CN117854713A

CN117854713A - Method for training traditional Chinese medicine syndrome waiting diagnosis model and method for recommending information

Info

Publication number: CN117854713A
Application number: CN202410256132.4A
Authority: CN
Inventors: 张建峰; 李劲松
Original assignee: Zhejiang Lab
Current assignee: Zhejiang Lab
Priority date: 2024-03-06
Filing date: 2024-03-06
Publication date: 2024-04-09
Anticipated expiration: 2044-03-06
Also published as: CN117854713B

Abstract

The specification discloses a method for training a traditional Chinese medicine diagnosis waiting model and a method for recommending information, which are characterized in that acquired inquiry vector data corresponding to historical traditional Chinese medicine inquiry text information is obtained through a preset language processing model, then the inquiry vector data is input into a prediction model to be trained, each syndrome label matched with the historical traditional Chinese medicine inquiry text information is determined through the prediction model and is used as prediction data corresponding to the historical traditional Chinese medicine inquiry text information, and the prediction model is trained with the aim of minimizing deviation between the prediction data and each actual syndrome label. And the trained prediction model is used for predicting the syndrome label aiming at the inquiry text information, and information recommendation is carried out according to the prediction result. The prediction result obtained after the prediction of the syndrome label of the inquiry text information by the method considers the relation and influence among the syndrome labels of different layers and the same layer, so that the accuracy of the syndrome obtained according to the prediction data is obviously improved.

Description

Method for training traditional Chinese medicine syndrome waiting diagnosis model and method for recommending information

Technical Field

The specification relates to the field of medical research, in particular to a method for training a traditional Chinese medicine syndrome diagnosis model and a method for recommending information.

Background

With the continuous development of neural network technology, artificial intelligence models are widely applied to various high-tech fields, and the capability of highly intelligent data processing brings convenience to various technical fields. For example, in the field of traditional Chinese medicine, syndrome differentiation can be performed based on four-diagnosis information of a patient by applying an artificial intelligence technology, so that an efficient medical auxiliary function is realized.

At present, in the method for determining the syndrome according to the four-diagnosis information of the traditional Chinese medicine, a mode of directly determining the final syndrome result through a trained neural network model is mainly adopted. However, this approach has certain limitations because it often ignores the interconnections and effects between different levels of syndromes. The diagnostic model is generally difficult to make accurate decisions by the training mode, so that the accuracy of syndrome results is low. Further, this also affects the reference value and practicality of the related recommendation information when information recommendation is performed later according to the syndrome result.

Therefore, how to accurately determine the corresponding syndrome labels according to the inquiry information is a problem to be solved urgently.

Disclosure of Invention

The specification provides a method for training a traditional Chinese medicine syndrome diagnosis model and a method for recommending information, so as to partially solve the problems existing in the prior art.

The technical scheme adopted in the specification is as follows:

the specification provides a method for training a traditional Chinese medicine syndrome waiting diagnosis model, which comprises the following steps:

acquiring historical Chinese medicine consultation text information;

inputting the historical traditional Chinese medicine consultation text information into a preset language processing model to obtain consultation vector data corresponding to the historical traditional Chinese medicine consultation text information;

inputting the inquiry vector data into a prediction model to be trained, so that the prediction model determines each evidence label matched with the historical traditional Chinese medicine inquiry text information obtained after multiple rounds of prediction according to the inquiry vector data and preset evidence labels, and as prediction data, determining intermediate vector data according to the inquiry vector data and the predicted evidence labels by the prediction model, fusing the intermediate vector data and the vector data of the sub-evidence labels corresponding to the predicted evidence labels to obtain fusion vector data, and determining the predicted evidence label matched with the historical traditional Chinese medicine inquiry text information from the sub-evidence labels corresponding to the predicted evidence labels according to the fusion vector data and the inquiry vector data;

And training the prediction model by taking the deviation between the minimum prediction data and each actual syndrome label corresponding to the historical traditional Chinese medicine consultation text information as an optimization target.

Optionally, the query vector data is input into a prediction model to be trained, so that the prediction model determines each syndrome label matched with the historical traditional Chinese medicine query text information obtained after multiple rounds of prediction according to the query vector data and preset each syndrome label, and the prediction model specifically includes:

when first-round prediction is executed, the inquiry vector data and preset root syndrome labels are input into a prediction model to be trained, so that the prediction model to be trained determines intermediate vector data during first-round prediction according to the inquiry vector data, the intermediate vector data during first-round prediction and the vector data of each root syndrome label are fused to obtain fusion vector data during first-round prediction, and a syndrome label matched with the historical traditional Chinese medicine inquiry text information is determined from each root syndrome label according to the fusion vector data during first-round prediction and the inquiry vector data.

Optionally, according to the fusion vector data and the inquiry vector data, determining a predicted syndrome label matching with the historical traditional Chinese medicine inquiry text information from sub-syndrome labels corresponding to the predicted syndrome labels, which specifically includes:

determining secondary fusion vector data according to the vector data of the sub-syndrome label corresponding to the predicted syndrome label and the fusion vector data;

and determining the predicted syndrome label which is matched with the historical traditional Chinese medicine consultation text information from the sub-syndrome label corresponding to the predicted syndrome label according to the secondary fusion vector data and the consultation vector data.

Optionally, determining secondary fusion vector data according to the vector data of the sub-syndrome label corresponding to the predicted syndrome label and the fusion vector data specifically includes:

determining the correlation degree between the vector data of the sub-syndrome label corresponding to the predicted syndrome label and the fusion vector data;

determining the weight corresponding to the vector data of the sub-syndrome label corresponding to the predicted syndrome label according to the correlation degree;

And according to the weight, fusing the vector data of the sub-syndrome label corresponding to the predicted syndrome label with the fused vector data to obtain secondary fused vector data.

according to the fusion vector data and the inquiry vector data, determining the matching degree between the historical traditional Chinese medicine inquiry text information and each sub-syndrome label as a first matching degree, and according to the inquiry vector data, determining the matching degree between the historical traditional Chinese medicine inquiry text information and each sub-syndrome label as a second matching degree;

and determining the predicted syndrome label which is matched with the historical traditional Chinese medicine inquiry text information from the sub-syndrome labels corresponding to the predicted syndrome labels according to the first matching degree and the second matching degree.

Optionally, training the prediction model with an optimization objective that minimizes a deviation between the prediction data and each actual syndrome label corresponding to the historical traditional Chinese medicine consultation text information, specifically includes:

Respectively determining deviation between each syndrome label matched with the historical traditional Chinese medicine inquiry text information and each actual syndrome label;

and determining total deviation according to the deviation between each syndrome label matched with the historical traditional Chinese medicine consultation text information and each syndrome label in the actual syndrome labels, and training the prediction model by taking the minimum total deviation as an optimization target.

The specification provides a method for recommending information, which comprises the following steps:

acquiring the text information of the traditional Chinese medicine inquiry of a user;

inputting the Chinese medicine inquiry text information into a prediction model, and obtaining prediction data corresponding to the Chinese medicine inquiry text information through the prediction model;

determining the syndrome data of the user according to each syndrome label contained in the prediction data;

and recommending information according to the syndrome data, wherein the prediction model is obtained by the traditional Chinese medicine syndrome diagnosis model training method.

The specification provides a device for training a traditional Chinese medicine syndrome waiting diagnosis model, which comprises:

the acquisition module is used for acquiring historical traditional Chinese medicine inquiry text information;

the text processing module is used for inputting the historical traditional Chinese medicine consultation text information into a preset language processing model so as to obtain consultation vector data corresponding to the historical traditional Chinese medicine consultation text information;

The prediction module is used for inputting the inquiry vector data into a prediction model to be trained, so that the prediction model determines each evidence label matched with the historical traditional Chinese medicine inquiry text information obtained after multiple rounds of prediction according to the inquiry vector data and preset evidence labels, and takes the evidence label matched with the historical traditional Chinese medicine inquiry text information as prediction data;

and the training module is used for training the prediction model by taking the deviation between the minimum prediction data and each actual syndrome label corresponding to the historical traditional Chinese medicine consultation text information as an optimization target.

