Method, device and equipment for mining reference keyword combination and medium

Abstract

The invention discloses a method, device and equipment for mining a reference keyword combination and a medium. The method for mining the reference keyword combination comprises the steps of obtaining a plurality of keywords used for sample classification in each sample, and obtaining a plurality of first keyword combinations based on the keywords; according to the occurrence frequency of each first keyword combination in the sample, obtaining a second keyword combination with the frequency greater than a first preset threshold value from the first keyword combinations; and obtaining the use frequency of each second keyword combination in the split node of the decision tree and the contribution degree of each second keyword combination participating in sample classification, and obtaining a reference keyword combination from the second keyword combinations according to the use frequency and the contribution degree. According to the method, the coverage of the reference keyword combination is improved, automatic recognition of the text report can be realized in a scene in which the reference keyword is used for text report recognition, and the accuracy of text report recognition is improved.

Description

Technical Field

[0001] The present application relates to the technical field of report identification, and in particular to a method, device, equipment and medium for mining reference keyword combinations. Background Art

[0002] Medical examination reports are an indispensable part of medical diagnosis and treatment. They are mainly composed of two parts: examination findings and examination conclusions. Examination findings cover the phenomena of human organs observed by medical equipment, while examination conclusions include the professional evaluation given by the physician based on the findings.

[0003] Although the analysis of medical image reports has achieved a certain degree of progress through automation technology, the analysis of text reports still relies heavily on manual work. However, manual analysis is inefficient in high-load or resource-constrained medical environments and may also lead to report analysis errors. Summary of the invention

[0004] Based on the above problems, the present application provides a method, device, equipment and medium for mining reference keyword combinations to improve the efficiency of text report recognition and improve the accuracy of text report recognition.

[0005] The embodiments of the present application disclose the following technical solutions:

[0006] In a first aspect, an embodiment of the present application provides a method for mining reference keyword combinations, the method comprising:

[0007] Acquire multiple keywords for classifying the samples in each sample, and obtain multiple first keyword combinations based on the keywords; wherein the first keyword combination includes at least two keywords;

[0008] According to the frequency of occurrence of each first keyword combination in the sample, obtaining a second keyword combination whose frequency is greater than a first preset threshold from the first keyword combination;

[0009] The usage frequency of each second keyword combination in the split node of the decision tree and the contribution of each second keyword combination to sample classification are obtained, and a reference keyword combination is obtained from the second keyword combination according to the usage frequency and contribution.

[0010] Optionally, the decision tree includes a first decision tree created according to the second keyword combination or a second decision tree obtained by optimizing the first decision tree; wherein the method of optimizing the first decision tree to obtain the second decision tree is as follows:

[0011] Marking the type of each sample according to the first decision tree and the second keyword combination, and obtaining a predicted value of the type of each sample based on the marking result;

[0012] When the difference between the predicted value and the corresponding actual value is greater than a second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain a second decision tree.

[0013] Optionally, the types of samples include first-type samples and second-type samples, and the types of each sample are marked according to the first decision tree and the second keyword combination, and a predicted value of the type of each sample is obtained based on the marking result, including:

[0014] Marking each sample as a first-category sample or a second-category sample according to the first decision tree and the second keyword combination;

[0015] Obtain the first proportion and the second proportion of the first type of samples and the second type of samples in the total samples respectively;

[0016] A predicted value of the type of each sample is obtained according to the first weight and the second weight.

[0017] Optionally, obtaining multiple keywords for classifying the samples in each sample, and obtaining multiple first keyword combinations based on the keywords, includes:

[0018] Obtain multiple keywords from each sample for classifying the sample;

[0019] The frequency of occurrence of each keyword in the sample is obtained, keywords with a frequency greater than a first preset threshold are obtained, and a plurality of first keyword combinations are obtained based on the keywords.

[0020] Optionally, the method further comprises:

[0021] Normalize the keywords.

[0022] Optionally, the sample includes a first type of text report and a second type of text report; wherein the first type of text report indicates that the physical indicators of the patient corresponding to the text report are normal, and the second type of text report indicates that the physical indicators of the patient corresponding to the text report are abnormal.

[0023] In a second aspect, an embodiment of the present application provides a device for mining reference keyword combinations, the device comprising: a first acquisition module, a second acquisition module and a third acquisition module;

[0024] A first acquisition module, used to acquire a plurality of keywords used to classify the samples in each sample, and obtain a plurality of first keyword combinations based on the keywords; wherein the first keyword combination includes at least two keywords;

[0025] A second acquisition module is used to acquire, from the first keyword combinations, second keyword combinations whose frequencies are greater than a first preset threshold value according to the frequencies of occurrence of each first keyword combination in the sample;

[0026] The third acquisition module is used to obtain the usage frequency of each second keyword combination in the split node of the decision tree, and the contribution of each second keyword combination to sample classification, and obtain a reference keyword combination from the second keyword combination according to the usage frequency and contribution.

