A data processing system for obtaining labels corresponding to medical record text

Abstract

The present invention relates to a data processing system for obtaining labels corresponding to medical record texts. The system comprises: a processor and a memory storing a computer program. When the computer program is executed by the processor, the following steps are implemented: obtaining a first sample medical record text set, obtaining an intermediate model and target parameters corresponding to the intermediate model, obtaining a first candidate label list and a first candidate priority list corresponding to a second sample medical record text list, obtaining a first specified priority, adjusting the target parameters according to a first instruction priority to obtain a target model, and obtaining a first target label list according to the target model. The present invention continuously adjusts the model based on the model output result, reduces the cost of obtaining labels, and improves the efficiency of obtaining labels corresponding to medical record texts. At the same time, target labels are obtained from two perspectives, namely, model and text characteristics, and different methods are used based on the consistency and inconsistency of the labels, so that the accuracy of the obtained labels corresponding to the medical record texts is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of text processing, and in particular to a data processing system for obtaining labels corresponding to medical record texts. Background Art

[0002] At present, with the increase in the number of patients, management models such as the development of disease pathways have become a popular research method. By determining the operation pathways corresponding to the diseases, the complexity of monitoring work can be reduced, work efficiency can be improved, and "same disease, same living" can be achieved as much as possible. However, before obtaining the disease pathway, it is necessary to obtain the corresponding operation information from the medical record text. How to efficiently and accurately obtain the corresponding labels from the medical record text has become a hot research direction.

[0003] In the prior art, the method for determining the corresponding label of medical record text is: obtaining medical record text, dividing the medical record text into several texts with the number of characters below a preset threshold, clustering the split texts, and manually labeling the medical record texts of each category after clustering, so as to obtain the label corresponding to the medical record text. In summary, the method for determining the corresponding label of medical record text has the following problems: manually labeling medical records increases the cost of obtaining labels and reduces the efficiency of obtaining the corresponding labels of medical record texts; the model is not utilized and the model is not continuously adjusted based on the output results of the model; it is limited to obtaining the target label from a single angle, which reduces the accuracy of obtaining the corresponding labels of the medical record texts. Summary of the invention

[0004] In view of the above technical problems, the technical solution adopted by the present invention is: a data processing system for obtaining labels corresponding to medical record texts, characterized in that the system comprises: a storage medium storing a preset label list, a processor, and a memory storing a computer program, wherein the preset label list includes a plurality of preset labels, and the preset labels are words corresponding to process nodes in the process of processing the user's body that has developed a disease. When the computer program is executed by the processor, the following steps are implemented:

[0005] S100, obtaining a first sample medical record text set according to a preset label list, wherein the first sample medical record text set includes a plurality of first sample medical record text lists, each preset label corresponds to a first sample medical record text list, the first sample medical record text list includes a plurality of first sample medical record texts, and the first sample medical record texts are medical record texts marked with preset labels.

[0006] S200, based on the first sample medical record text set, obtain an intermediate model and a target parameter η corresponding to the intermediate model, wherein the intermediate model is a model obtained by inputting the first sample medical record text set as a training set into a preset model for training.

[0007] S300, according to the intermediate model, obtain a first candidate label list C corresponding to the second sample medical record text list = {C1, ..., C j , ..., C m} and the first candidate priority list C corresponding to C 0 ={C 0 1, ..., C 0 j , ..., C 0 m}, C j is the first candidate label corresponding to the jth second sample medical record text, C 0 j C j The corresponding first candidate priority.

[0008] S400, according to C 0 and η, obtain the first specified priority λ 1 , where the first specified priority λ 1 Meet the following conditions:

[0009] λ 1 =λ 0 / m,λ 0 C 0 The number of first candidate priorities that is not less than η.

[0010] S500, when λ 1 <F 0 When , η = η - η0, and S2 to S4 are repeatedly executed until the preset cycle cutoff condition is met, wherein the preset cycle cutoff condition is: λ 1 ≥F 0 , η0 is the preset parameter threshold, F 0 is the preset priority threshold.

[0011] S600, when 1 ≥F 0 , obtain the target model, wherein the target model is an intermediate model with a target parameter of η.

[0012] S700, obtaining a target medical record text list L = {L1, ..., L g , ..., L h}, L g is the g-th target medical record text, g=1…h, h is the number of target medical record texts, wherein the target medical record text is the medical record text obtained by splitting the original medical record text of the label to be obtained according to the preset delimiter.

[0013] S800, input L into the target model, and obtain the first target label list L corresponding to L 1 ={L1 1, ..., L 1 g , ..., L 1 h}, L 1 g For L g The corresponding first target label.

[0014] Compared with the prior art, the present invention has obvious beneficial effects. By means of the above technical solution, the data processing system for obtaining the corresponding labels of medical record text provided by the present invention can achieve considerable technical progress and practicality, and has wide industrial utilization value, and has at least the following beneficial effects:

[0015] A data processing system for obtaining labels corresponding to medical record texts, the system comprising: a storage medium storing a preset label list, a processor, and a memory storing a computer program, wherein when the computer program is executed by the processor, the following steps are implemented: obtaining a first sample medical record text set according to the preset label list, obtaining an intermediate model and a target parameter corresponding to the intermediate model according to the first sample medical record text set, obtaining a first candidate label list corresponding to a second sample medical record text list and a first candidate priority list corresponding to the first candidate label list according to the intermediate model, obtaining a first designated priority according to the target parameter and the first candidate priority list, and updating the target parameter when the first instruction priority is less than a preset priority threshold. number, repeat the above steps until the preset loop end condition is met. When the first instruction priority is not less than the preset priority threshold, obtain the target model, obtain the target medical record text list, input the target medical record text list into the target model, and obtain the first target label list corresponding to the target medical record text list. The present invention uses a model to obtain labels corresponding to medical record texts, and continuously adjusts the model based on the model output results, thereby reducing the cost of obtaining labels and improving the efficiency of obtaining labels corresponding to medical record texts. At the same time, target labels are obtained from two perspectives, namely model and text characteristics, and different methods are used based on the consistency and inconsistency of labels, so that the accuracy of the obtained labels corresponding to medical record texts is higher.

