Inquiry information generation method and device, equipment and storage medium

Abstract

The application relates to the fields of artificial intelligence and digital medical treatment, and discloses a method and device for generating diagnosis information, equipment and a storage medium, the method comprises the following steps: acquiring a first symptom keyword and a first time; determining a second symptom keyword and a second time; determining the time relationship between the first time and the second time; and generating corresponding target diagnosis information based on the first symptom keyword, the second symptom keyword, the first time, the second time and the time relationship. In the foregoing manner, the embodiment determines the second symptom keyword and the second time with the highest correlation according to the first symptom keyword and the first time by using an information generation model, and then infers the user's symptoms and duration according to the time relationship between the first time and the second time, thereby solving the problem that the diagnosis information is inaccurate due to the fact that only a single symptom can be solved in a traditional medical treatment scenario and improving the accuracy of diagnosis information in robot-assisted diagnosis in the field of digital medical treatment.

Description

Technical Field

[0001] This application relates to the fields of artificial intelligence technology and digital medical technology, and in particular to a method, apparatus, device and storage medium for generating consultation information. Background Technology

[0002] With the development of artificial intelligence, machine learning technology is being successfully applied in more and more scenarios in the field of digital healthcare, such as telemedicine, intelligent diagnosis, and robot consultation.

[0003] In the field of digital healthcare, robot-assisted diagnosis is a common business scenario. Robots collect patient information by conversing with patients, replacing human experts. In this process, correctly understanding the semantic information of each interaction is crucial. Although current NLU (Natural Language Understanding) basic tasks such as named entity recognition have made some progress and basically meet industrial applications, there is still a gap in low-fault-tolerance application scenarios, especially in fields such as precision medicine.

[0004] In task-oriented multi-turn dialogue systems, slot-filling methods are generally used to drive the dialogue task. In traditional Chinese medicine scenarios, the main task of consultation is to collect patient symptom information, which includes continuous slots related to symptom duration. Traditional methods generally treat continuous slot values ​​as a special entity and use NER (Named Entity Recognition) technology for direct annotation, training, and recognition. However, this method can only solve scenarios where the sentence contains explicit slot values, and cannot solve scenarios where slot values ​​require simple reasoning to obtain. Therefore, how to improve the accuracy of consultation information in robot-assisted consultations in the field of digital healthcare has become an urgent technical problem to be solved. Summary of the Invention

[0005] This application provides a method, apparatus, device, and storage medium for generating consultation information, in order to improve the accuracy of consultation information in robot-assisted consultations in the field of digital healthcare.

[0006] Firstly, this application provides a method for generating consultation information, the method comprising:

[0007] Obtain the initial consultation information of the target user, and obtain the first symptom keyword and the first time corresponding to the first symptom keyword based on the initial consultation information;

[0008] Using a pre-trained information generation model, determine the second symptom keyword that is most relevant to the first symptom keyword and the second time corresponding to the second symptom keyword;

[0009] Determine the temporal relationship between the first time and the second time;

[0010] Based on the first symptom keywords, the second symptom keywords, the first time, the second time, and the time relationship, target consultation information corresponding to the target user is generated.

[0011] Secondly, this application also provides a medical consultation information generation device, the device comprising:

[0012] The first symptom keyword acquisition module is used to acquire the initial consultation information of the target user, and acquire the first symptom keyword and the first time corresponding to the first symptom keyword based on the initial consultation information;

[0013] The second symptom keyword acquisition module is used to determine the second symptom keyword that is most relevant to the first symptom keyword and the second time corresponding to the second symptom keyword by using a pre-trained information generation model.

[0014] A time relationship determination module is used to determine the time relationship between the first time and the second time;

[0015] The target consultation information generation module is used to generate target consultation information corresponding to the target user based on the first symptom keywords, the second symptom keywords, the first time, the second time, and the time relationship.

[0016] Thirdly, this application also provides an apparatus comprising a memory and a processor; the memory is used to store a computer program; the processor is used to execute the computer program and, when executing the computer program, implement the consultation information generation method as described above.

[0017] Fourthly, this application also provides a computer-readable storage medium storing a computer program that, when executed by a processor, causes the processor to implement the above-described method for generating consultation information.

