Doctor-patient experience optimization suggestion generation system driven by data mining technology

Abstract

The invention belongs to the technical field of medical information, and particularly relates to a doctor-patient experience optimization suggestion generation system driven by a data mining technology. The scheme comprises an acquisition module, an analysis module and a suggestion module, the acquisition module is used for collecting images, voices, questionnaires and doctor seeing information of patients; the analysis module performs emotion labeling on the comments and the face images by using an emotion analysis model, generates emotion vectors and stores related data in combination with a doctor seeing stage; the suggestion module is used for matching investigation questions according to the emotion vectors and the treatment stages, generating questionnaires after adjusting the number of the questions and sending the questionnaires to the patients. After the patient fills in the questionnaire, the collection module collects results and sends the results to the analysis module, and the results are integrated into promotion suggestions. According to the scheme, the problem that the accuracy of the patient satisfaction degree obtained through questionnaire survey result analysis is low is solved.

Description

Technical Field

[0001] This solution belongs to the field of medical information technology, and specifically involves a doctor-patient experience optimization suggestion generation system driven by data mining technology. Background Art

[0002] As the medical level of social development improves, along with the transformation of the medical model and the enhancement of patients' rights awareness, medical service regulators and providers pay more and more attention to listening to patients' opinions, and improve medical services and promote the harmonious development of doctor-patient relationship accordingly. Patient satisfaction is the comparison and evaluation of the expected quality of medical services by the patient based on his or her understanding of health and personal experience during the medical process.

[0003] The traditional method of doctor-patient satisfaction survey is generally in the form of a questionnaire, that is, patients fill out paper questionnaires or electronic questionnaires through patient terminals (usually mobile phones carried by patients) to convey their personal feelings during the medical process to the hospital. The traditional questionnaire survey method consumes a lot of manpower and material resources in the formulation, filling, collection and statistical analysis of questionnaires, and cannot timely discover and deal with the problems reflected by patients (that is, information feedback is delayed). The existing questionnaire survey method can collect, screen and count the results of patients' questionnaires in batches, solving the problem of inefficient statistical analysis. The existing questionnaire survey method can even set up multiple questionnaires, match more suitable questionnaires for patients according to the patient's treatment department, and obtain the patient's medical experience more accurately. In addition, the existing questionnaire survey method also includes predicting the patient's emotions during the visit by the patient's voice or expression during the visit, and then predicting the patient's doctor-patient satisfaction during the visit based on the patient's emotions during the visit, classifying patients more accurately according to their satisfaction, and matching more accurate questionnaires for patients.

[0004] However, even if the existing questionnaire survey method can segment patients, it is impossible to formulate a questionnaire that fully meets the needs of a certain patient (i.e., meets the patient's individual needs), and the efficiency of obtaining accurate doctor-patient satisfaction still needs to be improved. It is impossible to match sufficiently accurate survey questions to patients, making the patient satisfaction obtained from the analysis of questionnaire survey results inaccurate. Summary of the invention

[0005] The purpose of this solution is to provide a doctor-patient experience optimization suggestion generation system driven by data mining technology to solve the problem of low accuracy of patient satisfaction obtained from questionnaire survey results analysis.

[0006] In order to achieve the above objectives, this solution provides a data mining technology-driven doctor-patient experience optimization suggestion generation system, including: The collection module is used to collect the image information of the patient in the hospital monitoring, to collect the voice information input by the patient terminal and convert the voice information into textual comment data, to collect the results of the questionnaire uploaded by the patient terminal, and to collect the patient's medical information; An analysis module, used to use a preset emotion analysis model to annotate comment data as neutral emotion, positive emotion or negative emotion, and use the annotated result of the comment data as a first vector; used to use a preset emotion analysis model to identify and annotate the facial image of the patient in the image information as a neutral state, a positive state or a negative state, and use the annotated result of the image information as a second vector; the analysis module calculates the first vector and the second vector as the patient's emotion vector according to the acquisition time of the voice information and the image information; the analysis module combines the acquisition location of the image information with the medical consultation information to determine the patient's medical consultation stage, and associates and stores the emotion vector with the same acquisition time with the medical consultation stage, and the medical consultation stage includes a waiting stage, an examination stage and a face-to-face consultation stage; A suggestion module includes various survey topics; the suggestion module is used to obtain the associated stored treatment stage and emotion vector, match the survey topic in the survey topic as a recommended topic according to the emotion vector and the treatment stage, and then adjust the preset number of topics according to the value of the emotion vector, generate a questionnaire with the adjusted number of topics and the recommended topics and send it to the patient terminal; The acquisition module is also used to obtain the completion results of the questionnaire by the patient terminal, and send the completion results to the analysis module, and the analysis module is used to analyze and integrate the completion results into improvement suggestions.

[0007] The principle and technical effect of this solution are as follows: First, this solution collects voice and image information during the patient's visit, analyzes the patient's emotions and the visit stage corresponding to the patient's emotions, matches the patient with accurate survey questions according to the visit stage and emotions, and obtains the patient's accurate satisfaction through these questions. Compared with the existing classification of patients according to emotions, this solution can formulate personalized survey questions for patients, thereby improving the accuracy of doctor-patient satisfaction surveys. In addition, the number of survey questions filled in by the patient is determined according to the intensity of the patient's emotions (the numerical value of the emotion vector), ensuring the pertinence and effectiveness of the questionnaire, and further improving the accuracy and reliability of the survey results.

[0008] Secondly, this solution not only collects the results of traditional questionnaire filling, but also combines multiple data sources such as images and voice, integrates and analyzes these multi-dimensional data through data mining technology, comprehensively captures patients' feedback information, and analyzes patients' emotional states (such as positive, negative or neutral) in real time, and associates them with the stage of treatment (such as waiting, examination, face-to-face consultation), so as to quickly discover potential problems and avoid information feedback lag. This solution can also match patients with accurate questionnaires at any time during the patient's treatment process, which can not only timely obtain the patient's satisfaction during the treatment process, but also timely obtain the patient's needs, provide patients with solutions to problems, and provide patients with emotional care in a timely manner, thereby improving the patient's satisfaction during the treatment process.