[0016] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented in accordance with the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the following specifically cites a preferred embodiment and describes it in detail with the accompanying drawings as follows. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] Figure 1 A flowchart of an execution computer program of a data processing system for obtaining labels corresponding to medical record texts provided in Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0018] 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.

[0019] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchangeable where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product, or server that includes a series of steps or units is not necessarily limited to those steps or units clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products, or devices.

[0020] Embodiment 1

[0021] This embodiment provides a data processing system for obtaining labels corresponding to medical record texts, the system comprising: a storage medium storing a preset label list, a processor, and a memory storing a computer program, wherein the preset label list includes a plurality of preset labels, and the preset labels are words corresponding to process nodes in the process of processing the user's body that has a disease. When the computer program is executed by the processor, the following steps are implemented, such as: Figure 1 As shown:

[0022] S100, obtaining a first sample medical record text set according to a preset label list, wherein the first sample medical record text set includes a plurality of first sample medical record text lists, each preset label corresponds to a first sample medical record text list, the first sample medical record text list includes a plurality of first sample medical record texts, and the first sample medical record texts are medical record texts marked with preset labels.

[0023] Specifically, for example: preset labels such as neck examination, blood routine examination, biochemical examination, etc.

[0024] Specifically, the first sample medical record text is obtained by obtaining the medical record text from a database and marking the obtained medical record text with a corresponding preset label.

[0025] Furthermore, those skilled in the art know that any database in the prior art that can obtain medical record texts falls within the protection scope of the present invention and will not be described in detail here, such as a public medical database.

[0026] S200, based on the first sample medical record text set, obtain an intermediate model and a target parameter η corresponding to the intermediate model, wherein the intermediate model is a model obtained by inputting the first sample medical record text set as a training set into a preset model for training.

[0027] Specifically, the value range of η is 0.9 to 0.99, wherein those skilled in the art know that η can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0028] Specifically, those skilled in the art know that any method for model training in the prior art falls within the protection scope of the present invention and will not be described in detail here, for example: neural network model, LLM model and other preset models.

[0029] S300, according to the intermediate model, obtain a first candidate label list C corresponding to the second sample medical record text list = {C1, ..., C j , ..., C m} and the first candidate priority list C corresponding to C 0 ={C 0 1, ..., C 0 j , ..., C 0 m}, C j is the first candidate label corresponding to the jth second sample medical record text, C 0 j C j The corresponding first candidate priority.

[0030] Specifically, the second sample medical record text list includes a plurality of second sample medical record texts, wherein the second sample medical record texts are texts obtained by splitting the medical record texts obtained from the database and excluding preset tags.

[0031] Furthermore, the splitting method is splitting by periods or by verbs, which can be understood as: obtaining a long medical record text from a database, splitting the obtained medical record text into short texts one by one according to periods or verbs, and the second sample medical record text is short texts one by one.

[0032] Specifically, the first candidate label is a preset label corresponding to each second sample medical record text obtained by inputting the second sample medical record text set into the intermediate model.

[0033] Furthermore, the first candidate priority is the degree to which the first candidate label corresponding to each second sample medical record text obtained by inputting the second sample medical record text set into the intermediate model matches the preset label in the preset label list.

[0034] S400, according to C 0 and η, obtain the first specified priority λ 1 , where the first specified priority λ 1 Meet the following conditions:

[0035] λ 1 =λ 0 / m,λ 0 C 0 The number of first candidate priorities that is not less than η.

[0036] S500, when λ 1 <F 0 When , η = η - η0, and S2 to S4 are repeatedly executed until the preset cycle cutoff condition is met, wherein the preset cycle cutoff condition is: λ 1 ≥F 0 , η0 is the preset parameter threshold, F 0 is the preset priority threshold.

[0037] Specifically, the value range of η0 is 0.01 to 0.1, wherein those skilled in the art know that η0 can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0038] Specifically, the value range of F0 is 0.8 to 0.9, wherein those skilled in the art know that F0 can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here.

[0039] S600, when 1 ≥F 0 , obtain the target model, wherein the target model is an intermediate model with a target parameter of η.

[0040] S700, obtaining a target medical record text list L = {L1, ..., L g , ..., L h}, L g is the g-th target medical record text, g=1…h, h is the number of target medical record texts, wherein the target medical record text is the medical record text obtained by splitting the original medical record text of the label to be obtained according to the preset delimiter.

[0041] S800, input L into the target model, and obtain the first target label list L corresponding to L 1 ={L 11, ..., L 1 g , ..., L 1 h}, L 1 g For L g The corresponding first target label.