[0018] This application discloses a method, apparatus, device, and storage medium for generating consultation information. The method includes acquiring initial consultation information of a target user, and acquiring a first symptom keyword and a first time corresponding to the first symptom keyword based on the initial consultation information; determining a second symptom keyword with the highest relevance to the first symptom keyword and a second time corresponding to the second symptom keyword through a pre-trained information generation model; determining the temporal relationship between the first time and the second time; and generating target consultation information corresponding to the target user based on the first symptom keyword, the second symptom keyword, the first time, the second time, and the temporal relationship. Through the above method, this embodiment, using an information generation model, determines the second symptom keyword with the highest relevance and the second time based on the first symptom keyword and the first time, and then infers the user's symptoms and duration based on the temporal relationship between the first time and the second time. This solves the problem of inaccurate consultation information caused by only being able to address a single symptom in traditional medical scenarios, and improves the accuracy of consultation information in robot-assisted consultations in the digital healthcare field. Attached Figure Description

[0019] To more clearly illustrate the technical solutions of the embodiments of this application, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0020] Figure 1 A schematic flowchart illustrating a method for generating consultation information provided in the first embodiment of this application;

[0021] Figure 2 A schematic flowchart illustrating a method for generating consultation information provided in the second embodiment of this application;

[0022] Figure 3 A schematic flowchart illustrating a method for generating consultation information according to a third embodiment of this application;

[0023] Figure 4 A schematic flowchart illustrating a method for generating consultation information according to the fourth embodiment of this application;

[0024] Figure 5 This is a schematic diagram illustrating the principle of the preset information generation model architecture provided in the embodiments of this application;

[0025] Figure 6 A schematic flowchart illustrating a method for generating consultation information according to the fifth embodiment of this application;

[0026] Figure 7A schematic block diagram of a medical consultation information generation device provided for embodiments of this application;

[0027] Figure 8 A schematic block diagram of the structure of a device provided for an embodiment of this application. Detailed Implementation

[0028] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0029] The flowchart shown in the attached diagram is for illustrative purposes only and does not necessarily include all content and operations / steps, nor does it necessarily have to be performed in the described order. For example, some operations / steps can be broken down, combined, or partially merged, so the actual execution order may change depending on the actual situation.

[0030] It should be understood that the terminology used in this specification is for the purpose of describing particular embodiments only and is not intended to limit the scope of the application. As used in this specification and the appended claims, the singular forms “a,” “an,” and “the” are intended to include the plural forms unless the context clearly indicates otherwise.

[0031] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0032] Embodiments of this application provide a method, apparatus, device, and storage medium for generating consultation information. The consultation information generation method can be applied to a server. Through an information generation model, based on a first symptom keyword and a first time, it determines the most relevant second symptom keyword and second time. Then, based on the temporal relationship between the first and second times, it infers the user's symptoms and duration, improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field. The server can be a standalone server or a server cluster.

[0033] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Unless otherwise specified, the following embodiments and features can be combined with each other.

[0034] Please see Figure 1 , Figure 1This is a schematic flowchart illustrating a method for generating consultation information according to the first embodiment of this application. This method can be applied to a server and, through an information generation model, determines the most relevant second symptom keyword and second time based on a first symptom keyword and a first time. Then, based on the temporal relationship between the first and second times, it infers the user's symptoms and duration. This solves the problem of traditional medical scenarios failing to reflect the correlation between different symptoms, and improves the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0035] like Figure 1 As shown, the method for generating consultation information specifically includes steps S10 to S40.

[0036] Step S10: Obtain the initial consultation information of the target user, and obtain the first symptom keyword and the first time corresponding to the first symptom keyword based on the initial consultation information;

[0037] Step S20: Using a pre-trained information generation model, determine the second symptom keyword that is most relevant to the first symptom keyword and the second time corresponding to the second symptom keyword;

[0038] Step S30: Determine the time relationship between the first time and the second time;

[0039] Step S40: Based on the first symptom keyword, the second symptom keyword, the first time, the second time, and the time relationship, generate the target consultation information corresponding to the target user.

[0040] Specifically, the current response text is a medical text, which can be an electronic healthcare record, an electronic personal health record, including medical records, electrocardiograms, medical images, and other electronic records with archival value. The consultation text generation method involved in this embodiment can be applied to intelligent diagnosis and treatment or remote consultation.

