Recommended method for the problem, apparatus and system thereof, and electronic equipment, readable storage medium

Abstract

The present disclosure provides a question recommendation method, an apparatus, a system, an electronic device, and a non-transitory readable storage medium. The question recommendation method includes the steps of: acquiring a candidate question set for a user, the candidate question set including a plurality of candidate questions; acquiring user behavior data and acquiring a user interest parameter based on the user behavior data; acquiring at least one similarity feature between each candidate question in the plurality of candidate questions and the user interest parameter based on the user interest parameter and the plurality of candidate questions; sorting the plurality of candidate questions based on the user basic information, the plurality of candidate questions, and the at least one similarity feature to obtain a question sequence; and recommending at least one candidate question in the question sequence to the user based on the order of the question sequence.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to a question recommendation method, a question recommendation apparatus, a question recommendation system, an electronic device, and a non-transitory readable storage medium.

Background Art

[0002] With the rapid development of Internet technology, great convenience has been brought to people's lives. People can search for and browse information they are interested in through the Internet, and can also conduct online inquiries, online medical consultations and the like through the Internet. Over time, a large amount of content information has been accumulated on the Internet. Conventional search engines provide a large number of web page search results, which may also include lots of duplicate and irrelevant content, making it difficult for users to find information they are interested in or questions they want to inquire about in a short time.

Disclosure of the Invention

[0003] At least one embodiment of the present disclosure provides a question recommendation method, comprising: obtaining a candidate question set of a user including a plurality of candidate questions; obtaining user behavior data, and obtaining user interest parameters based on the user behavior data; obtaining at least one similarity feature between each candidate question in the plurality of candidate questions and the user interest parameter based on the user interest parameter and the plurality of candidate questions; sorting the plurality of candidate questions to obtain a question sequence based on user basic information, the plurality of candidate questions and the at least one similarity feature; and recommending at least one candidate question in the question sequence to the user based on the order of the question sequence.

[0004] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the step of sorting a plurality of candidate problems to obtain a problem sequence based on the user basic information, the plurality of candidate problems, and the at least one similarity feature includes using a sorting model to configure the user basic information, the plurality of candidate problems, and the at least one similarity feature as input feature vectors of the sorting model, obtaining a score corresponding to each candidate problem in the plurality of candidate problems, sorting the corresponding plurality of candidate problems according to the magnitude of the scores, and obtaining the problem sequence.

[0005] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the step of obtaining at least one similarity feature between each candidate problem in the plurality of candidate problems and the user interest parameter, based on the user interest parameter and the plurality of candidate problems, includes the step of obtaining at least one similarity feature between each candidate problem and the user interest parameter, based on the user interest parameter and the plurality of candidate problems, using at least one similarity matching model.

[0006] For example, in the problem-recommended method provided by at least one embodiment of the present disclosure, the at least one similarity matching model includes at least one of a cosine similarity model, a Gecart similarity model, an edit distance similarity model, a word movement distance similarity model, and a deep semantic matching similarity model.

[0007] For example, in the problem-solving method provided by at least one embodiment of the present disclosure, the sorting model includes a Wide&Deep model.

[0008] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the step of obtaining the user's candidate problem set includes accessing a data knowledge base containing a plurality of knowledge problem sets; obtaining user basic information and creating a user label set based on the user basic information; associating the user label set with the data knowledge base and obtaining the candidate problem set from the plurality of knowledge problem sets.

[0009] For example, in the problem-recommended method provided by at least one embodiment of this disclosure, the user label set includes a multilevel label set containing multilevel labels, where different levels of labels are of different types.

[0010] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the problem recommendation method is used to recommend a disease-related problem, and the multi-level label set is such that the first level label is an age range, the second level label is a time period, the third level label is a disease type, and the fourth level label is a complication.

[0011] For example, in the problem recommendation method provided by at least one embodiment of the present disclosure, each of the plurality of knowledge problem sets includes a standard problem, a standard answer corresponding to the standard problem, and an extended problem corresponding to the standard problem.

[0012] For example, in the problem recommendation method provided by at least one embodiment of the present disclosure, the step of obtaining the user's candidate problem set further includes the step of creating the data knowledge base.

[0013] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the step of creating the data knowledge base includes the steps of taking a dataset from a network, classifying the dataset according to intent to form the plurality of knowledge problem sets, and creating the data knowledge base.

[0014] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the problem recommendation method is used to recommend a problem relating to a disease, and the dataset consists of at least one of a physician-patient interview dataset, hot topics related to the disease, and sweepstakes related to the disease.

[0015] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the steps of associating the user label set with the data knowledge base and obtaining a candidate problem set from the plurality of knowledge problem sets include creating a mapping relationship between the user label set and standard problems in the data knowledge base, matching the user label set with standard problems in the data knowledge base, and constructing the candidate problem set with knowledge problem sets corresponding to the matched standard problems.

[0016] For example, in the problem recommendation method provided by at least one embodiment of the present disclosure, the step of obtaining the user's candidate problem set includes the step of obtaining a pre-stored user's candidate problem set.

[0017] For example, in a problem recommendation method provided by at least one embodiment of the present disclosure, the step of obtaining the user interest parameters based on the user behavior data includes the step of analyzing the user behavior data and converting the problems the user has clicked on, or the words and phrases the user is interested in, into the user interest parameters.

[0018] At least one embodiment of the present disclosure further provides a problem recommendation device comprising: a set acquisition circuit configured to acquire a user's candidate problem set comprising a plurality of candidate problems; an activity analysis circuit configured to acquire user activity data and user interest parameters based on the user activity data; a feature generation circuit configured to acquire at least one similarity feature between each of the plurality of candidate problems and the user interest parameters based on the user interest parameters and the plurality of candidate problems; a problem sorting circuit configured to sort the plurality of candidate problems and acquire a problem sequence based on user basic information, the plurality of candidate problems and the at least one similarity feature; and a recommendation circuit configured to recommend at least one candidate problem from the problem sequence to the user based on the order of the problem sequence.

[0019] For example, in a problem recommendation device provided by at least one embodiment of the present disclosure, the problem sorting circuit includes a problem sorting subcircuit configured to use a sorting model to configure the user basic information, the plurality of candidate problems and the at least one similarity feature as input feature vectors to the sorting model, obtain a score corresponding to each candidate problem in the plurality of candidate problems, sort the corresponding plurality of candidate problems according to the magnitude of the score, and obtain the problem sequence.

[0020] For example, in a problem recommendation device provided by at least one embodiment of the present disclosure, the set acquisition circuit includes a knowledge base access circuit configured to access a data knowledge base containing a plurality of knowledge problem sets; an information acquisition circuit configured to acquire user basic information and create a user label set based on the user basic information; and a candidate set generation circuit configured to associate the user label set with the data knowledge base and acquire a candidate problem set containing a plurality of candidate problems from the plurality of knowledge problem sets.

[0021] For example, in a problem recommendation device provided by at least one embodiment of the present disclosure, the set acquisition circuit further includes a knowledge base creation circuit configured to acquire (grab) a dataset from a network, classify the dataset according to intent to form the plurality of knowledge problem sets, and create the data knowledge base.

[0022] For example, in a problem recommendation device provided by at least one embodiment of the present disclosure, the candidate set generation circuit includes a candidate set generation subcircuit configured to create a mapping relationship between the user label set and the standard problems in the data knowledge base, to match the user label set to the standard problems in the data knowledge base, and to constitute the candidate problem set with a set of knowledge problems corresponding to the matched standard problems.

[0023] For example, in a problem recommendation device provided by at least one embodiment of the present disclosure, the behavior analysis circuit includes a behavior analysis subcircuit configured to analyze the user behavior data and convert the problems clicked by the user or words and phrases of interest to the user interest parameter.

[0024] For example, in a problem recommendation device provided by at least one embodiment of the present disclosure, the feature generation circuit includes a feature generation subcircuit configured to use at least one similarity matching model to obtain at least one similarity feature between each of the candidate problems and the user interest parameter, based on the user interest parameter and the plurality of candidate problems.

[0025] At least one embodiment of the present disclosure further provides a question recommendation system including a terminal and a question recommendation server. The terminal is configured to send request data to the question recommendation server, the question recommendation server is configured to, in response to the request data, obtain a set of candidate questions for a user, the set of candidate questions includes a plurality of candidate questions, obtain user behavior data, obtain a user interest parameter based on the user behavior data, obtain at least one similarity feature between each candidate question in the plurality of candidate questions and the user interest parameter based on the user interest parameter and the plurality of candidate questions, sort the plurality of candidate questions based on basic user information, the plurality of candidate questions and the at least one similarity feature to obtain a question sequence, and the terminal is further configured to display up to N candidate questions in the question sequence, wherein N is an integer greater than or equal to 1.

[0026] At least one embodiment of the present disclosure further provides an electronic device, including a processor, and a memory including one or more computer program modules, the one or more computer program modules are stored in the memory and configured to be executed by the processor, and the one or more computer program modules include instructions for executing the question recommendation method described in any one of the above embodiments.

[0027] At least one embodiment of the present disclosure further provides a non-transitory readable storage medium storing computer instructions, which, when executed by a computer instruction processor, executes the question recommendation method described in any one of the above embodiments.

[0028] In order to more clearly describe the technical solutions of the embodiments of the present disclosure, the drawings of the embodiments are briefly described below. Obviously, the drawings in the following description only relate to some embodiments of the present disclosure, and are not limitations on the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] [Figure 1A]This is an illustrative flowchart of a problem-solving method provided by at least one embodiment of the present disclosure. [Figure 1B] This is an illustrative flowchart of another problem-solving method provided by at least one embodiment of the present disclosure. [Figure 2A] This document shows a user interface for a platform according to at least one embodiment of the present disclosure. [Figure 2B] A schematic diagram is shown illustrating the creation of a user label set according to at least one embodiment of the present disclosure. [Figure 3A] A schematic diagram of a rules scheme between a user label set and a data knowledge base according to at least one embodiment of this disclosure is shown. [Figure 3B] This document shows another user interface for a platform according to at least one embodiment of the present disclosure. [Figure 4A] This is a schematic diagram of the Wide&Deep model provided by at least one embodiment of this disclosure. [Figure 4B] This document shows yet another user interface for a platform according to at least one embodiment of the present disclosure. [Figure 4C] This document shows yet another user interface for a platform according to at least one embodiment of the present disclosure. [Figure 5A] This is an illustrative flowchart of another problem-solving method provided by at least one embodiment of the present disclosure. [Figure 5B] This is a schematic block diagram of the problem-solving method shown in Figure 5A, provided by at least one embodiment of the present disclosure. [Figure 6] This is a schematic block diagram of a problem-recommended device provided by at least one embodiment of the present disclosure. [Figure 7] This is a schematic block diagram of the problem-solving system provided by at least one embodiment of the present disclosure. [Figure 8] This is a schematic block diagram of an electronic device provided by at least one embodiment of the present disclosure. [Figure 9]This is a schematic block diagram of a terminal provided in at least one embodiment of the present disclosure. [Figure 10] This is a schematic block diagram of a non-temporary readable storage medium provided in at least one embodiment of the present disclosure. [Figure 11] An exemplary scene diagram of the problem recommendation system provided by at least one embodiment of this disclosure is shown. [Modes for carrying out the invention]

[0030] Hereinafter, exemplary embodiments of the present invention will be described in detail with reference to the drawings in order to further clarify the object, technical means, and merits of the present invention. It will be clear that the embodiments described are only a selection of embodiments of the present invention and not all embodiments of the present invention. All other embodiments that can be obtained by those skilled in the art without creative effort based on the embodiments of the present disclosure described herein are within the scope of protection of this disclosure.

