Intelligent recommendation diagnosis method and system based on user behavior self-learning analysis

Abstract

The invention provides an intelligent recommendation diagnosis method based on user behavior self-learning analysis, and belongs to the field of diagnosis information query and retrieval, and the method comprises the following steps: querying according to an input keyword to obtain a recommendation diagnosis list matched with the keyword, calculating the weight of each diagnosis in the recommendation diagnosis list, and calculating the weight of each diagnosis in the recommendation diagnosis list; sorting the diagnoses based on the weights of the diagnoses to obtain a diagnosis list; according to the index data of the current patient, adjusting and optimizing the weight of each diagnosis in the diagnosis list based on the diagnosis index and diagnosis association knowledge base to obtain an adjusted and optimized diagnosis list; diagnosis needing to be opened is obtained based on the optimized diagnosis list, and complication diagnosis associated with the opened diagnosis is obtained based on historical patient data; recording a current selection result of the doctor and carrying out self-learning; the invention further provides an intelligent recommendation diagnosis method and system based on user behavior self-learning analysis, and doctors are helped to quickly and accurately query and retrieve diagnosis information in thousands of diagnoses.

Description

Technical Field

[0001] The present invention relates to the technical field of diagnostic information query and retrieval, and particularly to an intelligent recommendation diagnosis method and system based on self-learning analysis of user behavior. Background Art

[0002] In a Hospital Information System (HIS), a doctor's action of issuing a diagnosis for a patient is a very frequent behavior. When a doctor issues a diagnosis, how to accurately query diagnostic information among thousands of diagnoses. The traditional method is to retrieve and search by the diagnosis name or the pinyin code or Wubi code of the diagnosis name. The disadvantages of the traditional retrieval and query method are as follows: (1) The diagnosis scope issued by a single doctor is relatively small, but the retrieval scope is usually large, and the commonly used diagnostic information of the doctor cannot be quickly retrieved and located; (2) For some diagnosis names, there may be non-standard aliases in different places, and the doctor's inertial thinking may lead to the inability to retrieve the required diagnostic information; (3) Typing less or wrongly interrupting the name, polyphonic characters, typos, etc. may lead to the inability to retrieve the desired diagnostic information; (4) The patient already has a primary diagnosis, and it is impossible to quickly recommend possible complications or other related diagnoses based on the primary diagnosis.

[0003] In the prior art, the Chinese patent application for invention with publication number CN110136838A, "Data Matching Decision Method and System Based on Multi-Knowledge Base Reasoning", discloses a data matching decision method based on multi-knowledge base reasoning, including step S110: establishing a clinical auxiliary decision knowledge base, and the clinical auxiliary decision knowledge base includes at least three sub-knowledge bases: referral recommendation knowledge base, medication recommendation knowledge base, and risk assessment knowledge base; step S120: according to the data information to be matched, matching the knowledge information corresponding to the data information in each sub-knowledge base of the clinical auxiliary decision knowledge base; step S130: fusing the knowledge information matched by each sub-knowledge base according to a preset rule to form a final data matching decision, which solves the problems of the current clinical auxiliary decision system knowledge base being one-sided, poor in integration, and poor in implementation. However, this method requires an internal knowledge base and needs to perform preprocessing of data, and cannot predict and correct based on the retrieval operation behavior of the using user. Summary of the Invention

[0004] The technical problem to be solved by the present invention is how to help a doctor quickly and accurately query and retrieve diagnostic information among thousands of diagnoses when the doctor issues a diagnosis.

[0005] The present invention solves the above technical problem through the following technical solutions: An intelligent recommendation diagnosis method based on self-learning analysis of user behavior, the method includes the following steps:

[0006] Step 1: perform a search based on the input keyword to obtain a recommended diagnosis list matching the keyword, calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis based on its weight, and obtain a diagnosis list;

[0007] Step 2: According to the indicator data of the current patient, the weight of each diagnosis in the diagnosis list is adjusted based on the associated knowledge base of the diagnosis indicator and the diagnosis to obtain an adjusted diagnosis list;

[0008] Step 3: Obtain the diagnosis to be issued based on the optimized diagnosis list, and obtain the comorbidity diagnosis associated with the issued diagnosis based on historical patient data;

[0009] Step 4: Record the doctor's selection result and conduct self-learning.