[0009] Furthermore, the existing questionnaire survey method constantly increases the number of questionnaires to meet the needs of more segmented patients, and the cost of this method increases exponentially. This solution generates questionnaires for patients based on their specific conditions, and only needs to store the corresponding survey questions, without storing the questionnaires (the questionnaires are only suitable for the patients themselves, and are not stored for other patients to use), which greatly reduces the cost of this solution. In addition, when modifying and expanding the questionnaire survey questions, the traditional solution needs to check and modify the existing questionnaires one by one, while this solution only needs to check and modify the questions to modify the questionnaire, which further reduces the cost of this solution.

[0010] In summary, this solution solves the problem of low accuracy of patient satisfaction obtained through analysis of questionnaire survey results.

[0011] Furthermore, the analysis module is also used to clean the image information or the voice information, and to fill the missing values ​​of the image information or the voice information by using an interpolation method or a KNN algorithm during the cleaning process; The analysis module uses formula (1) to standardize the data. Formula (1) is as follows: (1), in, is the standardized data, is the original data, is the mean, is the standard deviation; When the analysis module classifies and encodes the image information or the voice information, the classified data is One-Hot encoded.

[0012] The interpolation method estimates missing values ​​based on the information of existing data points in the data set (data set of image information or voice information), which can better fit the overall distribution law of the data, and can also fill in the data based on the internal correlation, avoiding the deviation that may be introduced by simply filling the missing values ​​with fixed values ​​(such as the mean), so that the data can more truly reflect the actual situation. The KNN algorithm can find similar data points through the distance measurement between data points, so as to fill in the missing values ​​more reasonably, which is more advantageous than some simple filling methods. The use of One-Hot encoding can better capture the nonlinear relationship between categorical variables and target variables, enrich the feature representation of data, improve the model's understanding of categorical data and classification accuracy, and reduce model prediction errors caused by incorrectly encoding categorical data. In this solution, this method can more accurately identify and extract facial images (such as facial features) in image information and text features (text features with emotions, conjunctions, adverbs, nouns, etc.) in comment data, thereby laying a solid foundation for the acquisition of emotion vectors.

[0013] Furthermore, the analysis module is also used to clean the image information or voice information, and during the cleaning process, a missing value filling algorithm based on a long short-term memory network is used to fill in missing values ​​of the data according to the acquisition time; before the analysis module classifies and encodes the image information or voice information, the continuous image information or voice information is standardized using Z-Score or Min-Max according to the acquisition time.

[0014] Long short-term memory network (LSTM) can effectively handle long-term dependency problems in time series data, and is particularly suitable for processing dynamic data (such as real-time patient emotions). In the medical experience data, if the patient's emotional data is missing in a certain time period, LSTM can use the emotional data and other related features of the previous and subsequent time periods (such as the department of treatment, the stage of treatment, etc.) to predict the missing values ​​and improve the accuracy and completeness of the data; and LSTM can process variable-length input sequences and is suitable for dynamically changing medical experience data. For example, the patient's emotional changes and voice feature changes in different time periods, LSTM can capture these dynamic changes and generate more accurate missing value filling results. Z-Score standardization and Min-Max standardization can eliminate the dimensionality of the data, make data with different features comparable, and improve the accuracy and stability of the model. One-Hot encoding can convert categorical data into a numerical form that the model can handle, avoiding the order relationship between categories affecting the model performance. In this way, this solution can collect patients' emotional and behavioral data in real time, and quickly generate personalized questionnaires, so that patients' opinions and suggestions can be conveyed to the hospital in a timely manner.

[0015] Furthermore, it is characterized in that: the emotion analysis model also includes a model established by a decision tree, a random forest or a support vector machine, and the emotion analysis model is used to identify and classify the patient's emotions in the image data using the DBSCAN algorithm when marking the facial image of the patient in the image information as a neutral state, a positive state or a negative state; the emotion analysis model combines PCA to perform dimensionality reduction processing on the image data; when clustering, the emotion analysis model uses a dynamic cluster number determination method to optimize the setting of the cluster number, and the dynamic cluster number determination method includes the elbow rule or the silhouette coefficient; the analysis model obtains a second vector according to the cluster number processing, wherein the Euclidean distance is used to calculate the distance between samples, and the Euclidean distance calculation formula is shown in the following formula (2): (2); in, represents the Euclidean distance between two sample points, Represents the number of dimensions, that is, the number of features of the sample points, and Represents two sample points in In a dimension.

[0016] Furthermore, the sentiment analysis model includes a medical field sentiment dictionary. The sentiment analysis model combines the medical field sentiment dictionary to use TF-IDF and naive Bayes algorithm, BERT or GPT to perform sentiment classification on the comment data, and calculates the weights of different sentiments. Finally, multiple different sentiments are processed into the first vector. The TF-IDF calculation formula (3) is as follows: (3), in, is the weight of different emotions, According to formula (4), is the first vector, Through formula (5), formula (4) and formula (5) are as follows: (4), (5), in, is the total number of documents, To contain words The number of documents.

[0017] This method can effectively measure the importance of a word in the comment data, while taking into account the prevalence of the word in the entire document collection. For the sentiment classification of comment data, the Naive Bayes algorithm can determine whether the comment data is positive, negative or neutral based on the vocabulary in the comment and its TF-IDF weight. The Naive Bayes algorithm can quickly process a large amount of comment data and show good accuracy when processing natural language data with certain regularities. Even when there is some noise or irregular expressions in the comment data, the Naive Bayes algorithm can make relatively reasonable sentiment classification through probability calculation. BERT and GPT can capture the contextual relationship and long-distance dependency in the text, generate high-quality word embedding, and improve the accuracy of sentiment classification. Especially when dealing with complex medical comments and feedback, emotional information can be more accurately identified. In addition, BERT and GPT can handle complex language phenomena such as ambiguity and metaphor. In medical reviews, patients may use metaphors or vague expressions to describe their feelings. BERT and GPT can better understand these expressions and improve the accuracy of sentiment classification.

