entity.

[0027] Furthermore, the character relationship is extracted for each pair of entities based on the Bert pre-training model, and the relationship between the entities is output by combining the Bert pre-training model with the bidirectional GRU and Attention layers for recognition.

[0028] Furthermore, it also includes further extracting basic identity information of other persons involved in the interrogation record through the UIE model after extracting the person relationship for each pair of entities based on the Bert pre-training model.

[0029] Furthermore, the knowledge graph is improved by taking case categories associated with characters as classification information and taking cities associated with characters as regional concepts.

[0030] Furthermore, an interrogation information extraction system includes:

[0031] An acquisition module is used to acquire interrogation record data and extract text information;

[0032] The keyword matching module is used to match the keywords in the text information through regular expressions and extract the corresponding content after the successfully matched keywords in the text information;

[0033] The reference resolution module is used to resolve the reference of the interrogated person in the text information using the standard character matching method based on the extracted keywords and their corresponding contents; and to resolve the reference of other personal pronouns in the text information using the T5 pre-trained model;

[0034] The recognition module is used to use the BiLstm-CRF model to identify the aliases of all characters in the text information after the reference resolution, and to pair all the aliases of the characters in pairs;

[0035] The character relationship extraction module is used to extract the character relationship of each pair of entities based on the Bert pre-trained model;

[0036] A knowledge graph construction module is used to build a knowledge graph using characters and extracted character relationships as node relationships and characters’ basic information as node attributes;

[0037] The extraction module is used to extract the information of the person to be interrogated and his / her relationship network based on the knowledge graph.

[0038] Furthermore, a computer device for extracting interrogation information includes: a memory, a processor, and a computer program stored in the memory, and the processor implements the steps of the interrogation information extraction method when executing the computer program.

[0039] Furthermore, a readable storage medium is provided, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, they are used to execute the steps of the method for extracting interrogation information.

[0040] The present invention provides a method, system, device and storage medium for extracting interrogation information, which have the following features:

[0041] Beneficial effects:

[0042] The present invention performs reference resolution on the aliases of the interrogated person and other personal pronouns in the interrogation record, adopts the BiLstm-CRF model to identify the aliases of all characters, pairs all the aliases, extracts the character relationships for each pair of entities based on the Bert pre-training model, constructs the topological structure between nodes based on the character relationships extracted from the interrogation information, constructs a network with the basic information of the characters as node attributes, and constructs a character relationship knowledge graph, so that the interrogation information is easier to understand and analyze. In addition, the relationship between the persons involved in the case is judged and processed through the character relationship knowledge graph, which helps to grasp the key points of the case in the shortest time and facilitates the reasoning of the case. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 This is a framework flow chart of a method for extracting interrogation information in an embodiment of the present invention;

[0044] Figure 2 A specific flow chart of a method for extracting interrogation information in an embodiment of the present invention;

[0045] Figure 3 This is an example diagram of the contents of the interrogation record table extracted in the embodiment of the present invention;

[0046] Figure 4 This is an example diagram of a network structure generated in an embodiment of the present invention. DETAILED DESCRIPTION

[0047] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments.

[0048] The present invention provides a method for extracting interrogation information, comprising the following steps:

[0049] Step S1: Obtain the interrogation record form.

[0050] Step S2: Use regular expressions to match the contents between keywords in the interrogation record table and save them; the contents extracted from the interrogation record table in step S2 include: current time, location, interrogator, work unit, recorder, work unit, recorder, name, alias, gender, age, date of birth, education level, ID card name and number, work unit, current address, contact information, and registered permanent residence. In step S2, a keyword-based regular expression matching method is used. For direct extraction of certain formatted information in the interrogation record, the system implements regular expression matching extraction. In addition to the keyword extraction provided by the system itself, the system also allows users to customize and add keywords to cope with changes in formatted information in different interrogation records.