[0010] The present invention is based on the input path of the doctor's diagnosis and analyzes the data of each diagnosis prescribed by the doctor, so that the system can understand the doctor's diagnosis and prescription behavior habits. In the process of constantly learning the data generated by the doctor's daily use of the system to prescribe diagnoses, the system constantly corrects the doctor's individual data and gradually grows into a diagnosis recommendation system that suits the doctor's personality, assists the doctor in quickly prescribing diagnoses, and helps the doctor to quickly and accurately query and retrieve diagnostic information among thousands of diagnoses.

[0011] Preferably, the process of calculating the weight of each diagnosis in the recommended diagnosis list in step 1 includes:

[0012] Step 1.1, use the keyword matching algorithm to calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis from large to small according to the weight, and select the top M diagnoses;

[0013] Step 1.2: Use a similarity algorithm to calculate the similarity between each diagnosis and the keyword in the recommended diagnosis list, sort each diagnosis from large to small according to the similarity, and select the top N diagnoses;

[0014] Step 1.3: Summarize the first M diagnoses and the first N diagnoses to obtain a diagnosis list.

[0015] Preferably, the weight p of each diagnosis in the recommended diagnosis list in step 1.1 is calculated as follows:

[0016]

[0017] quantity sum >limit

[0018] Among them, quantity sum is the total number of diagnostic hits, and limit is the set limit.

[0019] Preferably, the total number of hits quantity for the diagnosis sum is calculated as follows:

[0020] Judge whether the number of hits in the recent one month is greater than the set limit limit. If so, the total number of hits quantity for the diagnosis sum is the number of hits in the recent one month. If not, calculate the number of hits after expanding the time factor. The calculation method is:

[0021]

[0022] where quanity i is the number of hits within i months, and timePower i is the time factor weight of the number of times this diagnosis is opened from the (i - 1)-th month to the i-th month;

[0023] Judge whether the number of hits is greater than the set limit limit. If so, the total number of hits quantity for the diagnosis sum is the number of hits. If not, query the number of times other doctors in the same department open the diagnosis to calculate the number of hits. The calculation method is:

[0024]

[0025] where quantity ij is the number of times the j-th doctor in the same department opens this diagnosis from the (i - 1)-th month to the i-th month, and timePower ij is the time factor value of the j-th doctor in the same department from the (i - 1)-th month to the i-th month, and otherDoctorPower j is the weight of the number of times the j-th doctor in the same department opens this diagnosis.

[0026] Preferably, step 2 includes:

[0027] Step 2.1, based on the association knowledge base between diagnostic indicators and diagnoses, obtain the set of indicators related to each diagnosis in the diagnosis list;

[0028] Step 2.2, according to the indicator data of the current patient, where the indicator data includes age, gender, physical signs, examination and test indicators, calculate the intersection quantity indicator sum of the indicator data and the indicator data of each diagnosis:

[0029] indicator sum = |A ∩ B|

[0030] where A represents the indicator data of each diagnosis, and B represents the actual indicator data of the current patient;

[0031] Step 2.3: Optimize the weight of each diagnosis in the diagnosis list to obtain an optimized diagnosis list. The weight of each diagnosis in the optimized diagnosis list is p * The calculation method is:

[0032] p * =p+p1

[0033] p1=0.01*inicator sum

[0034] p1 is the total weight of the indicators for each diagnosis in the diagnosis list.

[0035] The present invention also provides an intelligent recommendation diagnosis system based on user behavior self-learning analysis, the system comprising:

[0036] A diagnosis list acquisition module is used to query according to the input keyword, obtain a recommended diagnosis list matching the keyword, calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis based on its weight, and obtain a diagnosis list;

[0037] The diagnosis list tuning module is used to tune the weight of each diagnosis in the diagnosis list based on the indicator data of the current patient and the associated knowledge base of the diagnosis indicator and the diagnosis to obtain the tuned diagnosis list;

[0038] A diagnosis recommendation module, used to obtain the diagnosis to be prescribed based on the optimized diagnosis list, and obtain the comorbidity diagnosis associated with the prescribed diagnosis based on historical patient data;

[0039] The self-learning module is used to record the doctor's selection results and perform self-learning.

[0040] Preferably, the process of calculating the weight of each diagnosis in the recommended diagnosis list in the diagnosis list acquisition module includes:

[0041] Step 1.1, use the keyword matching algorithm to calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis from large to small according to the weight, and select the top M diagnoses;

[0042] Step 1.2: Use a similarity algorithm to calculate the similarity between each diagnosis and the keyword in the recommended diagnosis list, sort each diagnosis from large to small according to the similarity, and select the top N diagnoses;

[0043] Step 1.3: Summarize the first M diagnoses and the first N diagnoses to obtain a diagnosis list.

