An intelligent guide inquiry feedback processing method and system based on doctor-patient interaction

Abstract

The embodiment of the application discloses a kind of intelligent guide diagnosis inquiry feedback processing method and system based on doctor-patient interaction, the feature correlation mining between inquiry interaction text and inquiry disease knowledge point set is carried out to the intelligent guide diagnosis inquiry activity of target doctor-patient interaction process, and the patient end corresponding to the target doctor-patient interaction process is guided based on feature correlation mining result Diagnosis content information feedback, obtain the medical evaluation data of the patient end for the feedback of guide diagnosis content information, the patient emotion analysis is carried out to the medical evaluation data, and according to patient emotion analysis result, the doctor-patient interaction platform operated by the doctor-patient service system is used feedback, whereby the feature correlation between inquiry interaction text and inquiry disease knowledge point set is combined Intelligent guide diagnosis feedback can improve the reliability of guide diagnosis feedback, and further track subsequent patient emotion and use feedback, which can facilitate to provide reference information for subsequent development of doctor-patient interaction platform.

Description

Technical Field

[0001] The application relates to the field of smart healthcare service technology, specifically to a method and system for intelligent triage, inquiry, and feedback processing based on doctor-patient interaction. Background Technology

[0002] Rehabilitation therapy (such as acupuncture) refers to treatment methods that help restore physical and mental functional disorders or disabilities caused by factors such as injury, disease, and developmental defects to normal or near-normal. It is an important part of rehabilitation medicine. It can not only solve problems such as pain caused by musculoskeletal diseases, but also has significant effects on consciousness disorders, sensory abnormalities, and cognitive impairments caused by neurological diseases.

[0003] As rehabilitation therapy is increasingly used in acupuncture treatment for stroke patients and other critically ill patients, medical professionals have encountered new challenges. During the pre-treatment triage phase, patients with impaired consciousness, cognitive impairment, poor verbal communication, emotional control disorders, or sensory abnormalities are unable to self-examine or provide feedback, and effective communication is difficult. Therefore, intelligent triage and inquiry processes are needed to guide and inquire with the relevant guardians of such patients, thereby reducing communication time before formal rehabilitation treatment. However, the reliability of triage feedback in these technologies is poor, and there is a lack of follow-up patient feedback tracking processes. Summary of the Invention

[0004] This application provides a method and system for intelligent triage, inquiry, and feedback processing based on doctor-patient interaction.

[0005] In a first aspect, embodiments of this application provide an intelligent triage and inquiry feedback processing method based on doctor-patient interaction, applied to a doctor-patient service system, comprising:

[0006] The intelligent triage and inquiry activities of the target doctor-patient interaction process are subjected to feature correlation mining between the inquiry interaction text and the set of disease knowledge points in the inquiry, and the triage content information is fed back to the patient end corresponding to the target doctor-patient interaction process based on the feature correlation mining results.

[0007] Obtain the patient's evaluation data regarding the feedback from the triage information;

[0008] The patient sentiment analysis is performed on the medical evaluation data, and the user feedback is provided to the doctor-patient interaction platform running the doctor-patient service system based on the results of the patient sentiment analysis.

[0009] In one possible implementation of the first aspect, the step of performing feature correlation mining between the intelligent triage and inquiry activities of the target doctor-patient interaction process and the set of disease knowledge points in the inquiry, and providing triage content information feedback to the patient end corresponding to the target doctor-patient interaction process based on the feature correlation mining results, includes:

[0010] In the intelligent triage and inquiry activity of the target doctor-patient interaction process, multiple inquiry interaction texts are obtained, and the interaction keyword vectors of each inquiry interaction text in the multiple inquiry interaction texts are parsed respectively.

[0011] Obtain the set of disease knowledge points corresponding to the intelligent triage and inquiry activity, and analyze the medical knowledge features of each disease knowledge point in the set of disease knowledge points;

[0012] The interaction keyword vectors of the multiple query interaction texts and the corresponding medical knowledge features in the query disease knowledge point set are concatenated to generate a concatenated feature vector sequence.

[0013] Determine the first contextual features of the query interaction text corresponding to each interactive keyword vector in the intelligent triage query activity, and determine the second contextual features of the query disease knowledge point corresponding to each medical knowledge feature in the query disease knowledge point set;

[0014] By combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the spliced ​​feature vector sequence to generate the target spliced ​​feature vector sequence.

[0015] And by combining the target splicing feature vector sequence, a triage feedback result sequence corresponding to the intelligent triage inquiry activity and at least one inquiry feedback extension result corresponding to the triage feedback result sequence are generated;

[0016] Based on the triage feedback results and / or the inquiry feedback extension results, triage content information is fed back to the patient end corresponding to the target doctor-patient interaction process.

[0017] The step of combining the target concatenated feature vector sequence to generate a triage feedback result sequence corresponding to the intelligent triage inquiry activity includes:

[0018] The target concatenated feature vector sequence is loaded into the triage feedback prediction model, and the first triage feedback result of the triage feedback result sequence is output.

[0019] The medical knowledge features of the first triage feedback result in the sequence of triage feedback results;

[0020] Combining the medical knowledge features of the first triage feedback result in the triage feedback result sequence and its carrying state in the triage feedback result sequence, the target spliced ​​feature vector sequence is updated based on feature fusion;

[0021] The updated target splicing feature vector sequence is loaded into the triage feedback prediction model. The target splicing feature vector sequence is updated by traversing the medical knowledge features of each obtained triage feedback result until all triage feedback statements of the triage feedback result sequence are obtained.

[0022] In one possible implementation of the first aspect, the triage feedback prediction model includes a semantic editing unit and a feedback prediction unit, wherein loading the target concatenated feature vector sequence into the triage feedback prediction model and outputting the first triage feedback result of the triage feedback result sequence includes:

[0023] The target concatenated feature vector sequence is loaded into the semantic editing unit, and the semantic edit vector corresponding to the first triage feedback result in the triage feedback result sequence is output.

[0024] The first feedback prediction information is obtained by parsing the semantic edit vector of the first triage feedback result in the triage feedback result sequence through the feedback prediction unit. The first feedback prediction information includes the support of each triage feedback result in the triage feedback result sequence.

[0025] Based on the first feedback prediction information, the first triage feedback result in the triage feedback result sequence is determined.

[0026] In one possible implementation of the first aspect, the target spliced ​​feature vector sequence is updated based on feature fusion by combining the medical knowledge features of the first triage feedback result in the triage feedback result sequence and its carrying state in the triage feedback result sequence, including:

[0027] The interaction keyword vectors of the multiple query interaction texts, the medical knowledge features of the query disease knowledge point set, and the medical knowledge features of the first triage feedback result of the triage feedback result sequence are concatenated to update the concatenated feature vector sequence.

[0028] By combining the first context features corresponding to each interactive keyword vector, the second context features corresponding to each medical knowledge feature of the disease knowledge point set, and the second context features of the first triage feedback result of the triage feedback result sequence, feature nodes are assigned to each spliced ​​feature vector in the updated spliced ​​feature vector sequence to update the target spliced ​​feature vector sequence.

[0029] In one possible implementation of the first aspect, the method further includes:

[0030] When all the triage feedback statements in the triage feedback result sequence are obtained, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the intelligent triage inquiry editing vector and one semantic editing vector is determined as the inquiry disease editing vector;

[0031] Determine the correlation between the intelligent triage query editing vector and the query disease editing vector;

[0032] Based on the relevance, the correlation between the intelligent triage inquiry activity and the set of disease knowledge points inquired is determined;

[0033] Among them, a semantic editing vector serving as the intelligent triage inquiry editing vector is a semantic editing vector corresponding to the first inquiry interaction behavior feature distributed before the interaction keyword vectors of the multiple inquiry interaction texts, and a semantic editing vector serving as the inquiry disease editing vector is a semantic editing vector corresponding to the second inquiry interaction behavior feature distributed between the interaction keyword vectors of the multiple inquiry interaction texts and the various medical knowledge features of the inquiry disease knowledge point set.

