Adverse event extraction method and device, storage medium, and electronic device

Abstract

The present disclosure belongs to the field of medical data processing, and relates to a method and device for extracting adverse events, a storage medium, and an electronic device. The method includes: obtaining case text information corresponding to the case to be mined, and inputting the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to mark entities present in the case text information; identifying target entities corresponding to adverse events in the target entity sequence to extract adverse events present in the case to be mined. In the present disclosure, the case text information is input into a medical named entity model to obtain a target entity sequence corresponding to the case text information, so that adverse events in all cases to be mined can be comprehensively extracted, thereby avoiding the incomplete extraction caused by extracting adverse events by sampling in the prior art.

Description

Technical Field

[0001] The present disclosure relates to the field of medical data processing, and in particular to a method for extracting adverse events, an adverse event extraction device, a computer-readable storage medium, and an electronic device. Background Art

[0002] In medical research, extracting adverse events from cases and analyzing them can improve the medical quality of the hospital to a certain extent and further ensure the life and health of patients.

[0003] In the related technology, a large number of cases need to be manually reviewed in order to extract adverse events. Moreover, due to the huge amount of case data, only a large number of cases can be sampled and reviewed, and it is impossible to extract adverse events existing in all cases, which leads to the extracted adverse events being not comprehensive enough and difficult to fully reflect the medical quality of the hospital. In addition, if adverse events are to be comprehensively extracted, a large amount of manpower and time are required, which reduces the efficiency of mining adverse events.

[0004] In view of this, there is an urgent need in this field to develop a new method and device for extracting adverse events.

[0005] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0006] The purpose of the present disclosure is to provide a method for extracting adverse events, an apparatus for extracting adverse events, a computer-readable storage medium and an electronic device, thereby at least to a certain extent overcoming the problem of incomplete and inefficient extraction of adverse events caused by related technologies.

[0007] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0008] According to a first aspect of an embodiment of the present invention, a method for extracting adverse events is provided, the method comprising: obtaining case text information corresponding to a case to be mined, and inputting the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to label entities present in the case text information; and identifying a target entity corresponding to the adverse event in the target entity sequence to extract the adverse event present in the case to be mined.

[0009] In an exemplary embodiment of the present invention, the medical named entity model includes a first model and a second model; the step of inputting the case text information into the medical named entity model to obtain a target entity sequence corresponding to the case text information includes: converting the case text information into a character / word vector; inputting the character / word vector into the first model of the medical named entity model to obtain a label score corresponding to each character / word in the case text information and each entity type label; inputting the label score into the second model of the medical named entity model to obtain a target entity type label corresponding to the case text information; and identifying a target entity sequence corresponding to the case text information according to the target entity type label.

[0010] In an exemplary embodiment of the present invention, the first model includes a forward model and a reverse model; the step of inputting the character / word vector into the first model of the medical named entity model to obtain a label score corresponding to each character / word in the case text information and each entity type label, comprises: inputting the character / word vector into the forward model to obtain a first label score corresponding to each character / word in the case text information and each entity type label; inputting the character / word vector into the reverse model to obtain a second label score corresponding to each character / word in the case text information and each entity type label; and calculating the first label score and the second label score to obtain a label score corresponding to each character / word in the case text information and each entity type label.

[0011] In an exemplary embodiment of the present invention, obtaining the target annotation sequence corresponding to the case text information includes: determining all annotation sequences corresponding to the case text information, and based on the label scores, obtaining the total label scores corresponding to all the annotation sequences respectively; determining the state transfer matrix corresponding to the second model, and determining the state transfer probabilities corresponding to all the annotation sequences respectively from the state transfer matrix; wherein the state transfer matrix is ​​used to record the combination probabilities between different entity type labels; calculating the state transfer probabilities and the total label scores to obtain sequence scores corresponding to all the annotation sequences, and comparing the sequence scores to obtain a score comparison result; and determining the target annotation sequence among all the annotation sequences according to the score comparison result.

[0012] In an exemplary embodiment of the present invention, identifying the target entity corresponding to the adverse event in the target sequence includes: determining the entity type corresponding to the entity existing in the target entity sequence according to the target entity sequence; based on the entity type, converting the entity name of the entity into a standard entity name; traversing the standard entity name to identify the target entity corresponding to the adverse event.

[0013] In an exemplary embodiment of the present invention, converting the entity name of the entity into a standard entity name includes: determining a conversion rule corresponding to the entity type, and converting the entity name of the entity into a standard entity name according to the conversion rule.

[0014] In an exemplary embodiment of the present invention, traversing the standard entity names in the target entity sequence to identify the target entity corresponding to the adverse event includes: traversing the standard entity names to identify the event entity name corresponding to the adverse event; searching for the standard entity name having a positional order relationship with the event entity name to identify the time entity name and the level entity name corresponding to the same adverse event; and reorganizing the event entity name, the time entity name and the level entity name according to a preset combination order to obtain the target entity corresponding to the adverse event.

[0015] In an exemplary embodiment of the present invention, after extracting the adverse events existing in the cases to be mined, the method further includes: storing the target entities corresponding to different adverse events according to the event entity name, the time entity name and the level entity name.

[0016] According to a second aspect of an embodiment of the present invention, a device for extracting adverse events is provided, the device comprising: an input module, configured to obtain case text information corresponding to a case to be mined, and input the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to label entities present in the case text information; and an extraction module, configured to identify a target entity corresponding to the adverse event in the target entity sequence to extract the adverse event present in the case to be mined.

[0017] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a processor and a memory; wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method for extracting adverse events of any of the above exemplary embodiments is implemented.

[0018] According to a fourth aspect of an embodiment of the present invention, there is provided a computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method for extracting adverse events in any of the above exemplary embodiments is implemented.