[0044] Preferably, the weight p of each diagnosis in the recommended diagnosis list in step 1.1 is calculated as follows:

[0045]

[0046] quantity sum >limit

[0047] Among them, quantity sum is the total number of hits in the diagnosis, and limit is the set limit value.

[0048] Preferably, the total number of hits quantity sum of the diagnosis is calculated as follows:

[0049] Judge whether the number of hits in the recent one month is greater than the set limit value limit. If so, the total number of hits quantity sum of the diagnosis is the number of hits in the recent one month. If not, calculate the number of hits after expanding the time factor. The calculation method is as follows:

[0050]

[0051] Among them, quanity i is the number of hits within i months, and timePower i is the time factor weight of the number of times this diagnosis is opened from the (i - 1)-th month to the i-th month;

[0052] Judge whether the number of hits is greater than the set limit value limit. If so, the total number of hits quantity sum of the diagnosis is the number of hits. If not, query the number of times other doctors in the same department open the diagnosis to calculate the number of hits. The calculation method is as follows:

[0053]

[0054] Among them, quantity ij is the number of times the j-th doctor in the same department opens this diagnosis from the (i - 1)-th month to the i-th month, and timePower ij is the time factor value of the j-th doctor in the same department from the (i - 1)-th month to the i-th month, and otherDoctorPower j is the weight of the number of times the j-th doctor in the same department opens this diagnosis.

[0055] Preferably, the diagnosis list optimization module further includes:

[0056] An index acquisition unit, configured to acquire an index set related to each diagnosis in the diagnosis list based on the association knowledge base between the diagnosis index and the diagnosis;

[0057] An indicator hit count calculation unit, which is used to calculate the intersection quantity of the indicator data and the indicator data of each diagnosis according to the indicator data of the current patient, where the indicator data includes age, gender, physical signs, and examination and test indicators sum :

[0058] indicator sum = |A ∩ B|

[0059] where A represents the indicator data of each diagnosis, and B represents the actual indicator data of the current patient;

[0060] A weight tuning unit, which is used to tune the weights of each diagnosis in the diagnosis list to obtain a tuned diagnosis list, and the weights p of each diagnosis in the tuned diagnosis list * The calculation method is as follows:

[0061] p * = p + p1

[0062] p1 = 0.01 * indicator sum

[0063] p1 is the total indicator weight of each diagnosis in the diagnosis list.

[0064] The advantages provided by the present invention are as follows:

[0065] (1) Based on the input path opened by the doctor's diagnosis, the present invention analyzes the data opened by the doctor each time for diagnosis, enabling the system to understand the doctor's diagnosis-opening behavior habits. In the process of continuously learning the data generated by the doctor using the system to open diagnoses over time, the system continuously corrects the doctor's individual data and gradually grows into a diagnosis recommendation system that conforms to the doctor's personality, assisting the doctor to quickly open diagnoses and helping the doctor quickly and accurately query and retrieve diagnosis information among thousands of diagnoses.

[0066] (2) The present invention has a knowledge base of key physical signs, key chief complaints, and key examination and test indicators corresponding to the built-in diagnoses. After the doctor determines the diagnosis, if the current patient has inconsistent key indicators with those in the knowledge base, the system will automatically remind the doctor of which indicators of the patient do not conform to the indicators corresponding to the diagnosis, assisting the doctor in making a final diagnosis.

[0067] (3) After the main diagnosis is issued, according to the built-in diagnosis association knowledge base and combined with the patient's own situation, the present invention gives possible complication diagnoses. The doctor can quickly open associated diagnoses according to the recommended results, improving the diagnosis and treatment efficiency. Description of the Drawings

[0068] Figure 1 is a flowchart of an intelligent recommendation diagnosis method based on user behavior self-learning analysis provided for the embodiments of the present invention;

[0069] Figure 2 and Figure 3 are respectively partial flowcharts in the present invention Figure 1 . Specific embodiments

[0070] In order to make the objectives, technical solutions and advantages of the present invention clearer and more understandable, the following combines specific embodiments and refers to the accompanying drawings to clearly and completely describe the technical solutions of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without making creative efforts fall within the scope of protection of the present invention.