[0027] Optionally, the decision tree includes a first decision tree created by the creation module according to the second keyword combination, or a second decision tree obtained by optimizing the first decision tree by the optimization module;

[0028] The optimization module is used to mark the type of each sample according to the first decision tree and the second keyword combination, and obtain the predicted value of the type of each sample based on the marking result; when the difference between the predicted value and the corresponding actual value is greater than the second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain the second decision tree.

[0029] Optionally, the acquisition unit in the third acquisition module is specifically used to mark each sample as a first category sample or a second category sample according to the first decision tree and the second keyword combination; obtain the first proportion and the second proportion of the first category samples and the second category samples in the total samples respectively; and obtain the predicted value of the type of each sample according to the first proportion and the second proportion.

[0030] Optionally, the first acquisition module is specifically used to obtain multiple keywords in each sample for classifying the sample; obtain the frequency of occurrence of each keyword in the sample, obtain keywords with a frequency greater than a first preset threshold, and obtain multiple first keyword combinations based on the keywords.

[0031] Optionally, the device further comprises: a normalization module;

[0032] The normalization module is used to normalize keywords.

[0033] Optionally, the sample includes a first type of text report and a second type of text report; wherein the first type of text report indicates that the physical indicators of the patient corresponding to the text report are normal, and the second type of text report indicates that the physical indicators of the patient corresponding to the text report are abnormal.

[0034] In a third aspect, an embodiment of the present application provides a computer device, including a processor and a memory, the memory being used to store programs, instructions or codes, and the processor being used to execute the programs, instructions or codes in the memory to complete a method for mining reference keyword combinations as in any embodiment of the first aspect.

[0035] In a fourth aspect, an embodiment of the present application provides a computer-readable storage medium, characterized in that a computer program is stored therein, and the computer program is loaded by a processor to execute a method for mining reference keyword combinations as in any embodiment of the first aspect.

[0036] In the embodiment of the present application, keywords are first extracted from the sample data, and a first keyword combination is formed based on the keywords; then, a second keyword combination with a higher frequency is screened out through the frequency of occurrence of the first keyword combination in the sample; finally, a reference keyword combination is further screened and determined for the second keyword combination through a decision tree. In addition to the reference keyword combination described in the embodiment of the present application, the coverage of the reference keyword combination is improved through the steps of generation and optimization. In the scenario of using reference keywords for text report recognition, automatic recognition of text reports can be achieved, and the accuracy of text report recognition can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art 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 labor.

[0038] Figure 1 A flowchart of a method for mining reference keyword combinations provided in an embodiment of the present application;

[0039] Figure 2 A flowchart of optimizing a second keyword combination provided in an embodiment of the present application;

[0040] Figure 3 A flowchart of another method for mining reference keyword combinations provided in an embodiment of the present application;

[0041] Figure 4 A schematic diagram of a stop word set provided in an embodiment of the present application;

[0042] Figure 5 A schematic diagram of a punctuation mark set provided in an embodiment of the present application;

[0043] Figure 6 A schematic diagram of the structure of a device for mining reference keyword combinations provided in an embodiment of the present application;

[0044] Figure 7 A schematic diagram of the structure of a computer device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0045] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0046] The terms "first" and "second" in the specification and claims of this application are used to distinguish different objects rather than to describe a specific order of the objects. For example, the first keyword combination and the second keyword combination are used to distinguish different keyword combinations rather than to describe a specific order of the keyword combinations.

[0047] In the embodiments of the present application, words such as "exemplary" or "for example" are used to indicate examples, illustrations or descriptions. Any embodiment or design described as "exemplary" or "for example" in the embodiments of the present application should not be interpreted as being more preferred or more advantageous than other embodiments or designs. Specifically, the use of words such as "exemplary" or "for example" is intended to present related concepts in a specific way.

[0048] In the description of the embodiments of the present application, unless otherwise specified, "multiple" means two or more than two. For example, multiple processing units refer to two or more processing units, etc.; multiple elements refer to two or more elements, etc.

[0049] To facilitate understanding of the technical solution of the present application, the application scenarios of the embodiments of the present application are introduced below.

[0050] In the application scenario of text report recognition in medical examination reports, due to the presence of many abnormal expressions in text reports, the current database used for text report recognition is not suitable for text report recognition in medical examination reports, which leads to the current manual analysis of text reports. However, manual analysis is inefficient in high-load or resource-constrained medical environments and may also lead to report analysis errors.