The present specification provides a computer readable storage medium storing a computer program which when executed by a processor implements the method for training a traditional Chinese medicine diagnosis model and the method for recommending information.

The specification provides an electronic device, which comprises a memory, a processor and a computer program stored on the memory and capable of running on the processor, wherein the processor realizes the method for training the traditional Chinese medicine diagnosis model and the method for recommending information when executing the program.

According to the method, in the method for training the traditional Chinese medicine diagnosis waiting model and the method for recommending information, the acquired historical traditional Chinese medicine diagnosis text information is input into a preset language processing model, so that corresponding diagnosis vector data are obtained. And inputting the inquiry vector data into a prediction model to be trained, determining each syndrome label matched with the historical traditional Chinese medicine inquiry text information from preset syndrome labels through the prediction model, and taking the identification label as prediction data corresponding to the historical traditional Chinese medicine inquiry text information. And training the prediction model by taking the deviation between the minimized prediction data and each actual syndrome label as an optimization target. The trained prediction model is used for predicting the Chinese medicine inquiry text information without the confirmation of the syndrome label, so that the syndrome data is determined according to the prediction result, and information recommendation is carried out according to the syndrome data.

From the above, it can be seen that the method for training the traditional Chinese medicine diagnosis waiting model and the method for recommending information provided in the present specification can predict each syndrome label matched with the traditional Chinese medicine diagnosis text information by a layer-by-layer progressive multi-round prediction mode according to the diagnosis vector data corresponding to the traditional Chinese medicine diagnosis text information through the prediction model. According to the method, the predicted result obtained after the prediction of the syndrome label of the traditional Chinese medicine inquiry text information considers the relation and influence among the syndrome labels of different layers and the same layer, so that the accuracy of the syndrome data obtained according to the predicted data is greatly improved, and the recommended data when the information is recommended according to the syndrome data also accords with the real situation of a patient to whom the traditional Chinese medicine inquiry text information belongs, so that the method can provide help for the patient more accurately and effectively.

Drawings

The accompanying drawings, which are included to provide a further understanding of the specification, illustrate and explain the exemplary embodiments of the present specification and their description, are not intended to limit the specification unduly. In the drawings:

FIG. 1 is a flow chart of a method for training a traditional Chinese medicine diagnosis model provided in the present specification;

FIG. 2 is a schematic diagram of a prediction process for syndrome labels provided in the present disclosure;

FIG. 3 is a flow chart of a method for recommending information provided in the present specification;

FIG. 4 is a schematic diagram of a device for training a traditional Chinese medicine diagnosis waiting model provided in the present specification;

FIG. 5 is a schematic diagram of an information recommendation device provided in the present specification;

fig. 6 is a schematic structural view of an electronic device corresponding to fig. 1 and 3 provided in the present specification.

Detailed Description

For the purposes of making the objects, technical solutions and advantages of the present specification more apparent, the technical solutions of the present specification will be clearly and completely described below with reference to specific embodiments of the present specification and corresponding drawings. It will be apparent that the described embodiments are only some, but not all, of the embodiments of the present specification. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are intended to be within the scope of the present disclosure.

The following describes in detail the technical solutions provided by the embodiments of the present specification with reference to the accompanying drawings.

Fig. 1 is a schematic flow chart of a method for training a traditional Chinese medicine diagnosis waiting model provided in the present specification, which includes the following steps:

S101: acquiring historical Chinese medicine inquiry text information.

S102: and inputting the historical traditional Chinese medicine consultation text information into a preset language processing model to obtain consultation vector data corresponding to the historical traditional Chinese medicine consultation text information.

With the continuous progress and development of neural network technology, the high efficiency and convenience brought by the neural network technology enable the artificial intelligent model to be popularized and applied in various fields. For example, in the field of traditional Chinese medicine, a corresponding syndrome label can be determined according to inquiry text information through an artificial intelligent model, but in the current stage, the syndrome label is determined according to the traditional Chinese medicine inquiry information, the syndrome result of the inquiry information is directly determined mainly through a trained model, consideration and deep research on the relation among the same or different layers of syndromes are lacked, and the accuracy of the syndrome result obtained by the method is low. Therefore, it is important how to accurately predict the syndrome label for the inquiry information.

For this reason, the present specification provides a method for training and recommending information of a traditional Chinese medicine diagnosis model, wherein the execution subject adopted by the method provided by the present specification may be a terminal device such as a desktop computer, a notebook computer, or a server, and besides, the execution subject of the present specification may also be a subject in a software form, such as a client installed in the terminal device. For convenience of explanation, the method for training the provided traditional Chinese medicine diagnosis model and the method for recommending information will be explained only by taking the terminal equipment as an execution subject in the following description.

Based on the information, the terminal equipment applying the traditional Chinese medicine diagnosis model training method and the information recommendation method provided by the specification can predict corresponding diagnosis prediction data according to the acquired inquiry text information, so that corresponding information recommendation is performed according to the diagnosis prediction data. The actual scene of the terminal device when information recommendation is performed according to the prediction data can be determined according to actual requirements, for example, when a patient inputs inquiry text information to the terminal device, the terminal device can predict the prediction data of the syndrome of the patient according to the inquiry text information input by the patient, and provide relevant traditional Chinese medicine physique health information to the patient according to the prediction data, so that the patient can be helped to better understand the disease condition of the patient, and proper health management measures are adopted to promote the health literacy of the patient; for another example, when the doctor of the professional traditional Chinese medicine makes a consultation on the patient and then inputs the consultation text information to the terminal device, the terminal device can predict the prediction data of the syndrome of the patient according to the consultation text information input by the doctor of the professional, and further recommend the past and historically same type of symptoms cases, and corresponding symptoms prescriptions and treatment methods to the doctor of the professional according to the prediction data, so as to be referred by the doctor of the professional.

The method provided by the specification mainly comprises two stages, namely a model training stage and an actual application stage, wherein in the model training stage, terminal equipment can acquire historical traditional Chinese medicine consultation text information of a known evidence label prediction result, the historical traditional Chinese medicine consultation text information is input into a preset language processing model, and the language processing model is used for processing the historical traditional Chinese medicine consultation text information, so that consultation vector data corresponding to the historical traditional Chinese medicine consultation text information is obtained.

Specifically, the terminal device may obtain the inquiry text information input by the user through a preset man-machine information interaction interface, where the man-machine information interaction interface may be an information interaction interface in a software application or an information input interface in a web page form, and the specific form of the man-machine information interaction interface is not limited in the present specification, and may be adjusted according to the actual situation. The above-mentioned inquiry text information mainly includes four kinds of information (hope, smell, ask, cut) commonly used in the field of traditional Chinese medicine, specifically includes basic physiological information of the patient to whom the inquiry text information belongs, description information for the symptoms, medical history information of the patient, etc. In order to facilitate the description of the specific text form of the inquiry text information, the inquiry text information entered after the inquiry of the patient by the professional doctor of traditional Chinese medicine will be described below in one example.

"complaint: dysphagia and hiccup for more than 1 month. The diagnosis is as follows: the patients have emaciation, listlessness, pale complexion, continuous sound insulation, difficult swallowing after eating, spitting after eating, poor sleep, dry stool, 5-6 times of nocturia, enlarged tongue, cold obstruction under the tongue, slippery and astringent pulse and small left-hand beans. "

As can be seen from the inquiry text information in the above example, the inquiry text information obtained by the terminal device mainly includes text description for physiological conditions and symptoms of the patient, the "main complaint" is mainly self-description of adverse reactions of the patient, and the "diagnosis" is to observe and record the patient one by one from four diagnosis aspects in the field of traditional Chinese medicine through a professional doctor. The above examples are mainly for facilitating introduction and understanding of specific text forms and general text contents of the inquiry text information acquired by the terminal device, and the specific text forms of the inquiry text information are not limited in the present specification, and may be recorded in a free text form as in the above examples, or may be recorded in other manners such as filling in a specific form or selected field, and may be flexibly adjusted according to actual scenes and requirements.

