A method for processing target text

Abstract

The present invention relates to the field of electronic digital data processing technology, and in particular to a method for processing a target text. The method comprises: obtaining a target text A, a i is the i-th target text; get the label B of the target text, b in B i is a i Tags, b i Medium i,j is a i The jth label of A; Initialize the first variable k to 1; Determine whether the kth label of the target text can be used to determine the accuracy of A; If it cannot be used to determine the accuracy of A and k

Description

Technical Field

[0001] The present invention relates to the technical field of electronic digital data processing, and in particular to a method for processing a target text. Background Art

[0002] When the initial text stored in the database is processed into the target text for storage, the accuracy of the target text must be judged, that is, the consistency between the target text and the initial text must be judged. The higher the consistency between the target text and the initial text, the higher the accuracy of the target text; the lower the consistency between the target text and the initial text, the lower the accuracy of the target text. However, the number of target texts is often large. If the accuracy of the target text is judged by obtaining the accuracy of all target texts, the corresponding workload is too large. How to balance the accuracy of the target text obtained and the workload of the process of obtaining the accuracy of the target text is an urgent problem to be solved. Summary of the invention

[0003] The object of the present invention is to provide a method for processing a target text, which is used to reduce the workload of the process of obtaining the accuracy of the target text as much as possible under the premise of ensuring the accuracy of the obtained target text.

[0004] According to the present invention, a method for processing a target text is characterized by comprising the following steps:

[0005] S100, obtaining target text A, A={a 1 ,a 2 ,…,a i ,…,a n}, a i is the i-th target text, the value range of i is 1 to n, and n is the number of target texts; the target text is obtained by processing the initial text.

[0006] S200, obtain the label B of the target text, B={b 1 ,b 2 ,…,b i ,…,b n}, b i is a i Tags, b i =(b i,1 ,b i,2 ,…,b i,j ,…,b i,m ), b i,j is a i The jth label of b i,j has a higher priority than b i,j+1 The priority of j is in the range of 1 to m, where m is the number of preset tags.

[0007] S300, initialize the first variable k to 1.

[0008] S400, determine whether the k-th tag of the target text can be used to judge the accuracy rate of A.

[0009] S500, if the k-th tag of the target text cannot be used to judge the accuracy rate of A and k < m, then k = k + 1, repeat S400 until the k-th tag of the target text can be used to judge the accuracy rate of A or k = m.

[0010] S600, if the k-th tag of the target text can be used to judge the accuracy rate of A, then extract the target text from the n target texts according to the k-th tag of the target text to judge the accuracy rate of A.

[0011] The present invention has at least the following beneficial effects:

[0012] In the present invention, the number of target texts is n, and each of the n target texts corresponds to m tags. Among them, the priority of the first tag is higher than that of the second tag, the priority of the second tag is higher than that of the third tag, and so on; the present invention first judges whether the first tag can be used to judge the accuracy rate of the n target texts. If so, the accuracy rate of the n target texts is judged according to the first tag; if not, it is judged whether the second tag can be used to judge the accuracy rate of the n target texts; if the second tag can, the accuracy rate of the n target texts is judged according to the second tag; if not, it is judged whether the third tag can be used to judge the accuracy rate of the n target texts; and so on, until one of the tags can be used to judge the accuracy rate of the n target texts or none of the m tags can be used to judge the accuracy rate of the n target texts. In the case where one of the tags can be used to judge the accuracy rate of the n target texts, the target text is extracted from the n target texts according to this tag to judge the accuracy rate of the n target texts. Since the number of extractions is less than n, the present invention reduces the workload of the process of obtaining the accuracy rate of the target text; and since the present invention extracts the target text according to the tag that can be used to judge the accuracy rate of the n target texts, the present invention can also ensure the accuracy of determining the accuracy rate of the n target texts based on the extracted target text, thereby achieving the purpose of taking into account the accuracy of the obtained accuracy rate of the target text and the workload of the process of obtaining the accuracy rate of the target text. Description of the Drawings

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0014] Figure 1 A flowchart of a method for processing a target text provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0015] 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 only 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 those skilled in the art without creative work are within the scope of protection of the present invention.

[0016] According to the present invention, a method for processing a target text is provided, such as Figure 1 As shown, including:

[0017] S100, obtaining target text A, A={a 1 ,a 2 ,…,a i ,…,a n}, a i is the i-th target text, the value range of i is 1 to n, and n is the number of target texts; the target text is obtained by processing the initial text.

