Information extraction method, computer equipment and computer readable storage medium

Abstract

The invention provides an information extraction method, computer equipment and a computer readable storage medium. The method comprises the following steps: acquiring first text information; in response to the acquisition of the first text information, retrieving second text information associated with the first text information in a predetermined database; extracting third text information from the second text information; the current prompt word and the third text information serve as input of a large language model, and fourth text information is output through the large language model; determining the confidence of the fourth text information based on the text features of the fourth text information; if the confidence coefficient is lower than a preset threshold value, updating the current cue word according to a preset optimization mode until the confidence coefficient reaches the preset threshold value; and if the confidence reaches a preset threshold, outputting the current fourth text information. According to the technical scheme, the labor cost and the time cost are reduced, and the accuracy and the practicability of the adverse drug reaction information retrieval result are also improved.

Description

[Technical field]

[0001] The present application relates to the field of information processing technology, and in particular to an information extraction method, a computer device, and a computer-readable storage medium. [Background technology]

[0002] Medical literature is an important source of information on adverse drug reactions, which contains a large amount of data and cases on drug use, which directly or indirectly reflect the adverse drug reactions. In related technologies, by searching and viewing medical literature on a search website, you can timely understand the adverse reactions that occur during the use of drugs. However, this manual search method depends on the ability level of the searcher, and requires the searcher to browse a large amount of literature, which takes a long time and is inconvenient to search.

[0003] In related technologies, although adverse drug reaction information can be directly obtained by querying a large language model, the results obtained are not accurate compared to manual medical literature analysis.

[0004] Therefore, taking into account both the convenience and accuracy of drug adverse reaction information retrieval has become a technical problem that needs to be solved urgently. [Summary of the invention]

[0005] The embodiments of the present application provide an information extraction method, a computer device, and a computer-readable storage medium, which aim to solve the technical problem that the drug adverse reaction information retrieval method in the related art cannot simultaneously meet the dual requirements of retrieval convenience and retrieval accuracy.

[0006] In a first aspect, the embodiments of the present application provide an information extraction method, comprising:

[0007] Acquire first text information, wherein the first text information is a search keyword for medical literature describing adverse drug reactions;

[0008] In response to obtaining the first text information, searching a predetermined database for second text information associated with the first text information, wherein the second text information includes medical literature describing adverse drug reactions associated with the search keyword;

[0009] Extracting third text information from the second text information, wherein the third text information is text content describing the adverse drug reaction in the medical literature;

[0010] Using the current prompt word and the third text information as inputs of the large language model, and outputting fourth text information through the large language model, wherein the current prompt word is used to reflect the need for structural standardization processing of the third text information;

[0011] Determining the confidence of the fourth text information based on the text feature of the fourth text information, wherein the confidence is used to reflect the information quality of the fourth text information;

[0012] If the confidence level is lower than a predetermined threshold, updating the current prompt word according to a preset optimization method, and returning to the step of using the current prompt word and the third text information as inputs of the large language model until the confidence level reaches the predetermined threshold;

[0013] If the confidence reaches the predetermined threshold, the current fourth text information is output as the target output content, wherein the predetermined threshold is the confidence that the fourth text information can effectively describe the adverse drug reaction.

[0014] In one embodiment of the present application, optionally, extracting the third text information from the second text information includes:

[0015] Acquire semantic association information of the first text information, wherein the semantic association information includes synonyms and near synonyms of the search keyword and associated words of the adverse drug reaction corresponding to the search keyword;

[0016] Acquire a hit word matching the first text information and the semantic association information in the second text information;

[0017] The sentence where the hit word is located is set as the third text information.

[0018] In one embodiment of the present application, optionally, before taking the current prompt word and the third text information as inputs of the large language model, the method further includes:

[0019] Get the current prompt word, where:

[0020] The current prompt word includes respective acquisition instructions for a plurality of associated descriptive information related to the adverse drug reaction,

[0021] The multiple associated description information includes: the describer of the adverse drug reaction, the qualifications of the describer, the basis of description, the description language type, the output data structure, the name of the drug causing the adverse drug reaction, the drug production information, the individual information of the drug user and the medication performance information.

[0022] In one embodiment of the present application, optionally, the text features of the fourth text information include text fluency, number of duplicates to be removed, and invalid information ratio of each of the multiple types of associated description information;

[0023] The determining the confidence level of the fourth text information based on the text feature of the fourth text information includes:

[0024] Convert the text fluency, the number of times to be deduplicated, and the proportion of invalid information of each of the multiple types of associated description information into a first feature value, a second feature value, and a third feature value respectively;

[0025] Obtaining weights corresponding to text fluency, number of times to be deduplicated, and proportion of invalid information of each of the plurality of associated description information;

[0026] Based on the first eigenvalues ​​of each of the multiple types of associated description information and the weights of the text fluency of each of the multiple types of associated description information, a first confidence association matrix is ​​determined, and based on the second eigenvalues ​​and third eigenvalues ​​of each of the multiple types of associated description information, and the weights of the number of times to be deduplicated and the weights of the proportion of invalid information of each of the multiple types of associated description information, a second confidence association matrix is ​​determined, wherein the first confidence association matrix is ​​used to reflect the degree of influence of the text fluency on the information quality of the fourth text information, and the second confidence association matrix is ​​used to reflect the comprehensive degree of influence of the number of times to be deduplicated and the proportion of invalid information on the information quality of the fourth text information;

[0027] The confidence of the fourth text information is determined based on the first confidence association matrix and the second confidence association matrix.

[0028] In one embodiment of the present application, optionally, before determining the first confidence association matrix and determining the second confidence association matrix, the method further includes:

[0029] The first eigenvalue, the second eigenvalue, and the third eigenvalue are normalized.