Furthermore, after the terminal device obtains the historical traditional Chinese medicine inquiry text information, the historical traditional Chinese medicine inquiry text information can be input into a language processing model preset in the terminal device, and inquiry vector data corresponding to the historical traditional Chinese medicine inquiry text information can be obtained through the language processing model.

Specifically, the terminal device may perform word segmentation processing on the query text information through a preset language processing model, perform vector conversion on each word segment in the word segmentation result, obtain vector representations corresponding to each word segment, and then perform data combination on the vector representations of each word segment to obtain vector representations corresponding to the whole query text information. In order to facilitate explanation of a specific process of obtaining, by the terminal device, the query vector data corresponding to the query text information through the language processing model, the following description is given by referring to a specific process of determining, by the language processing model, the query vector data according to the query text information in an example.

Assume that the question text information acquired by the terminal device is "main complaint: headache, weakness and inappetence. Checking: a white coating and a wiry and thready pulse. The language processing module performs word segmentation on the whole inquiry text information, so that a word segmentation result is obtained: the following formula can be referred to for specific purposes, namely, the following formula is obtained by carrying out corresponding vector conversion on each word in the word segmentation result one by one according to the following conditions:

X

Wherein, X is a set of vector representations corresponding to each word segmentation result in the query text information, namely, "main complaint" in the above example: headache, weakness and inappetence. Checking: a white coating and a wiry and thready pulse. "set of corresponding vector representations".Vector representation corresponding to "main complaint">The vector representation corresponding to "headache" and so on, up to +.>The vector representation corresponding to "thin". After determining the vector representation of each word in the word segmentation result of the inquiry text information, the terminal equipment combines the vector representations of each word through a language processing model, thereby obtaining the overall inquiryVector representation of the diagnostic text information. By converting the text information into a data format in a vector form, semantic information in the inquiry text information is reserved, and meanwhile, the data format requirement required by model input is met when the prediction of the evidence label is carried out through the prediction model in the follow-up process.

It should be noted that, the language processing model preset in the terminal device may be a BERT language model as commonly used in the technical field of the present stage, or a model with the capability of converting text information into vector data such as a RoBERTa language model, which is not limited in the specification, and may be selected according to actual requirements and application scenarios.

S103: inputting the inquiry vector data into a prediction model to be trained, so that the prediction model determines each evidence label matched with the historical traditional Chinese medicine inquiry text information obtained after multiple rounds of prediction according to the inquiry vector data and preset evidence labels, and as prediction data, determining intermediate vector data according to the inquiry vector data and the predicted evidence labels by the prediction model, fusing vector data and vector data of sub-evidence labels corresponding to the predicted evidence labels to obtain fusion vector data, and determining the evidence label matched with the historical traditional Chinese medicine inquiry text information obtained by the round of prediction according to the fusion vector data and the inquiry vector data from the sub-evidence labels corresponding to the predicted evidence labels.

In the specification, the terminal device inputs the query vector data corresponding to the historical traditional Chinese medicine query text information into a prediction model to be trained, so as to predict each syndrome label matched with the historical traditional Chinese medicine query text information according to the query vector data and each preset syndrome label through the prediction model, and the predicted each syndrome label is used as the prediction data corresponding to the historical traditional Chinese medicine query text information.

It should be noted in advance that each preset syndrome label mentioned in the method is mainly obtained by screening from the national standard of the people's republic of China (GB/T16751.2-1997) to the traditional Chinese clinical diagnosis and treatment term syndrome section according to the traditional Chinese medical condition appearing in history and each actual case recorded in the case by a developer, the number and the content of the preset syndrome labels are not strictly limited, and the number and the content of the preset syndrome labels are flexibly modified according to actual requirements and scenes in the specification.

Regarding a specific screening process of preset syndrome labels, for example, assume that "3.2.6.1.2 wind-damp skin accumulation syndrome" in "clinical diagnosis and treatment terminology syndrome section" of traditional Chinese medicine is required to be used as one of preset syndrome labels, all disease labels on the upper layer of "3.2.6.1.2 wind-damp skin accumulation syndrome" are required to be used as preset syndrome labels together, namely "3.2.6.1 wind-damp external attack syndrome", "3.2.6 wind-damp syndrome" and "3.2 wind syndrome" are all used as preset syndrome labels, and a tree of syndrome label with tree structure for representing the preset syndrome labels is constructed by the method.

Wherein, the syndrome label 3.2 corresponding to the "3.2 wind syndrome" is used as a root syndrome label in the syndrome label tree, the root syndrome label in the specification can be understood as a disease representing a large class, and the syndrome label 3.2.6 corresponding to the "3.2.6 wind syndrome" is used as one of the child labels of the syndrome label 3.2, the relationship between the syndrome label 3.2 and the syndrome label 3.2.6 is that of the father and son syndrome labels, the syndrome label 3.2 is that of the father and the syndrome label 3.2.6 is that of the syndrome label 3.2; similarly, the syndrome label 3.2.6.1 corresponding to the 3.2.6.1 external wind-damp attack syndrome is also one of the child labels of the syndrome label 3.2.6, and similarly, there is a relationship between the syndrome label 3.2.6 and the syndrome label 3.2.6.1, namely, a parent-child syndrome label is also present, the syndrome label 3.2.6 is the parent of the syndrome label 3.2.6.1, and the syndrome label 3.2.6.1 is the child of the syndrome label 3.2.6.

Continuing with the above example, if it is assumed that, on the premise that the "3.2.6.1.2 wind-damp skin accumulation syndrome" in the "traditional Chinese medical science clinical diagnosis and treatment terminology syndrome section" is taken as one of the preset syndrome labels, the "3.2.6.1.4 wind-damp arthralgia syndrome" in the "traditional Chinese medical science clinical diagnosis and treatment terminology syndrome section" is also taken as one of the preset syndrome labels, then the syndrome label 3.2.6.1 corresponding to the "3.2.6.1 wind-damp external attack syndrome" is correspondingly added with one sub-syndrome label, namely the syndrome label 3.2.6.1.4 corresponding to the "3.2.6.1.4 wind-damp arthralgia syndrome", and the syndrome label 3.2.6.1.2 corresponding to the above-mentioned "3.2.6.1.2 wind-damp skin accumulation syndrome" is in a brother relation with the newly added syndrome label 3.2.6.1.4, namely the syndrome label 3.2.6.1.2 is the "brother" of the syndrome label 3.2.6.1.4, and the syndrome label 3.2.6.1.4 is the "brother" of the syndrome label 3.2.6.1.2. It should be further explained that, in order to facilitate the data processing in the subsequent prediction process, the specific data expression form of each syndrome label in the syndrome label tree for representing each preset syndrome label is a digital label corresponding to the syndrome text expression, that is, "3.2.6.1.2 wind-damp skin accumulation syndrome" corresponding to the syndrome label in the tree structure is "3.2.6.1.2".