[0041] In specific embodiments, in task-oriented multi-turn dialogue systems, a slot-filling-based method is generally used to drive the completion of dialogue tasks. In the context of Traditional Chinese Medicine (TCM), the main task of consultation is to collect patient symptom information, which includes continuous slots related to symptom duration, as shown in Table 1.

[0042] Table 1

[0043]

[0044] In traditional medical settings, if the symptom information provided by the target user only includes "sore throat" and the duration of "sore throat," it is often insufficient to provide targeted diagnostic information. However, this embodiment, in the field of digital healthcare, uses a pre-set model to determine the most relevant symptoms associated with "sore throat" and their corresponding durations, based on the information of "sore throat" and related symptoms. The system then generates consultation information for the target user based on "sore throat" and related symptoms, including both the symptoms and the inferred duration.

[0045] This embodiment discloses a method, apparatus, device, and storage medium for generating consultation information. The method includes acquiring initial consultation information of a target user, and acquiring a first symptom keyword and a first time corresponding to the first symptom keyword based on the initial consultation information; determining a second symptom keyword with the highest correlation to the first symptom keyword and a second time corresponding to the second symptom keyword through a pre-trained information generation model; determining the temporal relationship between the first time and the second time; and generating target consultation information corresponding to the target user based on the first symptom keyword, the second symptom keyword, the first time, the second time, and the temporal relationship. Through the above method, this embodiment, using an information generation model, determines the second symptom keyword with the highest correlation and the second time based on the first symptom keyword and the first time, and then infers the user's symptoms and duration based on the temporal relationship between the first time and the second time. This solves the problem that traditional medical scenarios cannot reflect the correlation between different symptoms, and improves the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0046] Please see Figure 2 , Figure 2 This is a schematic flowchart illustrating a method for generating consultation information according to a second embodiment of this application. This method can be applied to a server, using historical consultation dialogue information as a training set. By training the training set, the duration of each symptom keyword database is determined. A correspondence between symptom keywords and durations is established in a preset information generation model. Subsequently, after determining symptom keywords, the corresponding duration can be generated, improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0047] based on Figure 1 The illustrated embodiment, as Figure 2 As shown, steps S11 to S12 are included before step S20.

[0048] Step S11: Obtain historical consultation dialogue information and determine the symptom keyword database based on the historical consultation dialogue information;

[0049] Step S12: Train the symptom keyword library using a preset maximum likelihood function to determine the duration corresponding to the symptom keywords and generate the pre-trained information generation model.

[0050] Specifically, the maximum likelihood function is chosen as the loss function, and the system is trained on historical consultation dialogue information. The specific training process is as follows:

[0051] Input: Training data. Format: <historical consultation dialogue information, generated labels>

[0052] Output: The set of all parameters W for the model.

[0053] For wheels not yet at their maximum size:

[0054] Forbatchintrain_data

[0055] p = mode_netl(batch)

[0056] loss = Likelihood(label, p)

[0057] △w=d(loss) / d(w) (d(*) denotes differentiation)

[0058] w = w - △w

[0059] ifloss value is less than the stopping condition:

[0060] quit

[0061] ReturnW_trained

[0062] Endfinishtrain.

[0063] This embodiment discloses a method, apparatus, device, and storage medium for generating consultation information. The method includes acquiring historical consultation dialogue information and determining a symptom keyword library based on the historical consultation dialogue information; training the symptom keyword library using a preset maximum likelihood function to determine the duration corresponding to the symptom keywords, and generating the pre-trained information generation model. Through this method, this application uses historical consultation dialogue information as a training set, and by training the training set, determines the duration of each symptom keyword, establishing a correspondence between symptom keywords and durations in the preset information generation model. Subsequently, after determining symptom keywords, the corresponding durations can be generated, improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0064] Please see Figure 3 , Figure 3This is a schematic flowchart of a method for generating consultation information according to the third embodiment of this application. This method can be applied to a server and, through an information generation model, determines the most relevant second symptom keyword and second time based on the first symptom keyword and the first time. Then, based on the temporal relationship between the first and second times, it infers the user's symptoms and duration, thereby improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0065] based on Figure 2 The illustrated embodiment, as Figure 3 As shown, step S30 includes steps S301 to S302.