[0018] In this embodiment, the target text and the initial text are both known texts, and the target text is obtained by processing the initial text. The purpose of this embodiment includes determining the accuracy of the target text, that is, determining the consistency between the target text and the initial text.

[0019] As a specific implementation, the initial text is the patient's medical data stored in the first database, and the target text is the patient's medical data stored in the second database. The patient's medical data stored in the second database is obtained by parsing the patient's medical data stored in the first database. Optionally, the parsing process includes natural language processing (NLP) and regularization processing.

[0020] S200, obtain the label B of the target text, B={b 1 ,b 2 ,…,b i ,…,b n}, b i is a i Tags, b i =(b i,1 ,b i,2 ,…,b i,j ,…,b i,m ), b i,j is a i The jth label of b i,jhas a higher priority than b i,j+1 The priority of j is in the range of 1 to m, where m is the number of preset tags.

[0021] Optional, b i,j The priority of b is the manually preset priority. i,j The higher the priority of b i,j In b i The closer the position is to the front.

[0022] As a specific implementation, the target text is the patient's medical data, and the preset labels include the patient's gender, department and medical time, etc. The medical data includes the patient's name, age, department, medical time, medical medication information and examination information, etc.

[0023] S300: Initialize a first variable k to 1.

[0024] S400, determining whether the kth label of the target text can be used to determine the accuracy of A.

[0025] Optionally, S400 includes:

[0026] S410, extract q according to the kth label of the target text 1 The target texts are appended to the first text set H k,1 ,q 1 is the preset first extraction quantity, H k,1 is initialized to an empty set.

[0027] Optional, q 1 is the experience value, or q 1 =z 2 ×(0.5) 2 / (w 1 ) 2 , z is the quantile corresponding to the preset confidence level, w 1 is the preset first sampling error; when the preset confidence level is 95%, z = 1.96; when w 1 =10%, q 1 =96.

[0028] Optionally, S410 includes:

[0029] S411, divide the n target texts according to the kth label of the target text to obtain a division result G, G = {g 1 ,g 2 ,…,g x ,…,g y}, g x is the xth type of target text obtained by division, the value range of x is 1 to y, and y is the number of classes of the target text obtained by division; different gx The k-th label corresponding to the target text is different, and the same g x The k-th label corresponding to the target text in is the same.

[0030] As a specific implementation method, the target text is the patient's medical data, and the kth label is the department. Then, after dividing the n target texts, each category of target text obtained belongs to the target text of a department, different categories of target texts belong to the target texts of different departments, and the same category of target texts belongs to the target text of the same department.

[0031] S412, traverse G, get g x The corresponding number of extractions f x , f x =round(q 1 ×qua x / n),qua x g x The number of target texts included. round() is used to round the value to an integer.

[0032] S413, traverse G, from g x Extract f x Append the target text to H k,1 .

[0033] Those skilled in the art know that any extraction method in the prior art falls within the protection scope of the present invention. Optionally, the extraction method is random extraction or systematic extraction.

[0034] S414, if H k,1 The number of target texts included is q 1 , then H k,1 Determined as the final H k,1 ; if H k,1 The number of target texts included is greater than q 1 , then from H k,1 Random deletion (∑ y x＝1 f x )-q 1 target text, and delete the H k,1 Determined as the final H k,1 ; if H k,1 The amount of target text included is less than q 1 , then randomly select q from the first remaining target text 1 -(∑ y x＝1 f x ) target text is appended to H k,1 , and the added H k,1 Determined as the final Hk,1 ; The first remaining target text is the other target texts among the n target texts except the extracted target text.

[0035] In this embodiment, the first remaining target text is AH k,1 .

[0036] S420, obtain H k,1 The corresponding target text error rate l k,1 , l k,1 By comparing H k,1 The corresponding target text and initial text are obtained.

[0037] In this embodiment, each a i Includes several fields, l k,1 =u / u', u is H k,1 The total number of error fields in the corresponding target text, u' is H k,1 The total number of fields in the corresponding target text.

[0038] As a specific implementation, each a i Including t fields, u'=t×q 1 , u is H k,1 The sum of the number of error fields in each target text in H; optionally, manually determined H k,1 The number of error fields for each target text in .

[0039] S430, according to l k,1 Get the first confidence interval [l - k,1 ,l + k,1 ],l - k,1 According to l k,1 The error rate lower limit of A is obtained, l + k,1 According to l k,1 The obtained error rate upper limit of A.