[0071] Embodiment 1

[0072] As Figure 1 shown, also referring to Figure 2 and Figure 3 , this embodiment provides an intelligent recommendation diagnosis method based on self-learning analysis of user behavior, including the following steps:

[0073] Step 1: Query according to the input keyword. When the keyword is not found, the His system obtains a diagnosis list based on the total number of diagnoses issued in historical consultations. When the keyword is found, a recommended diagnosis list matching the keyword is obtained, the weights of each diagnosis in the recommended diagnosis list are calculated, and each diagnosis is sorted based on the weights of each diagnosis to obtain a diagnosis list;

[0074] The present invention performs machine learning analysis on historical diagnosis data every day according to the department's professional field. Based on the big data of historical patient diagnoses, the diagnosis data is analyzed in multiple dimensions including department specialty, hospital location, patient age, patient gender, patient occupation, and onset time to form the diagnosis distribution of different disease fields, different categories of patients, and different doctors within the hospital, which is stored in the big data search engine database to form a basic knowledge base. After a doctor receives a patient, based on the big data search engine, possible diagnoses are intelligently recommended according to the doctor and patient information. For example, when a pediatrician receives a child patient, the search engine brings in the doctor information, patient age, department visited, region, and season factors to recommend possible diagnoses such as upper respiratory tract infection and iodine deficiency, reducing the workload of the doctor's continuous retrieval and improving the consultation efficiency.

[0075] The process of calculating the weights of each diagnosis in the recommended diagnosis list includes:

[0076] Step 1.1: Calculate the weight p of each diagnosis in the recommended diagnosis list using the keyword matching algorithm, sort the diagnoses in descending order of weight, and select the top M diagnoses. In this embodiment, M is taken as 30, including the top 10 diagnoses with the highest matching degree with the input keywords, the top 10 with the highest similarity found based on diagnosis aliases, and the top 10 with the highest similarity calculated by retrieving keyword matches based on past error correction data; the calculation method of the weight p of each diagnosis in the recommended diagnosis list is as follows:

[0077]

[0078] quantity sum >limit

[0079] where quantity sum is the total number of hits of the diagnosis, and limit is the set limit value.

[0080] where the total number of hits quantity sum of the diagnosis is calculated as follows:

[0081] Judge whether the hit data volume in the recent one month is greater than the set limit value limit. If so, the total number of hits quantity sum of the diagnosis is the hit data volume in the recent one month. If not, calculate the hit data volume after expanding the time factor, and the calculation method is:

[0082]

[0083] where quantity i is the hit data volume within i months, and timePower i is the time factor weight of the number of times this diagnosis is opened from the (i - 1)-th month to the i-th month; the time factor weight of the number of times this diagnosis is opened within one month is 1.0, the time factor weight of the number of times this diagnosis is opened from one month to two months is 0.5, and the time factor weight of the number of times this diagnosis is opened from two months to three months is 0.2.

[0084] Judge whether this hit data volume is greater than the set limit value limit. If so, the total number of hits quantity sum of the diagnosis is this hit data volume. If not, query the diagnoses opened by other doctors in the same department to calculate the hit data volume, and the calculation method is:

[0085]

[0086] where quantity ijtimePower is the number of times the jth doctor in the same department prescribed this diagnosis from the i-1th month to the ith month. ij is the time factor value of the jth doctor in the same department from the i-1th month to the ith month, otherDoctorPower j is the weight of the number of times the jth doctor in the same department has prescribed this diagnosis, and the weight is 0.5.

[0087] Step 1.2, use a similarity algorithm to calculate the similarity between each diagnosis and the keyword in the recommended diagnosis list, sort each diagnosis from large to small according to the similarity, and select the top N diagnoses. In this embodiment, N is 10. The calculation process of the similarity algorithm is to perform similarity matching on the diagnosis field based on the input keyword. For example, if "atrioventricular nodal reentrant tachycardia" is input, the input diagnosis name is matched with the full amount of diagnostic information similarity. The matching result example is shown in Table 1:

[0088] Table 1 Matching result examples

[0089] Search content Matched content Similarity coefficient Atrioventricular nodal reentrant tachycardia Paroxysmal atrioventricular nodal reentrant tachycardia 0.854 Atrioventricular nodal reentrant tachycardia Double pathways within the atrioventricular node 0.316 Atrioventricular nodal reentrant tachycardia Triple pathways of the atrioventricular node 0.304

[0090] After the similarity coefficients are sorted, if the filtered similarity is lower than a limit value (eg, similarity limit value = 0.8), the first and second data diagnostic information in the above table will be recommended to the user.

[0091] Step 1.3: Summarize the first M diagnoses and the first N diagnoses to obtain a diagnosis list.