[0051] For example, "changes after central venous catheterization", "no obvious abnormalities in the heart and diaphragm of both lungs", and "the patient presents with multiple pneumonias in the lungs, accompanied by mild pleural effusion" cannot be identified through conventional databases.

[0052] To this end, an embodiment of the present application provides a method for mining reference keyword combinations. First, multiple keywords used to classify samples in each sample are obtained, and multiple first keyword combinations are obtained based on the keywords; wherein the first keyword combination includes at least two keywords; then, based on the frequency of occurrence of each first keyword combination in the sample, a second keyword combination with a frequency greater than a first preset threshold is obtained from the first keyword combination; finally, the usage frequency of each second keyword combination in the split node of the decision tree and the contribution of each second keyword combination to the sample classification are obtained, and a reference keyword combination is obtained from the second keyword combination based on the usage frequency and the contribution; wherein the reference keyword combination is used to identify text reports.

[0053] The present application scheme will be described below in conjunction with embodiments.

[0054] See also Figure 1 , which is a flow chart of a method for mining reference keyword combinations provided in an embodiment of the present application.

[0055] like Figure 1 As shown, the method includes:

[0056] S110: Acquire multiple keywords for classifying the samples in each sample, and obtain multiple first keyword combinations based on the keywords; wherein the first keyword combination includes at least two keywords.

[0057] In a possible implementation, in the embodiments of the present application, an unsupervised keyword extraction method or a supervised keyword extraction method may be used to extract keywords from samples.

[0058] For example, for the unsupervised keyword extraction method, firstly, candidate words are extracted from the sample; then, each candidate word is scored; and finally, the topK candidate words with the highest scores are output as keywords. Among them, the candidate words can be scored according to different algorithms, such as the term frequency-inverse document frequency (TF-IDF) algorithm, the TextRank algorithm, and the linear discriminant analysis (LDA) algorithm.

[0059] For example, for the supervised keyword extraction method, candidate words are first extracted from the sample; and then the candidate words are classified by the trained keyword extraction classifier. It should be understood that the training process of the keyword extraction classifier is a mature technical means and will not be described in detail here.

[0060] In a possible implementation, the samples described in the embodiment of the present application include a first type of text report (white sample) and a second type of text report (black sample). The first type of text report (white sample) indicates that the patient's physical indicators are normal, and the second type of text report (black sample) indicates that the patient's physical indicators are abnormal.

[0061] Exemplarily, the first type of text report (white sample) includes "changes after central venous catheterization", "no obvious abnormalities in both lungs, heart and diaphragm", "no dislocation of both knees and patellae", and "patent colorectal intestinal cavity, smooth mucosa, and clear vascular network". Correspondingly, the keywords in the first type of text report (white sample) can be "central venous catheterization", "postoperative", "changes", "both lungs, heart and diaphragm", "nothing", "obvious abnormalities", "both knees and patellae", "dislocation", "what was found", "colorectal", "intestinal cavity", "patent", "," "mucosa", "smooth", "vascular network" and "clear", etc.

[0062] Exemplarily, in the first type of text report (black sample), "The patient presented with symptoms of intermittent claudication and lower limb ischemia. Blood flow Doppler examination suggested possible peripheral arterial disease. It is recommended to perform lower limb arteriovenography to further confirm the diagnosis." and "The patient presented with multiple pneumonias in the lungs, accompanied by mild pleural effusion." Correspondingly, the keywords in the second type of text report (black sample) may be "patient", "presentation", "intermittent claudication", "and", "lower limb ischemia", "of", "symptoms", ".", "blood flow", "Doppler examination", "suggests", "for", "peripheral arterial disease", ".", "recommendation", "perform", "lower limb arteriovenography", "to", "further", "determine", "diagnosis", ".", "patient", "presentation", "multiple pneumonias in the lungs", ",","accompanied by", "mild", "pleural effusion".

[0063] In the embodiment of the present application, the keyword rules extracted from the black sample and the white sample are used in combination to enhance the recognition accuracy of the abnormal inspection report. In particular, in the recognition of implicit abnormal conclusions, it is possible to use the complex keyword combinations in the black sample, which are often impossible to capture through a single model or a fixed abnormal keyword network database.

[0064] Based on the keywords obtained in the aforementioned embodiments, in the embodiment of the present application, the keywords in the first type of text report (white sample) are combined to obtain the first keyword combination, and the keywords in the second type of text report (black sample) are combined to obtain the first keyword combination.