[0042] Specifically, in S800, L is obtained through the following steps: 1 g :

[0043] S801, input L into the target model, and obtain the candidate label list set P corresponding to L = {P1, ..., P g , ..., P h}, P g = {P g1 , ..., P gd , ..., P gv(g)} and P corresponding candidate matching list set Q = {Q1, ..., Q g , ..., Q h}, Q g = {Q g1 , ..., Q gd , ..., Q gv(g)}, P gd For L g The corresponding candidate label in the candidate label list is the dth candidate label, Q gd P gd The corresponding candidate matching degree, d = 1...v(g), v(g) is L g The number of candidate tags in the corresponding candidate tag list, wherein the candidate tags are preset tags obtained from the preset tag list based on the target model and matched with the target medical record text, and the candidate matching degree is the degree of matching between the target medical record text obtained based on the target model and the preset tags.

[0044] S803, when Q gd Q g When the candidate matching degree is the largest, obtain P gd For L 1 g .

[0045] Specifically, after step S800, the following steps are also included:

[0046] S1, process L and obtain the second target label list L corresponding to L 2 ={L 2 1, ..., L 2 g , ..., L 2 h}, L2 g For L g The corresponding second target label.

[0047] Specifically, in S1, L is obtained by the following steps: 2 g :

[0048] S11, obtain the target medical record text cluster list H = {H1, ..., H ω , ..., H ψ}, H ω is the ωth target medical record text cluster, ω=1…ψ, ψ is the number of target medical record text clusters.

[0049] Specifically, in S11, H is obtained through the following steps:

[0050] S111, obtain the sample medical record text vector set Y = {Y1, ..., Y ε , ..., Y z},Y ε is the th sample medical record text vector, ε=1…z, z is the number of sample medical record text vectors.

[0051] Specifically, the sample medical record text vector is a text vector obtained by inputting the sample medical record text into a preset natural language processing model. Those skilled in the art know that any method in the prior art for converting text into a vector through a natural language processing model falls within the scope of protection of the present invention and will not be repeated here.

[0052] Furthermore, the sample medical record text is a medical record text obtained from a database.

[0053] S112, obtain the preferred medical record text vector list set G = {G1, ..., G t , ..., G k1}, G t is the tth preferred medical record text vector list, t=1...k1, k1 is the number of preferred medical record text vector lists.

[0054] Specifically, the preferred medical record text vector list includes several preferred medical record text vectors.

[0055] Furthermore, the preferred medical record text vector is a text vector randomly obtained from Y.

[0056] Specifically, the initial preferred medical record text vector list includes only one preferred medical record text vector.

[0057] S113, according to Y and G, obtain the first similarity list set W corresponding to Y 1 = {W1 1, ..., W 1 ε , ..., W 1 z},W 1 ε = {W 1 ε1 , ..., W 1 εt , ..., W 1 εk1},W 1 εt Y ε With G t A first similarity between them; those skilled in the art know that any method for obtaining the similarity between vectors in the prior art belongs to the protection scope of the present invention and will not be described in detail here, such as a method for obtaining the first similarity such as a cosine similarity method.

[0058] S114, based on W 1 , get the first intermediate medical record text vector cluster list J 1 ={J 1 1, ..., J 1 t , ..., J 1 k1}, wherein, in S114, J is obtained by the following steps 1 t :

[0059] S1141, when W 1 ε W 1 When the first similarity is the smallest, Y ε Insert to G t middle;

[0060] S1143, according to G t , get the first center point vector list G corresponding to G 1 = {G 1 1, ..., G 1 t , ..., G 1 k1}, G 1 t G t The corresponding first center point vector.

[0061] Specifically, those skilled in the art know that any method for obtaining multiple vector center points in the prior art falls within the protection scope of the present invention and will not be described in detail herein, such as a method for obtaining the first center point vector such as obtaining an average vector.

[0062] S1145, repeat steps S1141 to S1143 to obtain J 1 t .

[0063] S115, repeating steps S112 to S114 to obtain a second intermediate medical record text vector cluster J = {J 1 , ..., J α , ..., J β}, J α ={J α 1, ..., J α c , ..., J α kα},J α c is the cth second intermediate medical record text vector cluster in the αth second intermediate medical record text vector cluster list, c=1…kα, kα is the number of second intermediate medical record text vector clusters in the αth second intermediate medical record text vector cluster list, α=1…β, β is the number of second intermediate medical record text vector clusters.

[0064] S116, according to J, obtain H, wherein in step S116, H is obtained by the following steps:

[0065] S1161, Get J α c The first specified medical record text vector KJ α c and J α c The corresponding second specified medical record text vector list LJ α c ={LJ α c1 , ..., LJ α cμ , ..., LJ α cζ}, L.J. α cμ For J α c The corresponding μth second designated medical record text vector, μ=1…ζ, ζ is the number of second designated medical record text vectors.

[0066] Specifically, the first specified medical record text vector is J α c Any second intermediate medical record text vector.

[0067] Furthermore, the second specified medical record text vector is J α c Any second intermediate medical record text vector except the first specified medical record text vector.

[0068] S1162, Get J α c The corresponding third specified medical record text vector list EJ α ={EJ α c1 , ..., E. J. α cσ , ..., E. J. α cρ}, E.J. α cσ For J α c The corresponding σth third designated medical record text vector, σ=1…ρ, ρ is the number of third designated medical record text vectors.

[0069] Specifically, the third specified medical record text vector is J α J α c Any second intermediate medical record text vector in any other second intermediate medical record text vector cluster.

[0070] S1163, according to KJ α c and LJ α c , get the first specified similarity list FJ α c ={FJ α c1 , ..., FJ α cμ , ..., FJ α cζ}, where FJ α cμ For KJ α c and LJ α cμ The similarity between .