[0092] Step 2: According to the indicator data of the current patient, the weight of each diagnosis in the diagnosis list is adjusted based on the associated knowledge base of the diagnosis indicator and the diagnosis to obtain an adjusted diagnosis list;

[0093] Step 2.1, based on the associated knowledge base of diagnostic indicators and diagnoses, obtain a set of indicators related to each diagnosis in the diagnosis list;

[0094] Step 2.2: Based on the indicator data of the current patient, including age, gender, physical signs, and examination and test indicators, calculate the number of intersections between the indicator data and the indicator data of each diagnosis. sum :

[0095] indicator sum =|A∩B|

[0096] Among them, A represents the indicator data of each diagnosis, and B represents the actual indicator data of the current patient;

[0097] Step 2.3: Optimize the weight of each diagnosis in the diagnosis list to obtain an optimized diagnosis list. The weight of each diagnosis in the optimized diagnosis list is p *The calculation method is as follows:

[0098] p * = p + p1

[0099] p1 = 0.01 * indicator sum

[0100] p1 is the total weight of indicators for each diagnosis in the diagnosis list.

[0101] Step 3: Obtain the diagnoses to be issued based on the optimized diagnosis list, and obtain the comorbidity diagnoses associated with the issued diagnoses based on historical patient data.

[0102] According to the historical discharged patients in the hospital and the analysis of multi-diagnosis patients, count the number of main diagnoses and corresponding secondary diagnoses issued, as shown in Table 2:

[0103] Table 2 Historical patient data of main diagnoses and corresponding secondary diagnoses

[0104] Primary diagnosis Secondary diagnosis Statistical frequency Hypertension Coronary atherosclerotic heart disease 1089 Hypertension Lower extremity arteriosclerosis 602 Hypertension Retinal disease examination 102

[0105] Concurrent diagnosis intelligent recommendation: When the diagnosis of "hypertension" has been issued for a patient, give a prompt, and recommend the diagnoses with higher counts such as "coronary atherosclerotic heart disease" and "lower extremity arteriosclerosis" sorted according to the statistical times based on the above analysis. Combine the doctor's associated diagnosis prescribing habits, patient age, gender, physical signs, examinations, and test data of this visit to calculate the weight scores, and recommend the top 10 relevant diagnoses with the highest weight scores to the doctor, record the doctor's associated diagnosis selection results, and the system self-learns to enhance the doctor's personal data.

[0106] Step 4: Record the doctor's selection results this time and perform self-learning. The present invention adopts an incremental recording retrieval process. After the doctor issues a diagnosis, record the search term and the finally selected diagnosis, store the data in the ElasticSearch search server, and form the doctor's own diagnosis behavior habits through machine learning.

[0107] For example: Search input content: slow → chronic → chronic parotid → chronic parotid gland → chronic parotitis

[0108] The recorded data is shown in Table 3:

[0109] Table 3 Recorded data

[0110] Search content Hit results Number of hits Slow Chronic parotitis 1 Chronic Chronic parotitis 1 Chronic parotid gland Chronic parotitis 1 Chronic parotid Chronic parotitis 1 Chronic parotitis Chronic parotitis 1

[0111] Summarize the recorded data in Table 3. When the following data is retrieved again, the matching quantity is incremented by 1

[0112] Slow Chronic parotitis

[0113] Then the recorded data in Table 3 is updated to Table 4:

[0114] The recorded data after the update in Table 4

[0115] Search content Hit results Number of hits Slow Chronic parotitis 2 Chronic Chronic parotitis 1 Chronic parotid gland Chronic parotitis 1 Chronic parotid Chronic parotitis 1 Chronic parotitis Chronic parotitis 1

[0116] Retrieval and recommendation sorting: Every time a doctor checks for "slow", the system will, based on the doctor's behavior habits, give priority to recommending "chronic mumps".

[0117] The present invention operates based on ICD10 diagnoses, in-hospital past patient diagnosis data, and doctors' past prescribed diagnosis data. Every time new data is generated, it will be written into the database in real time for reinforcement learning, achieving the goal of the more it is used, the more accurate the results are, and the more it is used, the more worry-free it is for doctors, becoming an intelligent diagnostic assistant for doctors, enabling doctors to focus more on the diagnosis and treatment itself and better serving patients. The present invention has the following advantages:

[0118] (1) Self-learning of doctors' diagnosis prescribing behavior habits: Traditional diagnosis prescribing is only based on the national diagnosis standard and does not consider the individual needs of doctors. Each diagnosis prescribing starts from scratch. The present invention takes into account that the diagnosis prescribing behavior of each doctor in different departments, different specialties, and with different professional capabilities is personalized. By using the current machine learning capabilities and analyzing the data of each doctor's diagnosis prescribing based on the input path of the doctor's diagnosis prescribing, the system understands the doctor's diagnosis prescribing behavior habits. In the process of continuously learning the data generated by the doctor's use of the system for diagnosis prescribing over time, the system continuously corrects the doctor's individual data and gradually grows into a diagnosis recommendation system that conforms to the doctor's personality, assisting the doctor to quickly prescribe diagnoses.