[0031] Unless otherwise specified, technical or scientific terms used in this disclosure have their ordinary meanings as understood by persons of ordinary skill in the art to which this disclosure belongs. The words “first,” “second,” and similar words used in this disclosure do not indicate order, quantity, or importance, but are used solely to distinguish different components. Similarly, similar words such as “one,” “single,” or “the relevant” do not imply a limit on number, but mean at least one entity. Similar words such as “contains” or “includes” mean that the element or object preceding the word is the same as the element or object listed after the word, and do not exclude other elements or objects. Similar words such as “connected” or “connected to each other” are not limited to physical or mechanical connections, but include electrical connections, which may be direct or indirect. “Up,” “down,” “left,” “right,” etc., are used solely to describe relative positional relationships, and if the absolute position of the described object changes, its relative positional relationship may change accordingly.

[0032] With economic development and improved living standards, more and more people are suffering from chronic diseases of varying degrees. Chronic diseases are a general term for diseases that are not contagious and accumulate over a long period of time, causing morphological damage. Common chronic diseases include cardiovascular diseases, cancer, diabetes, and chronic respiratory diseases, with cardiovascular diseases including hypertension, stroke, and coronary artery sclerosis. According to data, one of the causes of chronic diseases is an unhealthy lifestyle. For example, an unhealthy lifestyle includes an irrational diet, lack of exercise, tobacco use, and excessive alcohol use. Therefore, for patients with chronic diseases, in addition to providing medical treatment (e.g., drug therapy) for the patient's condition, doctors need to provide patients with rationalized advice on managing chronic diseases (e.g., dietary advice, exercise advice) as a means of supporting medical treatment to better control and prevent chronic diseases. With the development of network technology, smart healthcare offers more convenience. In particular, health knowledge regarding chronic disease prevention and self-monitoring can increase patients' awareness of their illnesses and enable more effective prevention and treatment. Therefore, a smart Q&A system for health education for chronic disease patients can efficiently and accurately resolve patients' questions and problems. At the same time, given the current shortage of medical resources in the country, it will reduce the burden on healthcare professionals and be useful for long-term health management.

[0033] The inventors of this disclosure have noticed that, currently, medical question-and-answer systems or online consultations rely on the symptoms described by the patient and the questions raised by the patient, and therefore, problems often arise where the system feedback answers and patient information are inappropriate or out of order due to the unclear expression from the patient. Furthermore, general medical question-and-answer systems provide targeted health knowledge without utilizing user information. Providing personalized and diverse services is difficult.

[0034] At least one embodiment of the present disclosure provides a problem recommendation method, a problem recommendation device, a problem recommendation system, an electronic device, and a non-temporary readable storage medium. The problem recommendation method includes the steps of obtaining a user's candidate problem set, the candidate problem set comprising a plurality of candidate problems; obtaining user behavior data and obtaining user interest parameters based on the user behavior data; obtaining at least one similarity feature between each candidate problem in the plurality of candidate problems and the user interest parameters based on the user interest parameters and the plurality of candidate problems; and sorting the plurality of candidate problems to obtain a problem sequence based on user basic information, the plurality of candidate problems and the at least one similarity feature.

[0035] The problem recommendation method provided by at least one embodiment of this disclosure not only effectively avoids the problem of inappropriate feedback responses caused by unclear patient expressions, but also enables users to more accurately grasp their own health knowledge by recommending problems in a personalized manner based on individual factors and characteristics (e.g., basic user information, user click behavior, browsing behavior, etc.). In at least some other embodiments, the problem recommendation method can further achieve an effect that better suits user needs and effectively improve the user experience by using a sorting model to enable problem recommendations to be correlated, personalized, and diversified, while also emphasizing the final feedback order.

[0036] The following examples and embodiments will not limit the description of the health management devices provided by the embodiments of this disclosure, and new examples and embodiments can be obtained by combining different features within these specific examples and embodiments in a non-conflicting manner, as described below, and all of these new examples and embodiments will also fall within the scope of protection of this disclosure.

[0037] Figure 1A is an illustrative flowchart of a problem-solving method provided by at least one embodiment of the present disclosure. Figure 1B is an illustrative flowchart of another problem-solving method provided by at least one embodiment of the present disclosure.

[0038] The problem recommendation method 10 provided by at least one embodiment of this disclosure can be applied to scenarios such as medical smart question and answer, online medical interviews, and health consultations, and can be applied, for example, to a chronic disease health knowledge smart question and answer system. For example, in one embodiment, as shown in Figure 1A, the problem recommendation method 10 may include the following operations. Step S100: Retrieve the user's candidate problem set, which includes multiple candidate problems. Step S140: Obtain user behavior data and obtain user interest parameters based on the user behavior data. Step S150: Based on the user interest parameter and multiple candidate problems, obtain at least one similarity feature between each candidate problem in the multiple candidate problems and the user interest parameter. Step S160: Sort the multiple candidate problems based on user basic information, multiple candidate problems, and at least one similarity feature to obtain a problem sequence. Step S170: Based on the order of the problem sequence, recommend at least one candidate problem from the problem sequence to the user.

[0039] For example, in one embodiment, as shown in Figure 1B, step S100 of the problem-recommended method 10 may specifically include steps S110-S130, and therefore the problem-recommended method 10 may specifically include steps S110-S170.

[0040] The following describes in detail the problem-solving method 10 provided by embodiments of the present disclosure, combining Figures 1A and 1B. For example, in one embodiment of the present disclosure, the problem-solving method 10 may specifically include steps S110-S170, as shown in Figure 1B. For example, steps S110-S170 may be executed sequentially or in any other adjusted order. For example, in one example, step S110 may be executed first, followed by step S120; in another example, step S120 may be executed first, followed by step S110. Furthermore, some or all of the operations in steps S110-S160 may be executed in parallel. For example, steps S110 and S120 may be executed in parallel. Embodiments of the present disclosure do not limit the execution order of each step and can be adjusted according to the actual situation. For example, steps S110-S160 may be implemented on a server or local end, and embodiments of the present disclosure do not limit this. For example, in some cases, implementing the problem-solving method 10 provided by at least one embodiment of the present disclosure may involve selectively performing some of the steps S110-S170, and may also involve performing some additional steps other than steps S110-S170, and the embodiments of the present disclosure are not specifically limited thereto.

[0041] The problem recommendation method 10 provided in this disclosure will be described in detail below with examples and drawings.

[0042] Step S110: Access a database knowledge base containing multiple knowledge question sets. For example, if question recommendation method 10 is applied to a chronic disease health knowledge smart question and answer system, the database knowledge base may be a database knowledge base related to chronic diseases (e.g., diabetes, hypertension, etc.) and may contain complete basic health knowledge related to chronic diseases. The database knowledge base may contain multiple knowledge question sets, for example, by classifying the basic health knowledge within the database knowledge base according to intent (e.g., diet, exercise, drugs, tests, complications, surgery, treatment, symptoms, etc.), and organizing (e.g., manually organizing) standard questions and corresponding extended questions and standard answers under each intent to constitute one knowledge question set.

[0043] For example, in at least one embodiment of the present disclosure, each of a set of knowledge questions may include a standard question, a standard answer corresponding to the standard question, and an extended question corresponding to the standard question. For example, in one example, if the user is diabetic and under the intention of “Diet,” the organized standard question may include “What foods should a diabetic person eat?” and the extended question corresponding to the standard question may include, for example, “What foods are good for diabetes?”, “What foods are suitable for diabetes?”, and “What foods should a diabetic person eat to lower their blood sugar?”. In this case, the set of knowledge questions under the “Diet” intention may include the above standard question, extended question, and the standard answers corresponding to them.

[0044] The standard problems and extended problems cited in the embodiments of this disclosure are merely illustrative and can be adjusted and updated according to the application scenario and medical practice. Similarly, the standard answers can also be adjusted and updated according to the application scenario and medical practice, and can also be adjusted and updated based on the language used (e.g., Chinese, English), and the embodiments of this disclosure do not specifically limit this.

[0045] For example, in the embodiments of this disclosure, the data knowledge base is a created knowledge database that may be pre-stored locally or on a server, or it may be created by a server, for example, or read from another device when the problem recommendation method 10 is implemented. The embodiments of this disclosure do not specifically limit this, and can be provided according to actual needs. A detailed explanation of how to create the data knowledge base is given below.

[0046] Step S120: Obtain basic user information and create a user label set based on the basic user information.

[0047] For example, in at least one embodiment of this disclosure, user basic information may include, for example, the user's age, gender, height, weight, waistline, and lifestyle. For example, in one example, if the user is diabetic, the user basic information may further include the type of diabetes, any previously diagnosed chronic complications, and any existing symptoms. For example, in another example, if the user is not diabetic, the user basic information may further include the user's medical history, fasting blood glucose level, or blood glucose level two hours after a meal. For example, in yet another example, if the user is hypertensive, the user basic information may further include the user's diastolic blood pressure, systolic blood pressure, type of parahypertensive syndrome, and symptoms. The embodiments of this disclosure are not specifically limited to the content included in user basic information and can be provided according to actual needs.

[0048] For example, in one instance, basic user information can be obtained from the user's online real-time information, such as, with permission, from online health file platforms, information management systems of medical institutions (e.g., hospitals or health checkup facilities) (e.g., laboratory information management systems), or electronic physical examination report digitization devices.

[0049] For example, in one instance, if a user is a registered user of a software platform (e.g., a health management platform), and the user proactively enters and saves their basic information (e.g., name, gender, age, height, weight, waistline, lifestyle, etc.) when registering, the user's basic information can be directly retrieved from a repository (e.g., backend) associated with that particular platform. For example, in another instance, if the user is not a registered user of that particular platform (health management platform), or if the user has not completed or saved their basic information on that particular platform, the user's basic information may be collected through a third-party platform (e.g., an information management system of a hospital or health checkup facility) or related electronic devices (e.g., a bracelet, smartwatch, etc.), and the embodiments of this disclosure are not specifically limited thereto and can be adapted to the actual situation. For example, in one instance, if a user is using a web page or platform for the first time, the user's basic information can also be retrieved based on personal information entered by the user in a text box, and the embodiments of this disclosure are not specifically limited thereto and can be adapted to the actual situation.

[0050] Figure 2A shows the user interface of a platform in one embodiment of the disclosure. As shown in Figure 2A, a user can fill in basic information (e.g., name, gender, age, height, weight, waistline, etc.) according to their circumstances, then compare it with specific lifestyle options provided in the user interface (e.g., frequent drinking (more than 3 times a week), smoking, eating too salty food, liking fried food, liking sweet food, frequently staying up late (average bedtime after midnight), not getting enough exercise, etc.), check the lifestyles that suit them, and save this basic user information to a database (e.g., backend) associated with the health management platform.