Specifically, when the terminal device predicts each syndrome label matched with the historical traditional Chinese medicine inquiry text information through the prediction model to be trained, the terminal device comprises a plurality of rounds of prediction processes, and the data processing process in each round of prediction process (comprising the second round of prediction) after the second round of prediction is started is completely the same except that the first round of prediction process is slightly different. For the prediction process of each round after the second round of prediction starts (including the second round of prediction), the terminal device may determine the corresponding intermediate vector data according to the prediction model according to the corresponding query vector data of the historical traditional Chinese medicine query text information and the predicted syndrome label, and specifically may refer to the following formula:

wherein,data representing a consultation vector corresponding to historical Chinese medical consultation text information->Vector data corresponding to the syndrome label predicted by the previous prediction is represented by +.>Representing the current turn, ++>Representing vector concatenation, calculated by the above first line formula>Data representing a query vector corresponding to the text information of a historical Chinese medical query>Vector data of already predicted syndrome tag +. >Data fusion is performed and then the +.A. is applied by the second line formula>Data processing is carried out by means of a weight matrix +.>And obtaining intermediate vector data capable of representing the relation between the inquiry vector data and the predicted syndrome label after the operation of the corresponding function>。

Next, the terminal device performs data fusion on the intermediate vector data and the vector data corresponding to each sub-syndrome label corresponding to the predicted syndrome label through the prediction model, so as to obtain fused vector data corresponding to the intermediate vector data under the round, and specifically, the following formula can be referred to:

wherein,intermediate vector data corresponding to the presentation vector data, < +.>Vector data for representing each sub-syndrome label corresponding to a syndrome label that has been predicted among preset individual syndrome labels, +.>Representing intermediate vector data +.>Corresponding fusion vector data. The specific calculation process of the formula is to use vector data of each sub-syndrome label corresponding to the predicted syndrome label>Intermediate vector data corresponding to the fused incoming diagnosis vector data +.>In (1) the intermediate vector data->Amplifying vector dimensions of (a) until vector data of all child syndrome labels are solicited, fusing the predicted syndrome labels serving as father syndrome labels with inquiry vector data, and then fusing the predicted syndrome labels with corresponding child syndrome labels to obtain fused vector data capable of representing relations between father and child syndrome labels >。

Then, the terminal device may determine, according to the fusion vector data and the vector data of each sub-syndrome label corresponding to the predicted syndrome label, a correlation between the vector data of the sub-syndrome label corresponding to the predicted syndrome label and the fusion vector data, and determine, according to the correlation, a weight occupied by the vector data of each sub-syndrome label in the fusion vector data, thereby performing data fusion again on the fusion vector data and the vector data of each sub-syndrome label corresponding to the predicted syndrome label according to the weight occupied by the vector data of each sub-syndrome label in the fusion vector data, so as to obtain secondary fusion vector data corresponding to the query vector data, and specifically may refer to the following formula:

wherein,fusion vector data representing correspondence of inquiry vector data,/->Indicating all sub-syndrome labels corresponding to the already predicted syndrome label ++>Indicating the predicted first ++of the syndrome label>Vector data of sub-syndrome label, +.>Is an attention function, which is used to represent the weight of the vector data of each sub-syndrome label corresponding to the predicted syndrome label in the fused vector data, and the following formula can be referred to specifically:

Wherein,indicating the predicted first ++of the syndrome label>Vector data of sub-syndrome tag +.>Correlation with corresponding fusion vector data, +.>I.e. +.>Indicate->Vector data of sub-syndrome tag +.>Weights occupied in corresponding fusion vector data, and weights +>In particular by the attention function in the first row of formulas aboveAcquisition is performed. In the whole process of determining the secondary fusion vector data corresponding to the inquiry vector data, the main purpose is that only the brother relation among all sub-syndrome labels in the fusion vector data obtained by simply fusing the intermediate vector data with the vector data of each sub-syndrome label is represented by applying a weight, so that the brother relation among all sub-syndrome labels and the importance degree of the follow-up prediction model can be better captured when predicting the syndrome label of the next round, and the brother relation mentioned here and how to screen preset syndromes according to the descriptionThe significance of the syndrome label 3.2.6.1.2 in the example mentioned in the label is the same as that represented by the sibling relationship of the syndrome label 3.2.6.1.4, so that, for example, the importance degree of the syndrome label 3.2.6.1.2 and the syndrome label 3.2.6.1.4 on the fusion vector data determined according to the parent syndrome label 3.2.6.1 and the inquiry vector data, that is, the weight ratio mentioned above, is confirmed by the above formula.

Further, after the terminal device determines the secondary fusion vector data corresponding to the inquiry vector data through the prediction model, the problem that the inquiry vector data has missing data influence or has reduced data duty ratio in the process of multiple calculation, so that deviation and the like occur in the subsequent prediction process is avoided, and the terminal device can re-fuse the inquiry vector data corresponding to the historical traditional Chinese medicine inquiry text information and the secondary fusion vector data corresponding to the inquiry vector data, wherein the following formula can be referred to specifically:

wherein,question vector data corresponding to the text information representing the historical Chinese medicine question,>representing inquiry vector dataCorresponding secondary fusion vector data, +.>Representing a weight matrix, +.>And representing final fusion vector data obtained by re-fusing the inquiry vector data corresponding to the history traditional Chinese medicine inquiry text information and the secondary fusion vector data corresponding to the inquiry vector data. The integral formula specifically represents the number of the secondary fusion vectors corresponding to the inquiry vector dataAccording to->Again with inquiry vector data->And fusion is carried out, so that the original inquiry vector data is not lost or reduced due to multiple data processing in the use process of the follow-up prediction process, and the accuracy of the prediction result in the follow-up prediction stage is improved.

After the terminal equipment determines final fusion vector data corresponding to the inquiry vector data through the prediction model, the terminal equipment can determine the first matching degree between the historical traditional Chinese medicine inquiry text information and each sub-syndrome label according to the final fusion vector data through the prediction model, and simultaneously determine the second matching degree between the historical traditional Chinese medicine inquiry text information and each sub-syndrome label according to the inquiry vector data. According to the first matching degree and the second matching degree of the historical traditional Chinese medicine inquiry text information, determining the predicted syndrome label which is matched with the historical traditional Chinese medicine inquiry text information from each sub-syndrome label corresponding to the predicted syndrome label, wherein the specific formula can be referred to as follows:

wherein,representing final fusion vector data corresponding to historical Chinese medicine consultation text information,/and>inquiry vector data corresponding to the text information representing the history of Chinese medicine inquiry, < >>For representing the predictive model according to the final fusion vector data +.>Determining a first matching degree between the historical Chinese medicine inquiry text information and each sub-syndrome labelThen it is indicated that the predictive model is based on the inquiry vector data +.>Determining a second matching degree between the historical Chinese medicine inquiry text information and each sub-syndrome label, and ++ >And->Weight matrix used in the process of determining the first matching degree and the second matching degree respectively>And->Bias terms used in the process of determining the first matching degree and the second matching degree respectively, +.>And->The weighing coefficients of the first matching degree and the second matching degree are respectively, the value range is between 0 and 1, specific numerical values can be actively set and modified, and the weighing coefficients are not +.>The specific numerical values of (2) are strictly limited, and corresponding adjustment can be made according to actual conditions and requirements.

Representing the final historical Chinese medicine inquiry text information and each sub-syndrome label obtained according to the first matching degree and the second matching degreeOverall degree of match between. The overall formula specifically represents +.>Determining a first matching degree of each sub-syndrome label and the historical Chinese medicine inquiry text information>Simultaneously according to the inquiry vector data +.>Determining a second matching degree of each sub-syndrome label and the historical Chinese medicine inquiry text informationBased on the weighing coefficient->Determining the total matching degree of each sub-syndrome label and the historical Chinese medicine inquiry text information>Furthermore, the terminal device can be adapted to the total degree of matching +. >Determining the predicted syndrome label which is matched with the historical Chinese medicine inquiry text information from each sub-syndrome label corresponding to the predicted syndrome label, specifically according to the total matching degree +.>And (3) judging the numerical value of the obtained data, and selecting the sub-label with the maximum total matching degree numerical value as a syndrome label matched with the historical Chinese medicine inquiry text information obtained by the round of prediction.