[0019] It can be seen from the above technical solutions that the adverse event extraction method, adverse event extraction device, computer storage medium and electronic device in the exemplary embodiments of the present invention have at least the following advantages and positive effects:

[0020] In the method and apparatus provided by the exemplary embodiments of the present disclosure, case text information is input into a medical named entity model to obtain a target entity sequence corresponding to the case text information. On the one hand, the manpower and time invested in extracting adverse events in the cases to be mined are reduced, thereby increasing the efficiency of extracting adverse events. On the other hand, adverse events in all cases to be mined can be comprehensively extracted, thereby avoiding the incomplete mining caused by extracting adverse events only by random sampling in the prior art.

[0021] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0023] Figure 1 A schematic diagram schematically illustrates a flow chart of a method for extracting adverse events in an embodiment of the present disclosure;

[0024] Figure 2 A schematic diagram schematically shows a flow chart of obtaining a target annotation sequence corresponding to case text information in a method for extracting adverse events in an embodiment of the present disclosure;

[0025] Figure 3 A schematic diagram schematically shows the structure of a medical named entity model in the method for extracting adverse events in an embodiment of the present disclosure;

[0026] Figure 4 A schematic diagram schematically shows a flow chart of obtaining a label score corresponding to each character / word in case text information and each entity type label in the method for extracting adverse events in an embodiment of the present disclosure;

[0027] Figure 5A schematic diagram schematically shows a flow chart of obtaining a target annotation sequence corresponding to case text information in a method for extracting adverse events in an embodiment of the present disclosure;

[0028] Figure 6 A schematic diagram of a process of identifying a target entity corresponding to an adverse event in a target annotation sequence in a method for extracting adverse events in an embodiment of the present disclosure is schematically shown;

[0029] Figure 7 A schematic diagram schematically shows a process of identifying a target entity corresponding to an adverse event in a method for extracting adverse events in an embodiment of the present disclosure;

[0030] Figure 8 A schematic diagram schematically shows a flow chart of a method for extracting adverse events in an application scenario in an embodiment of the present disclosure;

[0031] Fig. 9 The structure diagram of the medical named entity model in the application scenario in the embodiment of the present disclosure is schematically shown;

[0032] Fig.10 A schematic diagram schematically shows the structure of a device for extracting adverse events in an embodiment of the present disclosure;

[0033] Fig.11 An electronic device for extracting adverse events in an embodiment of the present disclosure is schematically shown;

[0034] Fig.12 A computer-readable storage medium for a method for extracting adverse events in an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0035] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0036] The terms "a", "an", "the" and "said" are used in this specification to indicate the presence of one or more elements / components / etc.; the terms "including" and "having" are used to express an open-ended inclusion and mean that additional elements / components / etc. may exist in addition to the listed elements / components / etc.; the terms "first" and "second" etc. are used only as labels and are not intended to limit the quantity of their objects.

[0037] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the drawings represent the same or similar parts, and their repeated descriptions will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities.

[0038] In view of the problems existing in the related art, the present disclosure proposes a method for extracting adverse events. Figure 1 A flow chart showing the method for extracting adverse events is shown in FIG. Figure 1 As shown, the medical data processing method comprises at least the following steps:

[0039] Step S110. Obtain case text information corresponding to the case to be mined, and input the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to annotate entities present in the case text information.

[0040] Step S120. Identify the target entity corresponding to the adverse event in the target entity sequence to extract the adverse events existing in the case to be mined.

[0041] In the method and apparatus provided by the exemplary embodiments of the present disclosure, case text information is input into a medical named entity model to obtain a target entity sequence corresponding to the case text information. On the one hand, the manpower and time invested in extracting adverse events in the cases to be mined are reduced, thereby increasing the extraction efficiency of adverse events. On the other hand, adverse events in all cases to be mined can be comprehensively extracted, thereby avoiding the incomplete extraction caused by extracting adverse events only by random sampling in the prior art.

[0042] The following is a detailed description of each step of the adverse event extraction method.

[0043] In step S110, case text information corresponding to the case to be mined is obtained, and the case text information is input into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to annotate entities present in the case text information.

[0044] In an exemplary embodiment of the present disclosure, a large amount of clinical trial data is required for the extraction of adverse events. The cases to be mined refer to cases that record a large amount of clinical trial data. Based on this, the case text information refers to the clinical trial data in text form recorded in the cases to be mined.

[0045] After obtaining the case text information recorded in the case to be mined, the case text information needs to be input into the medical named entity model to obtain a target entity sequence, which means that the target entity sequence is marked with entities corresponding to the case text information.

[0046] The medical named entity model refers to a model specifically used to identify entities with specific meaning or strong referentiality in the medical field. Specifically, it can be a model obtained after training with entities existing in a large number of adverse events. Furthermore, the entities marked in the target entity sequence refer to the content with specific meaning or strong referentiality identified from the case text information, and these entities are related to adverse events. For example, the entity can be a time entity, such as December 6, 2019, the entity can also be a trigger entity, such as occurrence, the entity can also be an adverse event name entity, such as headache, the entity can also be an adverse event level entity, such as level 1, and the entity can also be any content related to adverse events, with specific meaning or strong referentiality. This exemplary embodiment does not make any special limitations on this.

[0047] For example, case x to be mined may be the case corresponding to patient A, wherein case text information is recorded in case x to be mined. Specifically, the case text information is "Patient A participated in the clinical trial on December 6, 2019. During the clinical trial, the patient experienced adverse events such as nausea and vomiting on December 20, 2019, with a level 1 rating. After treatment, the patient recovered to normal levels. On February 9, 2020, he experienced back pain with a level 1 rating and cough with a level 2 rating."