[0030] In one embodiment of the present application, optionally, determining a first confidence association matrix based on the first eigenvalues ​​of the plurality of associated description information and the weights of the text fluency of the plurality of associated description information includes:

[0031] For each item of the associated description information, determine a product of the first characteristic value of the associated description information and a weight of the text fluency of the associated description information as a fourth characteristic value corresponding to the associated description information;

[0032] Based on the fourth eigenvalue of each item of the associated description information, determine a first confidence association matrix, wherein the element in the mth row and nth column of the first confidence association matrix is ​​the product of the fourth eigenvalue of the mth item of the associated description information and the fourth eigenvalue of the nth item of the associated description information;

[0033] The determining of a second confidence association matrix based on the second eigenvalue and the third eigenvalue of each of the multiple association description information, and the weight of the number of times to be deduplicated and the weight of the proportion of invalid information of each of the multiple association description information includes:

[0034] For each item of the associated description information, determine a weighted average of the second eigenvalue and the third eigenvalue of the associated description information as a fifth eigenvalue corresponding to the associated description information;

[0035] Based on the fifth eigenvalue of each item of the associated description information, a second confidence association matrix is ​​determined, wherein the element in the jth row and kth column of the second confidence association matrix is ​​the product of the fifth eigenvalue of the jth item of the associated description information and the fifth eigenvalue of the kth item of the associated description information.

[0036] In one embodiment of the present application, optionally, determining the confidence of the fourth text information based on the first confidence association matrix and the second confidence association matrix includes:

[0037] determining the rank of the first confidence association matrix as a first confidence association parameter, and

[0038] Determining a rank of the second confidence association matrix as a second confidence association parameter;

[0039] Obtaining the historical weighted average values ​​of the text fluency, the number of times to be deduplicated, and the proportion of invalid information;

[0040] determining a third confidence association parameter based on the first confidence association parameter and the historical weighted average of the text fluency, and

[0041] Determine the historical comprehensive weight corresponding to the second confidence association parameter based on the historical weight averages of the number of times to be deduplicated and the invalid information, and determine a fourth confidence association parameter based on the second confidence association parameter and the historical comprehensive weight;

[0042] The confidence of the fourth text information is determined based on the third confidence association parameter and the fourth confidence association parameter.

[0043] In one embodiment of the present application, optionally, updating the current prompt word according to a preset optimization method includes:

[0044] For each of the associated description information in the current prompt word,

[0045] Determine a replacement expression text for the acquisition instruction, and set the replacement expression text as the acquisition instruction, wherein the replacement expression text is a different expression content having the same semantics as the acquisition instruction, or

[0046] An additional description text is added to the acquisition instruction, wherein the additional description text is a subordinate description content that explains the semantics of the acquisition instruction.

[0047] In a second aspect, the embodiments of the present application provide an information extraction device, comprising:

[0048] A first text information acquisition unit, configured to acquire first text information, wherein the first text information is a search keyword for medical literature describing adverse drug reactions;

[0049] a second text information acquisition unit, configured to retrieve second text information associated with the first text information in a predetermined database in response to the acquisition of the first text information, wherein the second text information includes medical literature describing adverse drug reactions associated with the search keyword;

[0050] a third text information acquisition unit, configured to extract third text information from the second text information, wherein the third text information is text content describing the adverse drug reaction in the medical literature;

[0051] a fourth text information acquisition unit, configured to use the current prompt word and the third text information as inputs of a large language model, and output fourth text information through the large language model, wherein the current prompt word is used to reflect the need for performing structural standardization processing on the third text information;

[0052] a confidence calculation unit, configured to determine the confidence of the fourth text information based on the text features of the fourth text information, wherein the confidence is used to reflect the information quality of the fourth text information;

[0053] a prompt word optimization unit, configured to update the current prompt word according to a preset optimization method if the confidence level is lower than a predetermined threshold, and return to the step of using the current prompt word and the third text information as inputs of the large language model until the confidence level reaches the predetermined threshold;

[0054] The content output unit is used to output the current fourth text information as the target output content if the confidence reaches the predetermined threshold, wherein the predetermined threshold is the confidence level when the fourth text information can effectively describe the adverse drug reaction.

[0055] In one embodiment of the present application, optionally, the third text information acquisition unit is used to: acquire semantic association information of the first text information, wherein the semantic association information includes synonyms, near-synonyms of the search keyword and associated words of the adverse drug reaction corresponding to the search keyword; acquire hit words that match the first text information and the semantic association information in the second text information; and set the sentence containing the hit words as the third text information.

[0056] In one embodiment of the present application, optionally, the information extraction device further includes:

[0057] A prompt word acquisition unit is used to acquire the current prompt word before using the current prompt word and the third text information as inputs of the large language model, wherein the current prompt word includes acquisition instructions for each of multiple associated description information related to the adverse drug reaction, and the multiple associated description information includes: the descriptor of the adverse drug reaction, the qualifications of the descriptor, the basis of the description, the description language type, the output data structure, the name of the drug of the adverse drug reaction, the drug production information, the individual information of the drug user and the medication performance information.

[0058] In one embodiment of the present application, optionally, the text features of the fourth text information include text fluency, number of times to be deduplicated, and proportion of invalid information of each of the multiple types of associated description information.

[0059] In one embodiment of the present application, optionally, the confidence calculation unit includes:

[0060] A feature value conversion unit, used to convert the text fluency, the number of times to be deduplicated, and the proportion of invalid information of each of the plurality of associated description information into a first feature value, a second feature value, and a third feature value, respectively;

[0061] A weight acquisition unit, used to acquire weights corresponding to the text fluency, the number of times to be deduplicated, and the proportion of invalid information of each of the plurality of associated description information;

[0062] A matrix construction unit, used to determine a first confidence association matrix based on the first eigenvalues ​​of each of the multiple types of associated description information and the weights of the text fluency of each of the multiple types of associated description information, and to determine a second confidence association matrix based on the second eigenvalues ​​and third eigenvalues ​​of each of the multiple types of associated description information, and the weights of the number of times to be deduplicated and the weights of the proportion of invalid information of each of the multiple types of associated description information, wherein the first confidence association matrix is ​​used to reflect the degree of influence of the text fluency on the information quality of the fourth text information, and the second confidence association matrix is ​​used to reflect the comprehensive degree of influence of the number of times to be deduplicated and the proportion of invalid information on the information quality of the fourth text information;

[0063] An execution unit is used to determine the confidence of the fourth text information based on the first confidence association matrix and the second confidence association matrix.

[0064] In one embodiment of the present application, optionally, the information extraction device further includes:

[0065] A normalization processing unit is used to perform normalization processing on the first eigenvalue, the second eigenvalue and the third eigenvalue before determining the first confidence association matrix and determining the second confidence association matrix.