[0040] In this embodiment, l - k,1 = l k,1 -z×((1-u' / u 1 )×s 1 / u') 0.5 , l + k,1 = l k,1 +z×((1-u' / u 1 )×s 1 / u') 0.5 , z is the quantile corresponding to the preset confidence level, s1 H k,1 The corresponding error rate variance, u' is H k,1 The total number of fields in the corresponding target text, u 1 The total number of fields in the n target texts.

[0041] In this embodiment, s 1 H k,1 The error rate of each target text in l k,1 The error rate of a target text is the ratio of the number of error fields in the target text to the total number of fields in the target text.

[0042] S440, extract q according to the kth label of the target text 2 The target text is appended to the second text set H k,2 ,q 2 is the preset second extraction quantity, H k,2 is initialized to an empty set.

[0043] Optional, q 2 is the experience value, and q 2 >q 1 ; or, q 2 =z 2 × k,1 ×(1-l k,1 ) / (w 2 ) 2 , w 2 is the preset second sampling error, w 2 <w 1 .

[0044] In this embodiment, q is extracted based on the kth label of the target text. 2 The target text is appended to the second text set H k,2 In the process, from g x Extract from and append to H k,2 The number of target texts is round(q 2 ×qua x / n).

[0045] S450, get H k,2 The corresponding target text error rate l k,2 , l k,2 By comparing H k,2 The corresponding target text and initial text are obtained.

[0046] This embodiment obtains l k,2 The process of obtaining l is as follows: k,1 The process is similar and will not be described here.

[0047] S460, according to l k,2 Get the second confidence interval [l - k,2 ,l + k,2 ],l - k,2 According to l k,2 The error rate lower limit of A is obtained, l + k,2 According to l k,2 The obtained error rate upper limit of A.

[0048] In this embodiment, l - k,2 = l k,2 -z×((1-u 2 / u 1 )×s 2 / u 2 ) 0.5 , l + k,2 = l k,2 +z×((1-u 2 / u 1 )×s 2 / u 2 ) 0.5 , z is the quantile corresponding to the preset confidence level, s 2 H k,2 The corresponding error rate variance, u 2 H k,2 The total number of fields of the corresponding target text, u 1 is the total number of fields in the n target texts. 2 H k,2 The error rate of each target text in l k,2 The mean of the squares of the differences, optional, determined manually H k,2 The error rate of a target text is the ratio of the number of error fields in the target text to the total number of fields in the target text.

[0049] S470, according to [l - k,1 ,l + k,1 ] and [l - k,2 ,l + k,2 ] Determine whether the kth label of the target text can be used to determine the accuracy of A.

[0050] S500. If the k-th tag of the target text cannot be used to judge the accuracy rate of A and k < m, then k = k + 1, and repeat S400 until the k-th tag of the target text can be used to judge the accuracy rate of A or k = m.

[0051] S600. If the k-th tag of the target text can be used to judge the accuracy rate of A, then extract the target text from the n target texts according to the k-th tag of the target text to judge the accuracy rate of A.

[0052] In this embodiment, S600 further includes: if k = m and the k-th tag of the target text cannot be used to judge the accuracy rate of A, then use the random sampling or systematic sampling method to extract q 4 target texts from the n target texts to judge the accuracy rate of A, where q 4 is the preset fourth extraction quantity. As an optional specific implementation manner, q 4 is an empirical value, q 4 > q 3 and q 4 > v × n, where v is the preset proportionality coefficient.

[0053] As the first preferred specific implementation manner, S470 includes: if l -k,1 < l -k,2 and l + k,2 < l + k,1 , then judge that the k-th tag of the target text can be used to judge the accuracy rate of A; otherwise, judge that the k-th tag of the target text cannot be used to judge the accuracy rate of A. In the first preferred specific implementation manner, if the condition of l -k,1 < l -k,2 and l + k,2 < l + k,1 is not satisfied, it means that the confidence interval obtained by stratified sampling according to the k-th tag does not narrow with the increase of the sample size, and it is determined that the k-th tag cannot be used as the stratification factor for reference in stratified sampling. In this case, judge that the k-th tag of the target text cannot be used to judge the accuracy rate of A.