[0018] Furthermore, when the analysis module is used to calculate the emotion vector, the first vector and the second vector are merged into the emotion vector through principal component analysis and dimensionality reduction, and the emotion vector is then divided into positive, negative or neutral emotions, and the weight of the emotion vector is calculated based on the distribution and weight of the first vector and the second vector; the analysis module is also used to extract comment content related to negative emotions in the comment data based on specific keywords, and store the comment content and the emotion vector in association; when the suggestion module matches the survey topics as recommended topics based on the emotion vector and the medical stage, it first matches the same topics in the medical stage from the survey topics, and then obtains the comment content associated with the emotion vector of the negative emotion, screens the survey topics based on the comment content, and adjusts the preset number of topics based on the weight of the emotion vector, and extracts recommended topics from the screened survey topics based on the preset number of topics to generate a questionnaire.

[0019] The analysis module extracts negative emotion-related comments through specific keywords and stores them in association with the emotion vector. It can automatically and quickly filter out key information, thereby improving the speed of filtering and analyzing negative emotion information. In addition, the suggestion module dynamically adjusts the distribution and number of questions based on the total negative emotion value, emotion ratio, etc. This automatic adjustment mechanism makes the survey more targeted and avoids the inefficiency of designing fixed questionnaires. Traditional questionnaires are often fixed, but this solution can dynamically adjust the questionnaire based on the currently collected patient emotion information, making the survey more in line with the actual situation, improving the efficiency and pertinence of questionnaire design, and also improving the efficiency of subsequent data analysis, because more targeted questionnaires can collect more valuable data.

[0020] Furthermore, the acquisition module is also used to collect the location information of the patient terminal in real time. The natural language processing model classifies emotions into positive emotions, negative emotions, neutral emotions and mixed emotions. The acquisition module is also used to pass the text conversion results of the voice comment data to the analysis model module. After receiving the text conversion results, the analysis model module uses the natural language processing model to perform emotion classification on the text conversion results, and passes the text conversion results with emotion classification results of negative emotions or mixed emotions as negative comment data to the acquisition module. After receiving the negative comment data, the acquisition module obtains the patient's movement trajectory according to the location information of the patient terminal. If the movement trajectory does not overlap with the medical environment within a preset time , the negative comment data is ignored; if the moving trajectory overlaps with the medical environment within the preset time, and the moving range formed by the moving trajectory is smaller than the preset waiting range, the negative comment data is stored in the database as comment data related to the consultation time; if the moving range formed by the moving trajectory is not smaller than the preset waiting range and the moving distance formed by the moving trajectory is larger than the preset running range, the negative comment data is stored in the database as comment data related to the consultation process; if the moving range formed by the moving trajectory is not smaller than the preset waiting range and the moving distance formed by the moving trajectory is not larger than the preset running range, the negative comment data is stored in the database as comment data related to the consultation process.

[0021] Furthermore, when the analysis module combines the acquisition position of the image information with the medical information to determine the patient's medical stage, it is used to obtain the medical department, examination information and medical time in the medical information, and then obtain the acquisition position corresponding to the image information. When the position of the medical department is the same as the acquisition position of the image information and the medical information does not include the examination information, the medical stage is determined to be the waiting stage; when the medical information includes the examination information, the medical stage is determined to be the examination stage; when the position of the medical department is the same as the acquisition position of the image information and the appearance characteristics of the medical staff are identified in the image information, the medical stage is determined to be the face-to-face consultation stage; when the medical stage is the waiting stage, the analysis module uses the time when the acquisition position first overlaps with the hospital department layout corresponding to the medical department as the first time, uses the acquisition position acquired in the waiting stage as the moving trajectory, and uses the difference between the current time and the first time as the waiting time. The average waiting time of the medical department is stored in the analysis module. When the average waiting time is greater than the waiting time, the analysis module is used to calculate the medical department's medical information based on the difference between the average waiting time and the waiting time. and the number of questions adjusted according to the stored average answering time; when the emotion vector obtained by the analysis module is related to negative emotion or negative state, the first vector and the second vector are processed as negative emotion vectors, the collection time and collection position corresponding to the negative emotion vector are obtained, and the multiple collection positions are processed as movement trajectories according to the sequence of collection time, and if the movement trajectory does not overlap with the hospital department layout within the preset time, the negative comment data is ignored; if the movement trajectory overlaps with the hospital department layout within the preset time, and the movement range formed by the movement trajectory is smaller than the preset waiting range, the comment data and image data corresponding to the negative emotion vector are stored in the database; if the movement range formed by the movement trajectory is not smaller than the preset waiting range and the movement distance formed by the movement trajectory is larger than the preset running range, the negative comment data is stored in the database as the comment data related to the medical treatment process; if the movement range formed by the movement trajectory is not smaller than the preset waiting range and the movement distance formed by the movement trajectory is not larger than the preset running range, the negative comment data is stored in the database as the comment data related to the medical treatment process.