[0066] In one embodiment of the present application, optionally, the matrix construction unit is used to: for each item of the associated description information, determine the product of the first eigenvalue of the associated description information and the weight of the text fluency of the associated description information as the fourth eigenvalue corresponding to the associated description information; based on the fourth eigenvalue of each item of the associated description information, determine a first confidence association matrix, wherein the element in the mth row and nth column of the first confidence association matrix is ​​the product of the fourth eigenvalue of the mth item of the associated description information and the fourth eigenvalue of the nth item of the associated description information; for each item of the associated description information, determine the weighted average of the second eigenvalue and the third eigenvalue of the associated description information as the fifth eigenvalue corresponding to the associated description information; based on the fifth eigenvalue of each item of the associated description information, determine a second confidence association matrix, wherein the element in the jth row and kth column of the second confidence association matrix is ​​the product of the fifth eigenvalue of the jth item of the associated description information and the fifth eigenvalue of the kth item of the associated description information.

[0067] In one embodiment of the present application, optionally, the execution unit is used to: determine the rank of the first confidence association matrix as a first confidence association parameter, and determine the rank of the second confidence association matrix as a second confidence association parameter; obtain the historical weight mean of the text fluency, the number of times to be deduplicated, and the proportion of invalid information; determine a third confidence association parameter based on the first confidence association parameter and the historical weight mean of the text fluency, and determine the historical comprehensive weight corresponding to the second confidence association parameter based on the historical weight mean of the number of times to be deduplicated and the invalid information, and determine a fourth confidence association parameter based on the second confidence association parameter and the historical comprehensive weight; determine the confidence of the fourth text information based on the third confidence association parameter and the fourth confidence association parameter.

[0068] In one embodiment of the present application, optionally, the prompt word optimization unit is used to: for each acquisition indication of the associated descriptive information in the current prompt word, determine a replacement expression text for the acquisition indication, and set the replacement expression text as the acquisition indication, wherein the replacement expression text is a different expression content having the same semantics as the acquisition indication, or add additional expression text to the acquisition indication, wherein the additional expression text is a subordinate expression content that explains the semantics of the acquisition indication.

[0069] In a third aspect, an embodiment of the present application provides a computer device, comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are configured to execute the method described in the first aspect above.

[0070] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to execute the method described in the first aspect above.

[0071] The above technical solution aims at the technical problem that the drug adverse reaction information retrieval method in the related technology cannot simultaneously meet the dual requirements of retrieval convenience and retrieval accuracy. The search personnel only need to enter the search keywords describing the drug adverse reaction in the search tool to automatically perform multiple steps such as literature search, key information extraction, prompt word optimization and large language model output. In other words, after the search personnel enter the search keywords describing the drug adverse reaction in the search tool, they can directly obtain the search results output by the large language model with the optimal prompt words. Through this technical solution, the manual operation in the process of drug adverse reaction information retrieval is simplified, and the search results with standard data structure and high content quality can be obtained through extremely simple keyword input, which not only reduces the labor cost and time cost, but also improves the accuracy and practicality of the drug adverse reaction information retrieval results, which can provide effective support for drug research and clinical application, especially for the more urgent clinical treatment needs, which can reduce the time cost and the trial and error cost.

Brief Description of the Drawings

[0072] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0073] Figure 1 A flow chart of an information extraction method according to an embodiment of the present application is shown;

[0074] Figure 2 A schematic diagram of a process for retrieving adverse drug reaction information according to an embodiment of the present application is shown;

[0075] Figure 3 A block diagram of a computer device according to an embodiment of the present application is shown;

[0076] Figure 4 A block diagram of a computer device according to an embodiment of the present application is shown. [Specific implementation method]

[0077] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0078] Figure 1 A flow chart of an information extraction method according to an embodiment of the present application is shown.

[0079] like Figure 1 As shown, according to an embodiment of the present application, the information extraction method includes:

[0080] Step 102: Obtain first text information.

[0081] The first text information is a search keyword for medical literature describing adverse drug reactions.

[0082] Step 104: in response to obtaining the first text information, searching a predetermined database for second text information associated with the first text information.

[0083] The second text information includes medical literature describing adverse drug reactions associated with the search keyword.

[0084] For example, a searcher enters a search keyword of medical literature describing adverse drug reactions in a tool with a medical literature search function, and the backend of the tool can search for medical literature on adverse drug reactions associated with the search keyword in a predetermined database based on the search keyword. The number of medical literature in the search results is one or more.

[0085] Step 106: extract third text information from the second text information.

[0086] The third text information is the text content describing the adverse drug reaction in the medical literature. That is, the content related to the adverse drug reaction can be automatically extracted from the retrieved one or more medical literatures.

[0087] Step 108: Use the current prompt word and the third text information as inputs of a large language model, and output fourth text information through the large language model.

[0088] The current prompt word is used to reflect the need for structural standardization of the third text information. That is, the current prompt word is added before the extracted content related to the adverse drug reaction, and then input into the large language model, and the large language model automatically generates a result that meets the user's need for structural standardization based on the content related to the adverse drug reaction.

[0089] Step 110: Determine the confidence level of the fourth text information based on the text features of the fourth text information.

[0090] The confidence level is used to reflect the information quality of the fourth text information. The text feature of the fourth text information reflects the text performance of the fourth text information, and the confidence level is an evaluation of the text performance.

[0091] At this point, a predetermined threshold may be set as the confidence level when the fourth text information can effectively describe the adverse drug reaction. Optionally, the predetermined threshold is the minimum confidence level when the fourth text information can effectively describe the adverse drug reaction.

[0092] If the confidence level is lower than a predetermined threshold, it indicates that the information quality of the fourth text information is too low to effectively describe the adverse drug reaction. In this case, step 112 may be entered to obtain fourth text information with a higher confidence level, that is, a higher quality.

[0093] On the contrary, if the confidence reaches the predetermined threshold, it means that the information quality of the fourth text information is high enough to effectively describe the adverse drug reaction. At this time, step 114 can be entered to directly output the result.

[0094] Step 112, updating the current prompt word according to a preset optimization method, and returning to step 108 until the confidence reaches the predetermined threshold.

[0095] That is, the current prompt word can be continuously and automatically optimized until the data quality of the fourth text information obtained by using the current prompt word reaches a level that can effectively describe the adverse drug reaction.