[0034] In one possible implementation of the first aspect, determining the first triage feedback result in the triage feedback result sequence by combining the first feedback prediction information includes:

[0035] In the first feedback prediction information, the support scores are sorted in descending order;

[0036] The first N support levels in the descending order queue are determined, and the corresponding triage feedback results are used as the reference triage feedback result for the first triage feedback result in the triage feedback result sequence. The updated target concatenated feature vector sequence is loaded into the semantic editing unit, and the target concatenated feature vector sequence is updated based on the medical knowledge features of each obtained triage feedback result until all triage feedback statements in the triage feedback result sequence are obtained, including:

[0037] Based on the reference guidance feedback result of the first guidance feedback result, reference guidance feedback results are generated one by one for the remaining guidance feedback results;

[0038] By combining the reference guidance feedback results of each guidance feedback result in the guidance feedback result sequence, N guidance feedback result sequences are determined;

[0039] The method further includes:

[0040] For each of the N triage feedback result sequences, the following steps are performed:

[0041] When all the triage feedback statements in the triage feedback result sequence are obtained, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the disease inquiry editing vector and one semantic editing vector is determined as the triage feedback editing vector.

[0042] And determine the correlation between the disease inquiry edit vector and the triage feedback edit vector;

[0043] When the maximum relevance is greater than the threshold relevance, the triage feedback result sequence corresponding to that relevance is determined as the triage feedback result sequence corresponding to the intelligent triage inquiry activity; otherwise, it is determined that there is no triage feedback result sequence corresponding to the intelligent triage inquiry activity.

[0044] In one possible implementation of the first aspect, the method further includes:

[0045] For each of the N triage feedback result sequences, the following steps are performed:

[0046] When all the triage feedback statements in the triage feedback result sequence are obtained, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the intelligent triage inquiry editing vector, one semantic editing vector is determined as the inquiry disease editing vector, and one semantic editing vector is determined as the triage feedback editing vector;

[0047] And determine the correlation between the intelligent triage inquiry editing vector and the inquiry disease editing vector, and determine the correlation between the intelligent triage inquiry editing vector and the triage feedback editing vector;

[0048] When the correlation between the intelligent triage inquiry editing vector and the inquiry disease editing vector is less than a first set correlation and the correlation between the intelligent triage inquiry editing vector and the triage feedback editing vector is greater than a second set correlation, the triage feedback result sequence is determined as a triage feedback result sequence obtained only based on the intelligent triage inquiry activity.

[0049] In one possible implementation of the first aspect, the acquisition of the interactive keyword vector and the medical knowledge features, as well as the allocation of feature nodes, are implemented based on a feature node allocation model, and the method further includes:

[0050] The feature node allocation model, the semantic editing unit, and the feedback prediction unit are trained by combining a first triage feedback learning data sequence. The first triage feedback learning data sequence includes multiple first triage feedback learning data sets. Each first triage feedback learning data set includes a first triage inquiry activity to be learned, a first set of training inquiry disease knowledge points corresponding to the first triage inquiry activity to be learned, and a priori triage feedback result sequence corresponding to the first triage inquiry activity to be learned and the first set of training inquiry disease knowledge points. Training the feature node allocation model, the semantic editing unit, and the feedback prediction unit by combining the first triage feedback learning data sequence includes:

[0051] In at least one first triage feedback learning data point in the first triage feedback learning data sequence, for each first triage feedback learning data point, the following steps are performed:

[0052] Multiple first training query interaction texts are obtained from the first learning-to-learn triage query activity of the first triage feedback learning data, and the interaction keyword vectors of each training query interaction text in the multiple first training query interaction texts are parsed respectively.

[0053] Obtain the first training query disease knowledge point set corresponding to the first learning triage query activity, and analyze the medical knowledge features of each query disease knowledge point in the first training query disease knowledge point set, wherein each medical knowledge feature is consistent with the disease semantic direction of each interactive keyword vector;

[0054] At least one of the prior triage feedback results in the triage feedback result sequence is converted into a triage feedback observation vector to generate a triage feedback observation vector set. The medical knowledge features of each disease knowledge point in the triage feedback observation vector set are then analyzed. The medical knowledge features of each disease knowledge point in the triage feedback observation vector set are consistent with the disease semantic direction of each interactive keyword vector.

[0055] The interaction keyword vectors of the multiple first training query interaction texts, the medical knowledge features of the first training query disease knowledge point set, and the medical knowledge features of each query disease knowledge point in the triage feedback observation vector set are concatenated to generate a first training concatenated feature vector sequence.

[0056] Determine the first contextual features of the query interaction text corresponding to the interaction keyword vector of each first training query interaction text in the first learning guidance query activity; determine the second contextual features of the query disease knowledge point corresponding to each medical knowledge feature in the first training query disease knowledge point set in the first training query disease knowledge point set; and determine the second contextual features of the disease knowledge point corresponding to the medical knowledge feature of each query disease knowledge point in the guidance feedback observation vector set in the guidance feedback observation vector set.

[0057] By combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the first training spliced ​​feature vector sequence to generate the first training target spliced ​​feature vector sequence.

[0058] By combining the first training target with the concatenated feature vector sequence, at least one triage feedback observation vector is generated from the triage feedback observation vector set.

[0059] Obtain the learning cost loss1 between the at least one triage feedback observation vector and the actual triage feedback result;

[0060] The feature node allocation model, the semantic editing unit, and the feedback prediction unit are trained by combining at least the learning cost loss1.

[0061] In one possible implementation of the first aspect, training the feature node allocation model, the semantic editing unit, and the feedback prediction unit by combining the first triage feedback learning data sequence includes:

[0062] From the multiple semantic editing vectors generated by the semantic editing unit, determine one semantic editing vector as the intelligent triage inquiry editing vector, one semantic editing vector as the disease inquiry editing vector, and one semantic editing vector as the triage feedback editing vector;

[0063] By combining the correlation between the intelligent triage inquiry editing vector and the corresponding inquiry disease editing vector, the correlation between the intelligent triage inquiry editing vector and the non-corresponding inquiry disease editing vector, and the correlation between the inquiry disease editing vector and the corresponding intelligent triage inquiry editing vector, and the correlation between the inquiry disease editing vector and the non-corresponding intelligent triage editing vector, the learning cost loss2 is obtained.

[0064] By combining the correlation between the intelligent triage inquiry editing vector and the corresponding triage feedback editing vector, the correlation between the intelligent triage inquiry editing vector and the non-corresponding triage feedback editing vector, and the correlation between the triage feedback editing vector and the corresponding intelligent triage editing vector, as well as the correlation between the triage feedback editing vector and the non-corresponding intelligent triage editing vector, the learning cost loss3 is obtained.

[0065] By combining the correlation between the disease inquiry edit vector and the corresponding triage feedback edit vector, the correlation between the disease inquiry edit vector and the non-corresponding triage feedback edit vector, and the correlation between the triage feedback edit vector and the corresponding disease inquiry edit vector, and the correlation between the triage feedback edit vector and the non-corresponding disease inquiry edit vector, the learning cost loss4 is obtained.