[0048] Based on this, by inputting the above case text information into the medical named entity model, the target entity sequence will be obtained. Specifically, the target entity sequence is [[December 6, 2019][December 20, 2019][nausea][vomiting][level 1][February 9, 2020][back pain][level 1][cough][level 2]], and the content in [] is the entity existing in the case text information.

[0049] In an alternative embodiment, Figure 2 FIG. 1 shows a flow chart of obtaining a target annotation sequence corresponding to case text information in an adverse event extraction method, such as Figure 2As shown, the medical named entity model includes a first model and a second model, and the method includes at least the following steps: In step S210, the case text information is converted into a character / word vector.

[0050] Among them, the medical named entity model can specifically include a first model and a second model. After the case text information is input into the medical named entity model, the case text information needs to be converted first to obtain a character vector or a word vector corresponding to the case text information.

[0051] For example, the case text information x is "Patient A participated in a clinical trial on December 6, 2019. During the clinical trial, the patient experienced adverse events such as nausea and vomiting on December 20, 2019, with a level 1, and recovered to normal levels after treatment.", and then the word vector corresponding to the case text information x can be obtained. Specifically, the number of vectors included in the word vector is the same as the number of words in the case text information x, and each vector has a one-to-one correspondence with a word in the case text information x.

[0052] In step S220, the character / word vector is input into the first model layer of the medical named entity model to obtain a label score corresponding to each character / word in the case text information and each entity type label.

[0053] Among them, each entity has a corresponding entity type, and the entity type label refers to the identifier that describes the entity type. For example, there are 4 entity types, namely entity type x1, entity type x2, entity type x3 and entity type x4, then there may be 4 entity type labels, specifically including entity type label X1 corresponding to entity type x1, entity type label X2 corresponding to entity type x2, entity type label X3 corresponding to entity type x3, and entity type label X4 corresponding to entity type x4.

[0054] The first model can be a recursive neural network, specifically a long short-term memory network in the recursive neural network, which is used to output the label score corresponding to the character / word vector and each entity type label, and the label score refers to the probability that each component in the character / word vector belongs to the entity type corresponding to the entity type label.

[0055] For example, there are three entity type labels. The word vector [w0 w1 w2] is input into the first model, and the label scores of 1.5, 0.9, and 0.1 corresponding to the word vector w0 and the three entity type labels can be obtained respectively. Similarly, the label scores of 0.2, 0.4, and 0.1 corresponding to the word vector w1 and the three entity type labels can be obtained respectively. The label scores of 0.09, 0.02, and 0.03 corresponding to the word vector w2 and the three entity type labels can be obtained respectively.

[0056] In step S230, the label score is input into the second model of the medical named entity model to obtain the target entity type label corresponding to the case text information; wherein the second model is used to determine the target entity type label corresponding to each character / word based on the combination relationship between the entity type labels.

[0057] Among them, although the character / word vector is input into the first model, the label score corresponding to each character / word and each entity type label has been obtained, but the combination relationship between different entity type labels is not considered, that is, the combination relationship between different entity types is not considered. For example, according to the label score, it can be determined that the entity type label corresponding to the first component in the character / word vector is the label corresponding to the trigger word entity, and the entity type label corresponding to the second component in the character / word vector is still the label corresponding to the trigger word entity. Obviously, in practice, the entity after the trigger word entity should not still be the trigger word entity, and it is inaccurate to determine the entity type of each component only by the label score.

[0058] Based on this, the label scores need to be input into the second model to analyze the combination relationship between the description entity types that can be combined based on the first model, and then determine a more accurate target annotation sequence corresponding to the case text information.

[0059] For example, Figure 3 The schematic diagram of the structure of the medical named entity model is shown in FIG. Figure 3 As shown, information 310 is the character / word vector obtained after converting the case text information, model 320 is the first model, after passing through the first model, the label scores corresponding to each component in the character / word vector and each entity type label are obtained, model 330 is the second model, after the label scores are input into the second model 330, the target entity type label 340 corresponding to each component in the character / word vector will be determined among all entity type labels, and then the target annotation sequence corresponding to the case text information composed of the entity type labels 340 is obtained.

[0060] In step S240, a target entity sequence corresponding to the case text information is identified based on the target entity type label.

[0061] Among them, the target annotation sequence records the entity label type corresponding to each word / character in the case text information. The entity corresponding to the case text information can be identified through the entity label type, and the sequence composed of these entities is the target entity sequence.

[0062] For example, the determined target annotation sequence is [AABCO]. Assuming that O represents something that cannot be an entity, it can be determined that the target entity sequence corresponding to the target annotation sequence does not include the word in the case text information corresponding to O.

[0063] In this exemplary embodiment, the medical named entity model includes a first model and a second model. The first model is used to determine the label score corresponding to each component and the entity type label, and the second model is used to determine the target entity type label corresponding to each component based on the relationship between the entity type labels, thereby avoiding the situation where the combination relationship between the entity type labels is not considered, resulting in an incorrect target entity sequence.

[0064] In an alternative embodiment, Figure 4 The flowchart of the adverse event extraction method for obtaining the label scores of each word / term in the case text information and each entity type label is shown. The first model layer includes a forward model and a reverse model, such as Figure 4 As shown, the method includes at least the following steps: In step S410, the character / word vector is input into the forward model to obtain a first label score corresponding to each character / word in the case text information and each entity type label.

[0065] Among them, the first model specifically includes a forward model and a reverse model. After the case text information is input into the first model of the medical named entity model, the case text information is first converted into a character / word vector, and then the character / word vector is input into the forward model to obtain a first label score. The first label score can be used to determine the possibility that each character / word in the case text information belongs to each entity type.