[0096] Step 114: output the current fourth text information as target output content, wherein the predetermined threshold is the confidence level when the fourth text information can effectively describe the adverse drug reaction.

[0097] That is, if the data quality of the fourth text information obtained by using the current prompt word reaches a level that can effectively describe the adverse drug reaction, the current fourth text information can be directly output as the final search result for the adverse drug reaction.

[0098] like Figure 2 As shown, the search personnel enter the search keywords describing adverse drug reactions on the search tool page, and the search tool can automatically output the medical literature search results, and then extract the sentences related to adverse drug reactions in the medical literature, and add prompt words to the sentences. Examples of prompt words are as follows. Of course, the following is only one implementation method of the prompt words of this application. The prompt words can be set arbitrarily based on actual search needs and are not limited to the following examples.

[0099] ##Role:

[0100] Expert in extracting adverse drug reaction information;

[0101] ##Profile:

[0102] -author: Medical and health products project team;

[0103] -version:0.1;

[0104] -language:Chinese;

[0105] -description: You are an expert in extracting adverse drug reaction information, extracting adverse drug reaction information based on medical journal literature provided by users;

[0106] ##Goals:

[0107] 1. Extract items from the “Adverse Drug Reaction Report Form for Marketing Authorization Holders (Trial)”;

[0108] ##Constrains:

[0109] 1. The output is structured, with one line per item;

[0110] 2. Do not add or merge extraction items;

[0111] 3. Strictly follow the workflow;

[0112] ##Skills:

[0113] 1. Have rich professional knowledge in pharmacy;

[0114] 2. Have rich experience in filling out the "Adverse Drug Reaction Report Form for Marketing Authorization Holders (Trial)";

[0115] 3. Very familiar with the structure of medical journal literature;

[0116] ##Workflow:

[0117] 1. Extraction: name, gender, age, ethnicity / race, height, weight, medical record number / outpatient number, name of medical institution / business enterprise, previous history of adverse drug reactions and drug allergies, smoking, drinking, other allergy history, others (liver disease history, medical history, family history), disease name, etc.;

[0118] 2. Extraction: Name of the drug causing the adverse reaction;

[0119] 3. Based on the name of the drug that caused the adverse reaction, extract the following information: approval number, trade name, generic name, dosage form, specification, manufacturer, product batch number, route of administration, single dose, frequency of administration, start time of medication, end time of medication, total duration of medication, treatment of the disease, medication instructions, measures taken, adverse reaction terminology, time of occurrence, end time, duration, whether the reaction disappears or is alleviated after stopping the drug or reducing the dosage, whether the same reaction occurs again after the suspected drug is used again, etc.;

[0120] ##Initialization:

[0121] Opening remarks: Please upload journal articles.

[0122] Based on the above prompt words, the adverse drug reaction retrieval results for each medical document in the medical literature retrieval results are output, wherein an example of the adverse drug reaction information extraction result of a single document is as follows.

[0123] Literature: "Pharmaceutical care of a patient with severe adverse reactions to methotrexate in the treatment of ectopic pregnancy";

[0124] Extracted items:

[0125] Name: No specific name was provided;

[0126] Gender: Female;

[0127] Age: 20 years old

[0128] Ethnicity / race: Not provided;

[0129] Height: 160cm;

[0130] Weight: 45kg;

[0131] Medical record number / outpatient number: Not provided;

[0132] Name of medical institution / business enterprise: Anning First People's Hospital;

[0133] Previous history of adverse drug reactions and drug allergies: None;

[0134] Smoking: Not provided;

[0135] Alcohol: Not provided;

[0136] Other allergy history: None;

[0137] Others (liver disease history, medical history, family history): Not provided;

[0138] Disease name: ectopic pregnancy;

[0139] Name of drug with adverse reaction: methotrexate;

[0140] According to the name of the drug with adverse reactions, the extracted information is:

[0141] Approval number, trade name, generic name, dosage form, specification, manufacturer, product batch number: not provided;

[0142] Route of administration: intramuscular injection;

[0143] Single dose: 19 mg;

[0144] Dosing frequency: once a day (qd);

[0145] Starting time of medication: May 24, 2023;

[0146] Time of termination of medication: May 29, 2023 (9th day of hospitalization);

[0147] Total duration of medication: 6 days;

[0148] Diseases treated: ectopic pregnancy;

[0149] Instructions for use: Embryocidal treatment;

[0150] Measures to be taken: discontinue methotrexate and treat symptoms;

[0151] Adverse reaction terms: oral ulcer, liver damage, rash, bone marrow suppression;

[0152] Time of occurrence: May 28, 2023 (eighth day of hospitalization);

[0153] End time: Not provided;

[0154] Duration: Not provided;

[0155] Whether the reaction disappears or is alleviated after stopping the drug or reducing the dosage: Not provided;

[0156] Whether the same reaction will occur again if the suspected drug is used again: Not provided.

[0157] From this, we can see that the search personnel only need to enter the search keywords describing adverse drug reactions in the search tool to automatically perform multiple steps such as literature search, key information extraction, prompt word optimization, and large language model output. In other words, after the search personnel enter the search keywords describing adverse drug reactions in the search tool, they can directly obtain the search results output by the large language model with the optimal prompt words.

[0158] This technical solution simplifies manual operations in the process of adverse drug reaction information retrieval, and allows search results with standard data structure and high content quality to be obtained through minimal keyword input. This not only reduces labor and time costs, but also improves the accuracy and practicality of adverse drug reaction information retrieval results. It can provide effective support for drug research and clinical applications, especially for more urgent clinical treatment needs, and can reduce time costs and trial and error costs.

[0159] In one embodiment of the present application, step 106 includes: first obtaining semantic association information of the first text information, then obtaining a hit word that matches the first text information and the semantic association information in the second text information, and setting the sentence where the hit word is located as the third text information.

[0160] The semantic association information includes synonyms and near synonyms of the search keyword and the associated words of the adverse drug reaction corresponding to the search keyword. Thus, sentences related to adverse drug reactions can be extracted from medical literature to the greatest extent possible as basic search content in the input of the large language model.

[0161] In one embodiment of the present application, the current prompt word needs to be acquired before step 108 .