[0065] Exemplarily, the first keyword combination may be in the following combination forms:

[0066] The first keyword combination 1: {“multiple inflammations of the lungs”, “pleural effusion”};

[0067] First keyword combination 2: {“multiple inflammations of the lungs”, “mild”};

[0068] First keyword combination 3: {“mild”, “pleural effusion”};

[0069] The first keyword combination 4; {"multiple inflammations of the lungs", "mild", "pleural effusion"}.

[0070] S120: According to the frequencies of occurrence of each first keyword combination in the sample, obtaining second keyword combinations whose frequencies are greater than a first preset threshold from the first keyword combinations.

[0071] Among them, the frequency of the first keyword combination appearing in the sample can be expressed as the support of the first keyword; the first threshold described in the embodiment of the present application can be a minimum support threshold. If the support of the first keyword combination is greater than the minimum support threshold, it means that the frequency of the first keyword combination appearing in all samples (including white samples and black samples) is relatively high. It should be understood that the minimum support threshold described in the embodiment of the present application can be set or adjusted according to the needs of the user.

[0072] For ease of understanding, this step will be described below in conjunction with specific numerical values.

[0073] For example, if the first preset threshold (minimum support threshold) is 5%, the frequency (support) of the first keyword combination 1 is 10%, the frequency of the first keyword combination 2 is 3%, the frequency of the first keyword combination 3 is 12%, and the frequency of the first keyword combination 4 is 4%. Then the second keyword combination is {"multiple inflammations of the lungs", "mild"} and {"mild", "pleural effusion"}.

[0074] In addition, the present application also provides a support calculation formula, which is as follows:

[0075] Support(X) = (number of times the first keyword combination X appears in all text reports) / (total number of all text reports)

[0076] S130: Obtain the usage frequency of each second keyword combination in the split node of the decision tree and the contribution of each second keyword combination to sample classification, and obtain a reference keyword combination from the second keyword combination according to the usage frequency and contribution.

[0077] In one possible implementation, a first decision tree is created based on the second keyword combinations, the usage frequency of each second keyword combination in the split node of the first decision tree and the contribution of each second keyword combination to sample classification are obtained, and a reference keyword combination is obtained from the second keyword combinations based on the usage frequency and contribution.

[0078] In another possible implementation, the type of each sample can be labeled according to the first decision tree and the second keyword combination, and the prediction value of each type can be obtained based on the labeling result; when the difference between the predicted value and the corresponding actual value is greater than the second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain the second decision tree; the usage frequency of each second keyword combination in the split node of the second decision tree and the contribution of each second keyword combination to the sample classification are obtained, and a reference keyword combination is obtained from the second keyword combination according to the usage frequency and the contribution.

[0079] See also Figure 2 , which is a flow chart of a method for optimizing the second keyword combination provided in an embodiment of the present application.

[0080] like Figure 2 As shown, the method includes:

[0081] S1301: Create a first decision tree according to the second keyword combination.

[0082] Exemplarily, the second keyword combination is as follows:

[0083] The second keyword combination 1 {“intermittent claudication”, “lower limb ischemia”, “Doppler examination”, “peripheral arterial disease”, “lower limb angiography”};

[0084] The second keyword combination 2 {“multiple inflammations of the lungs”, “associated with”, “mild”, “pleural effusion”};

[0085] The second keyword combination 3 {“central venous catheter”, “postoperative”, “change”};

[0086] The second keyword combination 4 {“bilateral lungs, heart and diaphragm”, “not seen”, “obvious abnormality”};

[0087] The second keyword combination 5 {“bilateral patella”, “not seen”, “dislocation”};

[0088] The second keyword combination 6 {“colorectal”, “patent”, “mucosa”, “smooth”, “vascular network”, “clear”}.

[0089] Based on the above second keyword combination, the training data described in the embodiment of the present application is shown in the following Table 1:

[0090] Table 1

[0091]

[0092] The embodiment of the present application creates a first decision tree through the above training data.

[0093] S1302: Mark the type of each sample according to the first decision tree and the second keyword combination, and obtain a predicted value of the type of each sample based on the marking result.