[0066] Step S301: Based on the historical consultation dialogue information, extract the time-related words of the first symptom corresponding to the first symptom keyword and the second symptom corresponding to the second symptom keyword;

[0067] Step S302: Determine the time relationship between the first time and the second time based on the time-related words.

[0068] In a specific embodiment, the onset time of the first symptom and the second symptom is determined by historical consultation dialogue information. For example, in historical consultation dialogue information, the second symptom is often three days before or four days after the first symptom. In this case, "before" and "after" are used as time-related words, and "three days" and "four days" are used as the second time. Combined with the first symptom and the first time in the initial consultation information, the target consultation information is generated.

[0069] This embodiment discloses a method, apparatus, device, and storage medium for generating consultation information. The method includes extracting time-related words for a first symptom corresponding to a first symptom keyword and a second symptom corresponding to a second symptom keyword based on historical consultation dialogue information; and determining the temporal relationship between the first time and the second time based on the time-related words. Through this method, this embodiment, using an information generation model, determines the most relevant second symptom keyword and second time based on the first symptom keyword and the first time, and then infers the user's symptoms and duration based on the temporal relationship between the first time and the second time, thereby improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0070] Please see Figure 4 , Figure 4This is a schematic flowchart illustrating a method for generating consultation information according to the fourth embodiment of this application. This method can be applied to a server and, through an information generation model, determines the most relevant second symptom keyword and second time based on a first symptom keyword and a first time. Then, based on the temporal relationship between the first and second times, it infers the user's symptoms and duration. This solves the problem in traditional medical scenarios where the connection between the first and second symptoms cannot be reflected, thus improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0071] based on Figure 1 The illustrated embodiment, as Figure 4 As shown, step S20 includes steps S201 to S203.

[0072] Step S201: Obtain the relevance ranking table of the first symptom keywords, and determine the symptom keyword with the highest relevance as the second symptom keyword;

[0073] Step S202: Determine the second symptom keyword as the slot ID of the input layer of the information generation model;

[0074] Step S203: Based on the preset mapping relationship, determine the second time corresponding to the second symptom keyword, and determine the second time as the slot value of the input layer of the information generation model.

[0075] In a specific embodiment, Figure 5 This is a schematic diagram illustrating the architecture principle of the preset information generation model provided in the embodiments of this application.

[0076] Based on the seq2seq architecture, such as Figure 5 As shown, the encoder is Roberta, and the decoder is a transformerdecoder. Both the encoder and decoder consist of 12 layers of transformer blocks.

[0077] Encoder:

[0078] The encoder is continuously trained based on the Chinese pre-trained parameters of Roberta-base, and its structure is the same as that of Roberta-base.

[0079] Decoder:

[0080] The decoder is initialized using the decoder module of the Chinese pre-trained version of the transformer, also with 12 blocks.

[0081] Decoder input prompt template:

[0082] The template format is “[slot_n][=]__”. Here, [slot_n] represents a special character, i.e., the slot ID enclosed in square brackets and stored in a dictionary. Similarly, [=] represents a special character and is stored in a dictionary. These characters are trained together with other characters in the dictionary to learn the high-dimensional vector representation of the latent space. After training is complete, this template represents the semantics: “The value that should be filled in slot_n is __”. “__” represents the content generated by the decoder.

[0083] This embodiment discloses a method, apparatus, device, and storage medium for generating consultation information. The method includes obtaining a relevance ranking table of first symptom keywords, determining the symptom keyword with the highest relevance as the second symptom keyword; determining the second symptom keyword as the slot ID of the input layer of the information generation model; determining the second time corresponding to the second symptom keyword based on a preset mapping relationship, and determining the second time as the slot value of the input layer of the information generation model. Through the above method, this embodiment, using an information generation model, determines the second symptom keyword with the highest relevance and the second time based on the first symptom keyword and the first time, and then infers the user's symptoms and duration based on the temporal relationship between the first time and the second time, thereby improving the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0084] Please see Figure 6 , Figure 6 This is a schematic flowchart illustrating a method for generating consultation information according to the fifth embodiment of this application. This method can be applied to a server, whereby a target probability is determined from a probability distribution table using an information generation model and compared with a preset probability threshold. A second time interval is then determined based on the target probability. This solves the problem in traditional medical scenarios where the correlation between different symptoms cannot be reflected, and improves the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0085] based on Figure 4 The illustrated embodiment, as Figure 6 As shown, step S203 includes steps S2031 to S2033.