[0162] Specifically, the current prompt word refers to the acquisition instructions for each of the multiple associated descriptive information related to the adverse drug reaction, and the multiple associated descriptive information includes but is not limited to: the descriptor of the adverse drug reaction, the qualifications of the descriptor, the basis of the description, the description language type, the output data structure, the name of the drug of the adverse drug reaction, the drug production information, the individual information of the drug user and the medication performance information, and any other information related to the adverse drug reaction.

[0163] In one embodiment of the present application, the text features of the fourth text information include the text fluency, the number of duplicates to be removed, and the proportion of invalid information of each of the multiple associated description information. Among them, the higher the text fluency, the stronger the readability of the fourth text information and the higher the text quality; the fewer the number of duplicates to be removed, the less repeated content in the fourth text information, the smaller the proportion of invalid information, and the less redundancy in the fourth text information. Therefore, the fourth text information is more readable and the text quality is higher. Therefore, the text fluency, the number of duplicates to be removed, and the proportion of invalid information can be used as evaluation criteria for the content under each associated description information.

[0164] On this basis, the specific method of calculating the confidence of the fourth text information is as follows.

[0165] First, the information to be processed can be converted into numerical values ​​by information coding to facilitate subsequent calculations. Specifically, the text fluency, number of duplicates to be removed, and invalid information ratio of each of the various types of related description information can be converted into the first characteristic value, the second characteristic value, and the third characteristic value, respectively.

[0166] In addition, after encoding, the first eigenvalue, the second eigenvalue and the third eigenvalue may be normalized to convert the data under the three evaluation criteria to the same magnitude to avoid inaccurate calculation results due to different magnitudes.

[0167] Next, the weights corresponding to the text fluency, number of duplicates to be removed, and invalid information ratio of each of the multiple types of associated description information are obtained. Specifically, for each type of associated description information, its respective weight can be determined based on the text fluency, number of duplicates to be removed, and invalid information ratio in the historical search results of the associated description information, that is, the performance in the historical sample data.

[0168] Furthermore, based on the first eigenvalues ​​of each of the multiple associated description information and the weights of the text fluency of each of the multiple associated description information, a first confidence association matrix is ​​determined, and based on the second eigenvalues ​​and third eigenvalues ​​of each of the multiple associated description information, as well as the weights of the number of times to be deduplicated and the weights of the proportion of invalid information of each of the multiple associated description information, a second confidence association matrix is ​​determined.

[0169] The specific construction method of the first confidence association matrix is ​​as follows.

[0170] For each of the associated description information, the product of the first eigenvalue of the associated description information and the weight of the text fluency of the associated description information is determined as the fourth eigenvalue corresponding to the associated description information. The fourth eigenvalue reflects the influence level of the text fluency of the single associated description information on the overall text quality of the fourth text information.

[0171] Based on the fourth eigenvalue of each item of the associated description information, a first confidence association matrix is ​​determined, wherein the element in the mth row and nth column of the first confidence association matrix is ​​the product of the fourth eigenvalue of the mth item of the associated description information and the fourth eigenvalue of the nth item of the associated description information. In other words, the element in the mth row and nth column of the first confidence association matrix reflects the comprehensive influence level of the text fluency of each of the mth item of the associated description information and the nth item of the associated description information on the overall text quality of the fourth text information.

[0172] Therefore, the first confidence association matrix is ​​actually used to reflect the influence of the text fluency on the information quality of the fourth text information.

[0173] The specific construction method of the second confidence association matrix is ​​as follows.

[0174] For each of the associated description information, a weighted average of the second eigenvalue and the third eigenvalue of the associated description information is determined as the fifth eigenvalue corresponding to the associated description information. The number of times to be deduplicated and the proportion of invalid information represent the redundancy level, or the entropy value of the information, and the fifth eigenvalue reflects the influence of the redundancy level of a single associated description information on the overall text quality of the fourth text information.

[0175] Next, based on the fifth eigenvalue of each item of the associated description information, a second confidence association matrix is ​​determined, wherein the element in the jth row and the kth column of the second confidence association matrix is ​​the product of the fifth eigenvalue of the jth associated description information and the fifth eigenvalue of the kth associated description information. In other words, the element in the jth row and the kth column of the second confidence association matrix reflects the comprehensive influence level of the redundancy level of the jth associated description information and the kth associated description information on the overall text quality of the fourth text information.

[0176] Therefore, the second confidence association matrix is ​​actually used to reflect the comprehensive impact of the number of times to be deduplicated and the proportion of invalid information on the information quality of the fourth text information.

[0177] At this point, the confidence of the fourth text information may be determined based on the first confidence association matrix and the second confidence association matrix.

[0178] Specifically, the rank of the first confidence association matrix is ​​determined as the first confidence association parameter, and the rank of the second confidence association matrix is ​​determined as the second confidence association parameter, and then the historical weighted averages of the text fluency, the number of duplicates to be removed, and the invalid information ratio are obtained. The historical weighted averages mentioned here refer to the overall influence of the three text features, the text fluency, the number of duplicates to be removed, and the invalid information ratio, on the output result of the large language model.

[0179] Next, based on the first confidence association parameter and the historical weight average of the text fluency, the third confidence association parameter is determined, and based on the historical weight averages of the number of times to be deduplicated and the invalid information, the historical comprehensive weight corresponding to the second confidence association parameter is determined, and based on the second confidence association parameter and the historical comprehensive weight, the fourth confidence association parameter is determined.

[0180] Optionally, the third confidence association parameter is the product of the first confidence association parameter and the historical weighted average of the text fluency.

[0181] Optionally, the fourth confidence association parameter is the product of the second confidence association parameter and the historical comprehensive weight.

[0182] Finally, the confidence of the fourth text information is determined based on the third confidence association parameter and the fourth confidence association parameter.

[0183] Optionally, the mean of the third confidence association parameter and the fourth confidence association parameter is used as the confidence of the fourth text information, and the confidence reflects the average level of the influence of different text features of the fourth text information on the text quality of the fourth text information.

[0184] Optionally, the product of the third confidence association parameter and the fourth confidence association parameter is used as the confidence of the fourth text information. The confidence reflects the comprehensive influence of multiple text features of the fourth text information on the text quality of the fourth text information.

[0185] In one embodiment of the present application, step 112 includes: for each acquisition indication of the associated descriptive information in the current prompt word, determining a replacement expression text for the acquisition indication, and setting the replacement expression text as the acquisition indication, wherein the replacement expression text is a different expression content having the same semantics as the acquisition indication, or adding additional expression text to the acquisition indication, wherein the additional expression text is a subordinate expression content that explains the semantics of the acquisition indication.