[0066] Furthermore, the training of the feature node allocation model, the semantic editing unit, and the feedback prediction unit, in conjunction with at least the learning cost loss1, includes:

[0067] The feature node allocation model, the semantic editing unit, and the feedback prediction unit are trained by combining the weighted fusion values ​​of learning loss1, learning loss2, learning loss3, and learning loss4.

[0068] Secondly, embodiments of this application provide a doctor-patient service system, including:

[0069] processor;

[0070] A memory, wherein the memory stores a computer program, which, when executed, implements the intelligent triage and inquiry feedback processing method based on doctor-patient interaction as described in the first aspect.

[0071] Compared to existing technologies, this approach involves mining the feature correlation between the intelligent triage and inquiry activities of the target doctor-patient interaction process and the set of disease knowledge points inquired about. Based on the feature correlation mining results, triage content information is fed back to the patient end corresponding to the target doctor-patient interaction process. The patient's evaluation data on the feedback triage content information is obtained, and patient sentiment analysis is performed on the evaluation data. Based on the patient sentiment analysis results, usage feedback is provided to the doctor-patient interaction platform running the doctor-patient service system. By combining the feature correlation between the inquiry and inquiry text and the set of disease knowledge points inquired about for intelligent triage feedback, the reliability of triage feedback can be improved. Furthermore, by tracking subsequent patient sentiment and providing usage feedback, it is possible to provide reference information for the development of subsequent doctor-patient interaction platforms. Attached Figure Description

[0072] Figure 1A flowchart illustrating the steps of an intelligent triage and inquiry feedback processing method based on doctor-patient interaction, provided in an embodiment of this application;

[0073] Figure 2 The embodiments of this application provide for execution Figure 1 The diagram shows the structure of a medical service system based on an intelligent triage and inquiry feedback processing method for doctor-patient interaction. Detailed Implementation

[0074] 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. All other embodiments obtained by those skilled in the art in conjunction with the embodiments of this application without creative effort are within the scope of protection of this application.

[0075] Step 10: Perform feature correlation mining between the intelligent triage and inquiry activities of the target doctor-patient interaction process and the set of disease knowledge points in the inquiry, and provide triage content information feedback to the patient end corresponding to the target doctor-patient interaction process based on the feature correlation mining results.

[0076] Step 20: Obtain the patient's evaluation data on the feedback of the triage content information from the patient's end.

[0077] For example, patient evaluation data can include voice evaluation data and text evaluation data from patients' feedback on the guidance information.

[0078] Step 30: Perform patient sentiment analysis on the medical evaluation data, and provide feedback on the use of the doctor-patient interaction platform running the doctor-patient service system based on the results of the patient sentiment analysis.

[0079] In this embodiment, based on corpus-based analysis methods, machine learning techniques can be used to classify the emotions in the medical evaluation data, thereby generating patient emotion analysis results, such as anger, hatred, fear, interest, happiness, sadness, etc. Machine learning methods typically require first training a classification model to learn patterns in the training data, and then using the trained model to predict test data.

[0080] Based on the above steps, this embodiment mines the feature correlation between the inquiry interaction text and the set of disease knowledge points in the intelligent triage and inquiry activity of the target doctor-patient interaction process. Based on the feature correlation mining results, it provides triage content information feedback to the patient end corresponding to the target doctor-patient interaction process, obtains the patient's evaluation data on the feedback triage content information, performs patient sentiment analysis on the evaluation data, and provides usage feedback to the doctor-patient interaction platform running the doctor-patient service system based on the patient sentiment analysis results. Thus, combining the feature correlation between the inquiry interaction text and the set of disease knowledge points for intelligent triage feedback can improve the reliability of triage feedback. Furthermore, by tracking subsequent patient emotions and providing usage feedback, it can facilitate the provision of reference information for the development of subsequent doctor-patient interaction platforms.

[0081] Step S100: Obtain multiple inquiry interaction texts in the intelligent triage inquiry activity, and parse the interaction keyword vectors of each inquiry interaction text in the multiple inquiry interaction texts respectively.

[0082] In this embodiment, a single intelligent triage inquiry activity can consist of multiple inquiry interaction texts. These texts are triage inquiries initiated by patients on the medical service platform (such as triage inquiries for rehabilitation treatment). Each inquiry interaction text corresponds to a set of disease knowledge points. This sequence of disease knowledge points contains at least one disease knowledge point, which is used to mark the textual disease knowledge points of the inquiry interaction text.

[0083] For multiple inquiry interaction texts obtained during intelligent triage and inquiry activities, a word vector encoding network can be used to extract the interaction keyword vector for each inquiry interaction text. The interaction keyword vector is used to represent the feature vector of the corresponding inquiry interaction text. Each inquiry interaction text corresponds to one interaction keyword vector.

[0084] In some possible implementations, after extracting the interaction keyword vectors of each query interaction text through a word vector encoding network, based on the temporal continuity of each triage query text in the intelligent triage query activity, to reduce processing capacity, the extracted interaction keyword vectors of each query interaction text can be integrated. For example, the relevance of the interaction keyword vectors of two consecutive query interaction texts in the temporal dimension can be obtained. Then, the highly relevant interaction keyword vectors are integrated.

[0085] Step S200: Obtain the set of disease knowledge points corresponding to the intelligent triage inquiry activity, and analyze the medical knowledge characteristics of each disease knowledge point in the set of disease knowledge points.

[0086] As can be seen from step S100 above, each intelligent triage inquiry activity contains a corresponding set of inquiry disease knowledge points. The triage feedback results in the inquiry disease knowledge point set are mapped to medical knowledge features in another value domain. Each medical knowledge feature and each interactive keyword vector can be consistent with the disease semantic direction.

[0087] Step S300: Concatenate the interactive keyword vectors of multiple query interaction texts with the corresponding medical knowledge features in the query disease knowledge point set to generate a concatenated feature vector sequence.

[0088] The method provided in this application embodiment further includes the process of obtaining the first query interaction behavior features and the first query interaction behavior distribution corresponding to the first query interaction behavior, and obtaining the second query interaction behavior features and the second query interaction behavior distribution corresponding to the second query interaction behavior.

[0089] In step S300, a spliced ​​feature vector sequence is generated by concatenating the first query interaction behavior features, the interaction keyword vectors of multiple query interaction texts, the second query interaction behavior features, and the medical knowledge features of the query disease knowledge point set one by one.

[0090] Step S400: Determine the first contextual features of the query interaction text corresponding to each interactive keyword vector in the intelligent triage query activity, and determine the second contextual features of the query disease knowledge point corresponding to each medical knowledge feature in the query disease knowledge point set.

[0091] Step S500: Combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the spliced ​​feature vector sequence to generate the target spliced ​​feature vector sequence.

[0092] In step S500, by combining the first query interaction behavior distribution, the first context features corresponding to each determined interaction keyword vector, the second query interaction behavior distribution, and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the spliced ​​feature vector sequence, thus completing the position embedding process of the spliced ​​feature vector and generating the target spliced ​​feature vector sequence.

[0093] Step S600: Combine the target splicing feature vector sequence to generate a triage feedback result sequence corresponding to the intelligent triage inquiry activity and at least one inquiry feedback extension result corresponding to the triage feedback result sequence.

[0094] In some possible implementations, combining the target concatenation feature vector sequence to generate a triage feedback result sequence corresponding to the intelligent triage inquiry activity may specifically include:

[0095] Step S601: Load the target concatenated feature vector sequence into the triage feedback prediction model, and output the current triage feedback result of the triage feedback result sequence. The current triage feedback result of the obtained triage feedback result sequence is the first triage feedback result.