It should be further noted that, when the terminal device determines the second matching degree between the historical traditional Chinese medicine inquiry text information and each sub-syndrome label according to the inquiry vector data through the prediction model, the terminal device not only determines the second matching degree between each sub-syndrome label corresponding to the predicted syndrome label and the historical traditional Chinese medicine inquiry text information through the prediction model, but also determines the second matching degree between each sub-syndrome label corresponding to the predicted syndrome label in the above-mentioned syndrome label tree formed by preset each syndrome label and other syndrome labels in the same layer structure with the second matching degree between the sub-syndrome label corresponding to the predicted syndrome label and the historical traditional Chinese medicine inquiry text information, that is, the terminal device can determine the second matching degree between each sub-syndrome label corresponding to the predicted syndrome label and all the syndrome labels in the same layer structure where the syndrome label tree is located through the prediction model according to the inquiry vector data. The aim of the operation is to check the upper round of prediction results, namely, whether the standard syndrome label which is actually predicted and matched with the historical traditional Chinese medicine inquiry text information is one of the sub-syndrome labels corresponding to the predicted syndrome label is judged by determining the second matching degree of the sub-syndrome labels in the other syndrome labels of the same layer structure, if the second matching degree of one syndrome label in the other syndrome labels of the same layer structure is far higher than the second matching degree of the sub-syndrome labels, the terminal equipment can terminate the calculation of the round of prediction process through the prediction model, and return to the prediction process of the previous round for re-prediction.

It should be noted that, the prediction model mentioned in the above method is a method of performing data fusion on the intermediate vector data and the vector data corresponding to each sub-syndrome label corresponding to the predicted syndrome label through the prediction model, and optionally, a plurality of data fusion modes are also selected, for example, based on a large amount of historical traditional Chinese medicine four-diagnosis information as pre-training data, each syndrome label and each sub-syndrome label corresponding to the syndrome label are extracted, a syndrome label frequency co-occurrence graph is constructed, the syndrome label is encoded into the vector data by adopting a graph neural network technology, and the sub-syndrome label and the syndrome label are fused, so that the fused vector data mentioned in the above process can be obtained.

The above mentioned prediction process is a specific detail of each prediction link after the start of the second round of prediction, whereas the first round of prediction links, which are not mentioned, are different from the following respective prediction links. When determining the intermediate vector data in the first-round prediction link, since the predicted syndrome label does not exist in the first-round prediction, the terminal device directly determines the intermediate vector data in the first-round prediction link according to the query vector data corresponding to the historical traditional Chinese medicine query text information through the prediction model, and the following formula can be referred to specifically:

Wherein,data representing a consultation vector corresponding to historical Chinese medical consultation text information->The intermediate vector data representing the first-round prediction link is slightly different from the intermediate vector data determined in each prediction link after the second-round prediction is started, and when the intermediate vector data corresponding to the first-round prediction link is determined, since the predicted syndrome label does not exist, the predicted syndrome label is not needed and can not be fused with the inquiry vector data any more, and the data is directly based on the inquiry vector data>Intermediate vector data corresponding to the first-round prediction link can be directly determined.

In the prediction process of the first-round prediction link, after the intermediate vector data corresponding to the first-round prediction link is determined, as the predicted syndrome label does not exist, that is, the predicted syndrome label does not exist corresponding to each sub-syndrome label, the terminal equipment can directly fuse each root syndrome label in the mentioned syndrome label tree with the intermediate vector data corresponding to the first-round prediction link, so as to obtain fused vector data corresponding to the first-round prediction link. The specific calculation formula is the same as the calculation formula used in each prediction link after the second round of prediction is started, and will not be described here. Then, the terminal equipment can determine secondary fusion vector data corresponding to the first-round prediction according to the fusion vector data and the weight occupied by each root certificate label in the fusion vector data, and then fuse the secondary fusion vector data with inquiry vector data to obtain final fusion vector data. And then determining a first matching degree of each root syndrome label according to the final fusion vector data, and simultaneously determining a second matching degree of each root syndrome label according to the inquiry vector data, determining the root syndrome label matched with the history traditional Chinese medicine inquiry text information from each root syndrome label according to the first matching degree and the second matching degree, and taking the root syndrome label as a first-round predicted syndrome label, namely a predicted syndrome label at the beginning of a second-round prediction link, wherein the specific calculation formula is the same as the calculation formula used in each prediction link after the second-round prediction is started, and the description is omitted again.

Further, after the terminal device passes through the prediction model and a plurality of rounds of prediction links, each syndrome label matched with the historical traditional Chinese medicine inquiry text information is obtained, the terminal device can use each syndrome label as a prediction result corresponding to the historical traditional Chinese medicine inquiry text information, and the specific data form can refer to the following formula:

wherein,representing a set of predicted results corresponding to historical Chinese medical consultation text information, < >>Representing the +.f. corresponding to the text information of the historical Chinese medicine inquiry>And predicting results after multiple rounds of prediction. As can be seen from the specific text form of the above-mentioned historical Chinese medicine inquiry text information, the historical Chinese medicine inquiry text information contains multi-angle information of the patient and has a very wide generalization range, so that when the terminal equipment predicts the disease condition label of the patient according to the historical Chinese medicine inquiry text information, it may involve that one piece of the historical Chinese medicine inquiry text information may correspond to a plurality of different disease condition labels, or one piece of the historical Chinese medicine inquiry text information may involve that one disease condition contained in one piece of the historical Chinese medicine inquiry text information has different condition labels from different Chinese medicine angles, so that the terminal equipment may generate a plurality of predicted results after multi-round prediction according to the historical Chinese medicine inquiry text information, and form a set of the predicted results, namely >。

In the above formulaTo->The condition labels which are respectively predicted by each prediction link in the multiple prediction links and are matched with the historical traditional Chinese medicine inquiry text information are taken as an example of the '3.2.6.1.2 rheumatism skin accumulation syndrome', and the prediction result corresponding to the condition labels is that the terminal equipment determines that a patient possibly suffers from the rheumatism skin accumulation syndrome according to the historical traditional Chinese medicine inquiry text information>。

In order to facilitate description of the specific flow direction of the overall prediction process and the specific structure of the number of syndrome labels mentioned above, the following description will be made with a schematic diagram of the prediction process for the syndrome labels based on the above-mentioned examples of specific text forms for describing the inquiry text information, as shown in fig. 2.

Fig. 2 is a schematic diagram of a prediction process for syndrome labels provided in the present specification.

As shown in fig. 2, the terminal device may input the acquired inquiry text information into a preset language processing model, obtain inquiry vector data corresponding to the inquiry text information through the language processing model, then input the inquiry vector data into a prediction model, and obtain a prediction result corresponding to the inquiry text information from a syndrome label tree formed by preset each syndrome label through multiple rounds of prediction by the prediction model. As can be seen from fig. 2, the above-mentioned prediction results corresponding to the examples of the specific text forms for explaining the inquiry text information are:

；

Wherein, according to the records in the national standard of the people's republic of China (GB/T16751.2-1997) to the section of the Chinese medical clinical diagnosis and treatment term syndrome,the corresponding syndrome is named 3.9.1.1 syndrome of phlegm-qi accumulation and->The corresponding syndrome symptoms are named as 4.1.2.2 qi and blood stasis syndrome and->The corresponding symptoms are named as 4.1.5 qi-yin deficiency syndrome and->The corresponding syndrome is named 5.3.1.3 spleen deficiency. For convenience of explanation and understanding, only a few individual syndrome labels are selected in fig. 2 to form a syndrome label number, the root syndrome labels of the syndrome label tree are 3.9, 4.1 and 5.3 respectively as shown in fig. 2, when the terminal device performs the first round of prediction according to the inquiry text information through the prediction model, intermediate vector data directly obtained according to the inquiry text information is fused with each root syndrome label (3.9, 4.1 and 5.3 in fig. 2), so that root syndrome labels matched with the inquiry text information are identified from each root syndrome label, and if a plurality of matched root syndrome labels are identified, the root syndrome labels are sequentially used as predicted syndrome labels of the second round of prediction links to perform subsequent rounds of tests.