[0022] Further classification and storage of negative comment data based on the patient's movement trajectory can help hospitals gain a deeper understanding of the specific reasons for patient dissatisfaction. If the patient's movement trajectory does not overlap with the medical environment, it may mean that the patient's negative comments have little to do with the hospital's services. Ignoring this data can reduce unnecessary analysis work; when the movement trajectory overlaps with the medical environment and the movement range is less than the preset waiting range, the negative comments are stored as comment data related to the consultation time, which helps the hospital analyze and optimize the patient's waiting time and improve the patient's medical experience; for the case where the movement range formed by the movement trajectory is not less than the preset waiting range and the movement distance is greater than the preset running range, the negative comments are stored as comment data related to the consultation process, which can help hospitals discover problems in the consultation process, such as the inconvenience of patients running between different departments; and when the movement range formed by the movement trajectory is not less than the preset waiting range and the movement distance is not greater than the preset running range, the negative comments are stored as comment data related to the consultation process, which helps hospitals pay attention to the problems of patients in the specific consultation process, such as the examination process, treatment process, etc. When patients are waiting for treatment, the remaining waiting time is estimated based on the average waiting time and the time the patients have waited. The number of questions in the questionnaire accepted by the patients is adjusted according to the remaining waiting time, so that the patients can fill in as many questions as possible in the remaining waiting time. This can not only help the hospital collect patient satisfaction, but also help patients spend boring waiting time, and also reflect the hospital's care for patients and soothe their emotions, thereby improving patient satisfaction with the hospital. It can be seen that this processing flow can help medical institutions understand patient feedback more comprehensively and accurately.

[0023] Furthermore, it also includes a questionnaire adjustment module, which is used to calculate the weight values ​​of various major factors that affect the doctor-patient experience. The questionnaire adjustment module generates a questionnaire according to the types of major factors and the weight values ​​of each major factor, and adjusts the order of questions in the questionnaire according to the weight value order of each major factor.

[0024] First, the questionnaire adjustment module generates a questionnaire based on these key factors and their weights, avoiding the blind setting of questionnaire questions. Compared with general questionnaires, this questionnaire based on data analysis can focus more on issues that actually affect the experience. Secondly, the sorting of questionnaire questions is adjusted according to the weight values ​​of each major factor, so that more important questions can be ranked first. This sorting method helps respondents pay attention to key issues more quickly, and also makes the data collected by the questionnaire more valuable. During the survey, respondents tend to answer the previous questions more seriously. Putting important questions in front can improve the quality of key data. Furthermore, by generating and sorting questionnaire questions in a targeted manner, the interference of irrelevant questions on respondents can be reduced, and the response rate and quality of the questionnaire can be improved. Respondents will feel that the content of the questionnaire is closely related to their actual experience, so they are more willing to answer seriously. For hospitals or medical institutions, the data obtained based on this method can more accurately reflect the problems in the doctor-patient experience, and then formulate targeted improvement measures to improve the experience of both doctors and patients.

[0025] Furthermore, the acquisition module is also used to generate an instruction to investigate the medical treatment time and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical treatment time in the database. After receiving the instruction to investigate the medical treatment time, the questionnaire adjustment module generates a questionnaire to investigate the medical treatment time and sends it to the patient terminal; the acquisition module is also used to generate an instruction to investigate the medical treatment process and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical treatment process in the database. After receiving the instruction to investigate the medical treatment process, the questionnaire adjustment module generates a questionnaire to investigate the medical treatment process, obtains the location information of the patient terminal, and when the location information of the patient terminal does not change within the preset temporary stop time, sends the questionnaire to investigate the medical treatment process to the patient terminal; the acquisition module is also used to generate an instruction to investigate the medical treatment process and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical treatment process in the database. After receiving the instruction to investigate the medical treatment process, the questionnaire adjustment module generates a questionnaire to investigate the medical treatment process, obtains the location information of the patient terminal, and when the location information of the patient terminal does not overlap with the medical environment, sends the questionnaire to investigate the medical treatment process to the patient terminal First, after the collection module stores the negative comment data in the database according to different types (visit time, visit process, and visit process), the corresponding survey instructions are generated respectively, so that the subsequent questionnaires can be closely centered around specific issues. Secondly, for negative comments related to the visit time, the questionnaire generated by the questionnaire adjustment module will specifically ask about the visit time, which helps to dig deeper into the specific reasons that cause patients to be dissatisfied with the visit time; for questionnaires related to the visit process, when the patient terminal does not change within the preset stop time, it is sent, which usually means that the patient may be waiting or resting. At this time, they are more likely to have time and energy to fill in the questionnaire, thereby increasing the response rate of the questionnaire; for questionnaires related to the visit process, when the patient terminal position does not overlap with the medical environment, it is sent, which may mean that the patient has left the medical environment. At this time, they may be more willing to objectively feedback their experience during the visit process, which helps to improve the quality of the questionnaire response. Finally, this targeted and timely way of sending questionnaires avoids excessive disturbance to patients. The questionnaires received by patients are related to the negative experience they have previously reported, and they will feel that the hospital attaches importance to their opinions, thereby increasing the enthusiasm of patients to participate in the survey. In the long run, this interactive method will help establish a good doctor-patient communication mechanism and enhance patients' trust in medical institutions. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 Schematic diagram of the structure of the system for generating suggestions for optimizing the doctor-patient experience driven by data mining technology.

[0027] Figure 2 This is a flow chart of the analysis module processing comment data according to the movement trajectory in an embodiment of the present invention. DETAILED DESCRIPTION

[0028] The following will clearly and completely describe the concept and technical effects of the present invention in combination with the embodiments, so as to fully understand the purpose, features and effects of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of them. Based on the embodiments of the present invention, other embodiments obtained by those skilled in the art without creative work are all within the scope of protection of the present invention: like Figure 1 As shown in the figure, the doctor-patient experience optimization suggestion generation system driven by data mining technology includes: The collection module is used to collect the image information of the patient in the hospital monitoring, to collect the voice information input by the patient terminal and convert the voice information into textual comment data, to collect the results of the questionnaire uploaded by the patient terminal, and to collect the patient's medical information; An analysis module is used to use a preset emotion analysis model to annotate comment data as neutral emotion, positive emotion or negative emotion, and use the annotated result of the comment data as a first vector; used to use a preset emotion analysis model to identify and annotate the facial image of the patient in the image information as a neutral state, a positive state or a negative state, and use the annotated result of the image information as a second vector; the analysis module calculates the first vector and the second vector as the patient's emotion vector according to the acquisition time of the voice information and the image information; the analysis module combines the acquisition location of the image information with the medical consultation information to determine the patient's medical consultation stage, and associates and stores the emotion vector with the same acquisition time with the medical consultation stage, and the medical consultation stage includes the waiting stage, the examination stage and the face-to-face consultation stage; The suggestion module includes various survey topics. The suggestion module is used to obtain the associated stored consultation stage and emotion vector, match the survey topics in the survey topics as recommended topics according to the emotion vector and the consultation stage, and then adjust the preset number of topics according to the value of the emotion vector, generate a questionnaire with the adjusted number of topics and the recommended topics and send it to the patient terminal; The collection module is also used to obtain the completion results of the questionnaire at the patient terminal and send the completion results to the analysis module. The analysis module is used to analyze and integrate the completion results into improvement suggestions.