[0051] For example, a health management platform can connect to at least one (e.g., multiple) medical institutions and obtain basic information and test results from at least one (e.g., multiple) medical institutions of patients who have participated in physical characteristic examinations at these institutions. For example, a health management platform can further obtain basic user information from a smart device (e.g., smart measuring device, smart bracelet, smart watch, smart wear, etc.) and obtain at least one piece of patient physical characteristic data (e.g., pulse, body temperature, heart rate, respiration, electroencephalography, electrocardiogram, blood pressure, blood glucose, electromyography, etc.) detected by sensors in the smart device. For example, a health management platform can periodically (e.g., daily) obtain test results from multiple medical institutions of patients who have participated in physical characteristic examinations at these institutions and store them in advance in a database (or memory) associated with the health management platform. The embodiments of this disclosure do not specifically limit the source of user basic information and can be provided according to actual needs. For example, in several other cases, user basic information can also be filled out by the user online, for instance, by providing a corresponding webpage where the user fills in their information, the webpage sends the information entered by the user to a server, and the server can retrieve the user basic information after it has processed this information.

[0052] For example, in one case, after obtaining basic user information based on the above, a label set is created based on the obtained basic user information.

[0053] For example, a user's label may include the user's gender, name, age, etc. For example, in one example, the label may further include the name of a disease, such as type 1 diabetes, and in another example, the label may further include the name of a complication, such as diabetic foot.

[0054] The following describes the process of creating a label set based on acquired user basic information, combining it with Figure 2B. Figure 2B shows a schematic diagram of creating a user label set according to at least one embodiment of this disclosure. As shown in Figure 2B, the dashed box on the left is the user's portrait, i.e., user basic information. This user basic information includes "User is diabetic patient A", "Height 172cm", "Weight 81kg", "Male", "68 years old", "Memory impairment", "Long-term lower limb ulcers", "Thick and swollen toes", "Prominent big toe bone", "Frequent drinking and smoking", and "Low exercise". Next, labels are acquired by a method such as entity recognition, which includes acquiring labels using a method such as a conditional random field or deep learning. For example, the dashed box on the right is the acquired associated labels for the user, i.e., the user label set includes "Blood glucose", "Elderly", "Overweight", "Complications", "Recipes", "Neurological lesions", and "Diabetic foot".

[0055] The specific labels mentioned in the above description of the embodiments of this disclosure are illustrative and not limiting.

[0056] For example, in at least one embodiment of the present disclosure, a user label set may include a multilevel label set, and the multilevel label set includes multilevel labels, where different levels of labels are of different types.

[0057] For example, in at least one embodiment of this disclosure, nearly 200 basic rules can be created based on different ages, time of day, type of disease, presence or absence of symptoms, presence or absence of complications, etc., and then diverge to different types of diabetes, combinations of multiple symptoms, and combinations of multiple types of complications, so that a multi-level set of labels can cover the user's basic information and interests.

[0058] For example, in one case, a multilevel label set might have a first-level label for age range, a second-level label for time of day, a third-level label for disease type, and a fourth-level label for complications.

[0059] The embodiments described herein do not limit the specific rules for user label sets and can be established according to actual needs.

[0060] Step S130: Associate the user label set with the data knowledge base and obtain a candidate question set containing multiple candidate questions from multiple knowledge question sets.

[0061] For example, in at least one embodiment of this disclosure, a user label set created in step S120 can be matched with a data knowledge base accessed in step S110 to obtain a set of candidate problems based on the association of rules.

[0062] For example, in at least one embodiment of the present disclosure, step S130, which involves associating a user label set with a data knowledge base and obtaining a candidate problem set from a plurality of knowledge problem sets, may include first creating a mapping relationship between the user label set and standard problems in the data knowledge base, matching the user label set with standard problems in the data knowledge base, and then constructing a candidate problem set with knowledge problem sets corresponding to the matched standard problems.

[0063] Figure 3A shows a schematic diagram of a rules scheme between a user label set and a data knowledge base according to at least one embodiment of the present disclosure. The operation of step S130 will be described in detail below, using Figure 3A as an example and a basic rules scheme for diabetes.

[0064] As shown in Figure 3A, in at least one embodiment of the present disclosure, an age range is used as a first-level label to first directly associate with a user (e.g., a diabetic patient). For example, when a user associates with a first-level label, they can be associated with three categories: minors, newborns, and children aged 18 or under; adults aged between 18 and 59; and middle-aged and elderly people aged 60 or over.

[0065] For different age groups, combining the second-level label (i.e., different time periods of the day) and the third-level label (type of diabetes) results in different corresponding rule schemes.

[0066] For example, when labeling at each level, there are no restrictions on the order in which they are assigned. For instance, age, time, and type of diabetes can be single-selected and classified earlier, while types of complications can be multiple-selected, are relatively complex, and are placed later.

[0067] For example, in at least one embodiment of this disclosure, for children with type 1 and type 2 diabetes, during the breakfast period (e.g., 6:00-8:00), they may inquire about related issues such as recipe and nutritional suggestions and 2-hour post-meal blood glucose precautions. For example, in one example, for children with type 1 and type 2 diabetes, during the breakfast period (e.g., 6:00-8:00), the corresponding label may include "Type 1 and Type 2 Diabetes," "Children," and "6:00-8:00," and several corresponding candidate questions may include "What should children with diabetes eat for breakfast?", "What precautions should children with type 1 and type 2 diabetes take after eating early?", and "What devices should diabetic patients use to measure blood glucose?". For children with type 1 and type 2 diabetes, during the lunch period (e.g., 11:00-15:00), they may inquire about related issues such as recipe and nutritional suggestions, exercise suggestions, and precautions. For example, in one example, for children with type 1 and type 2 diabetes, during the lunchtime period (e.g., 11:00-15:00), the corresponding label might include "Type 1 and Type 2 Diabetes," "Children," and "11:00-15:00," and several corresponding candidate questions could include "What should children with diabetes eat for lunch?", "What precautions should children with type 1 and type 2 diabetes take after lunch?", and "What kind of exercise is suitable for children with type 1 and type 2 diabetes after lunch?". For children with type 1 and type 2 diabetes, during the dinnertime period (e.g., 17:00-20:00), they might inquire about related issues such as recipe and nutritional suggestions, exercise suggestions and precautions, and matters concerning medication use. For example, in one example, for children with type 1 and type 2 diabetes, the corresponding label for the evening meal period (e.g., 17:00-20:00) could include "Type 1 and Type 2 Diabetes," "Children," and "17:00-20:00," and several corresponding candidate questions could include: "What should children with type 1 and type 2 diabetes eat for dinner?", "What precautions should diabetic patients take after dinner?", "What kind of exercise is suitable for children with diabetes after dinner?", and "How should children with type 1 and type 2 diabetes take their medication at night?"Furthermore, the recommended problems cited in the embodiments of this disclosure are not limiting but merely illustrative.

[0068] For example, in at least one embodiment of this disclosure, adult patients with type 1 and type 2 diabetes may inquire about relevant issues such as recipe and nutritional suggestions and precautions for measuring blood glucose 2 hours after an early meal during the breakfast period (e.g., 6:00-8:00). For example, in one example, for adult patients with type 1 and type 2 diabetes, during the breakfast period (e.g., 6:00-8:00), the corresponding label may include "Type 1 and Type 2 Diabetes," "Adult," and "6:00-8:00," and several corresponding candidate questions may include "What should adult diabetic patients eat for breakfast?", "What precautions should adult diabetic patients take after an early meal?", and "Can adult diabetic patients drink milk in the morning?" During the lunch period (e.g., 11:00-15:00), they may inquire about relevant issues such as recipe and nutritional suggestions, exercise suggestions, and precautions. For example, in one example, for adult patients with type 1 and type 2 diabetes, the lunchtime period (e.g., 11:00-15:00) could include the labels "Type 1 and Type 2 Diabetes," "Adult," and "11:00-15:00," and several corresponding candidate questions could include "What should adult patients with type 1 and type 2 diabetes eat for lunch?", "What precautions should adult patients with diabetes take after lunch?", "What kind of exercise is suitable for adult patients with type 1 and type 2 diabetes after lunch?", and "How should adult patients with type 1 and type 2 diabetes take their medication at noon?". During the dinnertime period (e.g., 17:00-20:00), they might ask about related issues such as recipe and nutritional suggestions, exercise suggestions and precautions, precautions for blood glucose monitoring 2 hours after dinner, and matters concerning medication use.For example, in one example, for adult patients with type 1 and type 2 diabetes, the dinner period (e.g., 17:00-20:00) could be labeled with "Type 1 and Type 2 Diabetes," "Adult," and "17:00-20:00," and several candidate questions could include "What should adult patients with type 1 and type 2 diabetes eat for dinner?", "What precautions should adult patients with type 1 and type 2 diabetes take after dinner?", "What kind of exercise is suitable for adult patients with type 1 and type 2 diabetes after dinner?", and "How should adult patients with type 1 and type 2 diabetes take their medication at night?". It should be noted that the recommended questions listed in the embodiments of this disclosure are not limiting and are merely illustrative.

[0069] For example, in at least one embodiment of this disclosure, elderly diabetic patients and gestational diabetic patients may inquire about relevant issues such as recipe and nutritional suggestions and precautions for blood glucose measurement 2 hours after an early meal during the breakfast period (e.g., 6:00-8:00). For example, in one example, for elderly diabetic patients and gestational diabetic patients, during the breakfast period (e.g., 6:00-8:00), the corresponding labels may include "diabetes," "gestational diabetic," "elderly," and "6:00-8:00," and several corresponding candidate questions may include "What should elderly diabetic and gestational diabetic patients eat for breakfast?", "What precautions should elderly diabetic and gestational diabetic patients take after an early meal?", and "How should elderly diabetic and gestational diabetic patients take their medication in the morning?" During the lunch period (e.g., 11:00-15:00), they may inquire about relevant issues such as recipe and nutritional suggestions, precautions for blood glucose measurement 2 hours after lunch, exercise suggestions and precautions. For example, in one case, for elderly diabetic patients and gestational diabetic patients, the lunchtime period (e.g., 11:00-15:00) could include the labels "Diabetes," "Gestational Diabetes," "Elderly," and "6:00-8:00," and several corresponding candidate questions could include "What should elderly diabetic and gestational diabetic patients eat for lunch?", "What precautions should elderly diabetic and gestational diabetic patients take after lunch?", "What kind of exercise is suitable for elderly diabetic and gestational diabetic patients after lunch?", and "How should elderly diabetic and gestational diabetic patients take their medication at noon?". During the dinnertime period (e.g., 17:00-20:00), they might inquire about related issues such as recipe and nutritional suggestions, exercise suggestions and precautions, precautions for blood glucose monitoring 2 hours after dinner, and matters concerning medication use.For example, in one example, for elderly diabetic patients and gestational diabetic patients, the corresponding label for the evening meal period (e.g., 17:00-20:00) might include "diabetes," "gestational diabetic," "elderly," and "17:00-20:00," and several corresponding candidate questions could include: "What should elderly diabetic patients and gestational diabetic patients eat for dinner?", "What precautions should elderly diabetic patients and gestational diabetic patients take after dinner?", "What kind of exercise is suitable for elderly diabetic patients and gestational diabetic patients after dinner?", and "How should elderly diabetic patients and gestational diabetic patients take their medication at night?". It should be noted that the recommended questions listed in the embodiments of this disclosure are not limiting and are merely illustrative.

