Data processing method, device and computer-readable storage medium

Abstract

This application proposes a data processing method, device, and computer-readable storage medium, wherein the method includes: obtaining a user's status information, the status information including one or more initial status description data; determining first query information based on the status information, classification decision path information, and status frequency distribution information, wherein the first query information includes one or more recommended status description data; obtaining the user's response operation to the first query information, and determining the query result of the first query information based on the response operation; and determining the user's target classification result based on the status information and the query result. This application can be applied to various scenarios such as cloud technology, artificial intelligence, and smart healthcare, and can improve the processing efficiency of status data and enhance the accuracy and intelligence of classification.

Description

Technical Field

[0001] The present application relates to the field of computer technology, and in particular to a data processing method, device, and computer-readable storage medium. Background Art

[0002] With the continuous development and application of computer technology, data processing technologies are increasingly required in various scenarios, such as recommending user queries and classifying targets based on their status information. Currently, query recommendations often rely on questionnaire templates based on user status, while target classification often relies on manual inspection methods, which classify users based on the inspector's prior experience and subjective judgment, leading to potential misjudgments. These methods suffer from low efficiency in processing user status information, low target classification accuracy, and insufficient intelligence. Summary of the Invention

[0003] The present application provides a data processing method, device and computer-readable storage medium, which can improve the processing efficiency of user status information and enhance the accuracy and intelligence of target classification.

[0004] The present application provides a data processing method, the method comprising: obtaining user status information, the status information comprising one or more initial status description data;

[0005] Determining first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data;

[0006] Obtaining a response operation of the user to the first inquiry information, and determining an inquiry result of the first inquiry information according to the response operation;

[0007] The target classification result of the user is determined based on the status information and the query result.

[0008] The present application provides a data processing device, which includes:

[0009] An acquisition module is used to acquire user status information, wherein the status information includes one or more initial status description data;

[0010] a processing module, configured to determine first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data;

[0011] The processing module is further configured to obtain a response operation of the user to the first inquiry information, and determine an inquiry result of the first inquiry information according to the response operation;

[0012] The classification module is used to determine the target classification result of the user according to the status information and the query result.

[0013] The present application provides a computer device, comprising: a memory and a processor, wherein a data processing program is stored in the memory, and when the data processing program is executed by the processor, the data processing method is implemented.

[0014] The present application provides a computer-readable storage medium, wherein the computer storage medium stores a computer program, wherein the computer program includes program instructions, and the program instructions are executed by a processor to execute the data processing method.

[0015] The present application provides a computer program product, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, so that the computer device performs the above-mentioned data processing method.

[0016] After obtaining the user's status information, this application determines the recommended query information based on the classification decision path information and the status frequency distribution information, making the query information recommendation more accurate and intelligent. During classification, the terminal device automatically classifies the target user without human intervention, which can improve classification efficiency. Automatic recognition is not affected by subjective factors, which can improve the accuracy of target classification. By combining status information and query results to classify the target, the accuracy of target classification is further improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, for ordinary technicians in this field, other drawings can be obtained based on these drawings without any creative work.

[0018] Figure 1 is a schematic diagram of the architecture of a data processing system provided by an exemplary embodiment of the present application;

[0019] Figure 2 is a flowchart of a data processing method provided by an exemplary embodiment of the present application;

[0020] Figure 3 is a schematic diagram of a user interface of a data processing system provided by an exemplary embodiment of the present application;

[0021] Figure 4 This is a structural diagram of classification decision path information provided by an exemplary embodiment of the present application;

[0022] Figure 5 is a schematic diagram of state frequency distribution information provided by an exemplary embodiment of the present application;

[0023] Figure 6 is a flowchart of another data processing method provided by an exemplary embodiment of the present application;

[0024] Figure 7 This is a flowchart of a state classification provided by an exemplary embodiment of the present application;

[0025] Figure 8 This is a schematic diagram of the structure and workflow of a state classification model provided by an exemplary embodiment of the present application;

[0026] Figure 9 is a schematic block diagram of a data processing device provided by an exemplary embodiment of the present application;

[0027] Figure 10 It is a schematic block diagram of a computer device provided by an exemplary embodiment of the present application. DETAILED DESCRIPTION

[0028] The following will be combined with the drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of the embodiments. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of this application.

[0029] It should be noted that the terms "first" and "second" in the embodiments of this application are for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a technical feature designated as "first" or "second" may explicitly or implicitly include at least one such feature.

[0030] Artificial Intelligence (AI) refers to the theories, methods, technologies, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, perceive the environment, acquire knowledge, and use that knowledge to achieve optimal results. In other words, AI is a comprehensive technology within computer science that attempts to understand the essence of intelligence and produce new intelligent machines that can respond in a manner similar to human intelligence. AI also studies the design principles and implementation methods of various intelligent machines, enabling them to perceive, reason, and make decisions. AI technology is an interdisciplinary discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Basic AI technologies generally include sensors, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, machine learning, and deep learning. The solutions provided in the embodiments of this application involve natural language processing and machine learning, which are subcategories of AI technology. These two technologies are described below.

[0031] Natural language processing (NLP) is an important field in the fields of computer science and artificial intelligence. It studies various theories and methods that can enable effective communication between humans and computers using natural language. Natural language processing is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field will involve natural language, that is, the language people use in daily life, so it is closely related to the study of linguistics. Natural language processing technology generally includes text processing, semantic understanding, machine translation, robot question answering, knowledge graphs and other technologies. This application mainly relates to text processing technology in natural language processing technology. Specifically, the terminal device determines the user's standardized status information (that is, the initial status description data) by identifying the user's input information (including the user's personalized description of the status). Subsequently, operations such as query information recommendation and target classification can be performed based on the user's initial status description data to improve classification efficiency and accuracy.

[0032] Machine Learning (ML) is a multi-disciplinary interdisciplinary subject involving probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory and other disciplines. Machine learning specializes in how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning and other technologies. The present application mainly relates to artificial neural networks in machine learning technology. Specifically, the terminal device automatically recommends query information for the collected user status information through the artificial neural network, thereby obtaining a variety of user status information, and classifying targets based on the user's status information and query results and other data, and can automatically generate electronic classification reports or query reports, making target classification more intelligent and further improving the accuracy of classification.