[0051] Step S3: Train and fine-tune the Text-to-Text Transfer Transformer pre-trained model, and input a single-round dialogue question and answer based on it; use a generative method to resolve reference for other personal pronouns; while completing the reference resolution, the model rewrites the questions and negative words in the dialogue and generates a simple text summary; convert the single-round dialogue into a more structured declarative sentence.

[0052] Step S4: Use the BiLSTM-CRF model to identify all the names of people in the sentence; Based on the Bert pre-trained model, combine the bidirectional GRU and Attention layers for identification, and the output result format is: entity, entity, relationship; The training data generation module in steps S3 and S4 uses summary generation to define the training so as to achieve decomposition and simplification results close to natural language; The text data representation module is used to establish a pre-trained BERT, and at the same time modify the BERT tokenizer to imitate the T5 tokenizer to add the word segmentation function, and input the original text data into this tokenizer to obtain the word embedding representation of the text data; The reference decomposition and sentence simplification module establishes a pre-trained T5 model, and performs reference decomposition and sentence simplification on the input text in a generative way, and converts the single-round dialogue in the interrogation record into a structured declarative sentence.

[0053] In step S4, the person entity recognition model includes: a text data representation module, which is used to establish a pre-trained BERT, input the original text data into the pre-trained BERT, and obtain the word embedding representation of the text data; a label probability prediction module, which is used to use BiLstm to predict the probability of the input word embedding to a given label label, and output the Emission score; a maximum probability prediction label sequence prediction module, which uses conditional random fields to predict the maximum probability predicted label sequence that meets the label transfer constraint conditions for a given Emission score input. The person entity relationship extraction model includes: a text data representation module, which is used to establish a pre-trained BERT, input the original text data into the pre-trained BERT, and obtain the word embedding representation of the text data; a maximum probability relationship prediction module, which is used to predict the input word embedding in combination with the bidirectional GRU and Attention layer, and output the relationship with the maximum belonging probability of the input sentence.

[0054] In step S4, the extracted relationship categories will depend on the relationship categories of the training data set. Currently, there are 14 default categories, namely: {"unknown", "husband and wife", "parents", "siblings", "superiors and subordinates", "teachers and students", "friends", "classmates", "cooperation", "colleagues", "couples", "grandparents and grandchildren", "fellow students", "relatives"}, where "unknown" represents relationships other than the above relationships. Unknown relationships can be manually labeled, and then the labeled data can be fine-tuned on the pre-trained model to increase recognizable default relationships.

[0055] Step S5: The basic identity information of other persons involved in the case is further extracted from the generated simple text summary through the UIE model; In step S5, the basic identity information of other persons involved in the case is further extracted from the interrogation record through the UIE model, including but not limited to: basic personal information: date of birth, gender, age, place of birth, permanent residence, place of household registration, ID number, contact information, work unit, whether or not a NPC representative; personal preference information: hobbies, occupation, habits; case details: time of crime, place of crime, tools used in crime, motives for crime. The extraction results and the original text of the interrogation record will be stored in the database in sequence, and the database will establish tables including but not limited to: 'personnel relationship table', 'interrogation personnel table', 'case table', and 'attached file table'.

[0056] Step S6: Use all the character relationships extracted from the interrogation information to build the topological structure between nodes, and use the basic information of the characters as node attributes to build a network. Combined with the extracted case details, a systematic and structured character relationship network is built. Take cities across the country as regional concepts, compare the extracted character information with the birthplace, date of birth, permanent residence, and family member information contained in the personal household registration book, find out the information related to the person involved in the case and improve the network structure. In step S6, the relationship network built by the system will use the neo4j database to build a graph database independent of the basic database. The graph database provides visualization of the entire network, which is convenient for users to view the relationship links between each node and the relationship between them, and helps users discover potential connections.

[0057] Example:

[0058] 1. Obtain the relevant data of the interrogation record of a certain department; the system supports two data acquisition methods: the user directly specifies the storage path of the relevant data in the system or the user specifies the database where the data is stored.