[0070] For example, in at least one embodiment of this disclosure, patients with prediabetes may inquire about relevant issues such as recipe and nutritional suggestions and 2-hour post-meal blood glucose precautions during the breakfast period (e.g., 6:00-8:00). For example, in one example, for patients with prediabetes, during the breakfast period (e.g., 6:00-8:00), several corresponding candidate questions may include: "What should patients with prediabetes eat for breakfast?", "What precautions should patients with prediabetes take after an early meal?", and "How should patients with prediabetes take their medication in the morning?" During the lunch period (e.g., 11:00-15:00), they may inquire about relevant issues such as recipe and nutritional suggestions, exercise suggestions, and precautions. For example, in one example, for a patient in the early stages of diabetes, several candidate questions corresponding to the lunchtime period (e.g., 11:00-15:00) could include: "What should a patient in the early stages of diabetes eat for lunch?", "What precautions should a patient in the early stages of diabetes take after lunch?", and "How should a patient in the early stages of diabetes take their medication at noon?". During the dinnertime period (e.g., 17:00-20:00), they might inquire about related issues such as recipe and nutritional suggestions, exercise suggestions, and precautions. For example, in one example, for a patient in the early stages of diabetes, several candidate questions corresponding to the dinnertime period (e.g., 17:00-20:00) could include: "What should a patient in the early stages of diabetes eat for dinner?", "What precautions should a patient in the early stages of diabetes take after dinner?", "What kind of exercise is suitable for a patient in the early stages of diabetes after dinner?", and "How should a patient in the early stages of diabetes take their medication at night?". It should be noted that the recommended questions listed in the embodiments of this disclosure are not limiting and are merely illustrative.

[0071] For example, in at least one embodiment of this disclosure, a regular association is made between a first-level label (e.g., age group), a second-level label (e.g., time of day), and a third-level label (e.g., type of disease), with complications being the fourth-level label. According to clinical data, within about 10 years after the onset of the disease, 30% to 40% of diabetic patients develop at least one complication such as cardiovascular disease, renal disease, retinal lesions, nerve lesions, lower extremity vascular lesions, or diabetic foot.

[0072] For example, in some exemplary rule schemes, if no complications are diagnosed, problems related to one or more complications that are already present symptoms of the user are selected and recommended, and then these related problems are returned to a candidate problem set corresponding to the user, i.e., multiple candidate problems corresponding to the user are formed. For example, if no complications are diagnosed, the clinical data is combined with the user's already present symptoms to correspondingly recommend problems related to one or more complications. For example, in one example, high blood pressure, chest pain and palpitations, and chest tightness belong to the symptoms of cardiovascular disease complications. For example, if the user already has symptoms of high blood pressure and chest pain, the label set for the user can include the label "cardiovascular disease," and several problems related to cardiovascular disease complications such as "What are the symptoms of cardiovascular disease complications?", "Why do I experience palpitations and chest tightness?", and "What should I do if I experience palpitations and chest tightness?" can be recommended, and the embodiments of this disclosure are not specifically limited thereto. For example, in another example, foamy urine, difficulty urinating and edema of the lower extremities, and edema of the eyelids belong to the symptoms of diabetic nephropathy complications. For example, in another instance, blurred vision, decreased visual acuity, and blackout are considered retinal complications. For example, in yet another instance, unclear speech, memory loss, and persistent numbness, stabbing pain, and swelling in the limbs are considered neurological complications. For example, in yet another instance, weakness in the lower limbs, foot problems, and intermittent nocturnal pain in the lower limbs are considered lower limb vascular complications. For yet another instance, for example, long-term ulcers in the lower limbs and swelling of the fingers or toes, and a prominent big toe bone are considered diabetic foot complications.

[0073] For example, if a complication has already been diagnosed, relevant issues can be recommended based on that complication. For instance, if a user's label includes "cardiovascular disease," the issues matched to that label in the data knowledge base could include, "How are cardiovascular complications treated in diabetic patients?" or "What are the symptoms of cardiovascular complications in diabetic patients?" The embodiments of this disclosure are not limited to these examples.

[0074] Furthermore, the classification of various diseases and the classification of various symptoms in combination as described in the embodiments of this disclosure are solely for the purpose of illustrating how to create a mapping relationship between a user label set and a data knowledge base. In other words, the specific rule scheme described above is explained for the purpose of further explanation, and the specific disease classification and symptom analysis are adjusted and established based on a large amount of clinical data, such as the empirical judgment of experts, and the embodiments of this disclosure do not specifically limit this.

[0075] As can be seen by referring to Figure 3A, the above explanations all apply when the user selects that they have diabetes. If the user selects that they do not have diabetes, the operation can be performed as follows. For example, if the user meets at least three of the following conditions, such as "always sits still, has a history of diabetes in a first-degree relative, has a history of hypertension, dyslipidemia, impaired glucose regulation, a history of producing large children, is a woman with a history of gestational diabetes, and is a patient with atherosclerotic cardiovascular disease," and also meets one of the following conditions, such as "6.1 mmol / L < fasting blood glucose < 7 mmol / L, 7.8 mmol / L < 2 hours postprandial blood glucose < 11.1 mmol / L," then the system will match the user to the data knowledge base according to the rules for early diabetes and construct a set of candidate questions.

[0076] For example, in at least one embodiment of this disclosure, the user is a hypertensive patient. The rule scheme for a hypertensive patient differs from that for a diabetic patient. For example, in one example, the rule scheme for a hypertensive patient can directly determine pre-hypertension, mild hypertension, moderate hypertension, and severe hypertension based on the magnitude of diastolic and systolic blood pressure, and different rule schemes can be implemented. For example, in the rule scheme for a hypertensive patient, for the first level of labeling (i.e., age group), a special stage is set for elderly people aged 80 and over, which is classified as severe elderly hypertension, and blood pressure is monitored at all times. In addition, since diabetes is a common complication of hypertension, therefore, the rule scheme for a hypertensive patient also needs to monitor blood glucose at the same time as monitoring blood pressure at specific times.

[0077] The rule schemes provided by the embodiments of this disclosure are merely illustrative and do not limit the specific rule schemes provided by this disclosure; they can be adapted to actual needs.

[0078] For example, in the embodiments of this disclosure, after creating a mapping relationship between a user label set and standard questions in a data knowledge base, the user label set is matched to standard questions in the data knowledge base using methods such as entity recognition, keyword matching, and deep learning. Then, a set of knowledge questions corresponding to the matched standard questions is obtained from the data knowledge base to constitute a candidate question set. Entity recognition refers to recognizing entities with specific meanings within text, such as names of people, places, and times. Keyword matching methods include broad matching, exact matching, phrase matching, and negative matching. For example, in one example, if the text content of a label is "diabetes," it can be matched to recommendation questions in the data knowledge base that contain the three characters "diabetes," such as "Recipes for people with diabetes?", "What kind of exercise is suitable for people with diabetes?", and "Symptoms of people with diabetes?", but the embodiments of this disclosure are not limited to these examples.

[0079] For example, in one case, an elderly person over 60 with type 1 diabetes and complications might want to obtain corresponding recipes, blood glucose monitoring suggestions, and information about issues related to complications at noon. In the steps above, a corresponding user label set is created based on this user's basic information (i.e., the labels include "type 1 diabetes," "complications," "over 60," and "noon"), and this user label set is associated with a data knowledge base. A candidate question set containing multiple candidate questions is then obtained from the multiple knowledge question sets contained in the data knowledge base. For example, based on the user label set, a candidate question set can be constructed by matching multiple relevant candidate questions from the data knowledge base using methods such as entity recognition, keyword matching, or deep learning. For example, multiple candidate questions corresponding to the user in the example above could include: "What are three daily meal recipes for a diabetic?", "What kind of exercise is suitable for elderly diabetic patients?", "How often should a diabetic patient monitor their blood glucose daily?", "What are the symptoms of type 1 diabetes?", and "How are nerve lesions around diabetic complications treated?".

[0080] Therefore, by creating a mapping relationship between the user label set and the standard questions in the data knowledge base, a convenient condition is provided for quickly retrieving the corresponding knowledge question set (including standard questions, extended questions, and standard answers) from the data knowledge base using the user label set when creating subsequent rule associations.

[0081] Step S140: Obtain user behavior data and obtain user interest parameters based on the user behavior data.

[0082] For example, in at least one embodiment of the Disclosure, user activity data (e.g., user activity logs) can be obtained from software on a client terminal or web server, and user activity data can be collected in a customized manner, and the embodiments of the Disclosure are not limited thereto. For example, user activity data may include all activity data such as access, browsing, and clicking when a user accesses a website, that is, user activity data can provide feedback on the user's specific actions, such as which links the user clicked, which pages they opened, and which search terms they used. For example, in at least one embodiment of the Disclosure, user interest parameters can be obtained by analyzing this user activity data. For example, in one example, a user's feedback behavior may include explicit and implicit feedback behavior. For example, explicit feedback behavior includes the user clearly indicating feedback on an answer, for example, clearly selecting whether the answer is helpful or not. As shown in Figure 3B, in one example, a user submits a question on a specific software platform (e.g., a health management platform) such as "How often should a hypertensive patient monitor their blood pressure daily?", and after providing an answer to the question, the platform asks the user, "Is this answer helpful to you?". Based on the user's actions, such as clicking "yes" or "no," the platform can clearly understand the user's feedback on the answer and reflect their interests and concerns. Implicit feedback is indirect feedback that does not directly respond to the user's preferences, such as the frequency with which the user clicks "view" within a certain time period.For example, in one case, the Maximum Boundary Similarity (MMR) algorithm can be used to summarize health knowledge that a user is reading. The MMR algorithm extracts sentences from the document according to their importance to construct the summary, and the term frequency-inverse document frequency (TF-IDF) method is used to obtain high-frequency words within the summary. These high-frequency words (also called keywords) are important features that reflect the user's interests and concerns.

[0083] For example, in one instance, when a user uses a specific software platform (e.g., a health management platform) within a device for the first time, user behavior data can be obtained by analyzing the application program logs stored in the device. These logs may be logs stored after the device was started up this time, or logs stored after the device was last started up, and the embodiments of this disclosure are not specifically limited thereto and can be adapted to the actual circumstances.

[0084] For example, in one case, the user's clicked questions or frequently viewed words and keywords, which reflect the user's interests and concerns, are converted into user interest parameters. These user interest parameters may be numerical vectors and are used to reflect the user's interests and concerns. For example, the questions the user clicked or the words and phrases the user is interested in are embedded (Word Embedding) to generate embedding vectors of these user interest words and phrases, which constitute the "user interest parameters" mentioned above. Word Embedding can be understood as a mapping relationship, where a word in text space is mapped or embedded in another numerical vector space in a certain way. In other words, Word Embedding can represent vocabulary and complete sentences in vector form.

[0085] Step S150: Based on the user interest parameter and multiple candidate problems, obtain at least one similarity feature between each candidate problem in the multiple candidate problems and the user interest parameter.