[0033] With the advancement of AI research and technology, AI is being studied and applied in a variety of fields, such as smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, driverless cars, autonomous driving, drones, robots, smart healthcare, smart customer service, connected vehicles, autonomous driving, and smart transportation. As technology develops, AI will be applied in even more fields and play an increasingly important role.

[0034] When applied to scenarios such as intelligent diagnosis and treatment and intelligent inquiry, this application can effectively address bottlenecks in accompanying condition recommendation and target classification accuracy. In intelligent medical diagnosis, given a user's chief complaint, an interactive system simulating a doctor's consultation typically engages the user in interactive inquiries about related accompanying conditions, with the goal of classifying the condition. Ultimately, a deep machine learning model determines the user's target classification result based on the interaction content. Currently, most applications with diagnostic functionality fail to effectively integrate differential diagnosis into user interactions. Specifically, when simulating the consultation process, inquiries are focused on questions related to distinguishing suspicious conditions. Furthermore, when summarizing information after the consultation for diagnosis, i.e., target classification tasks, existing applications often rely on traditional machine learning classification models for classification. Deep machine learning, due to its powerful feature encoding capabilities, has been widely used in various scenarios across general fields. However, due to the serious nature of the medical field, the probabilities of models based on deep machine learning models do not truly reflect the model's certainty in target classification, reducing the model's credibility and thus limiting its application in internet hospitals. To introduce differential diagnosis functionality, the present invention utilizes a decision tree model to generate a decision tree, which serves as part of the basis for user interaction. In order to take into account both model credibility and differential diagnosis functions, the present invention proposes a Bayesian neural network based on MC-Dropout. On the one hand, this network enhances feature representation by encoding inputs through multiple nonlinear layers. On the other hand, MC-Dropout is applied to model parameters to increase uncertainty in the network. Thus, while outputting the probability of target classification, the corresponding uncertainty is additionally output, which adds a credibility measure to the model's prediction results.

[0035] This application can be well applied to health status assessment and intelligent inquiry scenarios. The health status assessment process usually starts with the user providing basic information (that is, the user's main complaint status). After the model learns the user's main complaint status, it starts multiple rounds of interaction with the user. Combined with the user's answer to the inquiry, it conducts inquiries and information collection on the relevant status. By integrating the relevant collected information, it can predict and inform the user's health status and provide medication or examination suggestions. The method proposed in the present invention can be seamlessly embedded in applications with auxiliary diagnosis and intelligent inquiry. For example, the AI inquiry product is mainly used to collect the user's status information. When the user completes the registration and waits for consultation offline, by accessing the AI inquiry applet, the user enters the main complaint status, gender, age, waiting status and other information. The product begins to inquire about the user's current status, past status, living habits, allergy history and other different ranges of questions in turn. Finally, all the information is summarized to automatically generate an inquiry report and push it to the diagnostician, so that the diagnostician can understand the user's status in advance and improve the efficiency of diagnosis. Among them, the current status involves multiple attributes of the main complaint status, including but not limited to location, characteristics, frequency, accompanying status, etc. The method proposed in the present invention can be used to assist in collecting user information along with status recommendations, and to classify user status based on all collected user information.

[0036] This application can also be applied to various scenarios such as cloud technology, artificial intelligence, and smart healthcare, performing state-based recommendation and target classification. In the cloud technology field, this application can store user query information, target classification results, and other data in intelligent inquiries on a cloud server. In addition to user status information, user query information can also include various data such as user personal information and initial / repeat inquiries, facilitating personalized inquiry services for different users. When a user's historical query information is needed, it can be directly obtained from the cloud server with the user's consent. In the field of artificial intelligence, with the user's consent, relevant research can be conducted on user query information to obtain better target classification strategies. It can also develop more intelligent application services based on the technology provided by this application. In the field of smart healthcare, this application can help diagnosticians conduct "pre-diagnosis" of users and provide them with a "pre-diagnosis report." Before the diagnostician meets the user, the diagnostician can obtain some basic information and conditions of the user in advance to assist in diagnosis and treatment. A similar "pre-diagnosis" process can also be reused in the user's follow-up visit scenario. By reminding users to follow doctor's orders and return for follow-up visits in a timely manner, and at the same time reminding the diagnostician to follow up on the user's condition, the diagnosis and treatment plan can be adjusted in a timely manner based on the follow-up results.

[0037] This application will be specifically described by the following examples:

[0038] See also Figure 1 , Figure 1 This is a schematic diagram of the architecture of a data processing system provided by an exemplary embodiment of the present application. Figure 1 As shown, the data processing system may specifically include a terminal device 101 and a server 102, and the terminal device 101 and the server 102 are connected via a network, for example, via a wireless network connection. Based on the data processing method proposed in this application, the terminal device 101 may collect the user's status information, and perform query information recommendation and target classification operations (the target operation is performed based on the user status information collected through multiple rounds of query interactions), and during the processing, the collected user status information, target classification results and other data are sent to the server 102 to facilitate subsequent management by the server 102; the server 102 may also perform query information recommendation and target classification operations. When the server 102 executes, the terminal device 101 may collect the user's input information and send the information to the server 102 for query information recommendation and target classification operations. The server returns the target classification results and other processing results to the terminal device 101, and then performs subsequent operations.

[0039] Specifically, the terminal device 101 can obtain the user's status information, wherein the status information includes one or more initial status description data; the terminal device 101 can send the user's status information to the server 102; the server 102 determines the first query information based on the user's status information, classification decision path information and status frequency distribution information, wherein the first query information includes one or more recommended status description data; the server 102 returns the first query information to the terminal device 101, and the terminal device 101 displays the first query information on the user interface; the terminal device 101 obtains the user's response operation to the first query information, and determines the query result of the first query information (that is, the result selected by the user) based on the response operation; the terminal device 101 sends the query result to the server 102, and the server 102 determines the user's target classification result based on the user's status information and the query result, and sends the target classification result to the terminal device 101; the terminal device 101 displays the target classification result on the user interface based on the received target classification result.

[0040] The terminal device 101 is also referred to as a terminal, user equipment (UE), access terminal, subscriber unit, mobile device, user terminal, wireless communication device, user agent, or user apparatus. The terminal device may be, but is not limited to, a smart home appliance, a handheld device with wireless communication capabilities (e.g., a smartphone or tablet), a computing device (e.g., a personal computer (PC)), an in-vehicle terminal, an intelligent voice interaction device, a wearable device, or other smart device.