[0066] For example, input the word vector [w0 w1 w2] into the forward model of the first model. Assuming that there are three entity type labels, namely X1, X2 and X3, we can get the first label score of 0.8 for the word in the case text information xx corresponding to the vector w0 and X1, the first label score of 0.05 for the word in the case text information xx corresponding to the vector w0 and X2, the first label score of 0.03 for the word in the case text information xx corresponding to the vector w0 and X3, and the first label score of 0.1 for the word in the case text information xx corresponding to the vector w1 and X1. , we can also get the first label score 0.02 corresponding to the word in the case text information xx corresponding to the vector w1 and X2, we can also get the first label score 0.05 corresponding to the word in the case text information xx corresponding to the vector w1 and X3, we can also get the first label score 0.5 corresponding to the word in the case text information xx corresponding to the vector w2 and X1, we can also get the first label score 0.06 corresponding to the word in the case text information xx corresponding to the vector w2 and X2, we can also get the first label score 0.03 corresponding to the word in the case text information xx corresponding to the vector w2 and X3.

[0067] In step S420, the character / word vector is input into the reverse model to obtain a second label score corresponding to each character / word in the case text information and each entity type label.

[0068] Among them, the order in which the character / word vectors are input into the reverse model is opposite to the order in which the character / word vectors are input into the forward model. If the first label score output by the forward model is used as the label score corresponding to each character / word and the entity type label, the accuracy of the label score may be reduced. Therefore, adding a reverse model on the basis of the forward model can further improve the accuracy of the final determined label score corresponding to each character / word and the entity type label.

[0069] The second label score is the probability that each character / word in the case text information belongs to each entity type determined by the reverse model.

[0070] For example, input the word vector [w0 w1 w2] into the reverse model of the first model. Assuming that there are three entity type labels, namely X1, X2 and X3, we can get the second label score of 0.7 for the word in the case text information xx corresponding to the vector w0 and X1, the second label score of 0.03 for the word in the case text information xx corresponding to the vector w0 and X2, the second label score of 0.02 for the word in the case text information xx corresponding to the vector w0 and X3, and the second label score of 0.05 for the word in the case text information xx corresponding to the vector w1 and X1. , we can also get the second label score 0.01 corresponding to the word in the case text information xx corresponding to the vector w1 and X2, we can also get the second label score 0.03 corresponding to the word in the case text information xx corresponding to the vector w1 and X3, we can also get the second label score 0.7 corresponding to the word in the case text information xx corresponding to the vector w2 and X1, we can also get the second label score 0.02 corresponding to the word in the case text information xx corresponding to the vector w2 and X2, we can also get the second label score 0.05 corresponding to the word in the case text information xx corresponding to the vector w2 and X3.

[0071] In step S430, the first label score and the second label score are calculated to obtain the label score corresponding to each character / word in the case text information and each entity type label.

[0072] Among them, after obtaining the first label score and the second label score, the first label score and the second label score can be calculated to obtain the label scores corresponding to each word / term in the case text information and each entity type label. Specifically, the first label score and the second label score can be averaged, or corresponding weight values ​​can be assigned to the first label score and the second label score, and then the first label score and the second label score can be calculated according to their respective weight values. The first label score and the second label score can also be calculated using a certain algorithm, or the first label score and the second label score can be calculated in any way. This exemplary embodiment does not make any special limitation on this.

[0073] For example, obtain the first label score corresponding to each word in the case text information and the entity type label, and obtain the second label score corresponding to each word in the case text information and the entity type label, then calculate the average of the first label score and the second label score, and determine the calculation result as the label score corresponding to each word in the case text information and each entity type label.

[0074] In this exemplary embodiment, the first model includes a forward model and a reverse model, and the label score is calculated by calculating the first label score output by the forward model and the second label score output by the reverse model, thereby improving the accuracy of the label score.

[0075] In an alternative embodiment, Figure 5 FIG. 1 shows a flow chart of obtaining a target annotation sequence corresponding to case text information in an adverse event extraction method, such as Figure 5 As shown, the method includes at least the following steps: In step S510, all the annotation sequences corresponding to the case text information are determined, and based on the label scores, the total label scores corresponding to all the annotation sequences are obtained.

[0076] Among them, assuming that there are three entity type labels, specifically, the three entity type labels are X1, X2 and X3, then each character in the case text information or each word in the case text information may belong to the entity type label X1, may also belong to the entity type label X2, or may also belong to the entity type label X3, and then according to the situation that each character in the case text information or each word in the case text information belongs to the three entity type labels, different annotation sequences can be formed.

[0077] Assuming that one of all the label sequences is [AAB], the total label score is the sum of the label scores of the first character in the case text information or the first word in the case text information belonging to the first entity type label, the label score of the second character in the case text information or the second word in the case text information belonging to the first entity type label, and the label score of the third character in the case text information or the third word in the case text information belonging to the second entity type label.

[0078] In step S520, a state transfer matrix corresponding to the second model is determined, and state transfer probabilities corresponding to all label sequences are determined from the state transfer matrix; wherein the state transfer matrix is ​​used to record the combination probabilities between different entity type labels.

[0079] Among them, the first model in the medical named entity model only determines the possibility that each word / character in the case text information belongs to each entity type label, but does not consider the combination relationship between entity type labels, that is, the combination relationship between different entity types is not considered. The state transfer matrix corresponding to the second model records the combination probability between different entity type labels. Assuming the annotation sequence is [AAB], the probability that the i position is an A entity type label and the i+1 position is still an A entity type label can be determined from the state transfer matrix. The probability that the i position is converted to an A entity type label and the i+1 position is a B entity type label can also be determined from the state transfer matrix.