Regarding the specific procedure of the prediction procedure and the specific use of the calculation formula, it can be embodied in the content shown in fig. 2 in such a way that For example, after the second-round prediction link starts, the terminal device may perform data fusion on the query vector data and the vector data corresponding to the first-round predicted syndrome label 3.9 through the prediction model, so as to obtain intermediate vector data. And then fusing the intermediate vector data with the vector data of each sub-syndrome label (3.9.1, 3.9.2) corresponding to the syndrome label 3.9 to obtain fused vector data. And then calculating the weight of each sub-syndrome label in the fusion vector data, determining secondary fusion vector data according to the fusion vector data and the weight of each sub-syndrome label, and fusing the secondary fusion vector data with the inquiry vector data corresponding to the inquiry text information to obtain final fusion vector data. According toFinally, the vector data and the inquiry vector data are fused to determine the total matching degree of each sub-syndrome label, so that a syndrome label 3.9.1 matched with inquiry text information in the layer structure is determined from each sub-syndrome label (3.9.1, 3.9.2) and used for the next round of prediction process.

In the above description about the prediction process, the terminal device not only determines the second matching degree of each sub-syndrome label corresponding to the predicted syndrome label and the historical traditional Chinese medicine inquiry text information through the prediction model, but also determines the second matching degree of each sub-syndrome label corresponding to the predicted syndrome label in the same layer structure in the above mentioned syndrome label tree formed by preset each syndrome label and the second matching degree of each sub-syndrome label corresponding to the predicted syndrome label and the historical traditional Chinese medicine inquiry text information, so as to verify the previous prediction result, which may be specifically represented in the content shown in fig. 2 and used again For example, assume that in the third round of prediction link prediction model, the syndrome label already predicted according to the second round of prediction link +.>The first matching degree of each sub-label of the sub-label 3.9.2.3 is judged to be the highest, but the prediction model judges that the sub-label is not the predicted syndrome label according to the inquiry vector data>The second matching degree of the syndrome label 3.9.1.1 of the sub-syndrome label is the highest in the same-layer structure of the syndrome label tree, and the prediction model directly stops the third-round prediction and returns to the second-round prediction link for re-prediction.

It should be noted that, the syndrome label tree composed of preset syndrome labels shown in fig. 2 is only for convenience of illustration, and in practical application, the syndrome label tree constructed according to the history cases and the medical experience of the professional physician is generally more complex and has wider coverage, and has more hierarchical structure, so that the structure size and scope of the syndrome label tree are not strictly limited in the present specification, and can be flexibly adjusted according to the practical needs and application scope.

S104: and training the prediction model by taking the deviation between the minimum prediction data and each actual syndrome label corresponding to the historical traditional Chinese medicine consultation text information as an optimization target.

In the present specification, the terminal device may train the prediction model by minimizing the deviation between the prediction data predicted by the prediction model to be trained and each actual syndrome label corresponding to the historical traditional Chinese medicine inquiry text information. The terminal equipment respectively determines the deviation between each syndrome label matched with the historical traditional Chinese medicine inquiry text information and each actual syndrome label in the prediction result, so as to determine the total deviation corresponding to the prediction data, and trains the prediction model with the minimum total deviation as a target.

When the prediction data corresponding to the inquiry text information includes a plurality of disease results, the terminal device calculates error values together with the syndrome labels in the same layer structure in the syndrome label tree in each syndrome label corresponding to different disease results, calculates the error value of each layer structure, and then calculates the total deviation corresponding to the prediction data. Continuing with the example of fig. 2, the terminal device obtains each prediction result through the prediction model、/>、/>、/>) The syndrome labels of each hierarchical structure in each prediction result are calculated to obtain error values, namely the +. >Chinese syndrome label 3.9->Chinese syndrome label 4.1->Middle syndrome label 4.1 +.>And 5.3, the error value of each corresponding actual syndrome label. Similarly, calculate +.>、/>、/>、/>And then carrying out numerical summation on the error values corresponding to all the layers in the prediction data to obtain the total error corresponding to the prediction data, and optimizing the prediction model according to the total error.

In the training process for the prediction model, the model can actively judge when the model is the prediction round of the final syndrome result according to the actual syndrome result after a certain round of training and optimization. Although the lowermost layer of syndrome labels exist in the constructed syndrome label tree, the actual prediction result may not be strictly directed to a certain syndrome label in the lowermost layer, the difference and diversity between different inquiry text information are very wide, the situation that a certain inquiry text information can be directly represented by a certain syndrome label in the upper layer of the lowermost layer may exist, and the prediction model trained at a certain stage can directly use the syndrome label which is not the lowermost layer as the final syndrome label corresponding to the inquiry text information.

Specifically, the prediction model can be trained according to a large amount of historical Chinese medicine consultation text information in a training stage, the specific layer number of different final syndrome labels corresponding to different text contents can be judged independently after training in a certain stage, the final syndrome labels are not strictly generated in each syndrome label at the bottommost layer in a syndrome label tree, more reasonable final syndrome labels can be predicted according to the actual condition of the text information, and the prediction model can be used for predicting the prediction result which accords with the actual symptoms of a patient more. In addition, in order to ensure format consistency of the subsequent prediction data, even if the final syndrome label corresponding to the text information is already predicted in a certain round in the prediction process, the prediction data of the same round as the final syndrome label is generated in the subsequent round. With the predicted data in FIG. 2 described aboveFor illustration, the inquiry text message and +.>Corresponding to '4.1.2.2 qi and blood stasis syndrome', when the inquiry text information is changed or modified, the name of the symptoms corresponding to the possibly changed inquiry text information becomes 4.1.2 qi and blood disorder syndrome, and the corresponding prediction data becomes +_qi->。

The method provided in the present specification is mainly divided into two phases, a model training phase and an actual application phase. The model training stage is mainly used for obtaining the prediction model with the capability of predicting the syndrome label after model training, so that in the actual application stage, data prediction aiming at the syndrome label can be performed aiming at the inquiry text information of the user, and information recommendation can be performed according to the prediction result.

Fig. 3 is a flow chart of a method for recommending information, which includes the following steps:

s301: and acquiring the text information of the traditional Chinese medicine inquiry of the user.

Along with the continuous development of artificial intelligence technology, the demand level of the neural network in each field is increased day by day, and the identification syndrome label part in the traditional Chinese medicine field at the present stage is popularized and applied, but the syndrome results of inquiry information are mostly directly identified through a trained model at present, and the interrelation and influence among different syndrome labels are not focused, so that the accuracy of the finally obtained syndrome results is lack of high guarantee. Therefore, it is important how to predict the corresponding syndrome label according to the higher accuracy of the inquiry text information, and further to make reasonable and practical information recommendation according to the prediction result.

For this reason, the present specification provides an information recommendation method, in which the execution subject adopted by the method provided in the present specification may be a server or a terminal device such as a desktop computer, a notebook computer, or the like, and in addition, the execution subject in the present specification may be a subject in the form of software, such as a client installed in the terminal device, or the like. For convenience of explanation, the method of providing information recommendation will be explained below with only the server as the execution subject.

Based on this, the actual scenario in which the server recommends information according to the prediction result may be dependent on the actual requirement, for example, the server may provide relevant health education information to the patient according to the prediction result, help them understand their own illness state better, or provide the practitioner with reference information such as similar cases in history and medication prescriptions according to the prediction result. In the present specification, the server may obtain the text information of the traditional Chinese medicine consultation input by the user, the specific text format of the text information of the traditional Chinese medicine consultation has been described in detail in the method for training the traditional Chinese medicine diagnosis waiting model, the specific text content may include basic physiological information of the patient to which the text information of the traditional Chinese medicine consultation belongs, description information for the symptoms, medical history information of the patient, and the like, and the specific text format may be a recording mode such as free text, a specific form, or filling in selected fields, etc., which is not limited in the present specification, and may be flexibly selected according to actual requirements.