[0119] (2) Intelligent diagnosis recommendation: Based on the doctor's personalized habits formed through self-learning, when prescribing a diagnosis each time, a relevant set of diagnosis results is quickly generated for the doctor to choose from.

[0120] (3) Error correction reminder for doctors' diagnosis prescribing: In traditional diagnosis prescribing, there are often mistakes in prescribing the wrong diagnosis, resulting in delays in the condition. The present invention has a knowledge base of key signs, key chief complaints, and key examination and test indicators corresponding to the diagnosis built in. After the doctor determines the diagnosis, if the current patient has key indicators that are inconsistent with those in the knowledge base, the system will automatically remind the doctor of which indicators of the patient do not conform to the indicators corresponding to the diagnosis, assisting the doctor in making a final diagnosis.

[0121] (4) Associative intelligent recommendation of related diagnoses for the prescribed diagnosis: In traditional diagnosis prescribing, after the main diagnosis of the patient is issued, if there are complications, the doctor needs to conduct another diagnosis search and prescription, which has no relevance and low efficiency. In the present invention, after the main diagnosis is issued, according to the built-in diagnosis association knowledge base and combined with the patient's own situation, possible complication diagnoses are given. The doctor can quickly prescribe related diagnoses based on the recommended results, improving the diagnosis and treatment efficiency.

[0122] (5)Self-learning of the latest diagnostic knowledge and standards in the industry: In traditional diagnostic order systems, when national or industry diagnostic standards change, the system cannot perceive it in a timely manner and needs to be upgraded and transformed to adapt. Doctors also need to spend time learning the new standards and specifications, which is time-consuming, laborious, and delays diagnosis and treatment. This invention uses machine learning. After the new standards and specifications are released and used in the hospital, relevant new data and the mapping relationship between new and old data are input into the system in a timely manner. The system combines the new data with the old data to form a new diagnostic knowledge base and updates the doctor's personal diagnostic data. When the doctor orders a diagnosis, the new specifications are recommended in a timely manner, and the corresponding old specifications are marked, enabling the doctor to switch seamlessly and learn the new specifications and standards in actual work without reducing the diagnosis and treatment efficiency.

[0123] Embodiment 2

[0124] This embodiment provides an intelligent recommendation diagnosis system based on self-learning analysis of user behavior, including:

[0125] A diagnosis list acquisition module, which is used to query according to the input keyword, obtain a recommended diagnosis list that matches the keyword, calculate the weights of each diagnosis in the recommended diagnosis list, and sort each diagnosis based on the weights of each diagnosis to obtain a diagnosis list.

[0126] The process of calculating the weights of each diagnosis in the diagnosis list acquisition module includes:

[0127] Step 1.1: Use a keyword matching algorithm to calculate the weights of each diagnosis in the recommended diagnosis list, sort each diagnosis from largest to smallest according to the weights, and select the top M diagnoses;

[0128] Among them, the calculation method of the weight p of each diagnosis in the recommended diagnosis list is:

[0129]

[0130] quantity sum >limit

[0131] Among them, quantity sum is the total number of hits of the diagnosis, and limit is the set limit value.

[0132] The total number of hits quantity of the diagnosis sum is calculated as follows:

[0133] Judge whether the hit data volume in the recent one month is greater than the set limit value limit. If so, the total number of hits quantity of the diagnosis sum is the hit data volume in the recent one month. If not, calculate the hit data volume after expanding the time factor. The calculation method is:

[0134]

[0135] Among them, i is the amount of hit data in i months, timePower i The time factor weight is the number of times the diagnosis is issued from month i-1 to month i;

[0136] Determine whether the hit data volume is greater than the set limit. If so, the total number of hits diagnosed is quantity. sum If not, query the diagnosis prescribed by other doctors in the same department to calculate the hit data volume. The calculation method is:

[0137]

[0138] Among them, quantity ij timePower is the number of times the jth doctor in the same department prescribed this diagnosis from the i-1th month to the ith month. ij is the time factor value of the jth doctor in the same department from the i-1th month to the ith month, otherDoctorPower j The weight of the number of times the jth doctor in the same department has prescribed this diagnosis.

[0139] Step 1.2: Use a similarity algorithm to calculate the similarity between each diagnosis and the keyword in the recommended diagnosis list, sort each diagnosis from large to small according to the similarity, and select the top N diagnoses.