[0086] For example, in at least one embodiment of the present disclosure, the step of obtaining at least one similarity feature between each candidate problem in a plurality of candidate problems and the user interest parameter, based on the user interest parameter and a plurality of candidate problems, may include the step of obtaining at least one similarity feature between each candidate problem and the user interest parameter, based on the user interest parameter and a plurality of candidate problems, using at least one similarity matching model.

[0087] For example, in at least one embodiment of the present disclosure, the at least one similarity matching model includes at least one of the following: a cosine similarity model, a Jaccard similarity model, an edit distance (Levenshtein) similarity model, a word movement distance (WMD) similarity model, and a deep semantic matching (DSSM) similarity model.

[0088] For example, in one example, given a user interest parameter A (e.g., a numerical vector), any candidate problem in the candidate set (e.g., a standard problem or an extended problem) is word-embedded to generate an embedding vector B for that candidate problem, and vector B is also a numerical vector. Word Embedding can be understood as a mapping relationship, where a word in text space can be mapped or embedded in another numerical vector space in a certain way; in other words, word Embedding can represent vocabulary and complete sentences in vector form. The user interest parameter A and a candidate problem B are input into several similarity models, and each of the multiple similarity models outputs a similarity feature between numerical vectors A and B, where a larger numerical value for the similarity feature indicates a closer relationship between the words and phrases corresponding to vector A and the words and phrases corresponding to vector B. Embodiments of this disclosure do not limit the number of similarity matching models used; for example, in one example, if five similarity matching models are used, there can be five similarity features between vectors A and B. For example, in one case, if three similarity matching models are used, there can be three similarity features between vectors A and B.

[0089] Below, we will briefly explain some of the similarity matching models proposed above.

[0090] (1) Cosine Similarity: Cosine similarity measures the magnitude of the difference between two individuals using the cosine value of the angle between the vectors. The closer the cosine value is to 1, the more similar the two vectors A and B are. Typically, cosine similarity (also called cosine distance) is calculated using the following formula.

[0091]

number

[0092] Furthermore, the similarity features output by the cosine similarity model are continuous.

[0093] (2) Jaccard distance: The Jaccard distance measures the degree of distinction between two sets using the ratio of all elements in different element stations within the two sets. It is expressed by the following formula, where J(A, B) is the Jaccard similarity coefficient.

[0094]

number

[0095] Furthermore, the similarity features output by the Gekart similarity model are continuous.

[0096] (3) Edit distance, also known as Levenshtein distance, refers to the minimum number of operations required to convert string A to string B using character operations. Allowed character operations include changing one character, inserting one character, and deleting one character. Generally, the smaller the edit distance of two strings, the more similar they are. If two strings are equal, their edit distance is 0. Also, the similarity features output by the edit distance similarity model are continuous.

[0097] (4) Word Movement Distance (WMD) measures the semantic similarity of documents by considering the similarity between two documents from the entire document and finding pairs of the sum of the minimum distances between all words in the two documents. The similarity features output by the WMD similarity model are continuous.

[0098] (5) DSSM model: DSSM is a deep semantic matching model that maps the two matched entities to a low-dimensional space and transforms the correlation problem into a distance of low-dimensional space vectors. This model can be used to predict the semantic similarity of two sentences and can also obtain a low-dimensional semantic vector representation of the sentences. Furthermore, the similarity features output by the DSSM model are discrete.

[0099] For example, in at least one embodiment of the present disclosure, the user interest parameters and multiple candidate problems can use the five similarity matching models described above simultaneously, and for each candidate problem among the multiple candidate problems, five similarity features corresponding to the five similarity matching models can be obtained between the user interest parameters.

[0100] Furthermore, the similarity matching models used in the embodiments of this disclosure are not limited to those described above; other similarity matching models may also be used, as long as they can achieve the same or similar technical effects, that is, as long as they can calculate the similarity between two vectors. The embodiments of this disclosure are not specifically limited in this regard. In addition, the embodiments of this disclosure are not limited in the number of similarity matching models used, and can be provided according to actual needs.

[0101] Step S160: Based on user basic information, multiple candidate problems, and at least one similarity feature, sort the multiple candidate problems to obtain a problem sequence.

[0102] For example, in at least one embodiment of the present disclosure, the step of sorting a plurality of candidate problems to obtain a problem sequence based on user basic information, a plurality of candidate problems, and at least one similarity feature includes using a sorting model to configure the user basic information, a plurality of candidate problems, and at least one similarity feature as input feature vectors of the sorting model, obtaining a score corresponding to each candidate problem in the plurality of candidate problems, and sorting the corresponding plurality of candidate problems by the magnitude of the score (e.g., in descending order of score) to obtain a problem sequence.

[0103] For example, in at least one embodiment of this disclosure, basic information of a user includes that the user is male (e.g., the corresponding discrete feature is "0") and has diabetes (e.g., the corresponding discrete feature is "1"), the Embedding vector corresponding to a candidate problem among several candidate problems is [0.3, 0.5, 0.6], and the similarity features between the candidate problem and the user interest parameter include that the cosine similarity is 0.85, the Jaccard distance is 0.91, the edit distance is 3, the WMD is 1.17, and the DSSM is 2. In such a case, if these are configured as the input feature vector of the sorting model, the vector is [0, 1, 0.3, 0.5, 0.6, 0.85, 0.91, 3, 1.17, 2]. Note that the feature data provided in this embodiment is merely illustrative, and the specific values ​​of the feature data can be set according to experimental results or actual circumstances, and the embodiments of this disclosure do not specifically limit this.

[0104] Here, basic user information, multiple candidate problems, and at least one similarity feature are constructed as the input feature vector of the sorting model, and problems are recommended in a personalized manner, taking into full consideration the individual factors of the user (e.g., basic user information, user behavior data, etc.), thereby enabling the user to more accurately grasp knowledge related to their health.

[0105] For example, in at least one embodiment of this disclosure, the sorting model used is the classic Wide&Deep model submitted by Google. In the above example, the vector [0, 1, 0.3, 0.5, 0.6, 0.85, 0.91, 3, 1.17, 2] is input to the Wide&Deep model as an input feature vector. The core idea of ​​this model is to combine the memory capabilities of a linear model with the generalization capabilities of a deep neural network model to reflect the fusion of correlation and diversity in the recommendation scenario. This model scores multiple candidate problems in a set of candidate problems and sorts the corresponding candidate problems in descending order of score to obtain a problem sequence. Then, depending on the actual needs, up to N problems in the problem sequence are presented to the user, where N is an integer greater than or equal to 1.

[0106] For example, in at least one embodiment of this disclosure, the score of a candidate problem can be expressed as a conditional probability, p(y|x), where y represents a label corresponding to a certain user action, for example, y=1 if the user clicks on the candidate problem, and y=0 if the user does not click on the candidate problem. Where x represents an input feature vector, the input feature vector x includes discrete features of the user basic information described above, a continuous embedding vector of the candidate problem itself, and at least one similarity feature (the similarity feature may be continuous or discrete) between the candidate problem and the user interest parameter. For example, the probability value P(y=1|x) output when y=1 can be used as the final score of the candidate problem, and this sorting model (i.e., Wide&Deep model) can output the score (i.e., probability value) of each candidate problem in the set of candidate problems.

[0107] The principle of the Wide&Deep model will be briefly explained below in conjunction with Figure 4A. Figure 4A is a schematic diagram of the Wide&Deep model provided by at least one embodiment of this disclosure.

[0108] As shown in Figure 4A, the Wide & Deep model consists of two parts: the Wide part (left side of Figure 4A) and the Deep part (right side of Figure 4A). The Wide & Deep model balances the memory and generalization capabilities of the Wide and Deep models. These two submodels require different input features.

[0109] Regarding the Wide part, the Wide model is a generalized linear model (for example, logistic regression), and the formula is as follows:

[0110]

number

[0111] In this structure, x represents the feature vector [x1, x2, x3…], w represents the parameter vector [w1, w2, w3…], b is the bias item, and y is the output label, which, after passing through the sigmoid function, outputs a probability value between 0 and 1. The input features used in this Wide section are discrete features, such as discrete features of user basic information and discrete similarity features.

[0112] Regarding the Deep part, the Deep model is a feedforward neural network. Generally, the input to a deep neural network model is continuous, dense features. Sparse, high-dimensional features need to be converted to low-dimensional, dense features via Embedding (down-dimensional), which are then used as input to the first hidden layer and updated by inverse training based on the final error (loss). The activation function f of the hidden layer typically uses the ReLU function to prevent gradient vanishing. Therefore, the input features used in the Deep part are continuous features, such as the Embedding vector and continuous similarity features of the candidate problem.

[0113] When training a model, the gradient is calculated based on the final error, and this is backpropagated to the two parts, Wide and Deep, to continuously update the parameters of each model and obtain the final model. It is important to note that training the Wide and Deep models simultaneously does not represent the fusion of the models, but rather the weighted sum of the results of the two models becomes the final prediction result.

[0114]

number

[0115] In the above Sigmoid function, W wide is the weight of the wide portion, x is the original feature vector, and φ(x) is the cross feature, for example, the new feature concatenated after one-hot encoding, W Deep is the weight output by the final activation layer of the deep neural network, l represents the hidden layer, f represents the activation function, a represents the input feature, and b is the bias item.

[0116] Model training uses associative training. Compared to integral learning, where a single model is trained independently and the models are fused only in the final prediction stage, associative training involves model fusion during the training phase, and training errors are simultaneously fed back to both the Wide and Deep models to update weights. Thus, the Wide model focuses on cross-multiplication of discrete features, performing nonlinear transformations on the original features to generate memory of how features interact with each other, while the Deep model focuses on generalization. Deep neural networks use low-dimensional dense features and require only a small number of feature processing steps, allowing them to better generalize combinations of features not present in the training samples, thereby improving the model's generalization ability. After model training is complete, it is deployed to the problem recommendation scenario.

[0117] Further information on the Wide&Deep model can be found in other relevant references, and the above description in this disclosure is merely illustrative.

[0118] For example, in one example of the present disclosure, the problem recommendation method 10 uses a sorting (i.e., Wide & Deep model) model to construct user basic information, a plurality of candidate problems, and at least one similarity feature as input feature vectors to the sorting model, and returns one score for each of the plurality of candidate problems included in the problem candidate set. The plurality of candidate problems are sorted according to the magnitude of the score to obtain a problem sequence. For example, in one example, the sorting is in descending order of score, in another example, the sorting is in ascending order of score, and other sorting orders are also possible, and the embodiments of the present disclosure are not specifically limited thereto.

[0119] Step S170: Based on the order of the problem sequence, recommend at least one candidate problem from the problem sequence to the user. For example, in one example, if the problem sequence is sorted in descending order of score, recommend up to N candidate problems from the problem sequence to the user, where N is an integer greater than or equal to 1. For example, select and recommend to the user, for example, the first five problems (which may be any other number) from the problem sequence as the final recommended problem.