[0096] For some possible implementations, the current triage feedback result of the triage feedback result sequence can be obtained based on the following method. For example, the triage feedback prediction model includes a semantic editing unit and a feedback prediction unit. The target concatenated feature vector sequence is loaded into the semantic editing unit to obtain the semantic edit vector corresponding to the current triage feedback result of the triage feedback result sequence, thus completing the encoding of the current triage feedback result. Then, the feedback prediction unit parses the semantic edit vector corresponding to the current triage feedback result of the triage feedback result sequence to obtain first feedback prediction information, completing the decoding. The first feedback prediction information includes the support level of each triage feedback result in the triage feedback result sequence. Then, the first feedback prediction information is used to determine the current triage feedback result of the triage feedback result sequence. For example, the feedback statement with the highest support level is determined in the first feedback prediction information, and the triage feedback result corresponding to that feedback statement is taken as the current triage feedback result of the triage feedback result sequence. Finally, a triage feedback result sequence corresponding to the intelligent triage inquiry activity is generated. Additionally, several feedback statements with high support can be identified from the initial feedback prediction information. The corresponding triage feedback results for these statements can then be designated as candidate triage feedback results for the triage feedback result sequence. Finally, multiple triage feedback result sequences corresponding to the intelligent triage inquiry activity will be generated.

[0097] For example, combining the first feedback prediction information, determining the first guidance feedback result in the guidance feedback result sequence includes: sorting the support in descending order in the first feedback prediction information; determining the top N support in the descending order queue, and using the guidance feedback results corresponding to the N support as reference guidance feedback results for the first guidance feedback result in the guidance feedback result sequence, where N≥1. The updated target concatenated feature vector sequence is loaded into the semantic editing unit, iterated until all guidance feedback statements in the guidance feedback result sequence are obtained, including: combining the reference guidance feedback results of the first guidance feedback result, generating reference guidance feedback results for the remaining guidance feedback results one by one; combining the reference guidance feedback results of each guidance feedback result in the guidance feedback result sequence, determining N guidance feedback result sequences. In other words, each reference guidance feedback result of the first guidance feedback result is loaded into the guidance feedback prediction model to obtain N reference guidance feedback results for the second guidance feedback result. This process yields N² reference guidance feedback results for the first and second guidance feedback results.

[0098] Step S602: Determine whether the current triage feedback result of the triage feedback result sequence obtained in step S601 is all triage feedback statements. If not, proceed to step S603; if yes, end.

[0099] Step S603: Generate the medical knowledge features of the current triage feedback result sequence.

[0100] Step S604: Combining the medical knowledge features of the current triage feedback result sequence and its carrying status in the triage feedback result sequence, update the target splicing feature vector sequence based on feature fusion.

[0101] For example, the current triage feedback result in the triage feedback result sequence is the first triage feedback result (the processing process for triage feedback results in other positions is the same). Combining the medical knowledge features of the first triage feedback result in the triage feedback result sequence and its carrying status within the sequence, the target concatenated feature vector sequence is updated based on feature fusion. This includes concatenating the interaction keyword vectors of multiple inquiry interaction texts, the medical knowledge features of the inquiry disease knowledge point set, and the medical knowledge features of the first triage feedback result in the triage feedback result sequence to update the concatenated feature vector sequence. Next, combining the first context features corresponding to each interaction keyword vector, the second context features corresponding to each medical knowledge feature of the inquiry disease knowledge point set, and the second context features of the first triage feedback result in the triage feedback result sequence, feature nodes are assigned to each concatenated feature vector in the updated concatenated feature vector sequence to update the target concatenated feature vector sequence, thus completing the update.

[0102] For other possible implementations, combining the medical knowledge features of the first triage feedback result in the triage feedback result sequence and its carrying state in the triage feedback result sequence, the target concatenated feature vector sequence is updated based on feature fusion. This includes: concatenating the interaction keyword vectors of multiple inquiry interaction texts, the medical knowledge features of the inquiry disease knowledge point set, and the medical knowledge features of the first triage feedback result in the triage feedback result sequence to update the concatenated feature vector sequence. Then, combining the individuals corresponding to each interaction keyword vector, the individuals corresponding to each medical knowledge feature of the inquiry disease knowledge point set, and the individuals corresponding to the inquiry disease knowledge point of the first triage feedback result in the triage feedback result sequence, individual embedding encoding is performed on each concatenated feature vector in the concatenated feature vector sequence to update the concatenated feature vector sequence, thus completing the update. Alternatively, by combining the individuals corresponding to each interactive keyword vector, the individuals corresponding to each medical knowledge feature of the disease knowledge point set, and the individuals corresponding to the first guidance feedback result of the guidance feedback result sequence, individual embedding encoding can be performed on each spliced ​​feature vector in the target spliced ​​feature vector sequence to update the target spliced ​​feature vector sequence.

[0103] Next, return to step S601, load the updated target splicing feature vector sequence into the triage feedback prediction model, traverse the target splicing feature vector sequence based on the medical knowledge features of each obtained triage feedback result, until all triage feedback statements of the triage feedback result sequence are obtained.

[0104] Specifically, when obtaining all the guidance feedback statements in the guidance feedback result sequence, one semantic editing vector is selected from multiple semantic editing vectors generated by the semantic editing unit as the intelligent guidance inquiry editing vector and one semantic editing vector is selected as the inquiry disease editing vector. For example, the semantic editing vector used as the intelligent guidance inquiry editing vector can be the semantic editing vector corresponding to the first inquiry interaction behavior feature distributed before the interaction keyword vectors of multiple inquiry interaction texts, and the semantic editing vector used as the inquiry disease editing vector is the semantic editing vector corresponding to the second inquiry interaction behavior feature between the interaction keyword vectors of multiple inquiry interaction texts and the various medical knowledge features of the inquiry disease knowledge point set. Next, the correlation between the intelligent guidance inquiry editing vector and the inquiry disease editing vector is determined. In some implementations, cosine distance can be used to obtain the correlation between the intelligent guidance inquiry editing vector and the inquiry disease editing vector. In this case, the correlation is located in [-1, 1], and the closer the correlation is to 1, the higher the correlation. Then, the correlation result between the intelligent guidance inquiry activity and the inquiry disease knowledge point set is determined by combining the correlation. Furthermore, the correlation between the obtained triage feedback result sequences and the set of disease knowledge points inquired can be further evaluated. When obtaining multiple triage feedback result sequences corresponding to the intelligent triage inquiry activity, the multiple triage feedback result sequences are filtered based on the correlation between the generated triage feedback result sequences and the set of disease knowledge points inquired, and triage feedback result sequences with low relevance are cleaned.

[0105] For example, for each of the N triage feedback result sequences, the following steps are performed: When obtaining all triage feedback statements in the triage feedback result sequence, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the inquiry disease editing vector and one semantic editing vector is determined as the triage feedback editing vector. For instance, one semantic editing vector as the inquiry disease editing vector can be the semantic editing vector corresponding to the second inquiry interaction behavior feature between the interaction keyword vectors distributed across multiple inquiry interaction texts and the various medical knowledge features of the inquiry disease knowledge point set, and one semantic editing vector as the triage feedback editing vector is the semantic editing vector corresponding to the second inquiry interaction behavior feature between the various medical knowledge features distributed across the inquiry disease knowledge point set and the various medical knowledge features of the triage feedback result sequence. Next, the correlation between the inquiry disease editing vector and the triage feedback editing vector is determined. In some implementations, the correlation between the inquiry disease editing vector and the triage feedback editing vector is obtained using vector cosine distance. When the maximum relevance is greater than the threshold relevance, the triage feedback result sequence corresponding to that relevance is determined to be the triage feedback result sequence corresponding to the intelligent triage inquiry activity; otherwise, it is determined that there is no triage feedback result sequence corresponding to the intelligent triage inquiry activity.