[0080] For example, all annotation sequences corresponding to the case text information include [AAA], [AAB], [AB A], [ABB], [B AA], [BBA], [BBB] and [B AB]. According to the label scores, the total label scores corresponding to the above 7 annotation sequences can be obtained. Specifically, taking [AA A] as an example, the total label score corresponding to [AAA] is the sum of the label score corresponding to the first word in the case text information and the A entity type label, the label score corresponding to the second word in the case text information and the A entity type label, and the label score corresponding to the third word in the case text information and the A entity type label.

[0081] In step S530, the state transition probability and the total label score are calculated to obtain sequence scores corresponding to all labeled sequences, and the sequence scores are compared to obtain a score comparison result.

[0082] Among them, the sequence score refers to the score obtained after calculating the state transition probability and the total label score. Specifically, the sum of the total label score and the state transition probability corresponding to the entity sequence can be used as the sequence score corresponding to the labeled sequence. Based on this, the sequence score corresponding to each labeled sequence can be obtained. The score comparison result is the result obtained after comparing these sequence scores.

[0083] For example, there are 4 labeled sequences, and according to the total label scores and state transition probabilities corresponding to these 4 labeled sequences, 4 sequence scores can be obtained, among which the sequence score 2.3 corresponds to the first labeled sequence, the sequence score 1.5 corresponds to the second labeled sequence, the sequence score 1.7 corresponds to the third labeled sequence, and the sequence score 2.5 corresponds to the fourth labeled sequence.

[0084] Based on this, the score comparison result obtained by comparing the sequence scores is that the sequence score corresponding to the fourth labeled sequence is greater than the sequence score corresponding to the first labeled sequence, the sequence score corresponding to the first labeled sequence is greater than the sequence score corresponding to the second labeled sequence, and the sequence score corresponding to the second labeled sequence is greater than the sequence score corresponding to the third labeled sequence.

[0085] In step S540 , a target entity sequence is determined from all the annotation sequences according to the score comparison result.

[0086] Among them, the target annotation sequence is the annotation sequence with the largest sequence score.

[0087] For example, the score comparison result obtained by comparing the sequence scores is that the sequence score corresponding to the fourth labeled sequence is greater than the sequence score corresponding to the first labeled sequence, the sequence score corresponding to the first labeled sequence is greater than the sequence score corresponding to the second labeled sequence, and the sequence score corresponding to the second labeled sequence is greater than the sequence score corresponding to the third labeled sequence. Based on this, the target labeled sequence is the first labeled sequence.

[0088] In this exemplary embodiment, the sequence score is obtained by calculating the total label score and the state transition probability. Since the state transition matrix records the combination probability between entity type labels, the relationship between one entity type and another entity type needs to be considered in the process of determining the target labeling sequence, which avoids the obtained target labeling sequence from not conforming to the actual language regulations and increases the accuracy of the target labeling sequence.

[0089] In step S120, target entities corresponding to adverse events are identified in the target entity sequence to extract adverse events present in the cases to be mined.

[0090] In an exemplary embodiment of the present disclosure, target entities corresponding to adverse events in case text information are identified based on target entity sequences, and then adverse events existing in the cases to be mined are extracted.

[0091] For example, the target entity sequence is [[2019-01-01][Vomiting][1][Level][2020-02-02][Surgery]], where the entity type of vomiting is the adverse event name entity. It can be identified that the target entity composed of the adverse events in the case to be mined is [[2019-01-01][Vomiting][1][Level]].

[0092] In an alternative embodiment, Figure 6 FIG. 1 shows a flow chart of identifying target entities corresponding to adverse events in a target annotation sequence in an adverse event extraction method, such as Figure 6 As shown, the method at least includes the following steps: In step S610, according to the target entity sequence, an entity type corresponding to an entity existing in the target entity sequence is determined.

[0093] The target entity sequence is composed of entities, specifically, it can be composed of multiple entities belonging to different entity types, and each entity type has a corresponding conversion rule. The entities belonging to the same entity type can be converted into entities of a unified format through the conversion rule.

[0094] For example, there are entities corresponding to the entity type of time entity, December 6 and 7, 2019, in the target entity sequence, and then the conversion rules corresponding to the time entity are determined. Specifically, the conversion rule is to convert the year and month to -. If "," is recognized, a new time node is generated.

[0095] For example, if the target entity sequence is [[2019-01-01][Vomiting][1][Level][2020-02-02][Surgery]], the entities that exist include 2019-01-01 with entity type as time entity, vomiting with entity type as trigger word entity, 1 and level with entity type as level entity, 2020-02-02 with entity type as time entity, and surgery with entity type as bad time name entity.

[0096] In step S620, based on the entity type, the entity name of the entity is converted into a standard entity name.

[0097] Among them, according to the conversion rules, the entity name of the entity can be converted into a standard entity name with a unified standard.

[0098] For example, Table 1 shows that the entity name of an entity is converted into a corresponding standard entity name according to the conversion rule.

[0099] Table 1 Entity names and standard entity names

[0100]

[0101] Based on this, if the target entity sequence is [[December 6, 2019][December 20, 2019][back pain][level 1][January 1, 2020][dizziness][level 2]], the entity names in the target entity sequence can be converted into standard entity names according to the conversion rules, and the converted target entity sequence is [[2019-12-6][2019-12-20][back pain][level 1][2020-1-1][dizziness][level 2]].

[0102] In step S630, the standard entity names are traversed to identify the target entity corresponding to the adverse event.

[0103] Among them, the standard entity names in the target entity sequence are traversed, and then the target entity corresponding to the adverse event can be identified.