[0086] Step S2031: Based on a preset algorithm, determine the probability distribution table of the slot ID, and take the highest probability in the probability distribution table as the target probability;

[0087] Step S2032: Compare the target probability with a preset probability threshold;

[0088] Step S2033: When the target probability is greater than the preset probability threshold, the slot value corresponding to the target probability is determined as the second time.

[0089] In a specific embodiment, in the digital healthcare scenario of this embodiment, the implementation method for determining the second time based on the prediction algorithm is as follows:

[0090] Input: Patient's script;

[0091] Output: Generated text;

[0092] Initialize w: Assign the trained W_trained value to the net model;

[0093] Begin:

[0094] W = W_trained.

[0095] Max_score_text / / Represents the generated text corresponding to the highest probability.

[0096] Max_score / / Represents the maximum probability

[0097] For slot in slot set:

[0098] slot_text, slot_score = mode_net(patient script)

[0099] If slot_score > max_scor, then max_score_text = slot_text, max_score = slot_score;

[0100] Return max_score_text.

[0101] End.

[0102] Specifically, based on the pre-acquired probability distribution table, the highest probability is selected and compared with a probability threshold. When the highest probability is greater than the preset probability threshold, the duration corresponding to the highest probability is used as the target slot value and output through the decoder.

[0103] This embodiment discloses a method, apparatus, device, and storage medium for generating consultation information. The method includes determining a probability distribution table of slot IDs based on a preset algorithm, and selecting the highest probability in the probability distribution table as the target probability; comparing the target probability with a preset probability threshold; and determining the slot value corresponding to the target probability as the second time when the target probability is greater than the preset probability threshold. Through this method, this embodiment determines the target probability from the probability distribution table and compares it with a preset probability threshold using an information generation model, and then determines the second time based on the target probability. This solves the problem that traditional medical scenarios cannot reflect the correlation between different symptoms, and improves the accuracy of consultation information in robot-assisted consultations in the digital healthcare field.

[0104] based on Figure 4 In the illustrated embodiment, step S201 includes:

[0105] Compare the highest correlation with a preset correlation threshold;

[0106] When the highest relevance is not less than the preset relevance threshold, the symptom keyword corresponding to the highest relevance is determined as the second symptom keyword.

[0107] Based on all the above embodiments, in this embodiment, the method for generating consultation information further includes:

[0108] Determine whether the second time includes an unknown UNK identifier;

[0109] If the second time includes the UNK identifier, then the second time is used as a negative sample for training the decoder in the information generation model.

[0110] Specifically, when inferring and predicting the symptoms of a target user using this embodiment, there may be situations where the symptom is not present in the training set corresponding to the historical consultation information. Based on this situation, negative samples are constructed to train on unknown situations, as follows:

[0111] Specifically, in actual TCM consultation scenarios, there may be situations where a patient's symptoms are not found in the historical consultation records. In such cases, it is necessary to construct negative samples to represent unknown items in the slot value range. If the current value of [slot_n] is not in the sentence input by the encoder, it is replaced with the special symbol "[unk]" to indicate that it is unknown. Similarly, the special symbol "[unk]" also represents a character in the dictionary.

[0112] In the training set, [slot_n] is randomly sampled, and the labels of the generated portions are replaced with the special symbol "[unk]". For example:

[0113] Positive example: I've been coughing for four days. [slot_1][=] Four days;

[0114] Negative sample: I've been coughing for four days. [slot_2][=][unk].

[0115] Please see Figure 7 , Figure 7 This is a schematic block diagram of a medical consultation information generation device provided in an embodiment of this application. This device is used to execute the aforementioned medical consultation information generation method. The medical consultation information generation device can be configured on a server.

[0116] like Figure 7 As shown, the consultation information generation device 400 includes:

[0117] The first symptom keyword acquisition module 10 is used to acquire the initial consultation information of the target user, and acquire the first symptom keyword and the first time corresponding to the first symptom keyword based on the initial consultation information;

[0118] The second symptom keyword acquisition module 20 is used to determine the second symptom keyword that is most relevant to the first symptom keyword and the second time corresponding to the second symptom keyword by using a pre-trained information generation model.