[0106] Higher relevance in the triage feedback result sequence indicates higher accuracy, representing the correlation between the set of disease knowledge points inquired and the triage feedback result sequence. This has an incentive effect on prior triage feedback result sequences and spurious triage feedback result sequences. By filtering the generated triage feedback result sequences based on the relevance between the set of disease knowledge points inquired and the triage feedback result sequence, and removing triage feedback result sequences with low relevance, the accuracy of the triage feedback result sequence can be increased.

[0107] Furthermore, based on the relevance between the intelligent triage inquiry activity and the triage feedback result sequence, as well as the relevance between the inquiry disease knowledge point set and the triage feedback result sequence, the triage feedback result sequence obtained solely through the intelligent triage inquiry activity can be determined. Specifically, for each of the N triage feedback result sequences, the following steps are performed: when obtaining all the triage feedback statements in the triage feedback result sequence, one semantic editing vector is selected from the multiple semantic editing vectors generated by the semantic editing unit as the intelligent triage inquiry editing vector, one semantic editing vector as the inquiry disease editing vector, and one semantic editing vector as the triage feedback editing vector. For example, a semantic edit vector serving as the intelligent triage inquiry editing vector could be a semantic edit vector corresponding to the first inquiry interaction behavior feature distributed before the interaction keyword vectors of multiple inquiry interaction texts. Similarly, a semantic edit vector serving as the inquiry disease editing vector could be a semantic edit vector corresponding to the second inquiry interaction behavior feature between the interaction keyword vectors of multiple inquiry interaction texts and the various medical knowledge features of the inquiry disease knowledge point set. Likewise, a semantic edit vector serving as the triage feedback editing vector could be a semantic edit vector corresponding to the second inquiry interaction behavior feature between the various medical knowledge features of the inquiry disease knowledge point set and the various medical knowledge features of the triage feedback result sequence. Next, the correlation between the intelligent triage inquiry editing vector and the inquiry disease editing vector, and the correlation between the intelligent triage inquiry editing vector and the triage feedback editing vector, are determined. For example, the correlation between the intelligent triage inquiry activity and the triage feedback result sequence, and the correlation between the inquiry disease editing vector and the triage feedback editing vector, can be obtained based on cosine distance. When the correlation between the intelligent triage inquiry editing vector and the disease inquiry editing vector is less than the first set correlation and the correlation between the intelligent triage inquiry editing vector and the triage feedback editing vector is greater than the second set correlation, the triage feedback result sequence is determined as a triage feedback result sequence obtained only based on the intelligent triage inquiry activity.

[0108] Step S700: Based on the triage feedback results and / or the extended inquiry feedback results, provide triage content information feedback to the patient end corresponding to the target doctor-patient interaction process.

[0109] The following describes the training methods for each module used in the generation process of the triage feedback result sequence in this application embodiment. A feature node allocation model, a semantic editing unit, and a feedback prediction unit are trained in conjunction with a first triage feedback learning data sequence. The first triage feedback learning data sequence includes multiple first triage feedback learning data sets. Each first triage feedback learning data set includes a first triage inquiry activity to be learned, a first set of training inquiry disease knowledge points corresponding to the first triage inquiry activity to be learned, and a priori triage feedback result sequence corresponding to the first intelligent triage inquiry activity and the first set of inquiry disease knowledge points.

[0110] In at least one first triage feedback learning data point in the first triage feedback learning data sequence, for each first triage feedback learning data point, the following steps are performed:

[0111] (1) Obtain multiple first training query interaction texts in the first learning query activity of the first triage feedback learning data, and parse the interaction keyword vectors of each training query interaction text in the multiple first training query interaction texts respectively.

[0112] (2) Obtain the first training query disease knowledge point set corresponding to the first learning guidance query activity, and analyze the medical knowledge features of each query disease knowledge point in the first training query disease knowledge point set, wherein each medical knowledge feature is consistent with the disease semantic direction of each interactive keyword vector.

[0113] (3) Convert at least one of the prior triage feedback results in the triage feedback result sequence into a triage feedback observation vector, generate a triage feedback observation vector set, and analyze the medical knowledge features of each query disease knowledge point in the triage feedback observation vector set respectively. The medical knowledge features of each query disease knowledge point in the triage feedback observation vector set are consistent with the disease semantic direction of each interactive keyword vector.

[0114] (4) The interaction keyword vectors of multiple first training query interaction texts, the medical knowledge features of the first training query disease knowledge point set, and the medical knowledge features of each query disease knowledge point in the triage feedback observation vector set are concatenated to generate the first training concatenated feature vector sequence.

[0115] (5) Determine the first context features of the first training query interaction text corresponding to the interaction keyword vector of each first training query interaction text in the first learning guidance query activity; determine the second context features of the query disease knowledge point corresponding to each medical knowledge feature in the first training query disease knowledge point set in the first training query disease knowledge point set; and determine the second context features of the query disease knowledge point corresponding to the medical knowledge feature of each query disease knowledge point in the guidance feedback observation vector set in the guidance feedback observation vector set.

[0116] (6) Combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the first training spliced ​​feature vector sequence to generate the first training target spliced ​​feature vector sequence.

[0117] (7) Combine the first training objective with the concatenated feature vector sequence to generate at least one guidance feedback observation vector in the guidance feedback observation vector set.

[0118] (8) Obtain the learning cost loss1 between at least one triage feedback observation vector and the actual triage feedback result.

[0119] (9) At least combine the learning cost loss1 to train the feature node allocation model, semantic editing unit and feedback prediction unit.

[0120] In constructing the learning value in this embodiment, in addition to the learning value loss1 between the triage feedback observation vector and the actual triage feedback result, multivariate (triple) learning value is also introduced between the intelligent triage inquiry activity and the set of inquired disease knowledge points, between the intelligent triage inquiry activity and the generated triage feedback result sequence, and between the set of inquired disease knowledge points and the generated triage feedback result sequence. Since the intelligent triage inquiry activity, the set of inquired disease knowledge points, and the generated triage feedback result sequence are all based on the same triage feedback prediction model to obtain a concatenated feature vector, the reason for introducing multivariate learning value is to make the triage feedback result sequence generated by the triage feedback prediction model as close as possible to the intelligent triage inquiry activity or the set of inquired disease knowledge points in the dimension of inquired disease.