[0140] Step 1.3: Summarize the first M diagnoses and the first N diagnoses to obtain a diagnosis list.

[0141] The diagnosis list tuning module is used to tune the weight of each diagnosis in the diagnosis list based on the current patient's indicator data and the associated knowledge base of the diagnosis indicator and diagnosis to obtain the tuned diagnosis list. The diagnosis list tuning module specifically includes:

[0142] An indicator acquisition unit, used to acquire an indicator set related to each diagnosis in the diagnosis list based on a knowledge base associated with the diagnosis indicator and the diagnosis;

[0143] The indicator hit count calculation unit is used to calculate the intersection number of the indicator data and the indicator data of each diagnosis based on the indicator data of the current patient, which includes age, gender, physical signs, and examination and inspection indicators. sum :

[0144] indicator sum =|A∩B|

[0145] Among them, A represents the indicator data of each diagnosis, and B represents the actual indicator data of the current patient;

[0146] A weight tuning unit, which is used to tune the weights of each diagnosis in the diagnosis list to obtain a tuned diagnosis list, and the weight p of each diagnosis in the tuned diagnosis list * is calculated as follows:

[0147] p * = p + p1

[0148] p1 = 0.01 * indicator sum

[0149] p1 is the total weight of indicators of each diagnosis in the diagnosis list.

[0150] A diagnosis recommendation module, which is used to obtain the diagnoses to be issued based on the tuned diagnosis list, and obtain the comorbidity diagnoses associated with the issued diagnoses based on the historical patient data.

[0151] A self-learning module, which is used to record the doctor's current selection result and perform self-learning.

[0152] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that: they can still modify the technical solutions recorded in the foregoing embodiments, or perform equivalent replacements on some of the technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent recommendation diagnosis method based on user behavior self-learning analysis, characterized by: The method comprises the following steps: Step 1: perform a search based on the input keyword to obtain a recommended diagnosis list matching the keyword, calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis based on its weight, and obtain a diagnosis list; Step 2: According to the indicator data of the current patient, the weight of each diagnosis in the diagnosis list is adjusted based on the associated knowledge base of the diagnosis indicator and the diagnosis to obtain an adjusted diagnosis list; Step 3: Obtain the diagnosis to be issued based on the optimized diagnosis list, and obtain the comorbidity diagnosis associated with the issued diagnosis based on historical patient data; Step 4: Record the doctor's selection result and conduct self-learning.

2. The intelligent recommendation diagnosis method based on user behavior self-learning analysis according to claim 1 is characterized by: The process of calculating the weight of each diagnosis in the recommended diagnosis list in step 1 includes: Step 1.1, use the keyword matching algorithm to calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis from large to small according to the weight, and select the top M diagnoses; Step 1.2: Use a similarity algorithm to calculate the similarity between each diagnosis and the keyword in the recommended diagnosis list, sort each diagnosis from large to small according to the similarity, and select the top N diagnoses; Step 1.3: Summarize the first M diagnoses and the first N diagnoses to obtain a diagnosis list.

3. The intelligent recommendation diagnosis method based on user behavior self-learning analysis according to claim 2 is characterized by: The weight p of each diagnosis in the recommended diagnosis list in step 1.1 is calculated as follows: quantitysu m >limit Among them, quantity m is the total number of diagnostic hits, and limit is the set limit.

4. The intelligent recommendation diagnosis method based on user behavior self-learning analysis according to claim 3 is characterized by: The total number of hits for the diagnosis m The calculation method is: Determine whether the hit data volume in the recent month is greater than the set limit. If so, the total number of hits diagnosed is quantitysu m The hit data volume in the recent month. If not, the hit data volume is calculated after expanding the time factor. The calculation method is: Among them, i is the amount of hit data in i months, timePower i The time factor weight is the number of times the diagnosis is issued from month i-1 to month i; Determine whether the hit data volume is greater than the set limit. If so, the total number of hits diagnosed is quantity. sum If not, query the diagnosis prescribed by other doctors in the same department to calculate the hit data volume. The calculation method is: Among them, quantity ij timePower is the number of times the jth doctor in the same department prescribed this diagnosis from the i-1th month to the ith month. ij is the time factor value of the jth doctor in the same department from the i-1th month to the ith month, otherDoctorPower j is the weight of the number of times the ,th doctor in the same department prescribes this diagnosis.