[0119] The time relationship determination module 30 is used to determine the time relationship between the first time and the second time.

[0120] The target consultation information generation module 40 is used to generate target consultation information corresponding to the target user based on the first symptom keywords, the second symptom keywords, the first time, the second time, and the time relationship.

[0121] Furthermore, the consultation information generation device also includes a pre-training module, specifically comprising:

[0122] The symptom keyword database determination unit is used to acquire historical consultation dialogue information and determine the symptom keyword database based on the historical consultation dialogue information.

[0123] The pre-training unit is used to train the symptom keyword library using a preset maximum likelihood function, determine the duration corresponding to the symptom keywords, and generate the pre-trained information generation model.

[0124] Furthermore, the time relationship determination module 30 includes:

[0125] The time-related word extraction unit is used to extract time-related words for the first symptom corresponding to the first symptom keyword and the second symptom corresponding to the second symptom keyword based on the historical consultation dialogue information.

[0126] A time relationship determination unit is used to determine the time relationship between the first time and the second time based on the time association words.

[0127] Furthermore, the second symptom keyword acquisition module 20 includes:

[0128] The relevance ranking table unit is used to obtain the relevance ranking table of the first symptom keywords and determine the symptom keyword with the highest relevance as the second symptom keyword.

[0129] The slot ID unit is used to determine the second symptom keyword as the slot ID of the information generation model input layer;

[0130] The slot value unit is used to determine the second time corresponding to the second symptom keyword based on a preset mapping relationship, and to determine the second time as the slot value of the information generation model input layer.

[0131] Furthermore, the slot value unit includes:

[0132] The target probability calculation subunit is used to determine the probability distribution table of the slot ID based on a preset algorithm, and to take the highest probability in the probability distribution table as the target probability.

[0133] A probability comparison subunit is used to compare the target probability with a preset probability threshold.

[0134] The second time determination subunit is used to determine the slot value corresponding to the target probability as the second time when the target probability is greater than the preset probability threshold.

[0135] Furthermore, the relevance ranking table unit includes:

[0136] The correlation comparison subunit is used to compare the highest correlation with a preset correlation threshold.

[0137] The second symptom keyword subunit is used to determine the symptom keyword corresponding to the highest relevance as the second symptom keyword when the highest relevance is not less than the preset relevance threshold.

[0138] Furthermore, the consultation information generation device also includes an UNK identifier module, specifically comprising:

[0139] UNK identifier determination unit, used to determine whether the second time includes an unknown UNK identifier;

[0140] A negative sample training unit is used to train the second time as a negative sample for the decoder in the information generation model if the second time includes the UNK identifier.

[0141] It should be noted that those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the above-described apparatus and modules can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0142] The aforementioned device can be implemented as a computer program, which can be used in, for example... Figure 8 It runs on the device shown.

[0143] Please see Figure 8 , Figure 8 This is a schematic block diagram illustrating the structure of a device according to an embodiment of this application. The device may be a server.

[0144] See Figure 8 The device includes a processor, memory, and network interface connected via a system bus, wherein the memory may include non-volatile storage media and internal memory.

[0145] Non-volatile storage media can store operating systems and computer programs. These computer programs include program instructions that, when executed, cause the processor to perform any method for generating medical information.

[0146] The processor provides computing and control capabilities to support the operation of the entire device.

[0147] Internal memory provides an environment for the execution of computer programs in non-volatile storage media. When the computer program is executed by the processor, it enables the processor to execute any method of generating medical information.

[0148] This network interface is used for network communication, such as sending assigned tasks. Those skilled in the art will understand that... Figure 8 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the device to which the present application is applied. Specific devices may include more or fewer components than those shown in the figure, or may combine certain components, or may have different component arrangements.

[0149] It should be understood that the processor can be a Central Processing Unit (CPU), but it can also be other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. Among these, a general-purpose processor can be a microprocessor or any conventional processor.