[0121] For example, the process of obtaining the aforementioned multivariate learning cost value may include: determining from multiple semantic editing vectors generated by the semantic editing unit one semantic editing vector as the intelligent triage inquiry editing vector, one semantic editing vector as the disease inquiry editing vector, and one semantic editing vector as the triage feedback editing vector. Then, combining the correlation between the intelligent triage inquiry editing vector and its corresponding disease inquiry editing vector, the correlation between the intelligent triage inquiry editing vector and non-corresponding disease inquiry editing vectors, and the correlation between the disease inquiry editing vector and its corresponding intelligent triage inquiry editing vector, and the correlation between the disease inquiry editing vector and non-corresponding intelligent triage editing vectors, the learning cost value loss2 is obtained and determined as the multivariate learning cost value between the intelligent triage inquiry activity and the set of disease knowledge points. The method for obtaining the learning cost value loss2 can refer to the following formula:

[0122] lossoHH2=G1+G2

[0123] G1=Max(H(V,K')+uH(V,K),0)

[0124] G2=Max(H(V',K)+uH(V,K),0)

[0125] Wherein, lossoHH2 is the learning cost lossoHHoHH2, V, K, and K' are respectively the intelligent triage inquiry activity, the prior inquiry disease knowledge point set (the inquiry disease knowledge point set corresponding to the intelligent triage inquiry activity), and the pseudo inquiry disease knowledge point set (the inquiry disease knowledge point set that does not correspond to the intelligent triage inquiry activity, i.e., the inquiry disease knowledge point set corresponding to other intelligent triage inquiry activities). K, V, and V' are respectively the inquiry disease knowledge point set, the prior intelligent triage inquiry activity (the intelligent triage inquiry activity corresponding to the inquiry disease knowledge point set), and the pseudo intelligent triage activity (the intelligent triage activity that does not correspond to the inquiry disease knowledge point set, i.e., the intelligent triage activity corresponding to other inquiry disease knowledge point sets). u is an adjustable parameter, which can be set according to actual conditions, such as 0.3. During training, a batch of triage feedback learning datasets is typically tested. For this batch of data, multiple sets of semantic edit vectors are obtained for intelligent triage inquiry activities, sets of disease knowledge points inquiries, and sequences of triage feedback results. Semantic edit vectors for the same set of intelligent triage inquiry activities, sets of disease knowledge points inquiries, and sequences of triage feedback results correspond; semantic edit vectors for different sets do not correspond. If, within a batch of triage feedback learning data, an intelligent triage inquiry edit vector has multiple non-corresponding disease edit vectors, the average correlation between the intelligent triage inquiry edit vector and each non-corresponding disease edit vector is calculated and denoted as K(V, K'). Similarly, if a disease edit vector has multiple non-corresponding intelligent triage inquiry edit vectors, the average correlation between the disease edit vector and each non-corresponding intelligent triage editing vector is calculated and denoted as K(V', K).

[0126] By combining the correlation between the intelligent triage inquiry editing vector and the corresponding triage feedback editing vector, the correlation between the intelligent triage inquiry editing vector and the non-corresponding triage feedback editing vector, and the correlation between the triage feedback editing vector and the corresponding intelligent triage inquiry editing vector, as well as the correlation between the triage feedback editing vector and the non-corresponding intelligent triage editing vector, the learning cost value lossoHH3 is obtained and used as the multivariate learning cost value between the intelligent triage inquiry activity and the triage feedback result sequence.

[0127] loss3 = G3 + G4

[0128] G3=Max(H(V,M')+uH(V,M),0)

[0129] G4=Max(H(V',M)+uH(V,M),0)

[0130] Where loss3 is the learning cost lossoHH3, V, M, and M' represent the intelligent triage inquiry activity, the prior intelligent triage feedback result sequence (the triage feedback result sequence corresponding to the intelligent triage inquiry activity), and the pseudo-intelligent triage feedback result sequence (the triage feedback result sequence that does not correspond to the intelligent triage inquiry activity, i.e., the triage feedback result sequence that corresponds to other intelligent triage inquiry activities), respectively. M, V, and V' represent the triage feedback result sequence, the prior intelligent triage inquiry activity (the intelligent triage inquiry activity corresponding to the triage feedback result sequence), and the pseudo-intelligent triage activity (the intelligent triage activity that does not correspond to the triage feedback result sequence, i.e., the intelligent triage activity that corresponds to other triage feedback result sequences), respectively, and u is the adjustment coefficient.

[0131] By combining the correlation between the disease inquiry edit vector and the corresponding triage feedback edit vector, the correlation between the disease inquiry edit vector and the non-corresponding triage feedback edit vector, and the correlation between the triage feedback edit vector and the corresponding disease inquiry edit vector, and the correlation between the triage feedback edit vector and the non-corresponding disease inquiry edit vector, the learning cost lossoHH4 is obtained as the multivariate learning cost between the disease inquiry knowledge point set and the triage feedback result sequence.

[0132] loss4 = G5 + G6

[0133] G5=Max(H(K,M')+uH(K,M),0)

[0134] G6=Max(H(K',M)+uH(K,M),0)

[0135] Where loss4 is the learning cost lossoHH4, K, M, and M' represent the set of disease knowledge points to be inquired, the prior triage feedback result sequence (the triage feedback result sequence corresponding to the set of disease knowledge points to be inquired), and the pseudo triage feedback result sequence (the triage feedback result sequence that does not correspond to the set of disease knowledge points to be inquired, i.e., the triage feedback result sequence that corresponds to other sets of disease knowledge points to be inquired). M, K, and K' represent the triage feedback result sequence, the prior set of disease knowledge points to be inquired (the ...

[0136] At least the learning cost value lossoHH1 is combined to train the feature node allocation model, semantic editing unit, and feedback prediction unit, including: combining the weighted fusion values ​​of learning cost value lossoHH1, learning cost value lossoHH2, learning cost value lossoHH3, and learning cost value lossoHH4 to train the feature node allocation model, semantic editing unit, and feedback prediction unit.

[0137] In some possible implementations, a pre-training process is included before training the feature node allocation model, semantic editing unit, and feedback prediction unit using the first triage feedback learning data sequence. The training process further includes: training the feature node allocation model and semantic editing unit based on the second triage feedback learning data sequence. The second triage feedback learning data sequence includes multiple second triage feedback learning data sets, each including a second triage inquiry activity to be learned and a set of second training inquiry disease knowledge points corresponding to that activity. The difference between the second and first triage feedback learning data sequences is that the second sequence is a triage feedback learning data sequence that is labeled with the learning basis data; that is, it does not have matching labels to annotate prior information.

[0138] When training the feature node allocation model and semantic editing unit based on the second triage feedback learning data sequence, for each second triage feedback learning data in at least one second triage feedback learning data in the second triage feedback learning data sequence, the following steps are performed:

[0139] (1) Obtain multiple second training inquiry interaction texts from the learning inquiry activities of the second triage feedback learning data, and parse the interaction keyword vectors of each training inquiry interaction text in the multiple second training inquiry interaction texts respectively.

[0140] (2) Obtain the second training query disease knowledge point set corresponding to the second learning guidance query activity, and analyze the medical knowledge features of each query disease knowledge point in the second training query disease knowledge point set respectively, wherein each medical knowledge feature is consistent with the disease semantic direction of each interactive keyword vector.

[0141] (3) The interaction keyword vectors of multiple second training query interaction texts and the medical knowledge features of the second training query disease knowledge point set are concatenated to generate a second training concatenated feature vector sequence.

[0142] (4) Determine the first context features of the query interaction text corresponding to the interaction keyword vector of each second training query interaction text in the second learning guidance query activity, and determine the second context features of the query disease knowledge point corresponding to each medical knowledge feature in the second training query disease knowledge point set.

[0143] (5) Combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the second training spliced ​​feature vector sequence to generate the second training target spliced ​​feature vector sequence.

[0144] A pre-training method is employed on a second set of triage feedback learning data sequences containing a large number of triage feedback results without annotations. This method learns the carrying state of the concatenated feature vectors of intelligent triage inquiry activities and the sets of inquiry disease knowledge points. Based on this, the concatenated feature vectors of intelligent triage inquiry activities and the concatenated feature vectors of inquiry disease knowledge point sets are more similar in tendency. Next, the second training target concatenated feature vector sequence is loaded into the semantic editing unit. From the multiple semantic editing vectors generated by the semantic editing unit, one semantic editing vector is determined as the intelligent triage inquiry editing vector, and another as the inquiry disease editing vector. Then, by combining the correlation between the intelligent triage inquiry editing vector and its corresponding inquiry disease editing vector, the correlation between the intelligent triage inquiry editing vector and non-corresponding inquiry disease editing vectors, and the correlation between the inquiry disease editing vector and its corresponding intelligent triage editing vector, and the correlation between the inquiry disease editing vector and non-corresponding intelligent triage editing vectors, a learning cost loss5 is obtained. Finally, the feature node allocation model and the semantic editing unit are trained using the learning cost loss5.