[0071] Specifically, the method for obtaining the first specified similarity is consistent with the method for obtaining the first similarity.

[0072] S1164, according to KJ α c and E.J. α , get the second specified similarity list BJ α = {BJ α c1 , ..., B.J. α cσ , ..., B.J. α cρ}, where BJ αcσ For KJ α c and E.J. α σ The similarity between .

[0073] Specifically, the method for obtaining the second specified similarity is consistent with the method for obtaining the first specified similarity.

[0074] S1165, when FRJ α c is the minimum value and FTJ α c When J is the maximum value, α The corresponding second intermediate medical record text vector cluster list is H, where FRJ α c =(FJ α c1 +……+FJ α cμ +……+FJ α cζ ) / ζ, FTJ α =(BJ α c1 +……+BJ α cσ +……+BJ α cρ ) / ρ.

[0075] Specifically, kα=k1+α.

[0076] In the above, the medical record text is clustered, and different methods are used for the initial clustering to obtain the clustering method with the smallest intra-cluster distance and the largest extra-cluster distance and the clustered medical record text, thereby improving the accuracy of obtaining the medical record text cluster.

[0077] S13, obtain the target keyword list set DH corresponding to H = {DH1, ..., DH ω , ..., DH ψ}, DH ω H ω Corresponding target keyword list.

[0078] Specifically, the target keyword list includes several target keywords, and the target keywords are keywords obtained from the target medical record text cluster. Those skilled in the art know that any method of obtaining keywords from text in the prior art falls within the protection scope of the present invention and will not be repeated here, for example: methods for obtaining keywords such as TF-IDF algorithm.

[0079] S15, according to DH and H, obtain the specified label list SH corresponding to H = {SH1, ..., SHω ,……,SH ψ}, SH ω H ω The corresponding specified label.

[0080] Specifically, in S15, SH is obtained through the following steps: ω :

[0081] S151, according to H ω and DH ω , get H ω The corresponding target text H 0 ω ={DH ω , H ω}.

[0082] S153, according to H 0 ω , get SH ω , among which, SH ω To H 0 ω Input the preset labels obtained in the target model.

[0083] S17, according to H, obtain L g The corresponding target similarity list HL g ={HL g1 , ..., HL gω , ..., HL gψ}, HL gω For L g The corresponding vector and H ω The similarity between corresponding vectors.

[0084] Specifically, those skilled in the art know that any method for converting text into a vector in the prior art falls within the protection scope of the present invention and will not be described in detail herein, for example, methods for converting text into a vector such as the Bert model.

[0085] Furthermore, those skilled in the art know that any method for obtaining the similarity between vectors in the prior art falls within the protection scope of the present invention and will not be described in detail herein, for example, methods for calculating the similarity between vectors such as cosine similarity.

[0086] S19, when HL gω HL g When the target similarity is the largest, obtain L 2 g SH ω .

[0087] In the above, the model is used to obtain the labels corresponding to the medical record text, and the model is judged based on the labels obtained by the model and the priorities corresponding to the labels, so as to obtain the target labels corresponding to the medical record text in different ways. When the model makes an inaccurate judgment on the labels corresponding to the medical record text, the target labels corresponding to the medical record text are obtained by combining clustering and text keywords, thereby improving the accuracy of obtaining the target labels.

[0088] S2, when L 1 g With L 2 g When consistent, obtain L 1 g For L g The corresponding target label.

[0089] S3, when L 1 g With L 2 g If inconsistent, obtain L g The corresponding target priority L 0 g .

[0090] Specifically, in S3, L is obtained through the following steps: 0 g :

[0091] S31, get L g The corresponding candidate label list P g = {P g1 , ..., P gd , ..., P gv(g)} and P g The corresponding candidate matching list Q g = {Q g1 , ..., Q gd , ..., Q gv(g)}, P gd For L g The corresponding candidate label in the candidate label list is the dth candidate label, Q gdw P gd The corresponding candidate matching degree, d = 1...v(g), v(g) is L g The number of candidate tags in the corresponding candidate tag list, wherein the candidate tags are preset tags obtained from the preset tag list based on the target model and matched with the target medical record text, and the candidate matching degree is the degree of matching between the target medical record text obtained based on the target model and the preset tags.

[0092] S33, according to P g and Q g , get L gThe corresponding target priority L 0 g , where L 0 g Meet the following conditions:

[0093]

[0094] S4, when L 0 g ≥E 0 When L 1 g For L g The corresponding target label, where E 0 is the preset target priority threshold.

[0095] S5, when L 0 g <E 0 When L 2 g For L g The corresponding target label.

[0096] As mentioned above, the model is used to obtain the labels corresponding to the medical record text, and the model is continuously adjusted based on the model output results, which reduces the cost of obtaining labels and improves the efficiency of obtaining labels corresponding to the medical record text. At the same time, the target labels are obtained from two perspectives, namely the model and text characteristics, and different methods are used based on the consistency and inconsistency of the labels, so that the accuracy of the obtained labels corresponding to the medical record text is higher.