S302: inputting the Chinese medicine inquiry text information into a prediction model, and obtaining prediction data corresponding to the Chinese medicine inquiry text information through the prediction model.

S303: and determining the syndrome data of the user according to each syndrome label contained in the prediction data.

In the specification, the server inputs the acquired traditional Chinese medicine inquiry text information into a pre-trained prediction model, and generates prediction data corresponding to the traditional Chinese medicine inquiry text information through the prediction model, so that the real syndrome data of the patient in the traditional Chinese medicine inquiry text information is determined according to each syndrome label in the prediction data.

When the prediction data generated by the prediction model corresponds to a plurality of syndrome data, the server feeds back the plurality of syndrome data to the user who inputs the Chinese medicine inquiry text information together, and carries out comprehensive recommendation in the follow-up information recommendation process.

S304: and recommending information according to the syndrome data, wherein the prediction model is obtained by the traditional Chinese medicine syndrome diagnosis model training method.

In the specification, the server can recommend more reasonable information to the user according to the syndrome data corresponding to the Chinese medicine inquiry text information, the language content of the Chinese medicine inquiry text information and the identity of the user inputting the Chinese medicine inquiry text information. For example, assuming that the text information of the traditional Chinese medicine inquiry is recorded and input by the patient, considering the uncertainty of the patient on the understanding of the traditional Chinese medicine terms and the understanding degree of the actual situation of the patient, the server can perform preliminary prediction according to the text information of the traditional Chinese medicine inquiry input by the patient, judge the possible symptoms of the patient according to the prediction result, provide relevant traditional Chinese medicine physique health information for the patient, help the patient understand the disease condition of the patient better, and take proper health management measures to promote the health literacy of the patient; then, supposing that the text information of the traditional Chinese medicine inquiry is input by a professional doctor after four diagnostic methods (inspection, smelling, asking and cutting) are performed on the patient, the server can predict the syndrome label according to the text information of the traditional Chinese medicine inquiry, display the prediction result to the professional doctor, and recommend the same case recorded in history, the medicine prescription and other referents to the professional doctor according to the prediction result so as to be referred by the professional doctor.

From the above, it can be seen that the method for training the traditional Chinese medicine diagnosis waiting model and the method for recommending information provided in the present specification can predict each syndrome label matched with the traditional Chinese medicine diagnosis text information by a layer-by-layer progressive multi-round prediction mode according to the diagnosis vector data corresponding to the traditional Chinese medicine diagnosis text information through the prediction model. The prediction result obtained after the prediction of the syndrome label of the traditional Chinese medicine inquiry text information by the method considers the connection and influence among the syndrome labels of different layers and the same layer, so that the accuracy of the syndrome label obtained according to the prediction data is higher, and in practical application, the recommendation information in the process of information recommendation according to the prediction data is more consistent with the real situation of a patient, thereby providing help for the patient more accurately and effectively.

The above is a method implemented by one or more of the present description, and based on the same thought, the present description further provides a device for training a corresponding traditional Chinese medicine diagnosis waiting model, as shown in fig. 4.

Fig. 4 is a schematic diagram of a device for training a traditional Chinese medicine diagnosis waiting model provided in the present specification, including:

an acquisition module 401, configured to acquire historical traditional Chinese medicine inquiry text information;

The text processing module 402 is configured to input the historical traditional Chinese medicine consultation text information into a preset language processing model, so as to obtain consultation vector data corresponding to the historical traditional Chinese medicine consultation text information;

the syndrome prediction module 403 is configured to input the query vector data into a prediction model to be trained, so that the prediction model determines, according to the query vector data and each preset syndrome label, each syndrome label that is obtained after multiple rounds of prediction and matches with the historical traditional Chinese medicine query text information, as prediction data, where, for each round of prediction, intermediate vector data is determined according to the query vector data and the predicted syndrome label by the prediction model, the intermediate vector data and vector data of sub-syndrome labels corresponding to the predicted syndrome label are fused to obtain fusion vector data, and according to the fusion vector data and the query vector data, a syndrome label that is obtained by the round of prediction and matches with the historical traditional Chinese medicine query text information is determined from the sub-syndrome label corresponding to the predicted syndrome label;

and the training module 404 is used for training the prediction model by taking the deviation between the minimum prediction data and each actual syndrome label corresponding to the historical traditional Chinese medicine consultation text information as an optimization target.

Optionally, the syndrome prediction module 403 is specifically configured to, when performing first-round prediction, input the query vector data and preset root syndrome labels into a prediction model to be trained, so that the prediction model to be trained determines intermediate vector data during first-round prediction according to the query vector data, fuses the intermediate vector data during first-round prediction and vector data of each root syndrome label to obtain fused vector data during first-round prediction, and determines a syndrome label matching with the historical traditional Chinese medicine query text information from each root syndrome label according to the fused vector data during first-round prediction and the query vector data.

Optionally, the syndrome prediction module 403 is specifically configured to determine secondary fusion vector data according to the vector data of the sub-syndrome label corresponding to the predicted syndrome label and the fusion vector data; and determining the predicted syndrome label which is matched with the historical traditional Chinese medicine consultation text information from the sub-syndrome label corresponding to the predicted syndrome label according to the secondary fusion vector data and the consultation vector data.

Optionally, the syndrome prediction module 403 is specifically configured to determine a correlation between vector data of a sub-syndrome label corresponding to the predicted syndrome label and the fused vector data; determining the weight corresponding to the vector data of the sub-syndrome label corresponding to the predicted syndrome label according to the correlation degree; and according to the weight, fusing the vector data of the sub-syndrome label corresponding to the predicted syndrome label with the fused vector data to obtain secondary fused vector data.

Optionally, the syndrome prediction module 403 is specifically configured to determine, according to the fusion vector data and the query vector data, a degree of matching between the historical traditional Chinese medicine query text information and each sub-syndrome label as a first degree of matching, and determine, according to the query vector data, a degree of matching between the historical traditional Chinese medicine query text information and each sub-syndrome label as a second degree of matching; and determining the predicted syndrome label which is matched with the historical traditional Chinese medicine inquiry text information from the sub-syndrome labels corresponding to the predicted syndrome labels according to the first matching degree and the second matching degree.

Optionally, the training module 404 is specifically configured to determine a deviation between each syndrome label matched with the historical traditional Chinese medicine inquiry text information and each actual syndrome label; and determining total deviation according to the deviation between each syndrome label matched with the historical traditional Chinese medicine consultation text information and each syndrome label in the actual syndrome labels, and training the prediction model by taking the minimum total deviation as an optimization target.

Based on the same thought, the present disclosure also provides a corresponding information recommendation device, as shown in fig. 5.

Fig. 5 is a schematic diagram of an apparatus for information recommendation provided in the present specification, including:

the acquisition module 501: the method comprises the steps of obtaining Chinese medicine inquiry text information of a user;

the prediction module 502: the method comprises the steps of inputting the traditional Chinese medicine inquiry text information into a prediction model, and obtaining prediction data corresponding to the traditional Chinese medicine inquiry text information through the prediction model;

the syndrome determination module 503: the method comprises the steps of determining the syndrome data of a user according to each syndrome label contained in the prediction data;

information recommendation module 504: and the prediction model is obtained by the method for training the traditional Chinese medicine syndrome diagnosis model.

The present specification also provides a computer-readable storage medium storing a computer program operable to perform a method of training a traditional Chinese medicine diagnosis model and a method of recommending information provided in fig. 1 and 3 described above.

The present specification also provides a schematic structural diagram of an electronic device corresponding to fig. 1 and 3 shown in fig. 6. At the hardware level, as shown in fig. 6, the electronic device includes a processor, an internal bus, a network interface, a memory, and a nonvolatile storage, and may of course include hardware required by other services. The processor reads the corresponding computer program from the nonvolatile memory into the memory and then runs the computer program to realize the method for training the traditional Chinese medicine diagnosis model shown in fig. 1 and the method for recommending information shown in fig. 3.