[0145] Combining the same inventive concept, combining Figure 2 As shown in the illustration, this application also provides a patient service system 100. The patient service system 100 can vary significantly due to different configurations or performance, and may include one or more central processing units (CPUs) 112 (e.g., one or more processors) and a memory 111. The memory 111 may be temporary or persistent storage. The program stored in the memory 111 may include one or more modules, each module including a series of instruction operations on the patient service system 100. Furthermore, the central processing unit 112 may be configured to communicate with the memory 111 and execute the series of instruction operations stored in the memory 111 on the patient service system 100.

[0146] The patient service system 100 may also include one or more power supplies, one or more communication units 113, one or more output interfaces, and / or one or more operating systems, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, etc.

[0147] In addition, this application embodiment also provides a storage medium for storing a computer program for executing the method provided in the above embodiment.

[0148] This application also provides a computer program product including instructions that, when run on a computer, cause the computer to perform the methods provided in the above embodiments.

[0149] Those skilled in the art will understand that all or part of the steps of the above method embodiments can be implemented by hardware related to program instructions. The aforementioned program can be stored in a computer-readable storage medium. When the program is executed, it performs the steps of the above method embodiments. The aforementioned storage medium can be at least one of the following media: read-only memory (ROM), RAM, magnetic disk, or optical disk, etc., and other media capable of storing program code.

[0150] It should be noted that the various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referred to mutually. The objectives of each embodiment are to describe the differences from other embodiments. In particular, for the device and system embodiments, since they are basically similar to the method embodiments, the description is relatively simple, and the relevant parts can be referred to the description of the method embodiments. The device and system embodiments described above are merely illustrative, and the units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of the solution in this embodiment according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0151] The above description is merely one specific embodiment of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the technical scope disclosed in this application should be included 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 for intelligent triage and inquiry feedback processing based on doctor-patient interaction, applied to a doctor-patient service system, characterized in that: include: The intelligent triage and inquiry activities of the target doctor-patient interaction process are subjected to feature correlation mining between the inquiry interaction text and the set of disease knowledge points in the inquiry, and the triage content information is fed back to the patient end corresponding to the target doctor-patient interaction process based on the feature correlation mining results. Obtain the patient's evaluation data regarding the feedback from the triage information; The patient sentiment analysis is performed on the medical evaluation data, and the user feedback is provided to the doctor-patient interaction platform running the doctor-patient service system based on the results of the patient sentiment analysis. The steps of performing feature correlation mining between the intelligent triage and inquiry activities of the target doctor-patient interaction process and the set of disease knowledge points in the inquiry, and providing triage content information feedback to the patient end corresponding to the target doctor-patient interaction process based on the feature correlation mining results, include: In the intelligent triage and inquiry activity of the target doctor-patient interaction process, multiple inquiry interaction texts are obtained, and the interaction keyword vectors of each inquiry interaction text in the multiple inquiry interaction texts are parsed respectively. Obtain the set of disease knowledge points corresponding to the intelligent triage and inquiry activity, and analyze the medical knowledge features of each disease knowledge point in the set of disease knowledge points; The interaction keyword vectors of the multiple query interaction texts and the corresponding medical knowledge features in the query disease knowledge point set are concatenated to generate a concatenated feature vector sequence. Determine the first contextual features of the query interaction text corresponding to each interactive keyword vector in the intelligent triage query activity, and determine the second contextual features of the query disease knowledge point corresponding to each medical knowledge feature in the query disease knowledge point set; By combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the spliced ​​feature vector sequence to generate the target spliced ​​feature vector sequence. And by combining the target splicing feature vector sequence, a triage feedback result sequence corresponding to the intelligent triage inquiry activity and at least one inquiry feedback extension result corresponding to the triage feedback result sequence are generated; Based on the triage feedback results and / or the inquiry feedback extension results, triage content information is fed back to the patient end corresponding to the target doctor-patient interaction process. The step of combining the target concatenated feature vector sequence to generate a triage feedback result sequence corresponding to the intelligent triage inquiry activity includes: The target concatenated feature vector sequence is loaded into the triage feedback prediction model, and the first triage feedback result of the triage feedback result sequence is output. The medical knowledge features of the first triage feedback result in the sequence of triage feedback results; Combining the medical knowledge features of the first triage feedback result in the triage feedback result sequence and its carrying state in the triage feedback result sequence, the target spliced ​​feature vector sequence is updated based on feature fusion; The updated target splicing feature vector sequence is loaded into the triage feedback prediction model. The target splicing feature vector sequence is updated by traversing the medical knowledge features of each obtained triage feedback result until all triage feedback statements of the triage feedback result sequence are obtained. Combining the medical knowledge features of the first triage feedback result in the triage feedback result sequence and its carrying state in the triage feedback result sequence, the target spliced ​​feature vector sequence is updated based on feature fusion, including: The interaction keyword vectors of the multiple query interaction texts, the medical knowledge features of the query disease knowledge point set, and the medical knowledge features of the first triage feedback result of the triage feedback result sequence are concatenated to update the concatenated feature vector sequence. By combining the first context features corresponding to each interactive keyword vector, the second context features corresponding to each medical knowledge feature of the disease knowledge point set, and the second context features of the first triage feedback result of the triage feedback result sequence, feature nodes are assigned to each spliced ​​feature vector in the updated spliced ​​feature vector sequence to update the target spliced ​​feature vector sequence.

2. The intelligent triage and inquiry feedback processing method based on doctor-patient interaction according to claim 1, characterized in that, The triage feedback prediction model includes a semantic editing unit and a feedback prediction unit, wherein the target concatenated feature vector sequence is loaded into the triage feedback prediction model, and the first triage feedback result of the triage feedback result sequence is output, including: The target splicing feature vector sequence is loaded into the semantic editing unit, and the semantic editing vector corresponding to the first triage feedback result in the triage feedback result sequence is output. The first feedback prediction information is obtained by parsing the semantic edit vector of the first triage feedback result in the triage feedback result sequence through the feedback prediction unit. The first feedback prediction information includes the support of each triage feedback result in the triage feedback result sequence. Based on the first feedback prediction information, the first triage feedback result in the triage feedback result sequence is determined.

3. The intelligent triage and inquiry feedback processing method based on doctor-patient interaction according to claim 1, characterized in that, The method further includes: When all the triage feedback statements in the triage feedback result sequence are obtained, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the intelligent triage inquiry editing vector and one semantic editing vector is determined as the inquiry disease editing vector; Determine the correlation between the intelligent triage query editing vector and the query disease editing vector; Based on the relevance, the correlation between the intelligent triage inquiry activity and the set of disease knowledge points inquired is determined; Among them, a semantic editing vector serving as the intelligent triage inquiry editing vector is a semantic editing vector corresponding to the first inquiry interaction behavior feature distributed before the interaction keyword vectors of the multiple inquiry interaction texts, and a semantic editing vector serving as the inquiry disease editing vector is a semantic editing vector corresponding to the second inquiry interaction behavior feature distributed between the interaction keyword vectors of the multiple inquiry interaction texts and the various medical knowledge features of the inquiry disease knowledge point set.