[0097] Specifically, after step S5, the following steps are also included:

[0098] S51, obtain the candidate object ID list A corresponding to the target disease type = {A1, ..., A i , ..., A n} and the candidate information list set E corresponding to A = {E1, ..., E i ,……,E n}, E i ={E i1 ,……,E ir ,……,E is(i)}, E ir ={E 1 ir , E 2 ir}, A i is the ID of the i-th candidate object, E 1 ir A i The first candidate information in the rth candidate information in the corresponding candidate information list, E 2 irA i The second candidate information in the rth candidate information in the corresponding candidate information list, i=1...n, n is the number of candidate object IDs, r=1...s(i), s(i) is A i The number of candidate information in the corresponding candidate information list, wherein the first candidate information is the time obtained from the initial medical record text corresponding to the candidate object ID, and the second candidate information is the target label obtained from the initial medical record text corresponding to the candidate object ID.

[0099] Specifically, the target disease is a single disease that will not cause complications, such as acute pancreatitis and other target diseases.

[0100] Specifically, the candidate object ID is a unique identifier that characterizes the candidate object, wherein the candidate object is an object suffering from a target disease.

[0101] Specifically, the candidate information includes first candidate information and second candidate information.

[0102] Specifically, the initial medical record text is the medical record text corresponding to the target disease obtained from the database. Those skilled in the art know that any database in the prior art that can obtain medical record text falls within the protection scope of the present invention and will not be repeated here, such as public medical databases.

[0103] Furthermore, the first candidate information is obtained by obtaining the time entity in the target medical record text through a NER model.

[0104] Specifically, the method for acquiring the second candidate information is consistent with the method for acquiring the target tag corresponding to the target medical record text, and can refer to steps S1 to S5 in this embodiment.

[0105] S52, according to A and E, obtain a candidate weight list M corresponding to the candidate label list = {M1, ..., M a , ..., M θ},M a is the candidate weight corresponding to the a-th candidate label, a=1……θ, θ is the number of candidate labels, where M a Meet the following conditions:

[0106] M a =M 0 a / n,M 0 a is the number of candidate object IDs corresponding to the candidate information whose second candidate information in the candidate information list in E is consistent with the candidate label.

[0107] Specifically, the candidate tag is the second candidate information obtained after deduplication processing is performed on the second candidate information obtained from each candidate information list.

[0108] S53, according to M, obtain the specified tag list V = {V1, ..., V e , ..., V f},V e is the e-th specified tag, e=1...f, f is the number of specified tags, where M a ≥F 1 When M a The corresponding candidate labels are inserted into V, F 1 is the preset weight threshold.

[0109] Specifically, the F 1 The value range of F is 0.3 to 0.5, wherein those skilled in the art know that F can be adjusted according to actual needs. 1 The selection of all falls within the protection scope of the present invention and will not be repeated here.

[0110] S54, according to E and V, obtain the target time period list T={T1, ..., T ab , ..., T cd The final label list set ST corresponding to T = {ST1, ..., ST ab , ..., ST cd} to obtain the operation path corresponding to the target disease, T ab is the (ab)th target time period corresponding to the target disease, ST ab T ab The corresponding final label list, wherein the final label list includes several final labels, the target time period is the first candidate information obtained from the candidate information list, and the final label is the type of operation required to be performed in the target time period based on the target disease, ab=1...cd, cd is the number of target time periods in the target time period list corresponding to the target disease.

[0111] Specifically, S54 also includes the following steps:

[0112] S541, according to E and V, obtain the specified time period list set N corresponding to V = {N1, ..., N e , ..., N f},N e = {N e1 , ..., N ev , ..., N eb(e)},N ev V eThe vth specified time period in the corresponding specified time period list, v = 1...b(e), b(e) is V e The number of specified time periods in the corresponding specified time period list, where b(e) meets the following conditions:

[0113] b(e)=T 0 e / ΔT,T 0 e V e The corresponding candidate object ID starts from the time of recording the medical record to the time of executing V e The longest time span that has passed, ΔT is the preset time span.

[0114] Specifically, the specified time period is a time period divided according to ΔT based on the first candidate information corresponding to each candidate object ID. Those skilled in the art know that ΔT can be selected according to actual needs, which falls within the protection scope of the present invention and will not be repeated here, for example: preset time spans such as half a day and one day.

[0115] Furthermore, the value range of ΔT is 1 day to 1.5 days.

[0116] Furthermore, the time span of each specified time period in the specified time period list corresponding to a specified tag is consistent, and there is no time interval between each specified time period.

[0117] Furthermore, N e1 The corresponding start time is 0.

[0118] S543, according to N, obtain the specified number list set N corresponding to N 0 = {N 0 1, ..., N 0 e , ..., N 0 f},N 0 e = {N 0 e1 , ..., N 0 ev , ..., N 0 eb(e)},N 0 ev N ev The corresponding specified quantity.

[0119] Specifically, the specified number is the number of candidate object IDs of the first candidate information obtained from E within the specified time period and the second candidate information corresponding to the first candidate information is consistent with the specified label; it can be understood as: taking the e-th specified label as an example, when the time span of the specified time period is days, the list of specified time periods corresponding to the e-th specified label is: the first day, the second day, the third day...the b-th day, and the specified number corresponding to the first day is the number of candidate object IDs that include this specified label in the second candidate information corresponding to the first candidate information when the first candidate information is on the first day in the entire medical record text record timeline.

[0120] S545, when N 0 e N does not exist in 0 e(v-1) <N 0 ev >N 0 e(v+1) When , ΔT=ΔT-ΔT 0 , S541 to S543 are repeatedly executed until a preset candidate cycle cutoff condition is met, wherein the preset candidate cycle cutoff condition is: N 0 e There is N 0 e(v-1) <N 0 ev >N 0 e(v+1) , N 0 e(v-1) V e The specified number corresponding to the (v-1)th specified time period in the corresponding specified time period list, N 0 e(v+1) V e The specified quantity corresponding to the (v+1)th specified time period in the corresponding specified time period list, ΔT 0 is the preset time span threshold.