[0041] Server 102 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers. It can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, content delivery networks (CDNs), and big data and artificial intelligence platforms.

[0042] It is understood that the system architecture diagram described in the embodiment of the present application is to more clearly illustrate the technical solution of the embodiment of the present application, and does not constitute a limitation on the technical solution provided by the embodiment of the present application. Figure 1 In addition to the three devices shown in FIG, more than three devices may be included; similarly, the server 102 includes Figure 1 In addition to the one server shown in , it can also be composed of multiple servers (that is, a server cluster). Those skilled in the art will know that with the evolution of system architecture and the emergence of new business scenarios, the technical solutions provided in the embodiments of this application are also applicable to similar technical problems.

[0043] See also Figure 2 , Figure 2 This is a flow chart of a data processing method provided by an exemplary embodiment of the present application, in which the method is applied to Figure 1 Taking the terminal device (the above-mentioned terminal device 101) as an example, the method may include the following steps:

[0044] S201: Acquire user status information, where the status information includes one or more initial status description data.

[0045] Specifically, the user to be detected is the target for target classification, and the user's status information is standard status data (i.e., initial status description data) derived from the original information provided by the user (i.e., status description information). This status data may include one or more initial status description data. This step is to obtain the original data for data processing operations. Subsequent steps are based on the status information obtained in this step.

[0046] In one embodiment, the status description information input by the user may be acquired through the terminal device, and the status information may be further obtained based on the status description information.

[0047] In one embodiment, the status information may include one or more initial status description data. For example, an initial status description data (for example, status A) input by the user on the terminal device may be used as the user's status information; or multiple initial status description data (for example, status A, status B, status C) input by the user may be used as the user's status information.

[0048] In one embodiment, the status information may include one or more of the user's behavioral status data, mood status information, and health status data.

[0049] In one embodiment, the status information can be determined based on the original information input by the user. The original input information of the user is usually non-standard and more colloquial, and it is impossible to directly apply the data for classification processing. Therefore, it is necessary to perform terminology standardization on the original information to obtain a unified grammatical expression result (that is, the user's status information) to facilitate subsequent analysis and processing. For example, when the user enters the main complaint status (for example: "the nose is very blocked, afraid of the cold"), the processing result (for example: nasal congestion, fear of the cold) can be obtained through terminology standardization. The data can then be used for subsequent processing to improve the accuracy of data analysis and classification.

[0050] In one embodiment, see Figure 3 After starting the inquiry, the user can input their own main complaint status. The terminal device can output the inquiry information based on the user's input information. The user can conduct multiple rounds of interaction by inputting information, selecting recommended status description data, etc. After the inquiry is completed, the terminal device can automatically generate an inquiry report to display the interactive content for easy viewing by the user. Figure 3 As shown in Figure c, Figure c includes an inquiry report 304, a modification function control 305, and a confirmation function control 306. The inquiry report 304 may include user information (e.g., female, 20 years old), user self-description (e.g., status A), inquiry time (e.g., 2021-8-30, 08:3), initial / repeated inquiry (e.g., initial), main complaint (e.g., status A has occurred for a week), current status (e.g., the user experienced status A a week ago, and also experienced status B, status C, status D, and status E), previous status (e.g., previously denied category A, denied category B, denied category C), other information (e.g., the user insists on exercising every day), classification results (e.g., category 1), etc. The modification function control 305 is used for the user to modify the content of the inquiry report 304; the confirmation function control 306 is used for the user to confirm the correctness of the content of the inquiry report 304. After the user triggers the confirmation function control 306 and obtains the user's consent, the terminal device can upload the inquiry report 304 to the user database to facilitate the user's next inquiry.

[0051] After obtaining the original information input by the user, the status information can be determined by the following methods:

[0052] (1) Display the user interface, which includes an information input area.

[0053] Specifically, multiple function options can be set to trigger the display of the user interface. When a trigger signal for the function option is received from the user, the user interface is displayed. The user interface includes an information input area 301, which can be used to receive the status description information input by the user. Figure 3 As shown in Figure a, after multiple rounds of interaction, the terminal device displays the inquiry information (for example, "How long has state A occurred?"), and the user can select the result through gestures and other operations in the information input area 301 according to their actual situation, and click the function control (for example, the "Confirm" button) to confirm the operation. Figure 3 As shown in Figure b, the user can also directly click on the multiple selection buttons provided by the terminal device to select the result. It should be noted that for the multiple selection buttons provided by the terminal device, the user can perform single-selection operations (for example, select "state C") or multiple-selection operations (for example, select "state D, state E"). In addition, the user interface can also include a conversation display area 302, which can be used to display the state description information entered by the user, as well as subsequent interactive data such as inquiry information, and can also include user guidance information to guide the user to perform related operations, etc. Figure 3 As shown in Figure a, the terminal device can display the inquiry process (e.g., "Is this your first inquiry?", "Yes") in the conversation display area 302. In addition, the conversation display area 302 also includes a modification control 303, which allows the user to modify the last interaction data. This prevents user input errors due to human factors and ensures the accuracy of status information.

[0054] In one embodiment, the user may also trigger the display of the user interface through voice, gestures, etc.

[0055] (2) Obtain the status description information entered by the user through the information input area.

[0056] (3) Call the terminology standardization model to process the state description information and obtain one or more initial state description data corresponding to the user.

[0057] In one embodiment, a standard term set and a learning model are stored in a term standardization model. The learning model is used to obtain the expression vector of the state description information and each term in the standard term set. By calculating the similarity between the state description information and the expression vector of each standard term, the standard term with a similarity higher than a similarity threshold or the highest similarity ranking is used as the term standardization processing result of the state description information (that is, the state information).

[0058] In one embodiment, a keyword extraction operation can be performed on the state description information through a terminology standardization model (for example, determining state keywords, affirmative / negative keywords, tone keywords, etc.), and then the extracted keywords can be analyzed through the terminology standardization model, and the initial state description data corresponding to the state description information can be determined in combination with a pre-set set of standard terms.

[0059] (4) Generate the user's status information based on one or more initial status description data.