In the 90 s of the 20 th century, improvements to one technology could clearly be distinguished as improvements in hardware (e.g., improvements to circuit structures such as diodes, transistors, switches, etc.) or software (improvements to the process flow). However, with the development of technology, many improvements of the current method flows can be regarded as direct improvements of hardware circuit structures. Designers almost always obtain corresponding hardware circuit structures by programming improved method flows into hardware circuits. Therefore, an improvement of a method flow cannot be said to be realized by a hardware entity module. For example, a programmable logic device (Programmable Logic Device, PLD) (e.g., field programmable gate array (Field Programmable Gate Array, FPGA)) is an integrated circuit whose logic function is determined by the programming of the device by a user. A designer programs to "integrate" a digital system onto a PLD without requiring the chip manufacturer to design and fabricate application-specific integrated circuit chips. Moreover, nowadays, instead of manually manufacturing integrated circuit chips, such programming is mostly implemented by using "logic compiler" software, which is similar to the software compiler used in program development and writing, and the original code before the compiling is also written in a specific programming language, which is called hardware description language (Hardware Description Language, HDL), but not just one of the hdds, but a plurality of kinds, such as ABEL (Advanced Boolean Expression Language), AHDL (Altera Hardware Description Language), confluence, CUPL (Cornell University Programming Language), HDCal, JHDL (Java Hardware Description Language), lava, lola, myHDL, PALASM, RHDL (Ruby Hardware Description Language), etc., VHDL (Very-High-Speed Integrated Circuit Hardware Description Language) and Verilog are currently most commonly used. It will also be apparent to those skilled in the art that a hardware circuit implementing the logic method flow can be readily obtained by merely slightly programming the method flow into an integrated circuit using several of the hardware description languages described above.

The controller may be implemented in any suitable manner, for example, the controller may take the form of, for example, a microprocessor or processor and a computer readable medium storing computer readable program code (e.g., software or firmware) executable by the (micro) processor, logic gates, switches, application specific integrated circuits (Application Specific Integrated Circuit, ASIC), programmable logic controllers, and embedded microcontrollers, examples of which include, but are not limited to, the following microcontrollers: ARC 625D, atmel AT91SAM, microchip PIC18F26K20, and Silicone Labs C8051F320, the memory controller may also be implemented as part of the control logic of the memory. Those skilled in the art will also appreciate that, in addition to implementing the controller in a pure computer readable program code, it is well possible to implement the same functionality by logically programming the method steps such that the controller is in the form of logic gates, switches, application specific integrated circuits, programmable logic controllers, embedded microcontrollers, etc. Such a controller may thus be regarded as a kind of hardware component, and means for performing various functions included therein may also be regarded as structures within the hardware component. Or even means for achieving the various functions may be regarded as either software modules implementing the methods or structures within hardware components.

The system, apparatus, module or unit set forth in the above embodiments may be implemented in particular by a computer chip or entity, or by a product having a certain function. One typical implementation is a computer. In particular, the computer may be, for example, a personal computer, a laptop computer, a cellular telephone, a camera phone, a smart phone, a personal digital assistant, a media player, a navigation device, an email device, a game console, a tablet computer, a wearable device, or a combination of any of these devices.

For convenience of description, the above devices are described as being functionally divided into various units, respectively. Of course, the functions of each element may be implemented in one or more software and/or hardware elements when implemented in the present specification.

It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The present description is described with reference to flowchart illustrations and/or block diagrams of methods, apparatus (systems) and computer program products according to embodiments of the specification. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

In one typical configuration, a computing device includes one or more processors (CPUs), input/output interfaces, network interfaces, and memory.

The memory may include volatile memory in a computer-readable medium, random Access Memory (RAM) and/or nonvolatile memory, such as Read Only Memory (ROM) or flash memory (flash RAM). Memory is an example of computer-readable media.

Computer readable media, including both non-transitory and non-transitory, removable and non-removable media, may implement information storage by any method or technology. The information may be computer readable instructions, data structures, modules of a program, or other data. Examples of storage media for a computer include, but are not limited to, phase change memory (PRAM), static Random Access Memory (SRAM), dynamic Random Access Memory (DRAM), other types of Random Access Memory (RAM), read Only Memory (ROM), electrically Erasable Programmable Read Only Memory (EEPROM), flash memory or other memory technology, compact disc read only memory (CD-ROM), digital Versatile Discs (DVD) or other optical storage, magnetic cassettes, magnetic tape magnetic disk storage or other magnetic storage devices, or any other non-transmission medium, which can be used to store information that can be accessed by a computing device. Computer-readable media, as defined herein, does not include transitory computer-readable media (transmission media), such as modulated data signals and carrier waves.

It should also be noted that the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or apparatus that comprises the element.

It will be appreciated by those skilled in the art that embodiments of the present description may be provided as a method, system, or computer program product. Accordingly, the present specification may take the form of an entirely hardware embodiment, an entirely software embodiment or an embodiment combining software and hardware aspects. Furthermore, the present description can take the form of a computer program product on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

The description may be described in the general context of computer-executable instructions, such as program modules, being executed by a computer. Generally, program modules include routines, programs, objects, components, data structures, etc. that perform particular tasks or implement particular abstract data types. The specification may also be practiced in distributed computing environments where tasks are performed by remote processing devices that are linked through a communications network. In a distributed computing environment, program modules may be located in both local and remote computer storage media including memory storage devices.

In this specification, each embodiment is described in a progressive manner, and identical and similar parts of each embodiment are all referred to each other, and each embodiment mainly describes differences from other embodiments. In particular, for system embodiments, since they are substantially similar to method embodiments, the description is relatively simple, as relevant to see a section of the description of method embodiments.

The foregoing is merely exemplary of the present disclosure and is not intended to limit the disclosure. Various modifications and alterations to this specification will become apparent to those skilled in the art. Any modifications, equivalent substitutions, improvements, or the like, which are within the spirit and principles of the present description, are intended to be included within the scope of the claims of the present description.

Claims

1. A method for training a traditional Chinese medicine syndrome waiting diagnosis model, comprising the following steps:

acquiring historical Chinese medicine consultation text information;

2. The method of claim 1, wherein the query vector data is input into a prediction model to be trained, so that the prediction model determines, according to the query vector data and each preset syndrome label, each syndrome label which is obtained after multiple rounds of prediction and matches with the historical traditional Chinese medicine query text information, as prediction data, and specifically includes:

3. The method according to claim 1, wherein determining, from sub-syndrome labels corresponding to the predicted syndrome labels, a syndrome label matching the historical traditional Chinese medicine consultation text information obtained by the round of prediction according to the fusion vector data and the consultation vector data specifically comprises:

4. The method of claim 3, wherein determining secondary fusion vector data based on vector data of sub-syndrome labels corresponding to the predicted syndrome labels and the fusion vector data, specifically comprises:

5. The method according to claim 1 or 3, wherein determining, according to the fusion vector data and the inquiry vector data, a predicted syndrome label matching the historical traditional Chinese medicine inquiry text information from sub-syndrome labels corresponding to the predicted syndrome labels, specifically includes:

6. The method according to claim 1, wherein training the prediction model with respect to minimizing deviation between the prediction data and each actual syndrome label corresponding to the historical traditional Chinese medicine consultation text information as an optimization objective specifically comprises:

7. An information recommendation method, comprising:

and recommending information according to the syndrome data, wherein the prediction model is obtained by the method of any one of claims 1-6.

8. A device for training a traditional Chinese medicine syndrome waiting diagnosis model, comprising:

9. A computer readable storage medium, characterized in that the storage medium stores a computer program which, when executed by a processor, implements the method of any of the preceding claims 1-7.

10. An electronic device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that the processor implements the method of any of the preceding claims 1-7 when executing the program.