[0059] If the specified database method is used, the user needs to specify the table name and field name where the relevant data is stored. The system will automatically link to the database and extract the relevant data.

[0060] 2. Data format preprocessing: Use text recognition models to extract text from image data, extract text content from PDF and Word files, and do not process ordinary text format files; automatically determine and preprocess the data format of interrogation records. If the data is detected to be in image format, use the ultra-lightweight PP-OCRv3 model in Chinese and English provided by Baidu Feijiang for text recognition to extract text from the image. If the data is detected to be a PDF or Word file, the corresponding file processing program of the system is automatically called to extract the interrogation record text from the file. The file processing program mainly uses Python third-party libraries Python-Docx and PyPDF2.

[0061] 3. Use regular expressions to match keywords in the interrogation records and save the content after the keywords; add keywords and default keywords according to user customization. Use Python's own standard library re to perform regular expression matching on the formatted information in the interrogation records, and extract the corresponding information after the original text for the successfully matched keywords. The keywords include but are not limited to: current time, location, interrogator, work unit, recorder, work unit, recorder, and name of the person being interrogated.

[0062] 4. Use the standard character matching method to resolve the interrogated person's pseudonym; based on the structured information extracted by the regular expression, use the standard character matching method to replace all the pseudonyms in the dialogue part of the interrogation content with the person's real name, as follows:

[0063] The "I" spoken by the interrogated person, the "you" spoken by the interrogated person, the personal pronoun "I" spoken by the interrogator, and the personal pronoun "you" spoken by the interrogator will be replaced by: the name of the person being interrogated, the name of the interrogator, the name of the interrogator, and the name of the person being interrogated.

[0064] 5. Construct other personal pronoun reference resolution and sentence simplification models based on the T5 pre-trained model; the pre-trained model T5 consists of two parts: Pre-training and Fine-tuning. In the Pre-training stage, an unsupervised task, namely Span-Mask Deno ising Objective fragment-mask denoising task, is performed when there is no labeled data in the text. In the Fine-tuning stage, T5 is initialized with the parameters of the pre-trained model, and then Fine-tuned with labeled data downstream.

[0065] Since the word segmentation mode of T5 itself is not suitable for the Chinese mode, the present invention optimizes it by modifying the tokenizer of BERT and imitating the tokenizer of T5 to add the word segmentation function.

[0066] Among them, BERT can use the full-word coverage Chinese BERT pre-training model jointly released by Harbin Institute of Technology and iFlytek.

[0067] 6. Use the BiLstm-CRF model to identify all the entities (namely, the names) of the characters in the sentence, and pair all the character entities in pairs; input the original text data into the pre-trained BERT to obtain the word embedding representation of the text data; use BiLstm to predict the probability of the given label label for the input word embedding, and output the Emission score; for the given Emission score input, use the conditional random field to predict the predicted label sequence with the maximum probability that meets the label transfer constraint, and then obtain all the character entities through the 'label-text' inverse mapping; in terms of model training, use the 'BIO' format to annotate the labels, Beginning indicates the beginning of an entity word, Inside indicates the middle of an entity word, and Outside indicates a non-entity word. Randomly divide the data into training set and test set, input the training set into the BiLstm-CRF model for training, and then use the test set to verify the performance of the model.

[0068] 7. For each pair of entities, character relationships are extracted based on the Bert pre-trained model; the model input format is: entity name, entity name, input text; the model will determine the relationship between the two specified entity names. Input the original text data into the pre-trained BERT to obtain the word embedding representation of the text data; predict the input word embedding combined with the bidirectional GRU and Attention layer, and output the relationship with the maximum belonging probability of the input sentence; in terms of model training, the model will use one-hot encoding for all training relationship labels to match the standard input format of the model. Randomly divide the data into training set and test set, input the training set into the GR U-Attention model for training, and then use the test set to verify the performance of the model.