5. The intelligent recommendation diagnosis method based on user behavior self-learning analysis according to claim 1 is characterized by: The step 2 comprises: Step 2.1, based on the associated knowledge base of diagnostic indicators and diagnoses, obtain a set of indicators related to each diagnosis in the diagnosis list; Step 2.2: Based on the indicator data of the current patient, including age, gender, physical signs, and examination and test indicators, calculate the number of intersections between the indicator data and the indicator data of each diagnosis. sum : indicator sum =|A∩B| Among them, A represents the indicator data of each diagnosis, and B represents the actual indicator data of the current patient; Step 2.3: Optimize the weight of each diagnosis in the diagnosis list to obtain an optimized diagnosis list. The weight of each diagnosis in the optimized diagnosis list is p * The calculation method is: p * =p+p1 p1=0.01*indicator sum p1 is the total weight of the indicators for each diagnosis in the diagnosis list.

6. An intelligent recommendation diagnosis system based on user behavior self-learning analysis, characterized by: The system comprises: A diagnosis list acquisition module is used to query according to the input keyword, obtain a recommended diagnosis list matching the keyword, calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis based on its weight, and obtain a diagnosis list; The diagnosis list tuning module is used to tune the weight of each diagnosis in the diagnosis list according to the indicator data of the current patient and the associated knowledge base of the diagnosis indicator and the diagnosis to obtain the tuned diagnosis list; A diagnosis recommendation module, used to obtain the diagnosis to be prescribed based on the optimized diagnosis list, and obtain the comorbidity diagnosis associated with the prescribed diagnosis based on historical patient data; The self-learning module is used to record the doctor's selection results and perform self-learning.

7. The intelligent recommendation diagnosis system based on user behavior self-learning analysis according to claim 6 is characterized by: The process of calculating the weight of each diagnosis in the recommended diagnosis list in the diagnosis list acquisition module includes: Step 1.1, use the keyword matching algorithm to calculate the weight of each diagnosis in the recommended diagnosis list, sort each diagnosis from large to small according to the weight, and select the top M diagnoses; Step 1.2: Use a similarity algorithm to calculate the similarity between each diagnosis and the keyword in the recommended diagnosis list, sort each diagnosis from large to small according to the similarity, and select the top N diagnoses; Step 1.3: Summarize the first M diagnoses and the first N diagnoses to obtain a diagnosis list.

8. The intelligent recommendation diagnosis system based on user behavior self-learning analysis according to claim 7 is characterized by: The weight p of each diagnosis in the recommended diagnosis list in step 1.1 is calculated as follows: quantity sum >limit Among them, quantity sum is the total number of diagnostic hits, and limit is the set limit.

9. The intelligent recommendation diagnosis system based on user behavior self-learning analysis according to claim 8 is characterized by: The total number of hits for the diagnosis sum The calculation method is: Determine whether the hit data volume in the recent month is greater than the set limit. If so, the total hit quantity of the diagnosis is sum The hit data volume in the recent month. If not, the hit data volume is calculated after expanding the time factor. The calculation method is: Among them, i is the amount of hit data within i months, timePower i The time factor weight is the number of times the diagnosis is issued from month i-1 to month i; Determine whether the hit data volume is greater than the set limit. If so, the total number of hits diagnosed is quantity. sum If not, query the diagnosis prescribed by other doctors in the same department to calculate the hit data volume. The calculation method is: Among them, quantity j timePower is the number of times the jth doctor in the same department prescribed this diagnosis from the i-1th month to the ith month. ij is the time factor value of the jth doctor in the same department from the i-1th month to the ith month, otherDoctorPower j is the weight of the number of times the ,th doctor in the same department prescribes this diagnosis.

10. The intelligent recommendation diagnosis system based on user behavior self-learning analysis according to claim 6, characterized in that: The diagnostic list tuning module also includes: An indicator acquisition unit, used to acquire an indicator set related to each diagnosis in the diagnosis list based on a knowledge base associated with the diagnosis indicator and the diagnosis; The indicator hit count calculation unit is used to calculate the intersection number of the indicator data and the indicator data of each diagnosis based on the indicator data of the current patient, which includes age, gender, physical signs, and examination and inspection indicators. sum : indicator sum =|A∩B| Among them, A represents the indicator data of each diagnosis, and B represents the actual indicator data of the current patient; The weight tuning unit is used to tune the weight of each diagnosis in the diagnosis list to obtain the tuned diagnosis list. The weight of each diagnosis in the tuned diagnosis list is p * The calculation method is: p * =p+p1 p1=0.01*indicator sum p1 is the total weight of the indicators for each diagnosis in the diagnosis list.

Images