[0094] Using the log odds statistic as the initial prediction, the steps are as follows:

[0095] 1) Calculate the probability that the classification is normal:

[0096] In the training data, the number of reports marked as normal is counted, set to p, and the total number of reports is N. The probability of the positive category and the probability of the negative category are expressed as follows:

[0097] P(positive category) = p / N = 4 / 6

[0098] P(negative category)=(Np) / N=2 / 6

[0099] 2) Calculate the log odds:

[0100] The logarithmic probability is a way to express the probability of the positive class relative to the negative class. It is the natural logarithm of the ratio of the probability of the positive class to the probability of the negative class. The logarithmic probability helps to map the probability to an infinite continuous space, which is beneficial for the subsequent gradient boosting algorithm. The logarithmic probability is identified as follows:

[0101] Log odds ln(odds) = ln(P(positive category) / P(negative category)) = ln(4 / 2)≈0.7

[0102] 3) Get the initial prediction value:

[0103] The initial prediction value is the starting point of the gradient boosting process. It represents the initial judgment of classifying any given inspection report as abnormal without any further information. The initial prediction value γ is expressed as follows:

[0104]

[0105] For ease of understanding, the relationship between the predicted results and the actual results of the first decision tree is shown in Table 2 below:

[0106] Table 2

[0107]

[0108] It should be noted that in the above Table 2, the true value of the abnormal text report is defined as "0", and the true value of the normal text report is defined as "1".

[0109] S1303: When the difference between the predicted value and the corresponding actual value is greater than a second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain a second decision tree.

[0110] In the embodiment of the present application, by optimizing the first decision tree, the difference between each predicted value and the corresponding actual value can be reduced.

[0111] It should be understood that in the embodiment of the present application, whether it is necessary to optimize the second keyword combination by optimizing the first decision tree is determined by the relationship between the residual (the difference between the predicted value and the corresponding actual value) and the second preset threshold. The second preset threshold can be set or modified according to the needs of the user.

[0112] As an example, the optimization of the first decision tree will be introduced in combination with the following formula:

[0113]

[0114] Among them, Residuali represents the current residual, Residual i+1 represents the next residual, and PreviousProobability i is the predicted value. Taking the examination report features "intermittent claudication", "lower limb ischemia", "Doppler examination", "peripheral arterial disease", and "lower limb angiography" in Table 2 above, when i=1, its residual Residual i=-0.7, PreviousProobability i=0.7, then Residual i+1≈-0.59; according to the residual value (Residual i+1≈-0.59), the first decision tree is optimized, and so on, until the residual value (absolute value) is less than the second preset threshold.

[0115] It should be noted that when the residual value of the inspection report feature is less than 0, the absolute value of the residual value must be less than the preset threshold.

[0116] S1304: Obtain the usage frequency of each second keyword combination in the split node of the first decision tree or the second decision tree, and the contribution of each second keyword combination to sample classification, and obtain a reference keyword combination from the second keyword combination according to the usage frequency and contribution.

[0117] The text report is identified by the obtained reference keyword combination.

[0118] Exemplarily, the keyword combination in the target text report is matched with the reference keyword combination. If the keyword combination in the black sample is matched and the match is successful, it means that there is an abnormality in the conclusion; if the keyword combination in the black sample is matched and the match is unsuccessful, it means that there is no abnormality in the conclusion.

[0119] In addition, reverse verification can also be introduced in the embodiments of the present application. When the conclusion indicates that the text reports an abnormality, the keyword rule matching in the white sample is used for reverse verification. If the verification is not normal, the output is abnormal. When the conclusion indicates that the text reports a normal state, the keyword rule matching in the white sample is used for reverse verification. If the verification is normal, the output is normal.

[0120] Use the keyword rule matching in the white sample for reverse verification. If the verification is normal, it will be marked as "pending review". Such a report will be submitted to clinical experts for final manual review. The system will automatically use these annotations as new training samples to further iterate and optimize the keyword rule set. In this embodiment of the present application, use the keyword rule matching in the white sample for reverse verification. If the verification is not normal, it will be marked as "pending review". Such a report will be submitted to clinical experts for final manual review. The system will automatically use these annotations as new training samples to further iterate and optimize the keyword rule set.

[0121] In an embodiment of the present application, by introducing reverse verification, when a report is initially marked as abnormal, the system will use another set of rules to perform a secondary check, thereby reducing misjudgments and ensuring that each report marked as abnormal has undergone a more rigorous confirmation process.

[0122] In the embodiment of the present application, keywords are first extracted from the sample data, and a first keyword combination is formed based on the keywords; then, a second keyword combination with a higher frequency is screened out through the frequency of occurrence of the first keyword combination in the sample; finally, a reference keyword combination is further screened and determined for the second keyword combination through a decision tree. In addition to the reference keyword combination described in the embodiment of the present application, the coverage of the reference keyword combination is improved through the steps of generation and optimization. In the scenario of using reference keywords for text report recognition, automatic recognition of text reports can be achieved, and the accuracy of text report recognition can be improved.

[0123] See also Figure 3 , which is a flowchart of another method for mining reference keyword combinations provided in an embodiment of the present application.

[0124] like Figure 3 As shown, the method includes:

[0125] S310: Obtain multiple keywords in each sample for classifying the sample.