[0150] In one embodiment, the processor is configured to run a computer program stored in memory to perform the following steps:

[0151] Obtain the initial consultation information of the target user, and obtain the first symptom keyword and the first time corresponding to the first symptom keyword based on the initial consultation information;

[0152] Using a pre-trained information generation model, determine the second symptom keyword that is most relevant to the first symptom keyword and the second time corresponding to the second symptom keyword;

[0153] Determine the temporal relationship between the first time and the second time;

[0154] Based on the first symptom keywords, the second symptom keywords, the first time, the second time, and the time relationship, target consultation information corresponding to the target user is generated.

[0155] In one embodiment, a pre-trained information generation model is used to determine the second symptom keyword that is most relevant to the first symptom keyword and the second time period corresponding to the second symptom keyword, for the purpose of:

[0156] Obtain historical consultation dialogue information, and determine a symptom keyword database based on the historical consultation dialogue information;

[0157] The symptom keyword database is trained by a preset maximum likelihood function to determine the duration corresponding to the symptom keywords and generate the pre-trained information generation model.

[0158] In one embodiment, determining the time relationship between the first time and the second time is used to achieve:

[0159] Based on the historical consultation dialogue information, extract the time-related words of the first symptom corresponding to the first symptom keyword and the second symptom corresponding to the second symptom keyword;

[0160] The temporal relationship between the first time and the second time is determined based on the time-related words.

[0161] In one embodiment, a pre-trained information generation model is used to determine the second symptom keyword that is most relevant to the first symptom keyword and the second time corresponding to the second symptom keyword, for the purpose of:

[0162] Obtain the relevance ranking table of the first symptom keywords, and determine the symptom keyword with the highest relevance as the second symptom keyword;

[0163] The second symptom keyword is determined as the slot ID of the input layer of the information generation model;

[0164] Based on a preset mapping relationship, the second time corresponding to the second symptom keyword is determined, and the second time is determined as the slot value of the input layer of the information generation model.

[0165] In one embodiment, based on a preset mapping relationship, the second time corresponding to the second symptom keyword is determined, and the second time is determined as the slot value of the input layer of the information generation model, for the purpose of:

[0166] Based on a preset algorithm, a probability distribution table of the slot ID is determined, and the highest probability in the probability distribution table is taken as the target probability.

[0167] The target probability is compared with a preset probability threshold.

[0168] When the target probability is greater than the preset probability threshold, the slot value corresponding to the target probability is determined as the second time.

[0169] In one embodiment, a relevance ranking table of the first symptom keywords is obtained, and the symptom keyword with the highest relevance is determined as the second symptom keyword, for the purpose of:

[0170] Compare the highest correlation with a preset correlation threshold;

[0171] When the highest relevance is not less than the preset relevance threshold, the symptom keyword corresponding to the highest relevance is determined as the second symptom keyword.

[0172] In one embodiment, the consultation information generation method is also used to implement:

[0173] Determine whether the second time includes an unknown UNK identifier;

[0174] If the second time includes the UNK identifier, then the second time is used as a negative sample for training the decoder in the information generation model.

[0175] The embodiments of this application also provide a computer-readable storage medium storing a computer program, the computer program including program instructions, and the processor executing the program instructions to implement any of the consultation information generation methods provided in the embodiments of this application.

[0176] The computer-readable storage medium can be an internal storage unit of the device described in the foregoing embodiments, such as the hard disk or memory of the device. The computer-readable storage medium can also be an external storage device of the device, such as a plug-in hard disk, Smart Media Card (SMC), Secure Digital (SD) card, or Flash Card equipped on the device.