[0186] In other words, different texts with the same semantics can be replaced, or additional text with more specific expressions can be added to improve the rationality of the prompt word, thereby increasing the accuracy of the model output result obtained through the prompt word.

[0187] Optionally, the setting of replacing the expression text and adding the expression text may be achieved through manual operation.

[0188] Optionally, a replacement expression text database and an additional expression text database may be set up, and a search tool may automatically obtain content matching the original acquisition instruction from the replacement expression text database and the additional expression text database as replacement expression text and additional expression text.

[0189] The present application embodiment provides an information extraction device, including:

[0190] A first text information acquisition unit, configured to acquire first text information, wherein the first text information is a search keyword for medical literature describing adverse drug reactions;

[0191] a second text information acquisition unit, configured to retrieve second text information associated with the first text information in a predetermined database in response to the acquisition of the first text information, wherein the second text information includes medical literature describing adverse drug reactions associated with the search keyword;

[0192] a third text information acquisition unit, configured to extract third text information from the second text information, wherein the third text information is text content describing the adverse drug reaction in the medical literature;

[0193] a fourth text information acquisition unit, configured to use the current prompt word and the third text information as inputs of a large language model, and output fourth text information through the large language model, wherein the current prompt word is used to reflect the need for performing structural standardization processing on the third text information;

[0194] a confidence calculation unit, configured to determine the confidence of the fourth text information based on the text features of the fourth text information, wherein the confidence is used to reflect the information quality of the fourth text information;

[0195] a prompt word optimization unit, configured to update the current prompt word according to a preset optimization method if the confidence level is lower than a predetermined threshold, and return to the step of using the current prompt word and the third text information as inputs of the large language model until the confidence level reaches the predetermined threshold;

[0196] The content output unit is used to output the current fourth text information as the target output content if the confidence reaches the predetermined threshold, wherein the predetermined threshold is the confidence level when the fourth text information can effectively describe the adverse drug reaction.

[0197] In one embodiment of the present application, optionally, the third text information acquisition unit is used to: acquire semantic association information of the first text information, wherein the semantic association information includes synonyms, near-synonyms of the search keyword and associated words of the adverse drug reaction corresponding to the search keyword; acquire hit words that match the first text information and the semantic association information in the second text information; and set the sentence containing the hit words as the third text information.

[0198] In one embodiment of the present application, optionally, the information extraction device further includes:

[0199] A prompt word acquisition unit is used to acquire the current prompt word before using the current prompt word and the third text information as inputs of the large language model, wherein the current prompt word includes acquisition instructions for each of multiple associated description information related to the adverse drug reaction, and the multiple associated description information includes: the descriptor of the adverse drug reaction, the qualifications of the descriptor, the basis of the description, the description language type, the output data structure, the name of the drug of the adverse drug reaction, the drug production information, the individual information of the drug user and the medication performance information.

[0200] In one embodiment of the present application, optionally, the text features of the fourth text information include text fluency, number of times to be deduplicated, and proportion of invalid information of each of the multiple types of associated description information.

[0201] In one embodiment of the present application, optionally, the confidence calculation unit includes:

[0202] A feature value conversion unit, used to convert the text fluency, the number of times to be deduplicated, and the proportion of invalid information of each of the plurality of associated description information into a first feature value, a second feature value, and a third feature value, respectively;

[0203] A weight acquisition unit, used to acquire weights corresponding to the text fluency, the number of times to be deduplicated, and the proportion of invalid information of each of the plurality of associated description information;

[0204] A matrix construction unit, used to determine a first confidence association matrix based on the first eigenvalues ​​of each of the multiple types of associated description information and the weights of the text fluency of each of the multiple types of associated description information, and to determine a second confidence association matrix based on the second eigenvalues ​​and third eigenvalues ​​of each of the multiple types of associated description information, and the weights of the number of times to be deduplicated and the weights of the proportion of invalid information of each of the multiple types of associated description information, wherein the first confidence association matrix is ​​used to reflect the degree of influence of the text fluency on the information quality of the fourth text information, and the second confidence association matrix is ​​used to reflect the comprehensive degree of influence of the number of times to be deduplicated and the proportion of invalid information on the information quality of the fourth text information;

[0205] An execution unit is used to determine the confidence of the fourth text information based on the first confidence association matrix and the second confidence association matrix.

[0206] In one embodiment of the present application, optionally, the information extraction device further includes:

[0207] A normalization processing unit is used to perform normalization processing on the first eigenvalue, the second eigenvalue and the third eigenvalue before determining the first confidence association matrix and determining the second confidence association matrix.

[0208] In one embodiment of the present application, optionally, the matrix construction unit is used to: for each item of the associated description information, determine the product of the first eigenvalue of the associated description information and the weight of the text fluency of the associated description information as the fourth eigenvalue corresponding to the associated description information; based on the fourth eigenvalue of each item of the associated description information, determine a first confidence association matrix, wherein the element in the mth row and nth column of the first confidence association matrix is ​​the product of the fourth eigenvalue of the mth item of the associated description information and the fourth eigenvalue of the nth item of the associated description information; for each item of the associated description information, determine the weighted average of the second eigenvalue and the third eigenvalue of the associated description information as the fifth eigenvalue corresponding to the associated description information; based on the fifth eigenvalue of each item of the associated description information, determine a second confidence association matrix, wherein the element in the jth row and kth column of the second confidence association matrix is ​​the product of the fifth eigenvalue of the jth item of the associated description information and the fifth eigenvalue of the kth item of the associated description information.

[0209] In one embodiment of the present application, optionally, the execution unit is used to: determine the rank of the first confidence association matrix as a first confidence association parameter, and determine the rank of the second confidence association matrix as a second confidence association parameter; obtain the historical weight mean of the text fluency, the number of times to be deduplicated, and the proportion of invalid information; determine a third confidence association parameter based on the first confidence association parameter and the historical weight mean of the text fluency, and determine the historical comprehensive weight corresponding to the second confidence association parameter based on the historical weight mean of the number of times to be deduplicated and the invalid information, and determine a fourth confidence association parameter based on the second confidence association parameter and the historical comprehensive weight; determine the confidence of the fourth text information based on the third confidence association parameter and the fourth confidence association parameter.