[0126] S320: Clean the keywords and perform normalization processing on the keywords.

[0127] For example, keywords are cleaned by the following formula:

[0128] [\text{cleaned text}(d)=\text{word segmentation}(d)-\text{stop word set}-\text{punctuation set}]

[0129] Among them, the stop word set and punctuation mark set are respectively as follows Figure 4 Neutralization Figure 5 shown.

[0130] like Figure 4 As shown, the stop word set includes but is not limited to the following: "it can be seen that", "of", "indeed", "words", "relatively speaking", "save", "look", "blink", "watch", "watch", "already", "already", "already", "already", "leave", "compete", "first" and "etc.

[0131] like Figure 5 As shown, the punctuation mark set includes but is not limited to the following: “?”, “、”, “.”, “””, “《》”, “!”, “;” and “:” punctuation marks.

[0132] In addition, the embodiment of the present application also normalizes the keywords to improve the standardization of the keywords and removes words related to the entity during the normalization process, such as: diseases, symptoms, and replacement related words.

[0133] For example, "peripheral arterial disease" can be normalized to "peripheral arterial disease"; "lower limb arteriovenography" can be normalized to "lower limb angiography" and "multiple inflammations of the lungs" can be normalized to "multiple inflammations of the lungs", etc.

[0134] The embodiment of the present application improves the semantic accuracy of keyword matching through normalization processing technology, thereby improving the accuracy of text report recognition.

[0135] S330: Obtain the frequency of occurrence of each keyword in the sample, obtain keywords whose frequency is greater than the first preset threshold, and obtain multiple first keyword combinations based on the keywords.

[0136] In the embodiment of the present application, by eliminating keywords whose frequency is less than the first preset threshold, the subsequent process of optimizing keyword combinations can be reduced, thereby improving efficiency. In addition, it is also beneficial to improve the accuracy of text report recognition.

[0137] S340: According to the frequencies of occurrence of each first keyword combination in the sample, obtaining second keyword combinations whose frequencies are greater than a first preset threshold from the first keyword combinations.

[0138] S350: Obtain the usage frequency of each second keyword combination in the split node of the decision tree, as well as the contribution of each second keyword combination to sample classification, and obtain a reference keyword combination from the second keyword combination based on the usage frequency and contribution; wherein the reference keyword combination is used to identify the text report.

[0139] In an embodiment of the present application, before generating the first keyword combination, a series of processing is performed on the keywords, such as cleaning, normalization, and screening, which improves the semantic accuracy of keyword matching and thereby improves the accuracy of text report recognition.

[0140] In addition, in an embodiment of the present application, for the same patient based on the same examination name, and after review, it is proved that the first examination report of this patient is abnormal and can be automatically marked as a black sample; when doctors or clinical staff mark a report that is not correctly identified during use, these annotations are automatically used as new training samples, and black samples and white samples are automatically marked, and the remaining samples are automatically identified and marked based on existing recognition methods.

[0141] Through the above method, combined with the automated learning mechanism, human intervention can be effectively reduced and automatic updating and optimization of keyword rules can be achieved.

[0142] See also Figure 6 , the figure is a device for mining reference keyword combinations provided by an embodiment of the present application, the device comprising: a first acquisition module 610, a second acquisition module 620 and a third acquisition module 630;

[0143] A first acquisition module 610 is used to acquire a plurality of keywords used to classify the samples in each sample, and obtain a plurality of first keyword combinations based on the keywords; wherein the first keyword combination includes at least two keywords;

[0144] A second acquisition module 620 is used to acquire, from the first keyword combinations, second keyword combinations whose frequencies are greater than a first preset threshold value according to the frequencies of occurrence of each first keyword combination in the sample;

[0145] The third acquisition module 630 is used to obtain the usage frequency of each second keyword combination in the split node of the decision tree, and the contribution of each second keyword combination to sample classification, and obtain a reference keyword combination from the second keyword combination according to the usage frequency and contribution.

[0146] Optionally, the decision tree includes a first decision tree created by the creation module according to the second keyword combination, or a second decision tree obtained by optimizing the first decision tree by the optimization module;

[0147] The optimization module is used to mark the type of each sample according to the first decision tree and the second keyword combination, and obtain the predicted value of the type of each sample based on the marking result; when the difference between the predicted value and the corresponding actual value is greater than the second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain the second decision tree.

[0148] Optionally, the acquisition unit 631 in the third acquisition module 630 is specifically used to mark each sample as a first category sample or a second category sample according to the first decision tree and the second keyword combination; obtain the first proportion and the second proportion of the first category samples and the second category samples in the total samples respectively; and obtain the predicted value of the type of each sample according to the first proportion and the second proportion.