[0069] 8. Further extract case information through UIE model; use Baidu UIE-General Information Extraction to automatically extract all case information. Users can add extraction targets on their own. Including but not limited to:

[0070] Basic personal information: contact information, work unit, whether a member of the National People’s Congress.

[0071] Personal preference information: hobbies, occupation, habits.

[0072] Case details: time and location of the crime.

[0073] 9. The extraction results are stored in the standard database in sequence; the system automatically links to the user database, and the extraction results and the original interrogation records are stored in the database in sequence. Multiple tables such as "personnel relationship table", "interrogation personnel table", "case table" and "attached file table" will be established in the database to facilitate subsequent management and viewing.

[0074] 10. Take characters as nodes, extract character relationships in the results as node relationships to build a topological structure between nodes, and use the basic information of characters in the results as node attributes to build a network; use the case categories associated with characters as classification information, and the cities across the country associated with characters as regional concepts to improve the network structure.

[0075] The neo4j database is used to build a graph database independent of the basic database, and the graph database provides visualization functions for the entire network.

[0076] The present invention realizes comprehensive processing and information extraction of interrogation records in various formats, improves user work efficiency, and provides users with flexible operating space to meet users' diverse needs for interrogation information processing. A highly structured network of relationships between people is constructed. This makes interrogation information easier to understand and analyze, helps to quickly and accurately understand the development of the case and the relationship between the people involved, and improves the efficiency of information integration and analysis. At the same time, the flexibility and comprehensiveness of the system provide users with a convenient and comprehensive information processing tool.

[0077] Based on the same inventive concept, the present invention proposes an interrogation information extraction system, comprising:

[0078] The acquisition module is used to obtain interrogation record data and extract text information.

[0079] The keyword matching module is used to match the keywords in the text information through regular expressions and extract the corresponding content after the successfully matched keywords in the text information.

[0080] The reference resolution module is used to resolve the reference of the person being interrogated in the text information based on the extracted keywords and their corresponding content using the standard character matching method; and to resolve the reference of other personal pronouns in the text information using the T5 pre-trained model.

[0081] The recognition module is used to use the BiLstm-CRF model to identify the aliases of all characters in the text information after reference resolution, and to pair all the aliases of characters in pairs.

[0082] The character relationship extraction module is used to extract the character relationship for each pair of entities based on the Bert pre-training model.

[0083] The knowledge graph construction module is used to build a knowledge graph using characters and extracted character relationships as node relationships, and basic character information as node attributes.

[0084] The extraction module is used to extract the information of the person to be interrogated and his / her relationship network based on the knowledge graph.

[0085] Based on the same inventive concept, the present invention also proposes a computer device for extracting interrogation information, comprising: a memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the interrogation information extraction method when executing the computer program.

[0086] Based on the same inventive concept, the present invention also proposes a readable storage medium, wherein the readable storage medium stores a computer program, wherein the computer program includes program instructions, and when the program instructions are executed by a processor, they are used to execute the steps of the method for extracting interrogation information.

[0087] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A method for extracting interrogation information, characterized in that: The following steps are involved: Obtain interrogation record data and extract text information; Match keywords in text information through regular expressions, and extract corresponding content after the successfully matched keywords in the text information; Based on the extracted keywords and their corresponding contents, the standard character matching method is used to resolve the pronouns of the person being interrogated in the text information; the T5 pre-trained model is used to resolve the pronouns of other personal pronouns in the text information; The BiLstm-CRF model is used to identify the aliases of all characters in the text information after the reference resolution, and all the aliases of characters are paired; Extract the person relationship for each pair of entities based on the Bert pre-trained model; A knowledge graph is constructed by using characters and extracted character relationships as node relationships, and basic character information as node attributes; Extract the information of the person to be interrogated and his / her relationship network based on the knowledge graph.