[0104] For example, after converting the entity name of the entity into a standard entity name, the target entity sequence obtained is [[2019-12-6][2019-12-20][back pain][level 1][2020-1-1][dizziness][level 2]], wherein the entity types corresponding to back pain and dizziness are adverse event name entities, and then the standardized entity name belonging to the time entity type corresponding to the adverse event name entity and the standardized entity name belonging to the adverse event level entity type corresponding to the adverse event name entity are identified to obtain the target entity corresponding to the adverse event. Specifically, the target entity is [[2019-12-20][back pain][level 1], [2020-1-1][dizziness][level 2]], and the target entity can also be [[back pain][<2019-12-20]>, <level 1>], [dizziness][<2020-1-1]>, <level 2>].

[0105] In this exemplary embodiment, the entity type corresponding to the entity in the target entity sequence is determined, and then the entity name of the entity is converted into a standard entity name, which reduces the complexity of subsequent identification of the target entity corresponding to the adverse event and improves the efficiency of identifying the target entity.

[0106] In an optional embodiment, converting the entity name of the entity into a standard entity name includes: determining a conversion rule corresponding to the entity type, and converting the entity name of the entity into the standard entity name according to the conversion rule.

[0107] The conversion rule refers to a rule corresponding to the entity type and used to convert the entity name into a standard entity name. Then, according to the conversion rule, the entity name of the entity can be converted into a standard entity name with a unified standard.

[0108] For example, Table 1 shows that the entity name of an entity is converted into a corresponding standard entity name according to the conversion rule.

[0109] In this exemplary embodiment, the entity name of the entity is converted into a standard entity name according to the standard conversion rule, which reduces the complexity of subsequent identification of the target entity corresponding to the adverse event and improves the efficiency of identifying the target entity.

[0110] In an alternative embodiment, Figure 7 A schematic diagram of the process of identifying the target entity corresponding to the adverse event in the adverse event extraction method is shown, such as Figure 7 As shown, the method at least includes the following steps: In step S710, traverse the standard entity names and identify the event entity names corresponding to the adverse events.

[0111] The event entity name is the entity name corresponding to the entity type of the adverse event name entity.

[0112] For example, if the target entity sequence is [[2019-12-6][2019-12-20][back pain][level 1][2020-1-1][dizziness][level 2]], the event entity names corresponding to the identified adverse events are back pain and dizziness.

[0113] In step S720, a standard entity name having a positional order relationship with the event entity name is searched to identify a time entity name and a level entity name respectively corresponding to the same adverse event.

[0114] Among them, taking the event entity name as the center, searching forward or backward for a standard entity name that has a positional order relationship with the event entity name, and then determining the time entity name and the level entity name corresponding to the event entity name, wherein the time entity name is the entity name belonging to the entity type of the time entity, and the level entity name is the entity name belonging to the entity type of the adverse event level entity.

[0115] For example, if the target entity sequence is [[2019-12-06][2019-12-20][back pain][level 1][2020-01-01][dizziness][level 2]], the event entity names corresponding to the identified adverse events are back pain and dizziness.

[0116] Based on this, the standard entity names that have a positional order relationship with the event entity name of back pain are found to be 2019-12-20 and Level 1. Obviously, 2019-12-20 belongs to the entity type of time entity, and Level 1 belongs to the entity type of adverse event level entity. Therefore, it can be determined that 2019-12-20 is the time entity name corresponding to the adverse event of back pain, and Level 1 is the level entity name corresponding to the adverse event of back pain. Similarly, 2020-01-01 is the time entity name corresponding to the adverse event of dizziness, and Level 2 is the level entity name corresponding to the adverse event of dizziness.

[0117] In step S730, the event entity name, the time entity name and the level entity name are reorganized according to a preset combination order to obtain a target entity corresponding to the adverse event.

[0118] Among them, the preset combination order is a pre-specified combination order of time entity names, event entity names and level entity names. Specifically, the preset combination order can be the time entity name, event entity name, level entity name, or the event entity name, time entity name, level entity name, or any combination order. This exemplary embodiment does not make any special limitation to this.

[0119] For example, the standard entity names that have a positional order relationship with the event entity name of back pain are found to be 2019-12-20 and level 1. Obviously, 2019-12-20 belongs to the entity type of time entity, and level 1 belongs to the entity type of adverse event level entity. Therefore, it can be determined that 2019-12-20 is the time entity name corresponding to the adverse event of back pain, and level 1 is the level entity name corresponding to the adverse event of back pain. Similarly, 2020-01-01 is the time entity name corresponding to the adverse event of dizziness, and level 2 is the level entity name corresponding to the adverse event of dizziness.

[0120] The preset combination order is event entity name, time entity name, level entity name. The target entity obtained by reorganizing the event entity name, time entity name and level entity name according to the preset combination order is [[back pain][2019-12-20][level 1], [dizziness][2020-01-01][level 2]].

[0121] In this exemplary embodiment, on the one hand, the event entity name is first identified, and then the time entity name and the level entity name that have a positional order relationship with the event entity name and belong to the same adverse event are found, thereby ensuring that the found time entity, event entity and level entity belong to the same adverse event; on the other hand, the event entity name, time entity name and level entity name are reorganized in a preset combination order to obtain the target entity corresponding to the adverse event, thereby improving the readability of the target entity.

[0122] In an optional embodiment, after extracting the adverse events present in the cases to be mined, the method further includes: storing the target entities corresponding to different adverse events according to the event entity name, time entity name and level entity name.

[0123] Among them, after obtaining the target entity, the target entity can be stored according to the event entity name, time entity name and level entity name. Specifically, the target entity can be stored in a table, and each row of the table corresponds to a different adverse event, and each column in the table stores the event entity name, time entity name and level entity name. The target entity can also be stored on a server according to the event entity name, time entity name and level entity name for extraction.