[0177] The above description is merely a specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in this application, and these modifications or substitutions should all be covered within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method of generating consultation information, characterized by, The method comprises: obtaining initial consultation information of a target user, and obtaining a first symptom keyword and a first time corresponding to the first symptom keyword based on the initial consultation information; determining, by a pre-trained information generation model, a second symptom keyword with the highest relevance to the first symptom keyword and a second time corresponding to the second symptom keyword; determining a time relationship between the first time and the second time; generating target consultation information corresponding to the target user based on the first symptom keyword, the second symptom keyword, the first time, the second time, and the time relationship; wherein, before determining, by the pre-trained information generation model, the second symptom keyword with the highest relevance to the first symptom keyword and the second time corresponding to the second symptom keyword, the method comprises: obtaining historical consultation dialogue information, and determining a symptom keyword library based on the historical consultation dialogue information; training the symptom keyword library by a pre-set maximum likelihood function to determine the duration of the symptom keyword, and generating the pre-trained information generation model; wherein, determining the time relationship between the first time and the second time comprises: extracting, based on the historical consultation dialogue information, a time association word between a first symptom corresponding to the first symptom keyword and a second symptom corresponding to the second symptom keyword; determining the time relationship between the first time and the second time based on the time association word; wherein, determining, by the pre-trained information generation model, the second symptom keyword with the highest relevance to the first symptom keyword and the second time corresponding to the second symptom keyword comprises: obtaining a relevance ranking table of the first symptom keyword, and determining the symptom keyword with the highest relevance as the second symptom keyword; determining the second symptom keyword as a slot ID of an input layer of the information generation model; determining the second time corresponding to the second symptom keyword based on a pre-set mapping relationship, and determining the second time as a slot value of the input layer of the information generation model; wherein, determining the second time corresponding to the second symptom keyword based on a pre-set mapping relationship, and determining the second time as a slot value of the input layer of the information generation model comprises: determining a probability distribution table of the slot ID based on a pre-set algorithm, and taking the highest probability in the probability distribution table as a target probability; comparing the target probability with a pre-set probability threshold; when the target probability is greater than the pre-set probability threshold, determining the slot value corresponding to the target probability as the second time.

2. The interrogation information generation method according to claim 1, characterized by, The method further comprises: determining whether the second time includes an unknown (UNK) identifier; ​ 3. The interrogation information generation method according to any one of claims 1 or 2, characterized by, ​ ​ If the second time includes the UNK identifier, the second time is trained as a negative sample of a decoder in the information generation model.

4. An interview information generation apparatus characterized by comprising: Comprise: The first symptom keyword acquisition module is used for acquiring initial inquiry information of a target user, and acquiring a first symptom keyword and a first time corresponding to the first symptom keyword based on the initial inquiry information; The second symptom keyword acquisition module is used for determining a second symptom keyword with the highest relevance to the first symptom keyword and a second time corresponding to the second symptom keyword through a pre-trained information generation model; The time relationship determination module is used for determining a time relationship between the first time and the second time; The target inquiry information generation module is used for generating target inquiry information corresponding to the target user based on the first symptom keyword, the second symptom keyword, the first time, the second time, and the time relationship; The inquiry information generation device further comprises a pre-training module, specifically comprising: The symptom keyword library determination unit is used for acquiring historical inquiry dialogue information, and determining a symptom keyword library based on the historical inquiry dialogue information; The pre-training unit is used for training the symptom keyword library through a preset maximum likelihood function, determining a duration corresponding to a symptom keyword, and generating the pre-trained information generation model; The time relationship determination module comprises: The time association word extraction unit is used for extracting a time association word of a first symptom corresponding to the first symptom keyword and a second symptom corresponding to the second symptom keyword based on the historical inquiry dialogue information; The time relationship determination unit is used for determining a time relationship between the first time and the second time based on the time association word; The second symptom keyword acquisition module comprises: The relevance ranking table unit is used for acquiring a relevance ranking table of the first symptom keyword, and determining a symptom keyword with the highest relevance as the second symptom keyword; The slot ID unit is used for determining the second symptom keyword as a slot ID of an input layer of the information generation model; The slot value unit is used for determining the second time corresponding to the second symptom keyword based on a preset mapping relationship, and determining the second time as a slot value of the input layer of the information generation model; The slot value unit comprises: The target probability calculation sub-unit is used for determining a probability distribution table of the slot ID based on a preset algorithm, and taking the highest probability in the probability distribution table as a target probability; The probability comparison sub-unit is used for comparing the target probability with a preset probability threshold; The second time determination sub-unit is used for determining the slot value corresponding to the target probability as the second time when the target probability is greater than the preset probability threshold.

5. An interview information generation apparatus characterized by comprising: The device comprises a memory and a processor; The memory is used for storing a computer program; The processor is used for executing the computer program and realizing the inquiry information generation method according to any one of claims 1 to 3 when executing the computer program.

6. A computer-readable storage medium, characterized in that, The computer readable storage medium stores a computer program, and the computer program is executed by the processor to enable the processor to implement the interrogation information generation method in any one of claims 1 to 3.

Images