[0149] Optionally, the first acquisition module 610 is specifically used to obtain multiple keywords in each sample for classifying the sample; obtain the frequency of occurrence of each keyword in the sample, obtain keywords with a frequency greater than a first preset threshold, and obtain multiple first keyword combinations based on the keywords.

[0150] Optionally, the device further comprises: a normalization module;

[0151] The normalization module is used to perform normalization processing on keywords.

[0152] Optionally, the sample includes a first type of text report and a second type of text report; wherein the first type of text report indicates that the physical indicators of the patient corresponding to the text report are normal, and the second type of text report indicates that the physical indicators of the patient corresponding to the text report are abnormal.

[0153] In addition, an embodiment of the present application provides a computer device, including a processor and a memory, wherein the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to complete the method of mining reference keyword combinations as described in any embodiment of the first aspect.

[0154] In addition, an embodiment of the present application provides a computer-readable storage medium, characterized in that a computer program is stored therein, and the computer program is loaded by a processor to execute the method for mining reference keyword combinations as described in any embodiment of the first aspect.

[0155] In practical applications, the computer-readable storage medium can adopt any combination of one or more computer-readable media. The computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. The computer-readable storage medium can be, for example, but not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples (non-exhaustive list) of computer-readable storage media include: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present embodiment, the computer-readable storage medium can be any tangible medium containing or storing a program, which can be used by an instruction execution system, device or device or used in combination with it.

[0156] Computer-readable signal media may include data signals propagated in baseband or as part of a carrier wave, which carry computer-readable program code. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than a computer-readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0157] The program code embodied on the computer readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wireline, optical fiber cable, RF, etc., or any suitable combination of the foregoing.

[0158] Computer program code for performing the operations of the present invention may be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user's computer, partially on the user's computer, as a separate software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In the case of a remote computer, the remote computer may be connected to the user's computer through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0159] like Figure 7As shown, a schematic diagram of the structure of a computer device provided in an embodiment of the present application. Figure 7 The computer device 12 shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0160] like Figure 7 As shown, the computer device 12 is in the form of a general-purpose computing device. The components of the computer device 12 may include, but are not limited to: one or more processors or processing units 16, a system memory 28, and a bus 18 that connects various system components (including the system memory 28 and the processing unit 16).

[0161] Bus 18 represents one or more of several types of bus structures, including a memory bus or memory controller, a peripheral bus, an accelerated graphics port, a processor or a local bus using any of a variety of bus architectures. By way of example, these architectures include, but are not limited to, an Industry Standard Architecture (ISA) bus, a Micro Channel Architecture (MAC) bus, an Enhanced ISA bus, a Video Electronics Standards Association (VESA) local bus, and a Peripheral Component Interconnect (PCI) bus.

[0162] The computer device 12 typically includes a variety of computer system readable media. These media can be any available media that can be accessed by the computer device 12, including volatile and non-volatile media, removable and non-removable media.

[0163] The system memory 28 may include computer system readable media in the form of volatile memory, such as random access memory (RAM) 30 and / or cache memory 32. The computer device 12 may further include other removable / non-removable, volatile / non-volatile computer system storage media. By way of example only, the storage system 34 may be used to read and write non-removable, non-volatile magnetic media ( Figure 7 not shown, usually called a "hard drive"). Although Figure 7 Not shown in the figure, a disk drive for reading and writing to a removable non-volatile disk (e.g., a "floppy disk"), and an optical disk drive for reading and writing to a removable non-volatile optical disk (e.g., a CD-ROM, a DVD-ROM, or other optical media) may be provided. In these cases, each drive may be connected to the bus 18 via one or more data medium interfaces. The memory 28 may include at least one program product having a set (e.g., at least one) of program modules that are configured to perform the functions of the various embodiments of the present invention.

[0164] A program / utility 40 having a set (at least one) of program modules 42 may be stored, for example, in the memory 28, such program modules 42 including, but not limited to, an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment. The program modules 42 generally perform the functions and / or methods of the embodiments described herein.

[0165] The computer device 12 may also communicate with one or more external devices 14 (e.g., keyboards, pointing devices, displays 24, etc.), one or more devices that enable a user to interact with the computer device 12, and / or any device that enables the computer device 12 to communicate with one or more other computing devices (e.g., network cards, modems, etc.). Such communication may be performed through an input / output (I / O) interface 22. Furthermore, the computer device 12 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) through a network adapter 20. Figure 7 As shown, the network adapter 20 communicates with other modules of the computer device 12 via the bus 18. It should be understood that although Figure 7 Not shown, other hardware and / or software modules may be used in conjunction with computer device 12, including but not limited to microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0166] The processor unit 16 executes various functional applications and data processing by running the programs stored in the system memory 28 to implement the method for mining reference keyword combinations provided in the embodiment of the present application.