[0060] S202 : Determine first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data.

[0061] Specifically, after obtaining the user's status information, the terminal device can determine the first query information based on the status information, classification decision path information, and status frequency distribution information. The first query information may include one or more recommended status description data, and the user can select the recommended status description data that matches their status from the one or more recommended status description data. This first query information is the information that the terminal device first queries the user. This step improves the correlation between the first query information and the status information, and improves the accuracy of subsequent target classification.

[0062] In one embodiment, after the terminal device obtains the initial state description data (for example: state A) input by the user, it can output query information related to the user state information, and the query information can include multiple recommended state description data (for example: state B, state C, state D, state E).

[0063] The above step S202 may include the following steps:

[0064] (1) Call the state classification model to classify the state information and obtain the user's first classification result.

[0065] Specifically, the state classification model is called to classify the state information, and the matching probability between the state information and multiple state category labels can be obtained. The first classification result of the state information is determined based on the matching probability. The first classification result is the result of the terminal device's first classification of the user.

[0066] In one embodiment, when the state classification model is called to classify the state information, the user's target classification probability and the confidence of the target classification can be obtained; then it is determined whether the first classification result or the number of classifications meets the preset conditions. If the preset conditions are met, the first classification result is used as the user's target classification result.

[0067] In one embodiment, the preset condition may be that the target classification probability is greater than a probability threshold, and the confidence is greater than a confidence threshold; the preset condition may also be that the number of classifications is greater than a number threshold.

[0068] (2) Determine the first query information based on the first classification result, the classification decision path information, and the state frequency distribution information.

[0069] A recommended state set is obtained based on the first classification result, classification decision path information, and state frequency distribution information, and query information is determined based on the recommended state set, so that the query information recommendation of this application takes into account the state classification function to a certain extent.

[0070] This step may also include the following steps:

[0071] 1) According to the first classification result, the classification decision path information is queried to obtain a first data set, which includes one or more candidate state description data.

[0072] Specifically, the classification decision path information includes the candidate state description data of each category in multiple categories and the conditions for the occurrence of the candidate state description data. By querying the classification decision path information through the first classification result, the first data set (that is, the data set of the existence state corresponding to the first classification result) can be obtained.

[0073] In one embodiment, if Figure 4 As shown, the classification decision path information includes multiple classification results (e.g., classification 1, classification 2, classification 3, classification 4, classification 5, classification 6, classification 7, classification 8, and classification 9) and the conditional probabilities of the occurrence states corresponding to the corresponding classification results. The multiple states corresponding to the first classification result (e.g., classification 3) can be queried in the state frequency distribution information, and the multiple states corresponding to the first classification result are used as the first data set.

[0074] In one embodiment, a classification decision path can be derived using a decision tree model. A decision tree is a predictive model that represents the mapping between object attributes and object values. A decision tree includes decision nodes, solution branches (several branches extending from a node, each representing a solution), state nodes (representing the expected values of different solutions), probability branches (branches extending from a state node, including the content and probability of that state), and outcome nodes (the gain or loss achieved by each solution in each state).

[0075] In one embodiment, the more layers a decision tree model has, the better it is. The more layers the model has, the more likely it is that the model will include some irrelevant details in the data into its learning, resulting in the model performing very well in the prediction results on the training set, but performing poorly on the test set, i.e., overfitting. Therefore, the decision tree algorithm can be processed through pruning operations. The pruning operation can use a pre-pruning operation to limit the growth of the tree by pre-setting the maximum depth of the generated tree, so that the decision tree model can obtain better prediction results. Specifically, pre-pruning is to stop the growth of the tree in advance during the process of generating the decision tree. The core idea is to calculate whether the current partitioning can bring about an improvement in the generalization ability of the model before expanding the nodes in the tree. If not, the subtree will no longer continue to grow. For example, for example, the depth of the specified tree (that is, Figure 4 The maximum number of layers in different states shown in the figure is 5, so the height of the trained decision tree is 5. Pre-pruning mainly establishes a stopping rule to limit the growth of the decision tree, reducing the risk of overfitting and also reducing the time of tree building. Pre-pruning has the following methods to determine when to stop the growth of the decision tree:

[0076] ① When the tree reaches a certain depth, stop the growth of the tree (that is, set the depth of the tree).

[0077] ② When the number of samples reaching the current node is less than a certain threshold (that is, no further division is needed), stop the tree growth.

[0078] ③ Calculate the accuracy improvement of the test set for each split. When it is less than a certain threshold (that is, the current split has little effect on improving the accuracy), no further expansion is carried out.

[0079] 2) Querying the state frequency distribution information according to the first classification result to obtain a second data set, the second data set including one or more candidate state description data.

[0080] Specifically, the state frequency distribution information includes multiple states of each category in multiple categories, and the multiple states are sorted according to the frequency of occurrence. By querying the state frequency distribution information through the first classification result, a second data set (that is, a data set of the existence states corresponding to the first classification result) can be obtained.

[0081] The state frequency distribution information includes the frequency distribution of the state description data corresponding to each category in a plurality of categories. In one embodiment, Figure 5As shown, the state frequency distribution information can be composed of two sets: classification and term frequency-inverse document frequency (TF-IDF). The classification set contains multiple classification results (e.g., classification 1, classification 2, classification 3, and classification 4), and the TF-IDF set contains the states corresponding to the multiple classification results. Through the first classification result (e.g., classification 3), the state frequency distribution information can be used to query the multiple states corresponding to the first classification result (e.g., state C, state I, state J, state F, state G, state H, state D, state E, state A, and state B). The multiple states corresponding to the first classification result are used as the second data set.

[0082] In one embodiment, the frequency of states corresponding to a given classification in a state dataset and the inverse document frequency of the given state in the state dataset can be calculated, multiplied together, and sorted in descending order to obtain a TF-IDF set. The given classification and TF-IDF are then mapped to obtain state frequency distribution information. The state dataset can be state data collected from a large number of users with their consent, or training data used for model training.

[0083] 3) Obtaining first query information by fusing the first data set and the second data set.