[0124] In this exemplary embodiment, the target entities corresponding to the adverse events are stored according to the event entity name, time entity name and level entity name. On the one hand, this helps to subsequently extract the target entity corresponding to a certain adverse event; on the other hand, different entity names corresponding to different adverse events can also be extracted according to finer dimensions, thereby improving the efficiency of subsequently extracting the target entity corresponding to the adverse event or extracting a certain entity name as needed.

[0125] In the method and apparatus provided by the exemplary embodiments of the present disclosure, case text information is input into a medical named entity model to obtain a target entity sequence corresponding to the case text information. On the one hand, the manpower and time invested in extracting adverse events in the cases to be mined are reduced, thereby increasing the efficiency of extracting adverse events. On the other hand, adverse events in all cases to be mined can be comprehensively extracted, thereby avoiding the incomplete mining caused by mining adverse events only by random sampling in the prior art.

[0126] The following is a detailed description of the method for extracting adverse events in the embodiment of the present disclosure in conjunction with an application scenario.

[0127] Figure 8 A flow chart of the method for extracting adverse events in an application scenario is shown, such as Figure 8 As shown, information 810 is case text information corresponding to the case to be mined, model 820 is a medical named entity model, sequence 830 is a target annotation sequence, sequence 840 is a target entity sequence, and sequence 850 is a target entity corresponding to an adverse event obtained by converting the entity names in the target entity sequence into standard entity names and reorganizing the event entity names, time entity names and level entity names in the target entity sequence in a preset combination order.

[0128] Fig. 9 The schematic diagram of the structure of the medical named entity model is shown in FIG. Fig. 9 As shown, vector 910 is the character / word vector obtained after converting the case text information, model 920 is the forward model in the first model, model 930 is the reverse model in the first model, value 940 is the label score corresponding to each character / word and each entity type label, model layer 950 is the second model layer, and the target annotation sequence is a sequence composed of target entity type labels 960.

[0129] In this application scenario, the case text information is input into the medical named entity model to obtain the target entity sequence corresponding to the case text information. On the one hand, it reduces the manpower and time invested in extracting adverse events in the cases to be mined, and increases the extraction efficiency of adverse events; on the other hand, it can comprehensively extract adverse events in all cases to be mined, avoiding the incomplete extraction caused by extracting adverse events only by random sampling in the existing technology.

[0130] In addition, in an exemplary embodiment of the present disclosure, a device for extracting adverse events is also provided. Fig.10 A schematic diagram of the structure of an adverse event extraction device is shown, such as Fig.10 As shown, the adverse event extraction device 1000 may include: an input module 1010 and an extraction module 1020 .

[0131] in:

[0132] The input module 1010 is configured to obtain case text information corresponding to the case to be mined, and input the case text information into the medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to mark entities existing in the case text information; the extraction module 1020 is configured to identify target entities corresponding to adverse events in the target entity sequence to extract adverse events existing in the case to be mined.

[0133] The specific details of the adverse event extraction device 1000 have been described in detail in the corresponding adverse event extraction method, so they will not be repeated here.

[0134] It should be noted that, although several modules or units of the extraction device 1000 for adverse events are mentioned in the above detailed description, such division is not mandatory. In fact, according to an embodiment of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0135] In addition, in an exemplary embodiment of the present disclosure, an electronic device capable of implementing the above method is also provided.

[0136] Refer to the following Fig.11 1100 according to this embodiment of the present invention will be described. Fig.11 The electronic device 1100 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.

[0137] like Fig.11As shown, the electronic device 1100 is in the form of a general computing device. The components of the electronic device 1100 may include, but are not limited to: the at least one processing unit 1110, the at least one storage unit 1120, a bus 1130 connecting different system components (including the storage unit 1120 and the processing unit 1110), and a display unit 1140.

[0138] The storage unit stores program codes, which can be executed by the processing unit 1110, so that the processing unit 1110 executes the steps according to various exemplary embodiments of the present invention described in the above “Exemplary Method” section of this specification.

[0139] The storage unit 1120 may include a readable medium in the form of a volatile storage unit, such as a random access memory unit (RAM) 1121 and / or a cache memory unit 1122 , and may further include a read-only memory unit (ROM) 1123 .

[0140] The storage unit 1120 may also include a program / utility 1124 having a set (at least one) of program modules 1125, such program modules 1125 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 the reality of a network environment.

[0141] Bus 1130 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0142] The electronic device 1100 may also communicate with one or more external devices 1170 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1100, and / or communicate with any device that enables the electronic device 1100 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1150. Furthermore, the electronic device 1100 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) via a network adapter 1160. As shown, the network adapter 1160 communicates with other modules of the electronic device 1100 via a bus 1130. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1100, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAI systems, tape drives, and data backup storage systems.

[0143] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the embodiment of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the embodiment of the present disclosure.

[0144] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a program product capable of implementing the above method of the present specification is stored. In some possible embodiments, various aspects of the present invention can also be implemented in the form of a program product, which includes a program code, and when the program product is run on a terminal device, the program code is used to enable the terminal device to perform the steps according to various exemplary embodiments of the present invention described in the above "Exemplary Method" section of the present specification.

[0145] refer to Fig.12 As shown, a program product 1200 for implementing the above method according to an embodiment of the present invention is described, which can adopt a portable compact disk read-only memory (CD-ROM) and include program code, and can be run on a terminal device, such as a personal computer. However, the program product of the present invention is not limited thereto. In this document, a readable storage medium can be any tangible medium containing or storing a program, which can be used by or in combination with an instruction execution system, an apparatus or a device.

[0146] The program product may use any combination of one or more readable media. The readable medium may be a readable signal medium or a readable storage medium. The readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples (non-exhaustive list) of readable storage media include: an electrical connection with one or more wires, a portable 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.