[0121] Specifically, ΔT 0 The value range is 3 hours to 6 hours.

[0122] S547, when N 0 e There is N 0 e(v-1) <N 0 ev >N 0 e(v+1) When V e and V e A corresponding key time period list, wherein the key time period list includes several key time periods.

[0123] Specifically, in S547, the key time period is obtained through the following steps:

[0124] S5471, according to N 0 ev , get V e Corresponding middle time period list TV e and TV e The corresponding intermediate priority list, wherein the intermediate time period list includes several intermediate time periods, the intermediate priority list includes several intermediate priorities, and the intermediate priority is V e The ratio between the specified number and n in the corresponding intermediate time period, where N 0 e(τ-1) >N 0 eτ <……<N 0 e(v-1) <N 0 ev >……>N 0 eδ <N 0 e(δ+1) Get TV e = {N eτ , ..., N e(v-1) , N ev , ..., N eδ},N eτ V e The τth specified time period in the corresponding specified time period list, N 0 eτ N eτ The corresponding specified number, N eδ V e The corresponding specified time period in the specified time period list, N 0 eδ N eδ The corresponding specified number, N 0 e(τ-1) V e The specified number corresponding to the (τ-1)th specified time period in the corresponding specified time period list, N 0 e(δ+1) V e The specified quantity corresponding to the (δ+1)th specified time period in the corresponding specified time period list.

[0125] S5473, when TV e When the intermediate priority in the corresponding intermediate priority list exceeds the preset intermediate priority threshold, the intermediate time period corresponding to the intermediate priority is obtained as V e The corresponding critical time period.

[0126] Specifically, the value range of the preset intermediate priority threshold is 0.3 to 0.5. Those skilled in the art know that the preset intermediate priority threshold can be selected according to actual needs, which falls within the protection scope of the present invention and will not be repeated here.

[0127] S549, the obtained V e Inserted into ST and V e The corresponding key time period list is inserted into T to obtain T and ST.

[0128] As mentioned above, for the same disease, consider obtaining the operation path corresponding to the disease based on historical data, determine the number of related objects by setting different time periods, and perform statistical analysis based on historical data to obtain representative operations for the same disease and the operation time periods corresponding to the operations, so that the accuracy of the operation path corresponding to the target disease obtained is higher.

[0129] The present embodiment provides a data processing system for obtaining labels corresponding to medical record texts. The system includes: a storage medium storing a preset label list, a processor, and a memory storing a computer program. When the computer program is executed by the processor, the following steps are implemented: according to the preset label list, a first sample medical record text set is obtained; according to the first sample medical record text set, an intermediate model and a target parameter corresponding to the intermediate model are obtained; according to the intermediate model, a first candidate label list corresponding to the second sample medical record text list and a first candidate priority list corresponding to the first candidate label list are obtained; according to the target parameter and the first candidate priority list, a first specified priority is obtained; when the first instruction priority is less than a preset priority threshold, New target parameters are set, and the above steps are repeated until the preset loop cutoff condition is met. When the first instruction priority is not less than the preset priority threshold, the target model is obtained, the target medical record text list is obtained, the target medical record text list is input into the target model, and the first target label list corresponding to the target medical record text list is obtained. The present invention uses a model to obtain labels corresponding to medical record texts, and continuously adjusts the model based on the model output results, thereby reducing the cost of obtaining labels and improving the efficiency of obtaining labels corresponding to medical record texts. At the same time, target labels are obtained from two perspectives, namely model and text characteristics, and different methods are used based on the consistency and inconsistency of labels, so that the accuracy of the obtained labels corresponding to medical record texts is higher.

[0130] Embodiment 2

[0131] This embodiment provides a system for obtaining labels corresponding to medical record texts, the system comprising: a storage medium storing a preset label list, a processor, and a memory storing a computer program, wherein the preset label list includes a plurality of preset labels, and the preset labels are words corresponding to process nodes in a process of processing a user's body that has a disease. When the computer program is executed by the processor, the following steps are implemented:

[0132] S100, obtaining a target medical record text list L = {L1, ..., L g , ..., L h}, L g is the g-th target medical record text, g=1…h, h is the number of target medical record texts, wherein the target medical record text is the medical record text obtained by splitting the original medical record text of the label to be obtained according to the preset delimiter.

[0133] Specifically, for example: preset labels such as neck examination, blood routine examination, biochemical examination, etc.

[0134] Specifically, those skilled in the art know that the preset delimiters can be selected according to actual needs, all of which fall within the protection scope of the present invention and will not be described in detail here, such as the preset delimiters such as period, semicolon, verb, etc.

[0135] S200, according to L, obtain the first target tag list L corresponding to L 1 ={L 1 1, ..., L 1 g , ..., L 1 h}, L 1 g For L g The corresponding first target label.

[0136] Specifically, in S200, L is obtained by the following steps: 1 g :

[0137] S201, obtaining a target model according to a preset tag list.

[0138] Specifically, the method for acquiring the target model is consistent with the method for acquiring the target model in the first embodiment, and reference may be made to steps S100 to S600 in the above embodiment.