[0054] In the first preferred specific implementation manner, extracting the target text from the n target texts according to the k-th tag of the target text to judge the accuracy rate of A includes: if l - k,1 < l - k,2 , l + k,2 < l + k,1 and l +k,2 -l - k,2 ≤Δl, or l - k,1 <l - k,2 , l + k,2 <l + k,1 , l + k,2 -l - k,2 >Δl and (l + k,1 -l - k,1 -l + k,2 +l - k,2 ) / ((q 2 -q 1 )×η))<β, then 1-l + k,2 Determined as the accuracy of A; if l - k,1 <l - k,2 , l + k,2 <l + k,1 , l + k,2 -l - k,2 >Δl and (l + k,1 -l - k,1 -l + k,2 +l - k,2 ) / ((q 2 -q 1 )×η))≥β, then extract more than q tags from n target texts according to the kth tag of the target text 2 The target text is used to determine the accuracy of A; Δl is the preset interval width, η is the preset difficulty coefficient for determining the accuracy of the target sample, 0<η<1, and β is the preset sampling loss threshold. The first preferred specific implementation method is to set l + k,1 -l - k,1 -l + k,2 +l - k,2 As the width reduction of the confidence interval corresponding to the second sampling, (q 2 -q1 )×η as the workload of the second sampling, through (l + k,1 -l - k,1 -l + k,2 +l - k,2 ) / ((q 2 -q 1 )×η)) to measure the loss of the second sampling, if (l + k,1 -l - k,1 -l + k,2 +l - k,2 ) / ((q 2 -q 1 )×η))<β, it is determined that the loss of continuing sampling is large, so no more sampling is done and 1-l + k,2 Determine the accuracy of A; if (l + k,1 -l - k,1 -l + k,2 +l - k,2 ) / ((q 2 -q 1 )×η))≥β, then it is determined that the loss of continuing sampling is small, and in l + k,2 -l - k,2 >Δl, continue sampling; this preferred specific implementation achieves the purpose of taking into account both the accuracy of the final A and the workload of continued sampling.

[0055] As a second preferred specific implementation, S470 includes:

[0056] S471, extract q according to the kth label of the target text 3 The target texts are appended to the third text set H k,3 ,q 3 is the preset third extraction quantity, H k,3 is initialized to an empty set.

[0057] Optional, q 3 is the experience value, and q 3 >q 2 ; or, q 3 =z 2 × k,2 ×(1-lk,2 ) / (w 3 ) 2 , w 3 is the preset third sampling error, w 3 <w 2 .

[0058] In a second preferred embodiment, q is extracted according to the kth label of the target text. 3 The target texts are appended to the third text set H k,3 In the process, from g x Extract from and append to H k,3 The number of target texts is round(q 3 ×qua x / n).

[0059] S472, Get H k,3 The corresponding target text error rate l k,3 , l k,3 By comparing H k,3 The corresponding target text and initial text are obtained.

[0060] The second preferred specific implementation method obtains l k,3 The process of obtaining l is as follows: k,1 The process is similar and will not be described here.

[0061] S473, according to l k,3 Get the third confidence interval [l - k,3 ,l + k,3 ],l - k,3 According to l k,3 The error rate lower limit of A is obtained, l + k,3 According to l k,3 The obtained error rate upper limit of A.

[0062] In a second preferred embodiment, - k,3 = l k,3 -z×((1-u 3 / u 1 )×s 3 / u 3 ) 0.5 , l + k,3 = l k,3 +z×((1-u 3 / u 1 )×s 3 / u 3 )0.5 , z is the quantile corresponding to the preset confidence level, s 3 H k,3 The corresponding error rate variance, u 3 H k,3 The total number of fields of the corresponding target text, u 1 is the total number of fields of n target texts. In the second preferred embodiment, s 3 H k,3 The error rate of each target text in l k,3 The mean of the squares of the differences, optional, determined manually H k,3 The error rate of a target text is the ratio of the number of error fields in the target text to the total number of fields in the target text.

[0063] S474, if l - k,1 <l - k,2 <l - k,3 And l + k,3 <l + k,2 <l + k,1 , then the kth label of the target text can be used to judge the accuracy of A; otherwise, the kth label of the target text cannot be used to judge the accuracy of A.