[0084] In one embodiment, an intersection extraction operation is performed on a first data set (e.g., state A, state C, state D) and a second data set (e.g., state C, state I, state J, state F, state G, state H, state D, state E, state A, state B) to obtain common data (e.g., state A, state C, state D), and then the common data is sorted (e.g., in descending order) according to the TF-IDF set corresponding to the classification in the state frequency distribution information, and the first query information is determined based on the sorted common data (e.g., state C, state D, state A).

[0085] In one embodiment, after sorting the shared data, a first query state set (i.e., one or more recommended state description data, such as state C and state A) is obtained, and a first query message is determined based on the first query state set. A selection condition can be set (e.g., the single recommended state description data with the highest probability or the top K recommended state description data), and the first query message (e.g., "Do you have state C?") is generated based on the recommended state description data that meets the condition.

[0086] S203: Obtain a response operation of the user to the first inquiry information, and determine an inquiry result of the first inquiry information according to the response operation.

[0087] Specifically, when the terminal device determines and outputs the first inquiry information, the user can select matching recommended state description data from one or more recommended state description data included in the first inquiry information according to their actual situation. When the user operates on the matching recommended state description data, the terminal device can obtain the user's response operation (for example, gesture operation, trigger operation, voice command, etc.) and determine the inquiry result based on the user's response operation. For example, when a part of the state description data (for example, state A, state C, state D) in the recommended state description data (for example, state A, state C) is selected, the selected recommended state description data is determined as the inquiry result.

[0088] S204: Determine the user's target classification result based on the status information and the query result.

[0089] Specifically, after obtaining the user's status information and query results in steps S201 to S203, the status classification model can be used to perform feature extraction on the status information and query results (that is, the model maps the status information and query results to a low-dimensional vector and enhances the feature representation capability through multiple nonlinear layers) to obtain feature representation information. Then, the feature representation information is classified to obtain the user's target classification result.

[0090] The specific implementation of step S204 may include the following steps:

[0091] (1) Call the state classification model to perform feature extraction on the state information and query results to obtain the user's feature representation information.

[0092] Specifically, the state information includes one or more initial state descriptions, and the query results include one or more recommended state descriptions. Both the initial state descriptions and the recommended state descriptions can be subjected to feature extraction using the state classification model to obtain corresponding feature representation information. Specifically, the user's feature representation information can be a feature vector representing the position of the corresponding user node in the state classification model network and its connection relationship with other nodes.

[0093] The specific implementation method of obtaining the user's feature representation information in this step will be described in detail in S604 of another embodiment and will not be repeated here.

[0094] (2) Call the state classification model to classify the feature representation information and obtain the user's target classification result.

[0095] Specifically, after obtaining the feature representation information of the user, the state classification model is called to classify the feature representation information, and multiple target classification labels and their corresponding target classification probabilities (that is, the probability distribution of the user in multiple categories) can be obtained. The terminal device can use the category corresponding to the highest target classification probability as the target classification result of the user. It should be noted that the target classification result cannot accurately represent the actual category of the user under actual circumstances. Therefore, the user can be classified multiple times until the target classification result meets the preset conditions, and the target classification result is used as the classification result of the user (that is, the target classification result).

[0096] In one embodiment, the method for obtaining a state classification model may include the following steps:

[0097] (1) Obtain training data, which includes multiple training samples and a classification label for each training sample. Each training sample includes one or more sample state description data.

[0098] (2) Use the initial neural network model to process each training sample and obtain the predicted classification results of each training sample.

[0099] (3) Adjust the model parameters of the initial neural network model according to the predicted classification results and classification labels of each training sample to obtain a state classification model.

[0100] In terms of inquiry information recommendation, commonly used status recommendations are usually in the form of templates such as questionnaires, and each status has a corresponding questionnaire template. As the number of states increases, state-based inquiries become diversified, which inevitably leads to an increase in questionnaire maintenance costs. By adopting the method proposed in the present invention, multiple recommended state description data can be obtained by training in a data set, greatly reducing the maintenance costs of existing questionnaires. In terms of target classification, the Bayesian deep network based on the dropout method can take into account the powerful feature learning ability of the neural network, and at the same time give the uncertainty of the model when outputting the target classification probability, making the target classification function more credible to a certain extent, providing an important method for de-blackboxing the model, and making the model more applicable.

[0101] After obtaining the user's status information, the embodiment of the present application determines the recommended query information based on the classification decision path information and the status frequency distribution information, making the query information recommendation more intelligent. During classification, the terminal device automatically performs the classification operation of the target user without the need for human participation, which can improve the classification efficiency. Automatic recognition is not affected by subjective factors, which can improve the accuracy of target classification. The status information is processed using the status classification model to obtain the first classification result, and then the target classification is performed in combination with the first classification result, the query result, etc., which further improves the accuracy of the target classification. Combined with the preset conditions for stopping classification, the classification efficiency is further improved.

[0102] See also Figure 6 , Figure 6 This is a flow chart of a data processing method provided by another exemplary embodiment of the present application, in which the method is applied to Figure 1 Taking the terminal device in FIG. 1 as an example, the method may include the following steps:

[0103] S601: Acquire user status information, where the status information includes one or more initial status description data.

[0104] S602: Determine first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data.

[0105] S603: Obtain a response operation of the user to the first inquiry information, and determine an inquiry result of the first inquiry information according to the response operation.

[0106] The method proposed in this embodiment is mainly aimed at state recommendation and prediction. By combining the classification decision path information and the state frequency distribution information to recommend the state, it can meet the state classification-oriented recommendation and integrate the classification function to a certain extent. The interaction process between the user and the terminal device is repeated until the target classification probability and confidence meet the preset conditions (for example: the target classification probability preset condition is 0.8, the confidence preset condition is 0.9) or exceeds the maximum dialogue discussion (for example, 10 rounds), the interaction process is terminated, and then the target classification result is output. Figure 7 As shown, Figure 7This is a flowchart of a state classification provided by an exemplary embodiment of the present application. First, the state description information of the user is obtained, and the state description information is standardized using a term standardization model to obtain state information; then, the state information is classified using a classification model (that is, state classification in the figure) to obtain a first classification result; it is determined whether the probability and confidence under the first classification result meet the preset conditions; when the preset conditions are not met, the candidate state description data is determined based on the first classification result, state information, classification decision path information, and state frequency distribution information, and the top K (i.e., TOP K, for example, K is 1) items in the candidate state description data are output as recommended states. Of course, the value of K can also be a value greater than or equal to 2, for example, K=3. Based on the user's interaction (answering yes / no, or entering status description information, etc.) and the data from the first classification, the state classification model is called to perform target classification again to obtain a second classification result; again, it is determined whether the probability and confidence under the second classification result meet certain conditions (that is, the probability is greater than the probability threshold P, and the confidence is greater than the confidence threshold U); if not, the above steps are repeated until the probability and confidence meet the above conditions or the number of conversations reaches the number threshold N, and the target classification result is output.