[0029] Among them, the patient's medical information includes the patient's name, medical time, medical department, examination results and other information.

[0030] The analysis module is also used to clean the image information or the voice information, and to fill the missing values ​​of the image information or the voice information by using the interpolation method or the KNN algorithm during the cleaning process; The data processing model uses formula (1) to standardize the data. Formula (1) is as follows: (1), in, is the standardized data, is the original data, is the mean, is the standard deviation; When the data processing model classifies and encodes image information or voice information, the classified data is One-Hot encoded.

[0031] The analysis module is also used to clean the image information or voice information, and during cleaning, the missing value filling algorithm based on the long short-term memory network is used to fill the missing values ​​of the data according to the acquisition time. Before the analysis module classifies and encodes the image information or voice information, the continuous image information or voice information is standardized using Z-Score or Min-Max according to the acquisition time.

[0032] Among them, the sentiment analysis model also includes a model established by a decision tree, a random forest or a support vector machine. The sentiment analysis model is used to identify and classify the patient's emotions in the image data using the DBSCAN algorithm when the facial image of the patient in the image information is labeled as a neutral state, a positive state or a negative state; the sentiment analysis model is combined with PCA to perform dimensionality reduction processing on the image data; When clustering, the sentiment analysis model uses a dynamic cluster number determination method to optimize the setting of the cluster number. The dynamic cluster number determination method includes the elbow rule or the silhouette coefficient. The analysis model obtains the second vector according to the cluster number processing, in which the Euclidean distance is used to calculate the distance between samples. The Euclidean distance calculation formula is shown in the following formula (2): (2); in, represents the Euclidean distance between two sample points, Represents the number of dimensions, that is, the number of features of the sample points, and Represents two sample points in In a dimension.

[0033] Among them, the sentiment analysis model includes a medical field sentiment dictionary. The sentiment analysis model combines the medical field sentiment dictionary and uses TF-IDF and Naive Bayes algorithm, BERT or GPT to classify the comment data, and calculate the weights of different sentiments. Finally, multiple different sentiments are processed into the first vector. The TF-IDF calculation formula (3) is as follows: (3), in, is the weight of different emotions, According to formula (4), is the first vector, Through formula (5), formula (4) and formula (5) are as follows: (4), (5), in, is the total number of documents, To contain words The number of documents.

[0034] Among them, when the analysis module is used to calculate the emotion vector, the first vector and the second vector are merged into the emotion vector through principal component analysis and dimensionality reduction, and the emotion vector is then divided into positive, negative or neutral emotions, and the weight of the emotion vector is calculated according to the distribution and weight of the first vector and the second vector; the analysis module is also used to extract the comment content related to negative emotions in the comment data according to specific keywords, and store the comment content and the emotion vector in association; When the suggestion module matches survey topics as recommended topics in the survey questions according to the emotion vector and the treatment stage, it first matches the same topics in the treatment stage from the survey topics, then obtains the comment content associated with the emotion vector of negative emotions, filters the survey topics according to the comment content, and adjusts the preset number of topics according to the weight of the emotion vector, and extracts recommended topics from the filtered survey topics according to the preset number of topics to generate a questionnaire.

[0035] Among them, Figure 2 As shown, the analysis module is also used to obtain the corresponding position information in the patient's negative emotion vector as the first position, and then generate a movement trajectory according to the first position. If the movement trajectory does not overlap with the medical environment within a preset time (generally 30 seconds, which can be set by the administrator), the negative comment data is ignored; if the movement trajectory overlaps with the medical environment within the preset time, and the movement range formed by the movement trajectory is smaller than the preset waiting range (generally 10m, which can be determined by the administrator according to the area occupied by the department), the negative comment data is stored in the database as comment data related to the consultation time; if the movement range formed by the movement trajectory is not less than the preset waiting range and the movement distance formed by the movement trajectory is greater than the preset running range (generally 30m, which can be determined by the administrator according to the distance from the examination office to the department), the negative comment data is stored in the database as comment data related to the consultation process; if the movement range formed by the movement trajectory is not less than the preset waiting range and the movement distance formed by the movement trajectory is not greater than the preset running range, the negative comment data is stored in the database as comment data related to the consultation process.

[0036] Among them, when the analysis module combines the collection location of the image information and the medical information to determine the patient's medical stage, it is used to obtain the medical department, examination information and medical time in the medical information, and then obtain the collection location corresponding to the image information. When the location of the medical department is the same as the collection location of the image information, and the medical information does not include examination information, the medical stage is determined to be the waiting stage; when the medical information includes examination information, the medical stage is determined to be the examination stage; when the location of the medical department is the same as the collection location of the image information and the appearance characteristics of the medical staff are identified in the image information, the medical stage is determined to be the face-to-face consultation stage.

[0037] When the consultation stage is the waiting stage, the analysis module takes the time when the collection position and the hospital department layout corresponding to the consultation department first overlap as the first time, takes the collection position obtained in the waiting stage as the movement trajectory, and takes the difference between the current time and the first time as the waiting time. The analysis module stores the average waiting time of the consultation department. The analysis module is used to adjust the number of questions according to the difference between the average waiting time and the waiting time and the stored average question-solving time when the average waiting time is greater than the waiting time.