[0064] In a second preferred specific implementation, the accuracy of A is determined by extracting the target text from n target texts according to the kth label of the target text, including: if l -k,2 <l -k,3 , l + k,3 <l + k,2 And l + k,3 -l -k,3 ≤Δl, or l -k,2 <l -k,3 , l + k,3 <l + k,2 And l + k,3 -l - k,3 >Δl and (l + k,2 -l - k,2 -l + k,3 +l k,-3) / ((q 3 -q 2 )×η))<β, then 1-l + k,3 Determined as the accuracy of A; if l - k,2 <l - k,3 , l + k,3 <l + k,2 And l + k,3 -l - k,3 >Δl and (l + k,2 -l - k,2 -l + k,3 +l k,-3 ) / ((q 3 -q 2 )×η))≥β, then extract more than q tags from n target texts according to the kth tag of the target text 3 The target text is used to determine the accuracy of A; Δl is the preset interval width, η is the preset difficulty coefficient for determining the accuracy of the target sample, 0<η<1, and β is the preset sampling loss threshold. The second preferred specific implementation method is to set l + k,2 -l - k,2 -l + k,3 +l k,-3 As the width reduction of the confidence interval corresponding to the third sampling, (q 3 -q 2 )×η as the workload of the third sampling, through (l + k,2 -l - k,2 -l + k,3 +l k,-3 ) / ((q 3 -q 2 )×η)) to measure the loss of the third sampling, if (l + k,2 -l - k,2 -l + k,3 +l k,-3 ) / ((q 3 -q 2 )×η))<β, it is determined that the loss of continuing sampling is large, so no more sampling is done and 1-l+ k,3 Determine the accuracy of A; if (l + k,2 -l - k,2 -l + k,3 +l k,-3 ) / ((q 3 -q 2 )×η))≥β, then it is determined that the loss of continuing sampling is small, and in l + k,3 -l -k,3 >Δl, continue sampling; the second preferred specific implementation achieves the purpose of taking into account both the accuracy of the final A and the workload of continuing sampling.

[0065] In this embodiment, the number of target texts is n, and each of the n target texts corresponds to m labels, wherein the priority of the first label is higher than the priority of the second label, the priority of the second label is higher than the priority of the third label, and so on; in this embodiment, it is first determined whether the first label can be used to determine the accuracy of the n target texts. If so, the accuracy of the n target texts is determined according to the first label; if not, it is determined whether the second label can be used to determine the accuracy of the n target texts; if the second label is, the accuracy of the n target texts is determined according to the second label; if not, it is determined whether the third label can be used to determine the accuracy of the n target texts; and so on, until one label can be used to determine the accuracy of the n target texts or none of the m labels can be used to determine the accuracy of the n target texts. In this embodiment, when a certain label can be used to judge the accuracy of n target texts, the target text is extracted from the n target texts according to the label to judge the accuracy of the n target texts. Since the number of extractions is less than n, the workload of the process of obtaining the accuracy of the target text is reduced in this embodiment. Moreover, the target text is extracted according to the label that can be used to judge the accuracy of the n target texts. Therefore, the accuracy of determining the accuracy of the n target texts according to the extracted target text is also ensured in this embodiment, thereby achieving the purpose of taking into account both the accuracy of the acquired target text and the workload of the process of acquiring the accuracy of the target text.

[0066] Although some specific embodiments of the present invention have been described in detail by way of example, it will be appreciated by those skilled in the art that the above examples are for illustration only and are not intended to limit the scope of the present invention. It will also be appreciated by those skilled in the art that various modifications may be made to the embodiments without departing from the scope and spirit of the present invention. The scope of the present invention is defined by the appended claims.

Claims

1. A method for processing a target text, characterized in that: It includes the following steps: S100, obtaining target text A, A={a1, a2, …, a i ,…,a n }, a i is the i-th target text, i ranges from 1 to n, and n is the number of target texts; the target text is obtained by processing the initial text; S200, obtain the label B of the target text, B={b1,b2,…,b i ,…,b n }, b i is a i Tags, b i =(b i,1 ,b i,2 ,…,b i,j ,…,b i,m ), b i,j is a i The jth label of b i,j has a higher priority than b i,j+1 The priority of j ranges from 1 to m, where m is the number of preset tags; S300, initialize the first variable k to 1; S400, determine whether the k-th tag of the target text can be used to judge the accuracy rate of A; S500, if the k-th tag of the target text cannot be used to judge the accuracy rate of A and k < m, then k = k + 1, repeat S400 until the k-th tag of the target text can be used to judge the accuracy rate of A or k = m; S600, if the k-th tag of the target text can be used to judge the accuracy rate of A, then extract the target text from n target texts according to the k-th tag of the target text to judge the accuracy rate of A; S400 includes: S410: extract q1 target texts according to the kth label of the target text and append them to the first text set H k,1 , q1 is the preset first extraction quantity, H k,1 is initialized to an empty set; S420, obtain H k,1 The corresponding target text error rate l k,1 , l k,1 By comparing H k,1 The corresponding target text and initial text are obtained; S430, according to l k,1 Get the first confidence interval [l - k,1 ,l + k,1 ],l - k,1 According to l k,1 The error rate lower limit of A is obtained, l + k,1 According to l k,1 The error rate upper limit of A obtained; S440: extract q2 target texts according to the kth label of the target text and append them to the second text set H k,2 , q2 is the preset second extraction quantity, H k,2 is initialized to an empty set; q2>q1; S450, get H k,2 The corresponding target text error rate l k,2 , l k,2 By comparing H k,2 The corresponding target text and initial text are obtained; S460, according to l k,2 Get the second confidence interval [l - k,2 ,l + k,2 ],l - k,2 According to l k,2 The error rate lower limit of A is obtained, l + k,2 According to l k,2 The error rate upper limit of A obtained; S470, according to [l - k,1 ,l + k,1 ] and [l - k,2 ,l + k,2 ] Determine whether the kth label of the target text can be used to determine the accuracy of A.