[0210] In one embodiment of the present application, optionally, the prompt word optimization unit is used to: for each acquisition indication of the associated descriptive information in the current prompt word, determine a replacement expression text for the acquisition indication, and set the replacement expression text as the acquisition indication, wherein the replacement expression text is a different expression content having the same semantics as the acquisition indication, or add additional expression text to the acquisition indication, wherein the additional expression text is a subordinate expression content that explains the semantics of the acquisition indication.

[0211] The information extraction device uses any of the solutions described in the above embodiments, and therefore has all the above technical effects, which will not be repeated here.

[0212] In addition, in one embodiment, the present application provides a computer device, which may be a server, and its internal structure diagram may be as follows: Figure 3 As shown. The computer device includes a processor, a memory, a network interface and a database connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile and / or volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external client via a network connection. When the computer program is executed by the processor, the method described in any of the above embodiments can be implemented.

[0213] In one embodiment, the present application further provides a computer device, which may be a client, and its internal structure diagram may be as follows: Figure 4As shown. The computer device includes a processor, a memory, a network interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The network interface of the computer device is used to communicate with an external server via a network connection. When the computer program is executed by the processor, the method described in any of the above embodiments can be implemented.

[0214] Any of the above-mentioned computer devices in the embodiments of the present application may exist in various forms, including but not limited to:

[0215] (1) Mobile communication devices: These devices are characterized by their mobile communication functions and their main purpose is to provide voice and data communications. These terminals include: smart phones (such as iPhone), multimedia phones, functional phones, and low-end phones.

[0216] (2) Ultra-mobile personal computer devices: These devices fall into the category of personal computers, have computing and processing capabilities, and generally also have mobile Internet access features. These terminals include: PDA, MID and UMPC devices, such as iPad.

[0217] (3) Portable entertainment devices: These devices can display and play multimedia content. They include audio and video players (such as iPods), handheld game consoles, e-books, as well as smart toys, wearable devices, and portable car navigation devices.

[0218] (4) Server: A device that provides computing services. The server consists of a processor, hard disk, memory, system bus, etc. The server is similar to a general computer architecture, but because it needs to provide highly reliable services, it has higher requirements in terms of processing power, stability, reliability, security, scalability, and manageability.

[0219] (5) Other electronic devices with data interaction functions.

[0220] In addition, an embodiment of the present application provides a computer-readable storage medium storing computer-executable instructions, wherein the computer-executable instructions are used to perform the following steps:

[0221] Acquire first text information, wherein the first text information is a search keyword for medical literature describing adverse drug reactions;

[0222] In response to obtaining the first text information, searching a predetermined database for second text information associated with the first text information, wherein the second text information includes medical literature describing adverse drug reactions associated with the search keyword;

[0223] Extracting third text information from the second text information, wherein the third text information is text content describing the adverse drug reaction in the medical literature;

[0224] Using the current prompt word and the third text information as inputs of the large language model, and outputting fourth text information through the large language model, wherein the current prompt word is used to reflect the need for structural standardization processing of the third text information;

[0225] Determining the confidence of the fourth text information based on the text feature of the fourth text information, wherein the confidence is used to reflect the information quality of the fourth text information;

[0226] If the confidence level is lower than a predetermined threshold, updating the current prompt word according to a preset optimization method, and returning to the step of using the current prompt word and the third text information as inputs of the large language model until the confidence level reaches the predetermined threshold;

[0227] If the confidence reaches the predetermined threshold, the current fourth text information is output as the target output content, wherein the predetermined threshold is the confidence that the fourth text information can effectively describe the adverse drug reaction.

[0228] It should be noted that the above functions or steps that can be implemented by the computer-readable storage medium or computer device can refer to the relevant description in the aforementioned method embodiment. In order to avoid repetition, they will not be described one by one here.

[0229] The above describes the technical solution of the present application in detail in combination with the accompanying drawings. Through the technical solution of the present application, the search personnel only need to enter the search keywords describing adverse drug reactions in the search tool to automatically perform multiple steps such as literature search, key information extraction, prompt word optimization, and large language model output. In other words, after the search personnel enter the search keywords describing adverse drug reactions in the search tool, they can directly obtain the search results output by the large language model through the optimal prompt words. Through this technical solution, the manual operation in the process of adverse drug reaction information retrieval is simplified, and the search results with standard data structure and high content quality can be obtained through extremely simple keyword input, which not only reduces labor costs and time costs, but also improves the accuracy and practicality of adverse drug reaction information retrieval results, which can provide effective support for drug research and clinical applications, especially for clinical treatment needs with higher urgency, which can reduce time costs and trial and error costs.

[0230] It should be understood that, although the terms first, second, etc. may be used to describe characteristic values ​​in the embodiments of the present application, these characteristic values ​​should not be limited to these terms. These terms are only used to distinguish characteristic values ​​from each other. For example, without departing from the scope of the embodiments of the present application, the first characteristic value may also be referred to as the second characteristic value, and similarly, the second characteristic value may also be referred to as the first characteristic value.

[0231] The word "if" as used herein may be interpreted as "at the time of" or "when" or "in response to determining" or "in response to detecting", depending on the context. Similarly, the phrases "if it is determined" or "if (stated condition or event) is detected" may be interpreted as "when it is determined" or "in response to determining" or "when detecting (stated condition or event)" or "in response to detecting (stated condition or event)", depending on the context.

[0232] The terms used in the embodiments of the present application are only for the purpose of describing specific embodiments, and are not intended to limit the present application. The singular forms "a", "said" and "the" used in the embodiments of the present application and the appended claims are also intended to include plural forms, unless the context clearly indicates other meanings.

[0233] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0234] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.

[0235] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to memory, storage, database or other media used in the embodiments provided in this application can include non-volatile and / or volatile memory. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM) or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. As an illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDRSDRAM), enhanced SDRAM (ESDRAM), synchronous link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM).

[0236] The embodiments described above are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. Such modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be included in the protection scope of the present invention.