[0120] Figure 4B shows a user interface diagram of recommended questions provided to a user by a platform (e.g., a health management platform). For example, in one example, as shown in Figure 4B, before the user enters the question they want to ask into the text input box, the health management platform already recommends five recommended questions (e.g., "Can hypertension be treated with surgery?", "What tests are performed for hypertension?", "What are the symptoms of hypertension?", "How often do hypertensive patients have their blood pressure checked daily?", and "What kind of exercise should hypertensive patients do?") based on the question recommendation method described above. Note that there are various ways in which the final recommended questions are provided to the user, and the embodiments of this disclosure are not specifically limited to these.

[0121] For example, in one embodiment, when a user enters a question they wish to inquire about into an input text box, the platform can match the text information entered by the user with relevant questions in its data knowledge base, for example, using entity recognition and keyword matching methods, and display these questions above the input text box in the user interface. For example, as shown in Figure 4C, when a user enters the three characters "hypertension" into the input text box, the platform can display recommended questions such as "Dietary therapy for hypertension," "Dietary precautions for hypertension," and "Recipes for hypertension during pregnancy" above the input text box. The embodiments of this disclosure are not specifically limited to this and can be adapted according to the actual situation.

[0122] In addition, depending on the actual situation, some of the steps S110-S170 can be selectively executed, and the embodiments of this disclosure are not specifically limited to this.

[0123] For example, in one embodiment, if the same user uses a platform multiple times but their basic information does not change, steps S110-S130 can be omitted, and the user's pre-stored candidate question set can be directly obtained. For example, in one example, when a user uses a platform (e.g., a health management platform) for the first time, the platform has already generated a corresponding candidate question set for that user and stored it in a database associated with the platform. When the user uses the platform a second time, there is no need to repeat steps S110-S130 to obtain the user's candidate question set. The objective of obtaining the user's candidate question set can be achieved by obtaining the candidate question set previously stored in the associated database. The embodiments of this disclosure are not specifically limited to this and can be adjusted according to the actual situation.

[0124] The problem recommendation method 10 provided by at least one embodiment of this disclosure not only effectively avoids the problem of inappropriate feedback responses caused by unclear expressions from patients, but also enables users to more accurately grasp knowledge related to their health by recommending problems in a personalized manner, taking into full consideration the individual factors of the user (e.g., user basic information, user behavior data, etc.), and by fusing multiple features such as user basic information and the outputs of various similarity matching models via a sorting model (i.e., a Wide & Deep model), the problem recommendation becomes correlated, personalized, and diverse, while also emphasizing the final feedback order, thereby achieving an effect that is more suited to the user's needs.

[0125] Figure 5A is an exemplary flowchart of another problem-solving method provided by at least one embodiment of the present disclosure. Figure 5B is a schematic block diagram of the problem-solving method shown in Figure 5A provided by at least one embodiment of the present disclosure. As shown in Figure 5, the problem-solving method 50 includes steps S510-S580. For example, the steps can be executed in the order S510-S580, or in any other adjusted order, for example, step S520 can be executed first and then step S530, or step S530 can be executed first and then step S520. Also, for example, some or all of the operations of steps S510-S580 can be executed in parallel, for example, steps S520 and S530 can be executed in parallel, and the embodiments of the present disclosure do not limit the execution order of each step and can be adjusted according to the actual situation. For example, steps S510-S580 can be implemented on a server or local end, and the embodiments of the present disclosure do not limit this. For example, in some cases, implementing the problem recommendation method 50 may involve selectively performing some of the steps S510-S580, or performing several additional steps other than steps S510-S580, and the embodiments of this disclosure are not specifically limited thereto.

[0126] Referring to Figures 5A and 1B, steps S520-S580 included in the problem recommendation method 50 shown in Figure 5A and steps S110-S170 included in the problem recommendation method 10 shown in Figure 1B are essentially the same. Therefore, the explanation of steps S520-S580 can be found in the related explanation of steps S110-S170 in Figure 1B, and the explanation will be omitted here.

[0127] In comparison with the problem recommendation method 10 shown in Figure 1B, Figure 5A further includes step S510, which is the step of creating a data knowledge base. For example, in one embodiment of the present disclosure, a data set can be created by ingesting a dataset from a network, classifying the dataset according to intent, and forming multiple knowledge problem sets.

[0128] For example, a web crawler can be used to retrieve datasets containing large amounts of data from, for instance, the internet network. A web crawler, also known as a web spider, is a program or script that automatically retrieves information from the World Wide Web according to certain rules.

[0129] For example, in one embodiment of the present disclosure, if the above-described problem recommendation method 50 is applied to a medical smart Q&A scenario, the problem recommendation method 50 is used to recommend a problem related to a disease, and the dataset can be obtained from at least one of the following: a medical interview dataset between doctors and patients on a network, hot questions related to the disease, or prize questions. For example, in one example, a prize question is a question that needs to be inquired about for a fee on a specific website (e.g., Xunyi Wenyao.com, 39 Health.com, etc.), and the embodiments of the present disclosure are not specifically limited thereto.

[0130] For example, the TF-IDF method can be used to extract high-frequency keywords from a dataset, such as "symptoms," "treatment," "blood sugar," "diet," "medication," "tests," "insulin," and "diabetic foot." Since large amounts of data from the dataset can be intentionally classified according to these high-frequency keywords, building a data knowledge base becomes easier. For example, a complete data knowledge base of basic health knowledge for chronic diseases such as diabetes and hypertension can be created by intentionally classifying the dataset using the deep learning algorithm Text-CNN and organizing standard questions and corresponding extended questions and answers under each intent (e.g., by manual organization). For example, in one example, multiple types of intents can be manually determined based on the extracted high-frequency keywords. For example, manual determination can be based on factors such as people's attention level and the frequency of keyword occurrence, and the embodiments of this disclosure are not specifically limited thereto. For example, in one example, the following intent categories: diet, exercise, medication, tests, complications, surgery, treatment, symptoms, etc., can be manually determined, and the embodiments of this disclosure are not specifically limited thereto. Next, the corresponding data (e.g., problems matching that type of intent) is manually organized under each type of intent, used as training data, and the model is trained using the deep learning algorithm Text-CNN. Then, a large amount of data from the dataset is input into the trained model, and the model outputs an intent category corresponding to each data point, thereby enabling intentional classification of large amounts of data. To improve the accuracy of intentional classification, the standard problems and corresponding extended problems and answers can be manually filtered, organized, and supplemented again under each intent. For example, when the problem recommendation method 50 is executed for the first time, the knowledge database is created by executing step S510, and the knowledge database is stored on a server, in memory, or in a database. Subsequently, when the problem recommendation method 50 is executed, step S510 can be omitted, and processing efficiency can be improved by directly accessing the knowledge database.For example, by performing step S510 or updating the knowledge database using other applicable methods, the knowledge database can be updated and optimized so that the candidate problems retrieved in subsequent steps are closer to the user's needs and closer to the current level of societal cognition.

[0131] The technical effects achieved by the problem recommendation method 50 shown in Figure 5A and the technical effects achieved by the problem recommendation method 10, which is explained by combining it with Figure 1B above, are similar and will not be explained here. For explanations of each block diagram shown in Figure 5B, please refer to the detailed explanations of each step in Figures 5A and 1B above and will not be explained here.

[0132] At least one embodiment of the present disclosure further provides a problem recommendation device. Figure 6 is a schematic block diagram of a problem recommendation device provided by at least one embodiment of the present disclosure. As shown in Figure 6, the problem recommendation device 60 includes a set acquisition module, an action analysis module, a feature generation module, a problem sorting module, and a recommendation module, which can be implemented in software, hardware, firmware, or any combination thereof, for example, as a set acquisition circuit 600, an action analysis circuit 640, a feature generation circuit 650, a problem sorting circuit 660, and a recommendation circuit 670, respectively.

[0133] For example, in one example, the set acquisition circuit 600 is configured to acquire a user's candidate problem set containing multiple candidate problems. For example, the set acquisition circuit 600 includes a knowledge base access circuit 610, an information acquisition circuit 620, and a candidate set generation circuit 630. For example, the knowledge base access circuit 610 is configured to access a data knowledge base, which contains multiple knowledge problem sets. For example, the information acquisition circuit 620 is configured to acquire basic user information and create a user label set based on the basic user information. For example, the candidate set generation circuit 630 is configured to associate the user label set with the data knowledge base and acquire a candidate problem set from multiple knowledge problem sets, which contains multiple candidate problems. For example, the behavior analysis circuit 640 is configured to acquire user behavior data and acquire user interest parameters based on the user behavior data. For example, the feature generation circuit 650 is configured to acquire at least one similarity feature between each candidate problem in the multiple candidate problems and the user interest parameter, based on the user interest parameter and the multiple candidate problems. For example, the problem sorting circuit 660 is configured to sort multiple candidate problems and obtain a problem sequence based on user basic information, multiple candidate problems, and at least one similarity feature. For example, the recommendation circuit 670 is configured to recommend at least one candidate problem from the problem sequence to the user based on the order of the problem sequence.

[0134] For example, the knowledge base access circuit 610, information acquisition circuit 620, candidate set generation circuit 630, behavior analysis circuit 640, feature generation circuit 650, problem sorting circuit 660, and recommendation circuit 670 all perform specific operations that can be referenced from the relevant descriptions of problem recommendation methods 10 and 50 provided in at least one embodiment of the present disclosure, which are omitted here.

[0135] For example, in at least one embodiment of the present disclosure, the problem sorting circuit 660 of the problem recommendation device 60 includes a problem sorting subcircuit 661. The problem sorting subcircuit 661 is configured to use a sorting model to configure user basic information, a plurality of candidate problems, and at least one similarity feature as input feature vectors to the sorting model, to obtain a score corresponding to each candidate problem in the plurality of candidate problems, and to sort the corresponding plurality of candidate problems according to the magnitude of the scores to obtain a problem sequence.

[0136] For example, the problem sorting subcircuit 661 is configured such that the specific operations it performs refer to the relevant descriptions of the problem recommendation methods 10 and 50 provided in at least one embodiment of the present disclosure, which are omitted here.

[0137] For example, in at least one embodiment of the present disclosure, the problem recommendation device 60 further includes a knowledge base creation circuit 601. The knowledge base creation circuit 601 is configured to create a data knowledge base that takes in a dataset from a network, classifies the dataset according to intent, and forms a set of knowledge problems.

[0138] For example, the knowledge base creation circuit 601 is configured such that the specific operations to be performed refer to the relevant description of the problem recommendation method 50 provided in at least one embodiment of the present disclosure, which is omitted here.

[0139] For example, in at least one embodiment of the present disclosure, the candidate set generation circuit 630 of the problem recommendation device 60 includes a candidate set generation subcircuit 631. The candidate set generation subcircuit 631 is configured to create a mapping relationship between a user label set and standard problems in a data knowledge base, to match the user label set to standard problems in the data knowledge base, and to construct a candidate problem set with a knowledge problem set corresponding to the matched standard problems.

[0140] For example, the candidate set generation subcircuit 631 is configured such that the specific operations it performs refer to the relevant descriptions of problem recommendation methods 10 and 50 provided in at least one embodiment of this disclosure, which are omitted here.

[0141] For example, in at least one embodiment of the present disclosure, the behavior analysis circuit 640 of the problem recommendation device 60 includes a behavior analysis sub-circuit 641. The behavior analysis sub-circuit 641 is configured to analyze user behavior data and convert the problems clicked by the user, or the words and phrases of interest to the user, into user interest parameters.