[0038] When the emotion vector obtained by the analysis module is related to negative emotion or negative state, the first vector and the second vector are processed as negative emotion vectors, the collection time and collection position corresponding to the negative emotion vector are obtained, and multiple collection positions are processed into movement trajectories according to the sequence of collection time. If the movement trajectory does not overlap with the hospital department layout within the preset time, the negative comment data is ignored; if the movement trajectory overlaps with the hospital department layout within the preset time, and the movement range formed by the movement trajectory is smaller than the preset waiting range, the comment data and image data corresponding to the negative emotion vector are stored in the database; if the movement range formed by the movement trajectory is not smaller than the preset waiting range and the movement distance formed by the movement trajectory is larger than the preset running range, the negative comment data is stored in the database as comment data related to the medical treatment process; if the movement range formed by the movement trajectory is not smaller than the preset waiting range and the movement distance formed by the movement trajectory is not larger than the preset running range, the negative comment data is stored in the database as comment data related to the medical treatment process.

[0039] It also includes a questionnaire adjustment module. The key factor identification model is also used to calculate the weight values ​​of the main factors that affect the doctor-patient experience. The questionnaire adjustment module generates a questionnaire based on the type of main factors and the weight value ratio of each main factor, and adjusts the order of questions in the questionnaire based on the weight value ranking of each main factor.

[0040] The collection module is also used to generate an instruction to investigate the medical treatment time and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical treatment time in the database. After receiving the instruction to investigate the medical treatment time, the questionnaire adjustment module generates a questionnaire to investigate the medical treatment time and sends it to the patient terminal.

[0041] The collection module is also used to generate an instruction to investigate the medical process and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical process in the database. After receiving the instruction to investigate the medical process, the questionnaire adjustment module generates a questionnaire to investigate the medical process, obtains the location information of the patient terminal, and sends a questionnaire to the patient terminal to investigate the medical process when the location information of the patient terminal does not change within the preset temporary stop time (generally 30 seconds, which can be adjusted by the administrator). Among them, when the patient runs to complete the medical process, most of the time when he does not move for 30 seconds is when he is queuing, seeing a doctor or undergoing an examination. When queuing, the probability of the patient playing with the mobile phone is high, and the willingness to fill in the questionnaire is high. At this time, sending the questionnaire to the patient can not only relieve the patient's boredom, but also express concern to the patient in time, so that the patient feels valued, thereby improving the patient's satisfaction. If the patient is seeing a doctor or undergoing an examination at this time, the questionnaire cannot attract the patient's attention. If the patient sees the questionnaire after the doctor or the examination, it can give the patient a feeling that the hospital is eager to know the patient's medical experience, so that the patient feels valued.

[0042] The collection module is also used to generate an instruction to investigate the medical treatment process and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical treatment process in the database. After receiving the instruction to investigate the medical treatment process, the questionnaire adjustment module generates a questionnaire to investigate the medical treatment process and obtains the location information of the patient terminal. When the location information of the patient terminal does not overlap with the medical environment, the questionnaire to investigate the medical treatment process is sent to the patient terminal.

[0043] In the specific implementation, a large general hospital introduced the doctor-patient experience optimization suggestion generation system driven by the data mining technology of the present invention to improve the patient's medical experience. Surveillance cameras were installed in key areas such as waiting areas, examination rooms, and consulting rooms in the hospital.

[0044] In the waiting area of ​​the Department of Cardiology, surveillance cameras collect facial images of patients in real time. When a patient, Mr. Zhang, was waiting for his consultation, his facial image was clearly captured, and the system recorded the time and location of the image acquisition. The hospital provides patients with a mobile application, through which patients can enter voice messages. While waiting, Mr. Zhang felt a little impatient, so he entered a voice message through the mobile application: "I have waited so long and it is not my turn yet. The efficiency is too low." The system converts this voice message into text comment data. The hospital sends questionnaires to patients regularly. After completing the consultation, Mr. Zhang received a questionnaire sent by the system. He carefully filled in the questions about his consultation experience, and the system collected his filling results. The hospital's information system is connected to the system to automatically collect the patient's consultation information. Mr. Zhang's consultation information includes his name, consultation time (10 am on March 2, 2007), consultation department (Cardiovascular Department), and examination results (preliminary diagnosis of arrhythmia).

[0045] The analysis module uses a preset sentiment analysis model to process Mr. Zhang's comment data. The sentiment analysis model combines the medical field sentiment dictionary and uses TF-IDF and naive Bayes algorithms to classify the sentiment of the comment data. The first vector is calculated according to formula (3), and Mr. Zhang's comment data is labeled as negative emotions, and the labeling result is used as the first vector. For Mr. Zhang's facial image, the sentiment analysis model uses the DBSCAN algorithm to identify and classify the patient's emotions in the image data. At the same time, PCA is combined to reduce the dimension of the image data, and the elbow rule is used to optimize the setting of the number of clusters during clustering. Finally, the image information is labeled as a negative state, and the labeling result is used as the second vector. According to the acquisition time of the voice information and image information, the analysis module reduces the dimension of the first vector and the second vector through principal component analysis to fuse them into Mr. Zhang's emotion vector. Then, the weight of the emotion vector is calculated according to the distribution and weight of the first vector and the second vector, and it is determined that Mr. Zhang's emotion vector is negative and has a high weight.

[0046] The analysis module combines the acquisition location of the image information with the medical information to determine Mr. Zhang's medical stage. Since Mr. Zhang is in the waiting area of ​​the Department of Cardiovascular Medicine, the medical information does not include examination information, so the medical stage is determined to be the waiting stage. At the same time, the emotion vectors with the same acquisition time are associated with the medical stage and stored. The analysis module cleans the image information and voice information. For possible missing values, a missing value filling algorithm based on the long short-term memory network is used for processing. Before classification and encoding, the continuous image information and voice information are standardized using Z-Score according to the acquisition time, and the classified data is One-Hot encoded.