4. The intelligent triage and inquiry feedback processing method based on doctor-patient interaction according to claim 2, characterized in that, The step of determining the first triage feedback result in the triage feedback result sequence by combining the first feedback prediction information includes: In the first feedback prediction information, the support scores are sorted in descending order; The first N support levels in the descending order queue are determined, and the corresponding triage feedback results are used as the reference triage feedback result for the first triage feedback result in the triage feedback result sequence. The updated target concatenated feature vector sequence is loaded into the semantic editing unit, and the target concatenated feature vector sequence is updated based on the medical knowledge features of each obtained triage feedback result until all triage feedback statements in the triage feedback result sequence are obtained, including: Based on the reference guidance feedback result of the first guidance feedback result, reference guidance feedback results are generated one by one for the remaining guidance feedback results; By combining the reference guidance feedback results of each guidance feedback result in the guidance feedback result sequence, N guidance feedback result sequences are determined; The method further includes: For each of the N triage feedback result sequences, the following steps are performed: When all the triage feedback statements in the triage feedback result sequence are obtained, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the disease inquiry editing vector and one semantic editing vector is determined as the triage feedback editing vector. And determine the correlation between the disease inquiry edit vector and the triage feedback edit vector; When the maximum relevance is greater than the threshold relevance, the triage feedback result sequence corresponding to the maximum relevance is determined as the triage feedback result sequence corresponding to the intelligent triage inquiry activity; otherwise, it is determined that there is no triage feedback result sequence corresponding to the intelligent triage inquiry activity.

5. The intelligent triage and inquiry feedback processing method based on doctor-patient interaction according to claim 4, characterized in that, The method further includes: For each of the N triage feedback result sequences, the following steps are performed: When all the triage feedback statements in the triage feedback result sequence are obtained, one semantic editing vector is determined from the multiple semantic editing vectors generated by the semantic editing unit as the intelligent triage inquiry editing vector, one semantic editing vector is determined as the inquiry disease editing vector, and one semantic editing vector is determined as the triage feedback editing vector; And determine the correlation between the intelligent triage inquiry editing vector and the inquiry disease editing vector, and determine the correlation between the intelligent triage inquiry editing vector and the triage feedback editing vector; When the correlation between the intelligent triage inquiry editing vector and the inquiry disease editing vector is less than a first set correlation and the correlation between the intelligent triage inquiry editing vector and the triage feedback editing vector is greater than a second set correlation, the triage feedback result sequence is determined as a triage feedback result sequence obtained only based on the intelligent triage inquiry activity.

6. The intelligent triage and inquiry feedback processing method based on doctor-patient interaction according to claim 2, characterized in that, The method for obtaining the interactive keyword vector and the medical knowledge features, as well as the feature node allocation based on the feature node allocation model, further includes: The feature node allocation model, the semantic editing unit, and the feedback prediction unit are trained by combining a first triage feedback learning data sequence. The first triage feedback learning data sequence includes multiple first triage feedback learning data sets. Each first triage feedback learning data set includes a first triage inquiry activity to be learned, a first set of training inquiry disease knowledge points corresponding to the first triage inquiry activity to be learned, and a priori triage feedback result sequence corresponding to the first triage inquiry activity to be learned and the first set of training inquiry disease knowledge points. Training the feature node allocation model, the semantic editing unit, and the feedback prediction unit by combining the first triage feedback learning data sequence includes: In at least one first triage feedback learning data point in the first triage feedback learning data sequence, for each first triage feedback learning data point, the following steps are performed: Multiple first training query interaction texts are obtained from the first learning-to-learn triage query activity of the first triage feedback learning data, and the interaction keyword vectors of each training query interaction text in the multiple first training query interaction texts are parsed respectively. Obtain the first training query disease knowledge point set corresponding to the first learning triage query activity, and analyze the medical knowledge features of each query disease knowledge point in the first training query disease knowledge point set, wherein each medical knowledge feature is consistent with the disease semantic direction of each interactive keyword vector; At least one triage feedback result in the prior triage feedback result sequence is converted into a triage feedback observation vector to generate a triage feedback observation vector set. The medical knowledge features of each disease knowledge point in the triage feedback observation vector set are then analyzed. The medical knowledge features of each disease knowledge point in the triage feedback observation vector set are consistent with the disease semantic direction of each interactive keyword vector. The interaction keyword vectors of the multiple first training query interaction texts, the medical knowledge features of the first training query disease knowledge point set, and the medical knowledge features of each query disease knowledge point in the triage feedback observation vector set are concatenated to generate a first training concatenated feature vector sequence. Determine the first contextual features of the query interaction text corresponding to the interaction keyword vector of each first training query interaction text in the first learning guidance query activity; determine the second contextual features of the query disease knowledge point corresponding to each medical knowledge feature in the first training query disease knowledge point set in the first training query disease knowledge point set; and determine the second contextual features of the disease knowledge point corresponding to the medical knowledge feature of each query disease knowledge point in the guidance feedback observation vector set in the guidance feedback observation vector set. By combining the first context features corresponding to each interactive keyword vector and the second context features corresponding to each medical knowledge feature, feature nodes are assigned to each spliced ​​feature vector in the first training spliced ​​feature vector sequence to generate the first training target spliced ​​feature vector sequence. By combining the first training target with the concatenated feature vector sequence, at least one triage feedback observation vector is generated from the triage feedback observation vector set. Obtain the learning cost loss1 between the at least one triage feedback observation vector and the actual triage feedback result; The feature node allocation model, the semantic editing unit, and the feedback prediction unit are trained by combining at least the learning cost loss1.

7. The intelligent triage and inquiry feedback processing method based on doctor-patient interaction according to claim 6, characterized in that, The step of training the feature node allocation model, the semantic editing unit, and the feedback prediction unit by combining the first triage feedback learning data sequence includes: From the multiple semantic editing vectors generated by the semantic editing unit, determine one semantic editing vector as the intelligent triage inquiry editing vector, one semantic editing vector as the disease inquiry editing vector, and one semantic editing vector as the triage feedback editing vector; By combining the correlation between the intelligent triage inquiry editing vector and the corresponding inquiry disease editing vector, the correlation between the intelligent triage inquiry editing vector and the non-corresponding inquiry disease editing vector, and the correlation between the inquiry disease editing vector and the corresponding intelligent triage inquiry editing vector, and the correlation between the inquiry disease editing vector and the non-corresponding intelligent triage editing vector, the learning cost loss2 is obtained. By combining the correlation between the intelligent triage inquiry editing vector and the corresponding triage feedback editing vector, the correlation between the intelligent triage inquiry editing vector and the non-corresponding triage feedback editing vector, and the correlation between the triage feedback editing vector and the corresponding intelligent triage editing vector, as well as the correlation between the triage feedback editing vector and the non-corresponding intelligent triage editing vector, the learning cost loss3 is obtained. By combining the correlation between the disease inquiry edit vector and the corresponding triage feedback edit vector, the correlation between the disease inquiry edit vector and the non-corresponding triage feedback edit vector, and the correlation between the triage feedback edit vector and the corresponding disease inquiry edit vector, and the correlation between the triage feedback edit vector and the non-corresponding disease inquiry edit vector, the learning cost loss4 is obtained. Furthermore, the training of the feature node allocation model, the semantic editing unit, and the feedback prediction unit, in conjunction with at least the learning cost loss1, includes: The feature node allocation model, the semantic editing unit, and the feedback prediction unit are trained by combining the weighted fusion values ​​of learning loss1, learning loss2, learning loss3, and learning loss4.

8. A doctor-patient service system, characterized in that, include: processor; A memory, wherein the memory stores a computer program, which, when executed, implements the intelligent triage and inquiry feedback processing method based on doctor-patient interaction as described in any one of claims 1-7.

Images