Claims

1. An information extraction method, characterized in that: include: Acquire first text information, wherein the first text information is a search keyword for medical literature describing adverse drug reactions; In response to the acquisition of the first text information, searching a predetermined database for second text information associated with the first text information, wherein the second text information includes medical literature describing adverse drug reactions associated with the search keyword; Extracting third text information from the second text information, wherein the third text information is text content describing the adverse drug reaction in the medical literature; Using the current prompt word and the third text information as inputs of the large language model, and outputting fourth text information through the large language model, wherein the current prompt word is used to reflect the need for structural standardization processing of the third text information; Determining the confidence of the fourth text information based on the text feature of the fourth text information, wherein the confidence is used to reflect the information quality of the fourth text information; If the confidence level is lower than a predetermined threshold, updating the current prompt word according to a preset optimization method, and returning to the step of using the current prompt word and the third text information as inputs of the large language model until the confidence level reaches the predetermined threshold; If the confidence reaches the predetermined threshold, the current fourth text information is output as the target output content, wherein the predetermined threshold is the confidence that the fourth text information can effectively describe the adverse drug reaction.

2. The information extraction method according to claim 1, characterized in that: The extracting the third text information from the second text information includes: Acquire semantic association information of the first text information, wherein the semantic association information includes synonyms and near synonyms of the search keyword and associated words of the adverse drug reaction corresponding to the search keyword; Acquire a hit word matching the first text information and the semantic association information in the second text information; The sentence where the hit word is located is set as the third text information.

3. The information extraction method according to claim 1, characterized in that: Before using the current prompt word and the third text information as inputs of the large language model, the method further includes: Get the current prompt word, where: The current prompt word includes respective acquisition instructions for a plurality of associated descriptive information related to the adverse drug reaction, The multiple associated description information includes: the describer of the adverse drug reaction, the qualifications of the describer, the basis of description, the description language type, the output data structure, the name of the drug causing the adverse drug reaction, drug production information, individual information of the drug user and medication performance information.

4. The information extraction method according to claim 3, characterized in that: The text features of the fourth text information include the text fluency, the number of duplicates to be removed, and the proportion of invalid information of each of the multiple related description information; The determining the confidence level of the fourth text information based on the text feature of the fourth text information includes: Convert the text fluency, the number of times to be deduplicated, and the proportion of invalid information of each of the multiple types of associated description information into a first feature value, a second feature value, and a third feature value respectively; Obtaining weights corresponding to text fluency, number of times to be deduplicated, and proportion of invalid information of each of the plurality of associated description information; Based on the first eigenvalues ​​of each of the multiple types of associated description information and the weights of the text fluency of each of the multiple types of associated description information, a first confidence association matrix is ​​determined, and based on the second eigenvalues ​​and third eigenvalues ​​of each of the multiple types of associated description information, and the weights of the number of times to be deduplicated and the weights of the proportion of invalid information of each of the multiple types of associated description information, a second confidence association matrix is ​​determined, wherein the first confidence association matrix is ​​used to reflect the degree of influence of the text fluency on the information quality of the fourth text information, and the second confidence association matrix is ​​used to reflect the comprehensive degree of influence of the number of times to be deduplicated and the proportion of invalid information on the information quality of the fourth text information; The confidence of the fourth text information is determined based on the first confidence association matrix and the second confidence association matrix.

5. The information extraction method according to claim 4, characterized in that: Before determining the first confidence association matrix and determining the second confidence association matrix, the method further includes: The first eigenvalue, the second eigenvalue, and the third eigenvalue are normalized.

6. The information extraction method according to claim 4, characterized in that: The determining of a first confidence association matrix based on the first eigenvalues ​​of the plurality of associated description information and the weights of the text fluency of the plurality of associated description information comprises: For each item of the associated description information, determine a product of the first characteristic value of the associated description information and a weight of the text fluency of the associated description information as a fourth characteristic value corresponding to the associated description information; Based on the fourth eigenvalue of each item of the associated description information, determine a first confidence association matrix, wherein the element in the mth row and nth column of the first confidence association matrix is ​​the product of the fourth eigenvalue of the mth item of the associated description information and the fourth eigenvalue of the nth item of the associated description information; The determining of a second confidence association matrix based on the second eigenvalue and the third eigenvalue of each of the multiple association description information, and the weight of the number of times to be deduplicated and the weight of the proportion of invalid information of each of the multiple association description information includes: For each item of the associated description information, determine a weighted average of the second eigenvalue and the third eigenvalue of the associated description information as a fifth eigenvalue corresponding to the associated description information; Based on the fifth eigenvalue of each item of the associated description information, a second confidence association matrix is ​​determined, wherein the element in the jth row and kth column of the second confidence association matrix is ​​the product of the fifth eigenvalue of the jth item of the associated description information and the fifth eigenvalue of the kth item of the associated description information.

7. The information extraction method according to claim 4, characterized in that: The determining the confidence of the fourth text information based on the first confidence association matrix and the second confidence association matrix includes: determining the rank of the first confidence association matrix as a first confidence association parameter, and Determining a rank of the second confidence association matrix as a second confidence association parameter; Obtaining the historical weighted average values ​​of the text fluency, the number of times to be deduplicated, and the proportion of invalid information; determining a third confidence association parameter based on the first confidence association parameter and the historical weighted average of the text fluency, and Determine the historical comprehensive weight corresponding to the second confidence association parameter based on the historical weight averages of the number of times to be deduplicated and the invalid information, and determine a fourth confidence association parameter based on the second confidence association parameter and the historical comprehensive weight; The confidence of the fourth text information is determined based on the third confidence association parameter and the fourth confidence association parameter.

8. The information extraction method according to any one of claims 1 to 7, characterized in that: The updating of the current prompt word according to a preset optimization method includes: For each of the associated description information in the current prompt word, Determine a replacement expression text for the acquisition instruction, and set the replacement expression text as the acquisition instruction, wherein the replacement expression text is a different expression content having the same semantics as the acquisition instruction, or An additional description text is added to the acquisition instruction, wherein the additional description text is a subordinate description content that explains the semantics of the acquisition instruction.

9. A computer device, characterized in that: include: at least one processor; and, a memory communicatively coupled to the at least one processor; The memory stores instructions executable by the at least one processor, wherein the instructions are configured to execute the method according to any one of claims 1 to 8.

10. A computer-readable storage medium, characterized in that: Computer executable instructions are stored, and the computer executable instructions are used to execute the method according to any one of claims 1 to 8.

Images