[0107] The specific implementation of steps S601 to S603 refers to the relevant description of steps S200 to S203 in the above embodiment, which will not be repeated here.

[0108] S604: Call the state classification model to perform feature extraction processing on the state information and the query result to obtain feature representation information of the user.

[0109] In one embodiment, if Figure 8 As shown, Figure 8 This is a processing flow chart of a state classification model provided by an exemplary embodiment of the present application. The state classification model proposed in this application consists of three parts: a representation layer, a Monte Carlo layer, and a classifier. First, the standardized state of the user (that is, state information, such as state A, state B) is used as the input of the state classification model. After the representation layer, the Monte Carlo layer (including multiple nonlinear layers), and the classifier of the state classification model, a variety of classification labels (for example: classification 1, classification 2, classification 3) and their corresponding probabilities (for example: the probability corresponding to classification 1 is 0.9, the probability corresponding to classification 2 is 0.6, and the probability corresponding to classification 3 is 0.02) and confidence (for example: the confidence corresponding to classification 1 is 0.92, the confidence corresponding to classification 2 is 0.86, and the confidence corresponding to classification 3 is 0.93).

[0110] In machine learning-based classification models, if the model has too many parameters and too few training samples, the trained model can easily overfit. Overfitting is a common problem when training neural networks. This problem manifests itself in: the model's loss function is small on the training data, resulting in high prediction accuracy; but on the test data, the loss function is large, resulting in low prediction accuracy. Overfitting is a common problem in many machine learning models. If a model overfits, the resulting model becomes almost unusable. To address overfitting, model ensembles are typically employed, training multiple models together. The time required to train the models becomes a significant issue, not only because training multiple models is time-consuming, but also because testing multiple models is time-consuming. The principle of dropout is to stop the activation of a neuron with a certain probability (e.g., P) during forward propagation. This allows the model to generalize more effectively because it is less dependent on certain local features. In the state classification model proposed in this application, the appropriate use of dropout can effectively mitigate overfitting, achieve the effect of ensemble learning, and achieve a certain degree of regularization.

[0111] The state classification model proposed in this application maps the user's main complaint state to a low-dimensional vector and enhances the feature representation capability through multiple nonlinear layers, thus having good representation learning capabilities. In addition, by applying Monte Carlo Dropout (MC-Dropout) to the model parameters, the model parameters are made to obey the probability distribution rather than fixed parameter values. For example, the network model of the state classification model includes 100 neurons, which can be divided into 50 neuron combinations (by discarding the first neuron and using the remaining 99 neurons as the first neuron combination; by discarding the second neuron and using the remaining 99 neurons as the second neuron combination, and so on, to obtain 50 groups of neuron combinations; or by using a random dropout method to finally obtain 50 groups of neuron combinations), thereby achieving the purpose of the probability distribution of the model parameters. MC-dropout is turned on during the training phase of the model and remains turned on during the testing phase. Given the user's main complaint symptoms, multiple calculations are performed, each time with different model parameters, to obtain multiple different outputs, and the mean and variance are calculated to measure the final prediction probability and uncertainty of the model. Therefore, the corresponding credibility is provided for the model to perform target classification. In order to integrate the state classification function, this application uses the decision tree model to explicitly learn the classification decision path information for state classification. By fusing the classification decision path information with the state frequency distribution information calculated based on TF-IDF, the state recommendation can take into account the state classification function to a certain extent.

[0112] The Monte Carlo layer of the state classification model interprets dropout as a Bayesian approximation of a Gaussian process. This allows the model to output both the target classification probability and the confidence level. This confidence level reflects the uncertainty of the model and serves as a reference for model optimization. Dropout is typically applied to model parameters during training to prevent overfitting. During model testing, dropout is disabled. However, Monte Carlo dropout (MC-Dropout) emphasizes that model parameters continue to be dropped during testing, meaning that parameters are dropped during both training and testing.

[0113] The classifier of the state classification model is used to convert the state classification task into a multi-label classification problem. In an exemplary classifier, the classifier can be composed of a linear layer, a dropout layer, and a sigmoid layer (an activation layer).

[0114] A first specific implementation of step S604 may include the following steps:

[0115] a1. Call the representation layer of the state classification model to perform feature extraction processing on each initial state description data included in the state information to obtain a feature vector of each initial state description data.

[0116] a2. Call the representation layer of the state classification model to perform feature extraction processing on each recommended state description data included in the query result to obtain a feature vector of each recommended state description data.

[0117] a3. Call the representation layer of the state classification model to fuse the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain the user's feature representation information.

[0118] After the feature representation information is obtained through the above method, the state classification model can be used to classify the feature representation information to obtain the target classification result.

[0119] The above method obtains user feature representation information based on the initial state description data and the recommended state description data. Furthermore, the state-limiting information corresponding to the initial state description data and the recommended state description data can be fully utilized to obtain multi-dimensional feature representation information. This multi-dimensional feature representation information can then be used for subsequent object classification, thereby improving classification accuracy.

[0120] In one embodiment, for a user, each state description data (including the above-mentioned initial state description data and recommended state description data) may correspond to a state limitation information. The above-mentioned state limitation information is used to indicate whether the state exists, specifically whether the user has the state corresponding to the state description data. The above-mentioned state information may also include the state limitation information of each initial state description data, that is, the state information specifically includes one or more initial state description data and the state limitation information of each initial state description data; the query result specifically includes each recommended state description data in one or more recommended state description data and the state limitation information of each recommended state description data.

[0121] On this basis, step S603 can also be implemented through the following steps:

[0122] b1. Obtaining the user's response operation to the first inquiry information.

[0123] b2. Determine the recommended status description data selected by the user based on the response operation;

[0124] b3. Determine the query result based on the recommended status description data selected by the user and the recommended status description data not selected by the user.