[0142] For example, the behavioral analysis subcircuit 641 is configured such that the specific operations to be performed refer to the relevant descriptions of the problem recommendation methods 10 and 50 provided in at least one embodiment of the present disclosure, which are omitted here.

[0143] For example, in at least one embodiment of the present disclosure, the feature generation circuit 650 of the problem recommendation device 60 includes a feature generation subcircuit 651. The feature generation subcircuit 651 is configured to use at least one similarity matching model to obtain at least one similarity feature between each candidate problem and the user interest parameter, based on the user interest parameter and a plurality of candidate problems.

[0144] For example, the feature generation subcircuit 651 is configured such that the specific operations it performs refer to the relevant descriptions of problem recommendation methods 10 and 50 provided in at least one embodiment of this disclosure, which are omitted here.

[0145] Furthermore, the set acquisition circuit 600, knowledge base access circuit 610, information acquisition circuit 620, candidate set generation circuit 630, behavior analysis circuit 640, feature generation circuit 650, problem sorting circuit 660, recommendation circuit 670 and feature generation sub-circuit 651, behavior analysis sub-circuit 641, candidate set generation sub-circuit 631, problem sorting sub-circuit 661 and knowledge base creation circuit 601 in the embodiments of this disclosure can be implemented by hardware such as a processor and controller, software capable of performing related functions, or a combination of both, and the embodiments of this disclosure are not limited to specific implementation methods.

[0146] In the embodiments of this disclosure, the problem recommendation device 60 may include more circuits, but it is not limited to the set acquisition circuit 600, knowledge base access circuit 610, information acquisition circuit 620, candidate set generation circuit 630, behavior analysis circuit 640, feature generation circuit 650, problem sorting circuit 660, recommendation circuit 670 and feature generation sub-circuit 651, behavior analysis sub-circuit 641, candidate set generation sub-circuit 631, problem sorting sub-circuit 661 and knowledge base creation circuit 601 described above. The embodiments of this disclosure may be determined according to actual needs, and are not limited thereto.

[0147] It can be understood that the problem recommendation device 60 provided by the embodiments of this disclosure can implement the above-described problem recommendation methods 10 and 50 and can achieve technical effects similar to those of the above-described problem recommendation methods 10 and 50, and therefore a detailed explanation is omitted here.

[0148] At least one embodiment of the present disclosure further provides a problem recommendation system. Figure 7 is a schematic block diagram of a problem recommendation system provided by at least one embodiment of the present disclosure. As shown in Figure 7, the problem recommendation system 70 includes a terminal 710 and a problem recommendation server 720, the terminal 710 and the problem recommendation server 720 being signal-connected. The terminal 710 is configured to send request data to the problem recommendation server 720. The problem recommendation server 720 is configured to respond to the request data by accessing a database containing multiple knowledge problem sets; retrieving a user's candidate problem set containing multiple candidate problems; retrieving user behavior data; retrieving user interest parameters based on the user behavior data; retrieving at least one similarity feature between each candidate problem in the multiple candidate problems and the user interest parameters based on the user interest parameters and the multiple candidate problems; and sorting the multiple candidate problems based on user basic information and the multiple candidate problems and at least one similarity feature to obtain a problem sequence. For example, the terminal 710 is further configured to display up to N candidate problems in the problem sequence, where N is an integer greater than or equal to 1.

[0149] For example, the problem server 720 may perform the above operations by referring to the problem recommendation methods 10 and 50 provided in at least one embodiment of this disclosure, which are omitted here from description.

[0150] For example, in one example, the terminal 710 included in the problem recommendation system 70 can be implemented as a client (e.g., a mobile phone, a computer, etc.), and the problem recommendation server 720 can be implemented as a server (e.g., a server).

[0151] For example, in one example, as shown in Figure 7, the problem recommendation system 70 may further include a knowledge base server 730 in which a data knowledge base is stored, in addition to the terminal 710 and the problem recommendation server 720. The knowledge base server 730 and the problem recommendation server 720 are signal-connected and configured to return data from the data knowledge base corresponding to the request information from the problem recommendation server 720 to the problem recommendation server 720 in response to the request information from the problem recommendation server 720. If the problem recommendation system 70 does not include the knowledge base server 730, the data in the data knowledge base may be stored directly in the problem recommendation server 720 or in other storage devices provided separately, and the embodiments of this disclosure are not specifically limited thereto.

[0152] The problem recommendation system 70 provided by at least one embodiment of this disclosure can implement the problem recommendation methods 10 and 50 provided by the above embodiment and can achieve similar technical effects to the problem recommendation methods 10 and 50 provided by the above embodiment, which are omitted here from further explanation.

[0153] At least one embodiment of the present disclosure further provides electronic equipment. Figure 8 is a schematic diagram of electronic equipment provided by at least one embodiment of the present disclosure. For example, as shown in Figure 8, the electronic equipment 80 includes a processor 810 and a memory 820. The memory 820 includes one or more computer program modules 821. The one or more computer program modules 821 are stored in the memory 820 and executed by the processor 810, and the one or more computer program modules 821 include instructions for performing any problem-solving method provided by at least one embodiment of the present disclosure, and when executed by the processor 810, can perform one or more steps of the problem-solving method provided by at least one embodiment of the present disclosure. The memory 820 and the processor 810 can be connected to each other via a bus system and / or other form of connection mechanism (not shown).

[0154] For example, the memory 820 and processor 810 can be located at the server end (or on the cloud side), for example, in the problem recommendation server 720 described above, and used to perform one or more steps of the problem recommendation method described in Figures 1A, 1B, and 5A.

[0155] For example, the processor 810 may be a central processing unit (CPU), a digital signal processor (DSP), or another type of processing unit having data processing and / or program execution capabilities, such as a field-programmable gate array (FPGA). For example, the central processing unit (CPU) may be an X86 or ARM architecture. The processor 810 may be a general-purpose processor or a dedicated processor, and can control other components of the electronic device 80 to perform desired functions.

[0156] For example, memory 820 may include any combination of one or more computer program products, and computer program products may include various forms of computer-readable storage media such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory (cache). Non-volatile memory may include, for example, read-only memory (ROM), hard disk, erasable programmable read-only memory (EPROM), portable compact disk read-only memory (CD-ROM), USB memory, flash memory, etc. One or more computer program modules 821 can be stored in the computer-readable storage media, and the processor 810 can execute one or more computer program modules 821 to realize various functions of the electronic device 80. The computer-readable storage media may also store various application programs and various data and various data used and / or generated by application programs. The specific functions and technical effects of the electronic device 80 can be found in the description of the recommended method above and are omitted here.

[0157] Figure 9 is a schematic block diagram of a terminal provided in at least one embodiment of the present disclosure. For example, in at least one embodiment of the present disclosure, the terminal is a display terminal 900, which can be applied, for example, to a problem recommendation method provided in an embodiment of the present disclosure. For example, the display terminal 900 can provide user access logs that reflect user behavior data (for example, access logs recorded by cookies, etc., when an application program such as a browser is executed within the system) and display at least one candidate problem recommended to the user. Note that the terminal shown in Figure 9 is a display terminal 900, which is merely one example and does not in any way limit the functions and scope of use of the embodiments of the present disclosure.

[0158] As shown in Figure 9, the display terminal 900 may include a processing unit (e.g., a central processor, graphics processor, etc.) 910, which performs various appropriate operations and processes based on programs stored in read-only memory (ROM) 920 or programs loaded from storage device 980 into random access memory (RAM) 930. The RAM 930 also stores various programs and data necessary for operating the display terminal 900. The processing unit 910, ROM 920, and RAM 930 are connected to each other via a bus 940. An input / output (I / O) interface 950 is also connected to the bus 940.

[0159] Typically, the following devices can be connected to the I / O interface 950, input devices 960 including, for example, a touch panel, touch tablet, keyboard, mouse, camera, microphone, accelerometer, gyroscope, etc., output devices 970 including, for example, a liquid crystal display (LCD), speaker, vibrator, etc., storage devices 980 including, for example, magnetic tape, hard disk, etc., and communication devices 990. The communication devices 990 can allow the display terminal 900 to communicate with other electronic devices wirelessly or wired to exchange data. Figure 9 shows a display terminal 900 with various devices, but it is not required to implement or have all of the devices shown, and the display terminal 900 may, alternatively, implement or have more or fewer devices.

[0160] At least one embodiment of the present disclosure further provides a non-temporary readable storage medium. Figure 10 is a schematic block diagram of a non-temporary readable storage medium 100 provided by at least one embodiment of the present disclosure. For example, as shown in Figure 10, the non-temporary readable storage medium 100 includes a computer program instruction 111 stored in the non-temporary readable storage medium. When the computer program instruction 111 is executed by the processor, one or more steps of the problem recommendation method 10 or 50 provided by at least one embodiment of the present disclosure are performed.

[0161] For example, the storage medium may be any combination of one or more computer-readable storage mediums, for instance, one computer-readable storage medium containing computer-readable program code for obtaining a user's candidate problem set; another computer-readable storage medium containing computer-readable program code for obtaining user behavior data and user interest parameters based on the user behavior data; yet another computer-readable storage medium containing computer-readable program code for obtaining at least one similarity feature between each candidate problem in the candidate problems and the user interest parameters based on the user interest parameters and the candidate problems; and one computer-readable storage medium containing computer-readable program code for sorting the candidate problems and obtaining a problem sequence based on user basic information, the candidate problems and at least one similarity feature. Of course, each of the above program codes can also be stored in the same computer-readable medium, and embodiments of this disclosure are not limited thereto. For example, when the program code is read by a computer, the computer can execute the program code stored in the computer storage medium, for example, by executing a problem recommendation method provided by any embodiment of this disclosure.

[0162] For example, the storage medium may include a smartphone memory card, a tablet computer storage device, a personal computer hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), portable compact disk read-only memory (CD-ROM), flash memory, or any combination of the above storage media, or other applicable storage media. For example, the readable storage medium may be memory 820 in Figure 8, and related explanations can be found in the above description, which are omitted here.

[0163] Furthermore, the storage medium 100 can be applied to the problem-recommended server 720, and a person skilled in the art may select it based on a specific scenario, but is not limited thereto.

[0164] Figure 11 shows an exemplary scene diagram of a problem recommendation system provided by at least one embodiment of the present disclosure. As shown in Figure 11, the problem recommendation system 300 may include a user terminal 310, a network 320, a server 330, and a database 340.

[0165] For example, the user terminal 310 may be the computer 310-1 or mobile terminal 310-2 shown in Figure 11. The user terminal may also be any other type of electronic device capable of receiving, processing, and displaying data, which may include, but is not limited to, desktop computers, laptop computers, tablet computers, smart home devices, wearable devices, in-vehicle electronic devices, medical electronic devices, etc.

[0166] For example, network 320 may be a single network or a combination of at least two different networks. For example, network 320 may include, but is not limited to, one or more combinations of local area networks, wide area networks, public networks, private networks, the Internet, and mobile communication networks.