[0139] S203, input L into the target model, and obtain the candidate label list set P corresponding to L = {P1, ..., P g , ..., P h}, P g = {P g1 , ..., P gd, ..., P gv(g)} and P corresponding candidate matching list set Q = {Q1, ..., Q g , ..., Q h}, Q g = {Q g1 , ..., Q gd , ..., Q gv(g)}, P gd For L g The corresponding candidate label in the candidate label list is the dth candidate label, Q gd P gd The corresponding candidate matching degree, d = 1...v(g), v(g) is L g The number of candidate tags in the corresponding candidate tag list, the target model is a model that matches the target medical record text with the preset tags in the preset tag list, the candidate tags are preset tags that are obtained from the preset tag list based on the target model and match the target medical record text, and the candidate matching degree is the degree of matching between the target medical record text obtained based on the target model and the preset tags.

[0140] S207, obtain L g The corresponding target label R g P gd , where Q gd Q g The maximum candidate match among .

[0141] S300, according to L, obtain a second target tag list L corresponding to L 2 ={L 2 1, ..., L 2 g , ..., L 2 h}, L 2 g For L g The corresponding second target label.

[0142] Specifically, the method for acquiring the second target tag is consistent with the method for acquiring the second target tag in the first embodiment, and reference may be made to S11 to S19 in the first embodiment.

[0143] S400, when L 1 g With L 2 g When consistent, obtain L 1 g For L g The corresponding target label.

[0144] S500, when L 1 g With L2 g If inconsistent, obtain L g The corresponding target tag, wherein L is obtained in S500 through the following steps g The corresponding target label:

[0145] S501, obtain a sample medical record text set DK = {DK1, ..., DK ε , ..., DK z}, DK ε is the εth sample medical record text, ε=1…z, z is the number of sample medical record texts.

[0146] Specifically, the sample medical record text is a medical record text obtained from a database.

[0147] Furthermore, those skilled in the art know that any database in the prior art that can obtain medical record texts falls within the protection scope of the present invention and will not be described in detail here, such as a public medical database.

[0148] S502, process DK using the first method to obtain a sample label list set corresponding to DK, wherein the sample label list set includes a plurality of sample label lists, the sample label list includes a plurality of sample labels, and each sample medical record text corresponds to a sample label list.

[0149] Specifically, the method for acquiring the sample label is consistent with the method for acquiring the first target label.

[0150] S503, obtain L g The corresponding candidate label pair GL g =(L 1 g-1 , L 1 g ), where L 1 g-1 For L g-1 The corresponding first target label.

[0151] S504, obtain GL according to the sample label list set g Corresponding key priority FL g , where FL g Meet the following conditions:

[0152] FL g =GL 0 g / z,GL 0 g Include GL in the sample label list set g The number of sample label lists.

[0153] S505, when FL g ≥FL 0 When L 1 g For L g The corresponding target label, FL 0 is the preset critical priority threshold.

[0154] S506, when FL g <FL 0 When L 2 g For L g The corresponding target label.

[0155] In the above, different methods are used to obtain the labels corresponding to the medical record text, and the characteristics of the model and the text are used to process the medical record text to obtain the corresponding labels, which improves the efficiency of obtaining the corresponding labels of the medical record text. When the obtained labels are inconsistent, the model is used to process the text connected before and after the medical record text to obtain several label pairs, and different methods are used based on the priority of the label pair to obtain the target label, so that the accuracy of the obtained medical record text corresponding labels is higher.

[0156] The present embodiment provides a system for obtaining labels corresponding to medical record texts, the system comprising: a processor and a memory storing a computer program, and when the computer program is executed by the processor, the following steps are implemented: obtaining a target medical record text list, processing the target medical record text list in a first manner, obtaining a first target label list corresponding to the target medical record text list, processing the target medical record text list in a second manner, obtaining a second target label list corresponding to the target medical record text list, when the first target label and the second target label are consistent, obtaining the first target label as the target label, when the first target label and the second target label are inconsistent, obtaining a sample medical record text set, processing the sample medical record text set in the first manner The sample label list set corresponding to the sample medical record text set is obtained, and the candidate label pair corresponding to the target medical record text is obtained. According to the sample label list set, the key priority corresponding to the candidate label pair is obtained. According to the key priority, the target label is obtained. This embodiment adopts different methods to obtain the label corresponding to the medical record text, and uses the characteristics of the model and the text to process the medical record text to obtain the corresponding label, which improves the efficiency of obtaining the corresponding label of the medical record text. When the obtained labels are inconsistent, the model is used to process the text connected before and after the medical record text to obtain several label pairs. Based on the priority of the label pair, different methods are used to obtain the target label, so that the accuracy of the obtained medical record text corresponding label is higher.