[0125] When a terminal device obtains a user's state description data (including the aforementioned initial state description data and recommended state description data), it typically obtains the state description data selected by the user and the state description data not selected by the user. The state description data selected by the user can be used to determine the user's target classification result. At the same time, the state description data not selected by the user can also be used to assist in determining the user's target classification result (for example, by limiting the classification decision path information through the state description data not selected by the user), thereby improving data utilization and enhancing the accuracy of target classification.

[0126] In one embodiment, the user's state information is diverse, and the state qualification information should include multiple dimensions to assist in determining the user's target classification results. Exemplarily, the user's state information can be mood state data. Then, the state qualification information can be used to indicate whether the state exists (that is, a yes / no keyword), and can also be used to indicate the tone state (that is, a tone keyword). The state description information is subjected to a keyword extraction operation (such as the above-mentioned yes / no keywords, tone keywords, etc.) through a term standardization model, and the extracted keywords are then analyzed through a term standardization model, and the initial state description data corresponding to the state description information is determined in combination with a pre-set set of standard terms.

[0127] On this basis, a second specific implementation of step S604 may include the following steps:

[0128] c1. Calling the representation layer of the state classification model to perform feature extraction processing on each initial state description data and state limitation information included in the state information to obtain a feature vector of each initial state description data.

[0129] c2. Call the representation layer of the state classification model to perform feature extraction processing on each recommended state description data and state limitation information included in the query result to obtain a feature vector of each recommended state description data.

[0130] c3. Call the representation layer of the state classification model to fuse the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain the user's feature representation information.

[0131] Compared to the first specific implementation of step S604, the feature vector of the state description data extracted in the second specific implementation includes two dimensions: one for state description information and the other for state qualification information. The state description information and state qualification information are then fused to obtain multi-dimensional feature representation information.

[0132] In one embodiment, a specific implementation method of obtaining the feature vector of each initial state description data in step c1 may include:

[0133] The representation layer of the state classification model is used to map the state description data into a low-dimensional vector S 状态 , which facilitates subsequent vector operations and processing. The input of the state classification model is standardized symptoms (including initial state description data, recommended state description data, etc.). In one embodiment, the representation layer can use a one-hot vector to represent the low-dimensional vector S for the state-limiting information corresponding to each state description data. 限定 The final state representation is obtained through the state description data and state qualification information. For example, when the recommended state description data is "state A, state C, state D", and "state A, state C" is selected, the query result can be expressed as "yes-state A, yes-state C, no-state D". The two types of data information, yes and no, can be represented by the vector [01,10] respectively, and the three types of data information, state A, state C, and state D, can be represented by the vector [100,010,001] respectively. Then, given composite data information (for example: yes-state C) can be represented by this one-hot vector (for example:

[01010] ).

[0134] If the user's complaint contains multiple statuses, the following formula is used:

[0135]

[0136] Where M is the number of initial state description data in the user state information, S 限定The value is 0 (representing: no, that is, it does not exist) or 1 (representing: yes, that is, it exists). The final state representation can be calculated using the above formula.

[0137] Similarly, the specific implementation method of obtaining the feature vector of each recommended status description data in step c2c can refer to the method of c1 above, and will not be repeated here.

[0138] S605: Call the state classification model to classify the feature representation information to obtain a second classification result of the user and obtain the number of classifications. The second classification result includes the user's target classification probability and the confidence of the target classification.

[0139] Specifically, if the first classification result does not meet the requirements, the user can continue to be classified based on the status information and query results to obtain a second classification result. During the continued classification process, the number of classifications can be obtained, which can be used as a conditional judgment basis for stopping the classification. The target classification probability and target classification confidence included in the second classification result can also be used as a conditional judgment basis for stopping the classification.

[0140] The above step S605 can be specifically implemented by the following steps:

[0141] (1) Multiple neuron combinations in the call state classification model are used to classify the feature representation information respectively to obtain multiple prediction results. Each of the multiple prediction results includes the probability distribution of the user in multiple categories.

[0142] Specifically, the network model of the state classification model includes multiple nonlinear layers, and each of the multiple nonlinear layers includes multiple neurons. A part of the neurons in the network model can be selected as a neuron combination. For example, the network model of the state classification model includes 100 neurons. If it is to be divided into 50 neuron combinations, the first neuron can be discarded and the remaining 99 neurons can be used as the first neuron combination; the second neuron can be discarded and the remaining 99 neurons can be used as the second neuron combination, and so on, to obtain 50 groups of neuron combinations. It should be noted that the above-mentioned method of discarding neurons can also adopt a random discarding method, and this application is not limited here.

[0143] Each of the multiple prediction results can be the probability distribution of the feature representation information generated by one of the neuron combinations across multiple categories. This allows us to obtain multiple probability distributions for each feature representation information across multiple categories, generated by different neuron combinations. Using the probability distributions for different categories, we can obtain the mean probability for each category, which serves as the data foundation for calculating the target classification probability and confidence.

[0144] In one embodiment, the state classification model can be equipped with the conditions to generate multiple neuron combinations by performing dropout processing on the model parameters. When performing target classification, K (for example, K=50) dropouts are randomly performed to obtain K neuron combinations (i.e., the sample is fed to the state classification model K times to obtain K output results). The K output results are then integrated and their probabilities are calculated by means and variances. The variance is used as the confidence of the model, and the disease label corresponding to the maximum probability mean is used as the final output result of the model (i.e., the target classification). Each sample model performs K forward calculations to predict the results because the parameters of the network model in the Bayesian network obey a specific probability distribution rather than a fixed value. When the model predicts the results, it should integrate the parameter distribution of the model. Since the neural network model usually has a large number of parameters, it is extremely difficult to integrate it in the real number domain. MC-Dropout is equivalent to random sampling from the variational distribution of the model parameters, which simplifies the integration operation and makes the model implementation easier.

[0145] In one embodiment, in addition to using MC-Dropout, the present application can also use variational Dropout, Gaussian distribution and other methods as an alternative to achieve the purpose of generating different neuron combinations. In the actual use of this application, the appropriate parameter distribution method should be selected according to the prediction results.