[0167] For example, server 330 may be a single server or a group of servers, and each server in the server group may be connected by a wired or wireless network. The wired network may use methods such as twisted pair, coaxial cable, or optical fiber transmission, and the wireless network may use, for example, 3G / 4G / 5G mobile communication networks, Bluetooth®, Zigbee. (Registered trademark)Alternatively, communication methods such as Wi-Fi may be used. This disclosure does not limit the type and functionality of the network. The single server group may be centralized, such as a data center, or distributed. The servers may be local or remote. For example, the server 330 may be a general-purpose server or a dedicated server, or it may be a virtual server or a cloud server, etc.

[0168] For example, the database 340 is used to store various data used, generated, and output from the operation of the user terminal 310 and the server 330. The database 340 can connect to or communicate with the server 330 or a part of the server 330 via the network 320, or it can connect to or communicate with the server 330 directly, or it can connect to or communicate with the server 330 in a combination of the above two ways. In some embodiments, the database 340 may be a standalone device. In another embodiment, the database 340 can be integrated into at least one of the user terminal 310 and the server 340. For example, the database 340 can be located in the user terminal 310 or in the server 340. Also, for example, the database 340 may be distributed, with a part of it located in the user terminal 310 and another part located in the server 340.

[0169] For example, in one example, first, a user terminal 310 (e.g., the user's mobile phone) transmits request data to a server 330 via a network 320 or other technology (e.g., Bluetooth communication, infrared communication, etc.). Next, the server 330 responds to the request data by obtaining a set of candidate problems for the user, which includes multiple candidate problems. Next, the server 330 obtains user behavior data and user interest parameters based on the user behavior data, for example, the user behavior data is transmitted from the user terminal 310 to the server 330 via the network 320. Next, the server 330 obtains at least one similarity feature between each candidate problem in the multiple candidate problems and the user interest parameter based on the user interest parameters and the multiple candidate problems. Next, the server 330 sorts the multiple candidate problems to obtain a problem sequence based on the user basic information, the multiple candidate problems, and at least one similarity feature, and then transmits up to N candidate problems in the display problem sequence to the user terminal 310 via the network 320 or other technology (e.g., Bluetooth communication, infrared communication, etc.). Finally, the user terminal 310 receives and displays up to N candidate problems from the server 330.

[0170] In this disclosure, the term "plural" means two or more unless otherwise specified.

[0171] A person skilled in the art will readily recall other ways of carrying out the disclosure after considering the specification and putting into practice the disclosures disclosed herein. The disclosure covers any variations, uses, or adaptations of the disclosure, and these variations, uses, or adaptations include common sense or customary art means known in the art that are not disclosed herein, in accordance with the general principles of the disclosure. The specification and examples are to be considered merely illustrative, and the true scope and spirit of the disclosure are shown by the following claims.

[0172] This disclosure is not limited to the exact structures described above and shown in the drawings, and can be modified and altered in various ways without departing from its scope. The scope of this disclosure is limited only by the appended claims.

Claims

1. A question recommendation method performed by a processor, Steps include obtaining a user's set of candidate questions, which may include multiple candidate questions, The steps include acquiring user behavior data and acquiring user interest parameters based on the user behavior data, A step of obtaining at least one similarity feature between each candidate question in the plurality of candidate questions and the user interest parameter, based on the user interest parameter and the plurality of candidate questions. A step of sorting the multiple candidate questions based on user basic information, the multiple candidate questions, and the at least one similarity feature to obtain a question sequence, The step of recommending to the user at least one candidate question from the question sequence based on the order of the question sequence, The step of sorting the plurality of candidate questions and obtaining a question sequence based on the user basic information, the plurality of candidate questions, and the at least one similarity feature is: The process includes the steps of using a sorting model to configure the user basic information, the plurality of candidate questions, and the at least one similarity feature as input feature vectors for the sorting model, obtaining a score corresponding to each of the plurality of candidate questions, sorting the plurality of candidate questions according to the magnitude of the scores, and obtaining the question sequence. Among these, the aforementioned sorting model combines the memory capacity of a linear model with the generalization capacity of a deep neural network model. A method for recommending questions characterized by the following features.

2. The step of obtaining at least one similarity feature between each candidate question in the plurality of candidate questions and the user interest parameter, based on the user interest parameter and the plurality of candidate questions, The process includes the step of using at least one similarity matching model to obtain at least one similarity feature between each of the candidate questions and the user interest parameter, based on the user interest parameter and the plurality of candidate questions. The method for recommending questions as described in feature 1.

3. The aforementioned at least one similarity matching model includes at least one of a cosine similarity model, a Gecart similarity model, an edit distance similarity model, a word movement distance similarity model, and a deep semantic matching similarity model. The method for recommending questions as described in feature 2.

4. The aforementioned sorting model includes the Wide & Deep model. The method for recommending questions as described in feature 1.

5. The step of obtaining the user's suggested question set is: Steps include accessing a data knowledge base containing multiple knowledge question sets, The steps include obtaining the user basic information and creating a user label set based on the user basic information, The steps include associating the user label set with the data knowledge base and obtaining the candidate question set from the plurality of knowledge question sets, A method for recommending questions according to any one of features 1 to 4.

6. The user label set includes a multilevel label set containing multilevel labels, where labels of different levels are of different types. The method for recommending questions as described in feature 5.

7. The aforementioned question recommendation method is used to recommend questions related to illness. The aforementioned multi-level label set includes, in which the first level label is an age range, the second level label is a time period, the third level label is a type of disease, and the fourth level label is a complication. The method for recommending questions as described in feature 6.

8. Each of the aforementioned sets of knowledge questions is: A standard question, a standard answer corresponding to the standard question, and an extended question corresponding to the standard question, A method for recommending questions according to any one of features 5 to 7.

9. The step of obtaining the user's suggested question set is: The step further includes creating the aforementioned data knowledge base, The method for recommending questions according to any one of features 5 to 8.

10. The step of creating the aforementioned data knowledge base is: The steps include: ingesting a dataset from a network, classifying the dataset according to intent, forming the plurality of knowledge question sets, and creating the data knowledge base; The question recommendation method described in feature 9.

11. The aforementioned question recommendation method is used to recommend questions related to illness. The dataset consists of at least one of the following: a patient interview dataset between a doctor and a patient; hot questions related to the disease; and prize questions related to the disease. The method for recommending questions as described in feature 10.

12. The step of associating the user label set with the data knowledge base and obtaining a candidate question set from the multiple knowledge question sets is: The steps include creating a mapping relationship between the user label set and the standard questions in the data knowledge base, matching the user label set to the standard questions in the data knowledge base, and constructing the candidate question set with the knowledge question set corresponding to the matched standard questions. The method for recommending questions as described in feature 8.

13. The step of obtaining the user's suggested question set is: The step includes obtaining a pre-stored set of candidate questions for the user, A method for recommending questions according to any one of features 1 to 4.

14. The step of obtaining the user interest parameters based on the user behavior data is: The process includes analyzing the user behavior data and converting the questions clicked by the user, or the words and phrases that the user is interested in, into user interest parameters. A method for recommending questions according to any one of features 1 to 13.

15. A device that recommends asking questions, A set acquisition circuit configured to acquire a user's set of candidate questions, which includes multiple candidate questions, An action analysis circuit configured to acquire user action data and user interest parameters based on the user action data, A feature generation circuit is configured to obtain at least one similarity feature between each candidate question in the plurality of candidate questions and the user interest parameter, based on the user interest parameter and the plurality of candidate questions. A question sorting circuit configured to sort the plurality of candidate questions and obtain a question sequence based on user basic information, the plurality of candidate questions, and the at least one similarity feature, A recommendation circuit is configured to recommend to the user at least one candidate question from the question sequence based on the order of the question sequence, The aforementioned question sorting circuit is, The system includes a question sorting subcircuit configured to use a sorting model to configure the user basic information, the plurality of candidate questions, and the at least one similarity feature as input feature vectors to the sorting model, obtain a score corresponding to each of the plurality of candidate questions, sort the plurality of candidate questions according to the magnitude of the score, and obtain the question sequence. Among these, the aforementioned sorting model combines the memory capacity of a linear model with the generalization capacity of a deep neural network model. A question recommendation device characterized by the following features.

16. The aforementioned set acquisition circuit is, A knowledge base access circuit configured to access a database knowledge base containing multiple knowledge question sets, An information acquisition circuit configured to acquire basic user information and create a user label set based on the basic user information, The system includes a candidate set generation circuit configured to associate the user label set with the data knowledge base and to obtain a candidate question set containing multiple candidate questions from the multiple knowledge question sets, The question recommendation device according to feature 15.

17. The aforementioned set acquisition circuit is, The system further includes a knowledge base creation circuit configured to take a dataset from a network, classify the dataset according to intent, form the plurality of knowledge question sets, and create the data knowledge base. The question recommendation device according to feature 16.

18. The aforementioned candidate set generation circuit is: The candidate set generation subcircuit is configured to create a mapping relationship between the user label set and the standard questions in the data knowledge base, match the user label set to the standard questions in the data knowledge base, and construct the candidate question set with the knowledge question set corresponding to the matched standard questions, The question recommendation device according to feature 16 or 17.

19. The aforementioned behavioral analysis circuit is, Includes an activity analysis subcircuit configured to analyze the user activity data and convert the questions clicked by the user, or the words and phrases of interest to the user, into user interest parameters. A question recommendation device according to any one of claims 15 to 18.

20. The feature generation circuit is, The feature generation subcircuit is configured to use at least one similarity matching model to obtain at least one similarity feature between each of the candidate questions and the user interest parameter, based on the user interest parameter and the plurality of candidate questions. A question recommendation device according to any one of claims 15 to 19.

21. It is a question recommendation system, Including terminals and recommended question servers, The terminal is configured to send the request data to the query recommendation server. The server recommended for the aforementioned question is: In response to the aforementioned request data, Retrieve the user's suggested question set, which includes multiple suggested questions. Obtain user behavior data, and obtain user interest parameters based on the said user behavior data. Based on the user interest parameter and the plurality of candidate questions, at least one similarity feature is obtained between each candidate question in the plurality of candidate questions and the user interest parameter. The system is configured to sort the multiple candidate questions and obtain a question sequence based on user basic information, the multiple candidate questions, and the at least one similarity feature, The terminal is further configured to display up to N candidate questions in the question sequence, where N is an integer of 1 or more. The step of sorting the plurality of candidate questions and obtaining a question sequence based on the user basic information, the plurality of candidate questions, and the at least one similarity feature is: The process includes the steps of using a sorting model to configure the user basic information, the plurality of candidate questions, and the at least one similarity feature as input feature vectors for the sorting model, obtaining a score corresponding to each of the plurality of candidate questions, sorting the plurality of candidate questions according to the magnitude of the scores, and obtaining the question sequence. Among these, the aforementioned sorting model combines the memory capacity of a linear model with the generalization capacity of a deep neural network model. A question recommendation system characterized by the following features.

22. It is an electronic device, Processor and A memory containing one or more computer program modules, The one or more computer program modules are stored in the memory, The processor is configured to be executed by the one or more computers The program module includes instructions for performing the question recommendation method described in any one of claims 1 to 14. An electronic device characterized by the following features.

23. A non-temporary, readable storage medium on which computer instructions are stored, When the computer instruction is executed by the processor, the question recommendation method described in any of claims 1 to 14 is performed. A non-temporary, readable storage medium characterized by the following features.

Images