[0157] 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 data processing system for obtaining labels corresponding to medical record texts, characterized in that: The system includes: a storage medium storing a preset tag list, a processor, and a memory storing a computer program, wherein the preset tag list includes a plurality of preset tags, and the preset tags are words corresponding to process nodes in the process of treating the user's body that has a disease. When the computer program is executed by the processor, the following steps are implemented: S100, obtaining a first sample medical record text set according to a preset label list, wherein the first sample medical record text set includes a plurality of first sample medical record text lists, each preset label corresponds to a first sample medical record text list, the first sample medical record text list includes a plurality of first sample medical record texts, and the first sample medical record texts are medical record texts marked with preset labels; S200, obtaining an intermediate model and a target parameter η corresponding to the intermediate model according to the first sample medical record text set, wherein the intermediate model is a model obtained by inputting the first sample medical record text set as a training set into a preset model for training; S300, according to the intermediate model, obtain the first candidate label list C={C1, ..., C j , ..., C m } and the first candidate priority list C corresponding to C 0 ={C 0 1, ..., C 0 j , ..., C 0 m }, C j is the first candidate label corresponding to the jth second sample medical record text, C 0 j C j The corresponding first candidate priority; S400, according to C 0 and η, obtain the first specified priority λ 1 , where the first specified priority λ 1 Meet the following conditions: λ 1 =λ 0 / m,λ 0 C 0 The number of first candidate priorities that is not less than η in ; S500, when λ 1 <F 0 When η=η-η0, S200-S400 are repeatedly executed until the preset cycle cutoff condition is met, wherein the preset cycle cutoff condition is: λ 1 ≥F 0 , η0 is the preset parameter threshold, F 0 is a preset priority threshold; S600, when 1 ≥F 0 , obtaining a target model, wherein the target model is an intermediate model with a target parameter of η; S700, obtain the target medical record text list L={L1, ..., L g , ..., L h }, L g is the g-th target medical record text, g=1…h, h is the number of target medical record texts, wherein the target medical record text is the medical record text obtained by splitting the original medical record text of the label to be obtained according to the preset delimiter; S800, input L into the target model, and obtain the first target label list L corresponding to L 1 ={L 1 1, ..., L 1 g , ..., L 1 h }, L 1 g For L g The corresponding first target tag, wherein L is obtained by the following steps in S800 1 g : S801, input L into the target model, and obtain the candidate label list set P corresponding to L = {P1, ..., P g , ..., P h } and P corresponding candidate matching list set Q = {Q1, ..., Q g , ..., Q h }, where P g ={P g1 , ..., P gd , ..., P gv（g） }, Q g ={Q g1 , ..., Q gd , ..., Q gv（g） }, P gd For L g The corresponding candidate label in the candidate label list is the d-th candidate label, Q gd P gd The corresponding candidate matching degree, d=1...v(g), v(g) is L g The number of candidate tags in the corresponding candidate tag list, wherein the candidate tags are preset tags that match the target medical record text obtained from the preset tag list based on the target model, and the candidate matching degree is the matching degree between the target medical record text obtained based on the target model and the preset tags; S803, when Q gd Q g When the candidate matching degree is the largest, obtain P gd For L 1 g ; After step S800, the following steps are also included: S1, process L and obtain the second target label list L corresponding to L 2 ={L 2 1, ..., L 2 g , ..., L 2 h }, L 2 g For L g The corresponding second target tag, wherein S1 also includes the following steps: S11, obtaining a target medical record text cluster list H = {H1, ..., H ω , ..., H ψ }, H ω is the ωth target medical record text cluster, ω=1…ψ, ψ is the number of target medical record text clusters; S13, obtain the target keyword list DH corresponding to H = {DH1, ..., DH ω , ..., DH ψ }, DH ω H ω Corresponding target keywords; S15, according to DH and H, obtain the specified label list SH corresponding to H = {SH1, ..., SH ω ,……,SH ψ }, SH ω H ω The corresponding designated tag, wherein in S15, SH is obtained by the following steps ω : S151, according to H ω and DH ω , get H ω The corresponding target text H 0 ω ={DH ω , H ω }; S153, according to H 0 ω , get SH ω , among which, SH ω To H 0 ω Input the preset labels obtained in the target model; S17, according to H, obtain L g The corresponding target similarity list HL g ={HL g1 , ..., HL gω , ..., HL gψ }, HL gω For L g The corresponding vector and H ω The similarity between corresponding vectors; S19, when HL gω HL g When the target similarity is the largest, obtain L 2 g SH ω ; S2, when L 1 g With L 2 g When consistent, obtain L 1 g For L g The corresponding target label; S3, when L 1 g With L 2 g If inconsistent, obtain L g The corresponding target priority L 0 g , where L is obtained in S3 through the following steps 0 g : S31, get L g The corresponding candidate label list P g ={P g1 , ..., P gd , ..., P gv（g） } and P g The corresponding candidate matching list Q g ={Q g1 , ..., Q gd , ..., Q gv（g） }, P gd For L g The corresponding candidate label in the candidate label list is the d-th candidate label, Q gdw P gd The corresponding candidate matching degree, d=1...v(g), v(g) is L g The number of candidate tags in the corresponding candidate tag list, wherein the candidate tags are preset tags obtained from the preset tag list based on the target model and matched with the target medical record text, and the candidate matching degree is the matching degree between the target medical record text obtained based on the target model and the preset tags; S33, according to P g and Q g , get L g The corresponding target priority L 0 g , where L 0 g Meet the following conditions: ； S4, when L 0 g ≥E 0 When L 1 g For L g The corresponding target label, where E 0 is the preset target priority threshold; S5, when L 0 g <E 0 When L 2 g For L g The corresponding target label.

2. The data processing system for obtaining labels corresponding to medical record texts according to claim 1, characterized in that: The value range of η is 0.9~0.

99.

3. The data processing system for obtaining labels corresponding to medical record texts according to claim 1, characterized in that: The first candidate label is a preset label corresponding to each second sample medical record text obtained by inputting the second sample medical record text set into the intermediate model.

4. The data processing system for obtaining labels corresponding to medical record texts according to claim 1, characterized in that: The first candidate priority is the degree to which the first candidate tag corresponding to each second sample medical record text obtained by inputting the second sample medical record text set into the intermediate model matches the preset tag in the preset tag list.

5. The data processing system for obtaining labels corresponding to medical record texts according to claim 1, characterized in that: The value range of η0 is 0.01~0.1.

Images