2. The method for extracting interrogation information according to claim 1, characterized in that: The method of obtaining the interrogation record data and extracting the text information specifically includes the following steps: Automatically determine and pre-process the data format of interrogation records; If the data is detected to be in image format, the PP-OCRv3 model is used for text recognition to extract the text in the image; If the data is detected as a PDF or Word file, the interrogation record text is automatically extracted from the file.

3. The method for extracting interrogation information according to claim 1, characterized in that: The method of using the T5 pre-trained model to resolve the reference of other personal pronouns in the text information specifically includes the following steps: Establish a reference resolution and sentence simplification model based on the T5 pre-trained model; the reference resolution and sentence simplification model includes a training data generation module, a text data representation module and a reference resolution and sentence simplification module; The coreference resolution and sentence simplification model is defined and trained in the form of summary generation by a training data generation module; Modify the tokenizer of the Bert pre-trained model through the text data representation module to imitate the tokenizer of the T5 pre-trained model and add the word segmentation function. Input the original text data into the tokenizer to obtain the word embedding representation of the text data. The coreference resolution and sentence simplification module is used to perform coreference resolution and sentence simplification on the input text in a generative manner, converting the single-round dialogue in the interrogation record into structured declarative sentences.

4. The method for extracting interrogation information according to claim 1, characterized in that: The BiLstm-CRF model is used to identify the aliases of all characters in the text information after the reference resolution, and all the aliases of the characters are paired, which includes the following steps: Based on the pre-trained BERT, the original text data is input into the pre-trained BERT to obtain the word embedding representation of the text data; Use BiLSTM to predict the probability of the given label of the input word embedding and output EmissionScore; the label is annotated in BIO format, B represents the beginning of an entity word, I represents the middle of an entity word, and O represents a non-entity word; For a given Emission Score, a conditional random field is used to predict the predicted label sequence with the maximum probability that meets the label transfer constraints; the output is a BIO label sequence, and the entire continuous BIO label sequence is scanned, and the word between each B and the last I after it is a target entity.

5. The method for extracting interrogation information according to claim 1, characterized in that: The Bert pre-trained model is used to extract the relationship between each pair of entities. The Bert pre-trained model is used as a basis, combined with the bidirectional GRU and Attention layers for recognition, and the relationship between the entities is output.

6. The method for extracting interrogation information according to claim 1, characterized in that: It also includes further extracting basic identity information of other persons involved in the interrogation records through the UIE model after extracting the person relationship for each pair of entities based on the Bert pre-trained model.

7. The method for extracting interrogation information according to claim 1, characterized in that: It also includes using case categories associated with people as classification information and cities associated with people as regional concepts to improve the knowledge graph.

8. An interrogation information extraction system, characterized in that: include: An acquisition module is used to acquire interrogation record data and extract text information; The keyword matching module is used to match the keywords in the text information through regular expressions and extract the corresponding content after the successfully matched keywords in the text information; A reference resolution module is used to resolve the reference of the interrogated person in the text information using a standard character matching method based on the extracted keywords and their corresponding contents; Use the T5 pre-trained model to resolve other personal pronouns in text information; The recognition module is used to use the BiLstm-CRF model to identify the aliases of all characters in the text information after the reference resolution, and to pair all the aliases of the characters in pairs; The character relationship extraction module is used to extract the character relationship of each pair of entities based on the Bert pre-trained model; A knowledge graph construction module is used to build a knowledge graph using characters and extracted character relationships as node relationships and characters’ basic information as node attributes; The extraction module is used to extract the information of the person to be interrogated and his / her relationship network based on the knowledge graph.

9. A computer device for extracting interrogation information, characterized in that: include: A memory, a processor, and a computer program stored in the memory, wherein the processor implements the steps of the interrogation information extraction method according to any one of claims 1 to 7 when executing the computer program.

10. A readable storage medium, characterized in that: The readable storage medium stores a computer program, and the computer program includes program instructions. When the program instructions are executed by a processor, they are used to execute the steps of the method for extracting interrogation information according to any one of claims 1 to 7.

Images