[0047] The suggestion module obtains the associated stored treatment stage (waiting stage) and emotion vector (negative and with high weight). Match the questions related to the waiting stage from the various survey questions, such as "How satisfied are you with the waiting time?" "Does the environment in the waiting area make you feel comfortable?", etc. At the same time, obtain the comment content associated with Mr. Zhang's negative emotion vector "I have waited for so long and it is not my turn yet, the efficiency is too low", and further filter the survey questions based on this content to remove questions that are not related to the waiting time. According to the weight of Mr. Zhang's emotion vector, adjust the preset number of questions. Due to the high weight, the number of survey questions is appropriately increased to more fully understand the reasons for his dissatisfaction. Based on the adjusted number of questions and the screened recommended questions, generate a questionnaire for Mr. Zhang and send it to his mobile terminal.

[0048] After receiving the questionnaire, Mr. Zhang filled it out carefully and submitted the results. The collection module obtained his filled-in results and sent them to the analysis module. The analysis module integrated and analyzed Mr. Zhang's filled-in results. For example, Mr. Zhang clearly stated in the questionnaire that the long waiting time was the main reason for his dissatisfaction. Based on this result, the analysis module generated improvement suggestions, such as suggesting that the hospital increase the number of doctors in the department, optimize the registration and queuing system, etc., and fed back these suggestions to the hospital management so that the hospital can take corresponding improvement measures.

[0049] The analysis module obtains the corresponding position information in Mr. Zhang's negative emotion vector as the first position, and generates his movement trajectory during the waiting stage. Since Mr. Zhang has been active in the waiting area, his movement trajectory overlaps with the layout of the hospital department, and the movement range is less than the preset waiting range (10m), so his negative comment data is stored in the database as comment data related to the consultation time. The acquisition module generates an instruction to investigate the consultation time and passes it to the questionnaire adjustment module. After receiving the instruction, the questionnaire adjustment module generates a questionnaire to investigate the consultation time and sends it to Mr. Zhang's mobile terminal. At this time, Mr. Zhang is queuing in the waiting area and has the time and willingness to fill out the questionnaire. Filling out the questionnaire not only relieves his boredom, but also makes him feel the care of the hospital and improves his satisfaction.

[0050] The above is only an embodiment of the present invention, and the common knowledge such as the known specific structure and characteristics in the scheme is not described in detail here. It should be pointed out that for those skilled in the art, several deformations and improvements can be made without departing from the structure of the present invention, which should also be regarded as the protection scope of the present invention, and these will not affect the effect of the implementation of the present invention and the practicality of the patent. The scope of protection required by this application shall be based on the content of its claims, and the specific implementation methods and other records in the specification can be used to interpret the content of the claims.

Claims

1. A data mining technology-driven doctor-patient experience optimization suggestion generation system, characterized by: include: The collection module is used to collect the image information of the patient in the hospital monitoring, to collect the voice information input by the patient terminal and convert the voice information into textual comment data, to collect the results of the questionnaire uploaded by the patient terminal, and to collect the patient's medical information; An analysis module, used to use a preset emotion analysis model to annotate comment data as neutral emotion, positive emotion or negative emotion, and use the annotated result of the comment data as a first vector; used to use a preset emotion analysis model to identify and annotate the facial image of the patient in the image information as a neutral state, a positive state or a negative state, and use the annotated result of the image information as a second vector; the analysis module calculates the first vector and the second vector as the patient's emotion vector according to the acquisition time of the voice information and the image information; the analysis module combines the acquisition location of the image information with the medical consultation information to determine the patient's medical consultation stage, and associates and stores the emotion vector with the same acquisition time with the medical consultation stage, and the medical consultation stage includes a waiting stage, an examination stage and a face-to-face consultation stage; A suggestion module includes various survey topics; the suggestion module is used to obtain the associated stored treatment stage and emotion vector, match the survey topic in the survey topic as a recommended topic according to the emotion vector and the treatment stage, and then adjust the preset number of topics according to the value of the emotion vector, generate a questionnaire with the adjusted number of topics and the recommended topics and send it to the patient terminal; The acquisition module is also used to obtain the completion results of the questionnaire by the patient terminal, and send the completion results to the analysis module, and the analysis module is used to analyze and integrate the completion results into improvement suggestions.

2. The data mining technology-driven doctor-patient experience optimization suggestion generation system according to claim 1, characterized in that: The analysis module is also used to clean the image information or the voice information, and to fill in missing values ​​of the image information or the voice information using an interpolation method or a KNN algorithm during cleaning; The analysis module uses formula (1) to standardize the data. Formula (1) is as follows: （1）， in, is the standardized data, is the original data, is the mean, is the standard deviation; When the analysis module classifies and encodes the image information or the voice information, the classified data is One-Hot encoded.

3. The data mining technology-driven doctor-patient experience optimization suggestion generation system according to claim 1, characterized in that: The analysis module is also used to clean the image information or voice information, and during the cleaning process, fill in missing values ​​of the data using a missing value filling algorithm based on a long short-term memory network according to the acquisition time; Before the analysis module classifies and encodes the image information or the voice information, the continuous image information or the voice information is standardized according to the acquisition time using Z-Score or Min-Max.

4. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 3 or claim 2, characterized in that: The emotion analysis model also includes a model established by a decision tree, a random forest or a support vector machine. The emotion analysis model is used to identify and classify the patient's emotions in the image data using the DBSCAN algorithm when the facial image of the patient in the image information is marked as a neutral state, a positive state or a negative state; the emotion analysis model combines PCA to perform dimensionality reduction processing on the image data; The sentiment analysis model uses a dynamic cluster number determination method to optimize the setting of the cluster number when clustering, and the dynamic cluster number determination method includes the elbow rule or the silhouette coefficient; the analysis model obtains the second vector according to the cluster number processing, wherein the Euclidean distance is used to calculate the distance between samples, and the Euclidean distance calculation formula is shown in the following formula (2): （2）； in, represents the Euclidean distance between two sample points, Represents the number of dimensions, that is, the number of features of the sample points, and Represents two sample points in In a dimension.

5. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 4 is characterized by: The sentiment analysis model includes a medical field sentiment dictionary. The sentiment analysis model combines the medical field sentiment dictionary to use TF-IDF and naive Bayes algorithm, BERT or GPT to classify the comment data, and calculate the weights of different sentiments. Finally, multiple different sentiments are processed into the first vector. The TF-IDF calculation formula (3) is as follows: （3）， in, are the weights of different emotions, According to formula (4), is the first vector, Through formula (5), formula (4) and formula (5) are as follows: （4）， （5）， in, is the total number of documents, To contain words The number of documents.

6. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 5, characterized in that: When the analysis module is used to calculate the emotion vector, the first vector and the second vector are merged into the emotion vector through principal component analysis and dimensionality reduction, and the emotion vector is then divided into positive, negative or neutral emotions, and the weight of the emotion vector is calculated according to the distribution and weight of the first vector and the second vector; the analysis module is also used to extract the comment content related to negative emotions in the comment data according to specific keywords, and store the comment content and the emotion vector in association; When the suggestion module matches survey topics as recommended topics in the survey topics according to the emotion vector and the treatment stage, it first matches the same topics in the treatment stage from the survey topics, then obtains the comment content associated with the emotion vector of negative emotions, filters the survey topics according to the comment content, and adjusts the preset number of topics according to the weight of the emotion vector, and extracts recommended topics from the filtered survey topics according to the preset number of topics to generate a questionnaire.

7. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 6, characterized in that: The analysis module is also used to obtain the position information corresponding to the patient's negative emotion vector as the first position, and then generate a movement trajectory according to the first position. If the movement trajectory does not overlap with the medical environment within a preset time, the negative comment data is ignored; if the movement trajectory overlaps with the medical environment within the preset time, and the movement range formed by the movement trajectory is smaller than the preset waiting range, the negative comment data is stored in the database as the comment data related to the medical treatment time; if the movement range formed by the movement trajectory is not smaller than the preset waiting range and the movement distance formed by the movement trajectory is greater than the preset running range, the negative comment data is stored in the database as the comment data related to the medical treatment process; If the moving range formed by the moving trajectory is not less than the preset waiting range and the moving distance formed by the moving trajectory is not greater than the preset running range, the negative comment data is stored in the database as comment data related to the medical consultation process.

8. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 7, characterized in that: When the analysis module combines the collection position of the image information with the medical information to determine the patient's medical stage, it is used to obtain the medical department, examination information and medical time in the medical information, and then obtain the collection position corresponding to the image information. When the location of the medical department is the same as the collection position of the image information, and the medical information does not include examination information, the medical stage is determined to be the waiting stage; when the medical information includes examination information, the medical stage is determined to be the examination stage; when the location of the medical department is the same as the collection position of the image information, and the appearance features of the medical staff are identified in the image information, the medical stage is determined to be the face-to-face consultation stage; When the consultation stage is the waiting stage, the analysis module takes the time when the collection position and the hospital department layout corresponding to the consultation department overlap for the first time as the first time, takes the collection position obtained in the waiting stage as the movement trajectory, and takes the difference between the current time and the first time as the waiting time. The analysis module stores the average waiting time of the consultation department. The analysis module is used to adjust the number of questions according to the difference between the average waiting time and the waiting time and the stored average time for answering questions when the average waiting time is greater than the waiting time; When the emotion vector obtained by the analysis module is related to negative emotion or negative state, the first vector and the second vector are processed as negative emotion vectors, the acquisition time and acquisition position corresponding to the negative emotion vector are obtained, and multiple acquisition positions are processed into movement trajectories according to the order of the acquisition time. If the movement trajectory does not overlap with the hospital department layout within the preset time, the negative comment data is ignored; if the movement trajectory overlaps with the hospital department layout within the preset time, and the movement range formed by the movement trajectory is smaller than the preset waiting range, the comment data and image data corresponding to the negative emotion vector are stored in the database; if the movement range formed by the movement trajectory is not smaller than the preset waiting range and the movement distance formed by the movement trajectory is greater than the preset running range, the negative comment data is stored in the database as comment data related to the medical treatment process; If the moving range formed by the moving trajectory is not less than the preset waiting range and the moving distance formed by the moving trajectory is not greater than the preset running range, the negative comment data is stored in the database as comment data related to the medical consultation process.

9. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 8, characterized in that: It also includes a questionnaire adjustment module, which is used to calculate the weight values ​​of the main factors that affect the doctor-patient experience. The questionnaire adjustment module generates a questionnaire according to the types of main factors and the weight values ​​of each main factor, and adjusts the order of questions in the questionnaire according to the weight value order of each main factor.

10. The data mining technology driven doctor-patient experience optimization suggestion generation system according to claim 9, characterized in that: The collection module is also used to generate an instruction to investigate the visit time and pass it to the questionnaire adjustment module after storing the negative comment data as the comment data related to the visit time in the database. After receiving the instruction to investigate the visit time, the questionnaire adjustment module generates a questionnaire to investigate the visit time and sends it to the patient terminal; The collection module is also used to generate an instruction to investigate the medical process and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical process in the database. After receiving the instruction to investigate the medical process, the questionnaire adjustment module generates a questionnaire to investigate the medical process, obtains the location information of the patient terminal, and sends the questionnaire to investigate the medical process to the patient terminal when the location information of the patient terminal does not change within the preset temporary stop time; The collection module is also used to generate an instruction for investigating the medical treatment process and pass it to the questionnaire adjustment module after storing the negative comment data as comment data related to the medical treatment process in the database. After receiving the instruction for investigating the medical treatment process, the questionnaire adjustment module generates a questionnaire for investigating the medical treatment process and obtains the location information of the patient terminal. When the location information of the patient terminal does not overlap with the medical environment, the questionnaire for investigating the medical treatment process is sent to the patient terminal.

Images