[0167] It should be noted that each embodiment in this specification is described in a progressive manner, and the same or similar parts between the embodiments can be referred to each other, and each embodiment focuses on the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the partial description of the method embodiments. The device and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components indicated as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0168] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed in the present application should be included in the protection scope of the present application. Therefore, the protection scope of the present application should be based on the protection scope of the claims.

Claims

1. A method for mining reference keyword combinations, characterized in that: The method comprises: Acquire multiple keywords for classifying the samples in each sample, and obtain multiple first keyword combinations based on the keywords; wherein the first keyword combination includes at least two of the keywords; According to the frequency of occurrence of each of the first keyword combinations in the sample, obtaining a second keyword combination whose frequency is greater than a first preset threshold from the first keyword combination; The usage frequency of each second keyword combination in the split node of the decision tree and the contribution of each second keyword combination to the sample classification are obtained, and a reference keyword combination is obtained from the second keyword combination according to the usage frequency and the contribution.

2. The method according to claim 1, characterized in that The decision tree includes a first decision tree created according to the second keyword combination, or a second decision tree obtained by optimizing the first decision tree; wherein the second decision tree is obtained by optimizing the first decision tree as follows: Marking the type of each of the samples according to the first decision tree and the second keyword combination, and obtaining a predicted value of the type of each of the samples based on the marking result; When the difference between the predicted value and the corresponding actual value is greater than a second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain the second decision tree.

3. The method according to claim 2, characterized in that The types of the samples include first-type samples and second-type samples, and the marking of the types of the samples according to the first decision tree and the second keyword combination, and obtaining a predicted value of the type of the samples based on the marking result, include: Marking each of the samples as the first category of samples or the second category of samples according to the first decision tree and the second keyword combination; Respectively obtaining a first proportion and a second proportion of the first type of samples and the second type of samples in the total samples; A predicted value of the type of each of the samples is obtained according to the first weight and the second weight.

4. The method according to any one of claims 1 to 3, characterized in that: The obtaining of a plurality of keywords for classifying the samples in each sample, and obtaining a plurality of first keyword combinations based on the keywords, includes: Obtaining multiple keywords from each sample for classifying the sample; The occurrence frequency of each of the keywords in the sample is obtained, the keywords whose frequency is greater than the first preset threshold are obtained, and a plurality of first keyword combinations are obtained based on the keywords.

5. The method according to any one of claims 1 to 3, characterized in that: The method further comprises: The keywords are normalized and cleaned.

6. The method according to any one of claims 1 to 3, characterized in that: The sample includes a first type of text report and a second type of text report; wherein, the first type of text report indicates that the physical indicators of the patient corresponding to the text report are normal, and the second type of text report indicates that the physical indicators of the patient corresponding to the text report are abnormal.

7. A device for mining reference keyword combinations, characterized in that: The device comprises: a first acquisition module, a second acquisition module and a third acquisition module; The first acquisition module is used to acquire a plurality of keywords used to classify the samples in each sample, and obtain a plurality of first keyword combinations based on the keywords; wherein the first keyword combination includes at least two of the keywords; The second acquisition module is used to acquire, from the first keyword combinations, second keyword combinations whose frequencies are greater than a first preset threshold according to the frequencies of occurrence of each of the first keyword combinations in the samples; The third acquisition module is used to obtain the usage frequency of each second keyword combination in the split node of the decision tree, and the contribution of each second keyword combination to the sample classification, and obtain a reference keyword combination from the second keyword combination based on the usage frequency and the contribution.

8. The device according to claim 7, characterized in that The decision tree includes a first decision tree created by a creation module according to the second keyword combination, or a second decision tree obtained by optimizing the first decision tree by an optimization module; The optimization module is used to mark the type of each sample according to the first decision tree and the second keyword combination, and obtain a predicted value of the type of each sample based on the marking result; When the difference between the predicted value and the corresponding actual value is greater than a second preset threshold, the first decision tree is optimized based on the difference between each predicted value and the corresponding actual value to obtain a second decision tree.

9. A computer device, characterized in that: It includes a processor and a memory, wherein the memory is used to store programs, instructions or codes, and the processor is used to execute the programs, instructions or codes in the memory to complete the method for mining reference keyword combinations as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that: A computer program is stored, and the computer program is loaded by a processor to execute the method for mining reference keyword combinations as described in any one of claims 1 to 6.

Images