2. The target text processing method according to claim 1, characterized in that: S470 includes: If l -k,1 <l -k,2 And l + k,2 <l + k,1 , then the kth label of the target text can be used to judge the accuracy of A; otherwise, the kth label of the target text cannot be used to judge the accuracy of A.

3. The target text processing method according to claim 1, characterized in that: S470 includes: S471, extract q3 target texts according to the kth label of the target text and add them to the third text set H k,3 , q3 is the preset third extraction quantity, H k,3 is initialized to an empty set; q3>q2; S472, Get H k,3 The corresponding target text error rate l k,3 , l k,3 By comparing H k,3 The corresponding target text and initial text are obtained; S473, according to l k,3 Get the third confidence interval [l - k,3 ,l + k,3 ],l - k,3 According to l k,3 The error rate lower limit of A is obtained, l + k,3 According to l k,3 The error rate upper limit of A obtained; S474, if l - k,1 <l - k,2 <l - k,3 And l + k,3 <l + k,2 <l + k,1 , then the kth label of the target text can be used to judge the accuracy of A; otherwise, the kth label of the target text cannot be used to judge the accuracy of A.

4. The method for processing a target text according to claim 1, characterized in that: Each i Includes several fields, l k,1 =u / u', u is H k,1 The total number of error fields in the corresponding target text, u' is H k,1 The total number of fields in the corresponding target text.

5. The method for processing target text according to claim 1, characterized in that: l - k,1 = l k,1 -z×((1-u' / u1)×s1 / u') 0.5 , l + k,1 = l k,1 +z×((1-u' / u1)×s1 / u') 0.5 , z is the quantile corresponding to the preset confidence level, s1 is H k,1 The corresponding error rate variance, u' is H k,1 The total number of fields of the corresponding target text, u1 is the total number of fields of n target texts.

6. The method for processing a target text according to claim 1, characterized in that: S600 further includes: if k = m and the k-th tag of the target text cannot be used to judge the accuracy rate of A, then use the random sampling or systematic sampling method to extract q4 target texts from n target texts to judge the accuracy rate of A, where q4 is the preset fourth extraction quantity.

7. The method for processing a target text according to claim 1, characterized in that: S410 includes: S411, divide the n target texts according to the kth label of the target text to obtain a division result G, G = {g1, g2, ..., g x ,…,g y }, g x is the xth type of target text obtained by division, the value range of x is 1 to y, and y is the number of classes of the target text obtained by division; different g x The k-th label corresponding to the target text is different, and the same g x The k-th label corresponding to the target text in is the same; S412, traverse G, get g x The corresponding number of extractions f x , f x =round(q1×qua x / n),qua x g x The number of target texts included, round() is rounded to the nearest integer; S413, traverse G, from g x Extract f x Append the target text to H k,1 ; S414, if H k,1 The number of target texts included is q1, then H k,1 Determined as the final H k,1 ; if H k,1 If the number of target texts included is greater than q1, then k,1 Random deletion (∑ y x＝1 f x )-q1 target text, and delete the H k,1 Determined as the final H k,1 ; if H k,1 The number of target texts included is less than q1, then q1-(∑ y x＝1 f x ) target text is appended to H k,1 , and the added H k,1 Determined as the final H k,1 ; The first remaining target text is the other target texts among the n target texts except the extracted target text.

Images