[0147] Computer readable signal media may include data signals propagated in baseband or as part of a carrier wave, in which readable program code is carried. 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. Readable signal media may also be any readable medium other than a 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.

[0148] The program code embodied on the readable medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0149] Program code for performing the operations of the present invention may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In cases involving a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., through the Internet using an Internet service provider).

[0150] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The specification and examples are to be considered as exemplary only, and the true scope and spirit of the present disclosure are indicated by the claims.

Claims

1. A method for extracting adverse events, characterized in that: The method comprises: Acquire case text information corresponding to the case to be mined, and input the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to annotate entities present in the case text information; the entities annotated in the target entity sequence refer to content with specific meaning or strong referentiality identified from the case text information, and the entities are related to adverse events; Identifying a target entity corresponding to the adverse event in the target entity sequence to extract the adverse event present in the case to be mined; The step of identifying a target entity corresponding to the adverse event in the target entity sequence comprises: According to the target entity sequence, an entity type corresponding to the entity in the target entity sequence is determined; each entity type has a corresponding conversion rule, and the conversion rule is used to convert entities belonging to the same entity type into entities of a unified format; Based on the entity type, the entity name of the entity is converted into a standard entity name; the converting the entity name of the entity into the standard entity name includes: determining a conversion rule corresponding to the entity type, and converting the entity name of the entity into the standard entity name according to the conversion rule; Traversing the standard entity names to identify a target entity corresponding to the adverse event; The traversing the standard entity names to identify a target entity corresponding to the adverse event includes: Traversing the standard entity names, and identifying the event entity name corresponding to the adverse event; Searching for the standard entity name having a positional order relationship with the event entity name to identify the time entity name and the level entity name respectively corresponding to the same adverse event; The event entity name, the time entity name and the level entity name are reorganized according to a preset combination order to obtain a target entity corresponding to the adverse event.

2. The method for extracting adverse events according to claim 1, characterized in that: The medical named entity model includes a first model and a second model; The step of inputting the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information includes: Convert the case text information into word vectors; Inputting the character / word vector into the first model of the medical named entity model to obtain a label score corresponding to each character / word in the case text information and each entity type label; Inputting the label score into the second model of the medical named entity model to obtain a target entity type label corresponding to the case text information; wherein the second model is used to determine the target entity type label corresponding to each of the characters / words according to the combination relationship between the entity type labels; According to the target entity type label, a target entity sequence corresponding to the case text information is identified.

3. The method for extracting adverse events according to claim 2, characterized in that: The first model includes a forward model and a reverse model; The step of inputting the character / word vector into the first model of the medical named entity model to obtain a label score corresponding to each character / word in the case text information and each entity type label, comprises: Inputting the character / word vector into the forward model to obtain a first label score corresponding to each character / word in the case text information and each entity type label; Inputting the character / word vector into the reverse model to obtain a second label score corresponding to each character / word in the case text information and each entity type label; The first label score and the second label score are calculated to obtain label scores corresponding to each character / word in the case text information and each entity type label.

4. The method for extracting adverse events according to claim 2, characterized in that: The step of obtaining a target entity sequence corresponding to the case text information includes: Determine all entity sequences corresponding to the case text information, and based on the label scores, obtain total label scores corresponding to all the entity sequences respectively; Determine a state transfer matrix corresponding to the second model, and determine the state transfer probabilities corresponding to all the entity sequences respectively from the state transfer matrix; wherein the state transfer matrix is ​​used to record the combination probabilities between different entity type labels; Calculating the state transition probability and the total label score to obtain sequence scores corresponding to all the entity sequences, and comparing the sequence scores to obtain a score comparison result; According to the score comparison result, a target entity sequence is determined from all the entity sequences.

5. The method for extracting adverse events according to claim 1, characterized in that: After extracting the adverse events present in the cases to be mined, the method further comprises: The target entities corresponding to different adverse events are stored according to the event entity name, the time entity name and the level entity name.

6. A device for extracting adverse events, characterized in that: include: An input module is configured to obtain case text information corresponding to the case to be mined, and input the case text information into a medical named entity model to obtain a target entity sequence corresponding to the case text information; wherein the target entity sequence is used to annotate entities present in the case text information; the entities annotated in the target entity sequence refer to content with specific meaning or strong referentiality identified from the case text information, and the entities are related to adverse events; An extraction module is configured to identify a target entity corresponding to the adverse event in the target entity sequence, so as to extract the adverse event existing in the case to be mined; The extraction module is configured as follows: According to the target entity sequence, an entity type corresponding to the entity in the target entity sequence is determined; each entity type has a corresponding conversion rule, and the conversion rule is used to convert entities belonging to the same entity type into entities of a unified format; Based on the entity type, the entity name of the entity is converted into a standard entity name; the converting of the entity name of the entity into the standard entity name is configured to: determine a conversion rule corresponding to the entity type, and convert the entity name of the entity into the standard entity name according to the conversion rule; Traversing the standard entity names to identify a target entity corresponding to the adverse event; The traversing the standard entity names to identify the target entity corresponding to the adverse event is configured as follows: Traversing the standard entity names, and identifying the event entity name corresponding to the adverse event; Searching for the standard entity name having a positional order relationship with the event entity name to identify the time entity name and the level entity name respectively corresponding to the same adverse event; The event entity name, the time entity name and the level entity name are reorganized according to a preset combination order to obtain a target entity corresponding to the adverse event.

7. An electronic device, characterized in that: include: processor; A memory, configured to store executable instructions of the processor; The processor is configured to execute the adverse event extraction method according to any one of claims 1 to 5 by executing the executable instructions.

8. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the method for extracting adverse events according to any one of claims 1 to 5 is implemented.

Images