[0146] In one embodiment, uncertainty in the model can be introduced into the dataset. This uncertainty reflects the impact of the data on model performance, provides a reference for dataset selection and model optimization, and improves model robustness and accuracy. For example, this uncertainty can be introduced into the dataset by randomly adding noise to the feature vectors processed by the representation layer, or by applying a Gaussian distribution to the data at the model's representation layer or output layer.

[0147] (2) Determine the second classification result of the user based on multiple prediction results.

[0148] Determining the second classification result of the user based on multiple prediction results can be specifically achieved by the following steps:

[0149] 1) Determine the probability mean of each category in multiple categories based on multiple prediction results.

[0150] 2) Obtain the target classification with the largest corresponding probability mean among multiple classifications, and use the probability mean of the target classification as the target classification probability.

[0151] 3) Determine the probability variance corresponding to the target classification based on the probability mean of the target classification and multiple prediction results, and use the probability variance as the confidence of the target classification.

[0152] Specifically, this step takes the target category with the largest probability mean among multiple categories as the processing object, and calculates the variance of multiple probability items obtained by different neuron combinations in this category. The variance calculation formula is as follows:

[0153]

[0154] Where D is the variance, n is the number of item probabilities, x i is the item probability, and x is the mean probability.

[0155] According to the above formula, the probability variance can be obtained, and the probability variance is used as the confidence of the target classification.

[0156] In one embodiment, assuming that the multiple categories include three categories, such as category 1, category 2, and category 3; and K = 3, three corresponding prediction results are obtained, such as prediction result 1, prediction result 2, and prediction result 3. Each prediction result includes the user's probability of being in the three categories respectively. For example, prediction result 1 includes category 1 -0.85, category 2 -0.55, and category 3 -0.01; prediction result 2 includes category 1 -0.9, category 2 -0.6, and category 3 -0.02; and prediction result 3 includes category 1 -0.95, category 2 -0.65, and category 3 -0.03. Based on the above three prediction results, the predicted probabilities of the user in category 1 are 0.85, 0.9, and 0.95; the predicted probabilities of category 2 are 0.55, 0.6, and 0.65; and the predicted probabilities of category 3 are 0.01, 0.02, and 0.03. The mean probability of each of the multiple categories can be determined based on the prediction results. For example, by adding the three predicted probabilities for category 1 and taking the average, the mean probability of category 1 is 0.9 (the calculation process is (0.85+0.9+0.95) / 3=0.9). Similarly, the mean probability of category 2 is 0.6, and the mean probability of category 3 is 0.02. By comparing the mean probabilities of categories 1, 2, and 3, the mean probability of category 1 is the maximum. Then, category 1 is used as the target category, and the mean probability of the target category is used as the target category probability. For example, the target category probability corresponding to category 1 is 0.9. Then, the probability variance corresponding to the target classification is determined based on the probability mean of the target classification and multiple prediction results. Here, the probability mean of classification 1 is the largest, which is 0.9. The predicted probabilities of classification 1 are 0.85, 0.9, and 0.95. The above formula (2) is used to calculate the variance of 0.85, 0.9, and 0.95, and the probability variance corresponding to classification 1 is 0.04082. 0.04082 is used as the confidence level of classification 1.

[0157] S606: Determine whether the second classification result or the number of classifications meets a preset condition.

[0158] In one embodiment, the second classification result includes the user's target classification probability and the confidence level of the target classification. The above-mentioned preset condition can be that the target classification probability is greater than a probability threshold and the confidence level is greater than a confidence threshold; the preset condition can also be that the number of classifications is greater than a number threshold.

[0159] S607. If the preset conditions are met, the second classification result is used as the user's target classification result. The preset conditions include that the target classification probability is greater than the probability threshold and the confidence is greater than the confidence threshold; or, the preset conditions include that the number of classifications is greater than the number threshold.

[0160] In one embodiment, the target classification method provided in the above embodiment is repeated. When the target classification probability is greater than a probability threshold (e.g., 0.8) and the confidence level is greater than a confidence level threshold (e.g., 0.9), the interaction process terminates. Or, when the number of classifications exceeds a threshold (e.g., 10), the interaction process terminates. This classification result is then used as the user's final target classification result.

[0161] S608. If the preset conditions are not met, the second query information is determined based on the second classification result, the classification decision path information, and the state frequency distribution information; the query result is updated based on the user's response operation to the second query information to obtain an updated query result; and the user's target classification result is determined based on the state information and the updated query result.

[0162] Specifically, when the preset conditions are not met, the user can be queried again according to the method of steps S202 to S202 and steps S604 to S607 provided in the above embodiment to obtain the result of the inquiry, and the result can be combined with the inquiry result before the inquiry as a query result set (that is, an updated inquiry result is obtained), and then the target classification result of the user after this inquiry is determined based on the status information and the updated inquiry result (based on the method of steps S604 to S605 above), until the above preset conditions are met, the inquiry is stopped, and the target classification result is output.

[0163] The embodiment of the present application obtains characteristic information of state information and query results through a state classification model, and performs parameter distribution operations on the model so that the model can obtain model confidence while outputting the prediction probability of state information, and also introduces the uncertainty of the model on the data set, and reflects the impact of data on model performance by applying interference at the model representation layer and output layer, providing a reference for data set selection and model optimization, and improving the robustness and accuracy of the model. This method can obtain the user's personalized original input information, and can also obtain multiple state description data of the user, thereby improving applicability and intelligence. While obtaining the user's state description data, its state limitation information is obtained, and the fusion processing obtains multi-dimensional feature representation information, and the target is classified based on the multi-dimensional feature representation information, further improving the accuracy of the target classification.

[0164] See also Figure 9 , Figure 9 : is a schematic block diagram of a data processing device provided in an embodiment of the present application. The data processing device may specifically include:

[0165] An acquisition module 901 is configured to acquire user status information, wherein the status information includes one or more initial status description data;

[0166] A processing module 902 is configured to determine first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data;

[0167] The processing module 902 is further configured to obtain a response operation of the user to the first inquiry information, and determine an inquiry result of the first inquiry information according to the response operation;

[0168] The classification module 903 is configured to determine the target classification result of the user according to the status information and the query result.

[0169] Optionally, when the processing module 902 is used to determine the first query information based on the state information, the classification decision path information, and the state frequency distribution information, it is specifically used to:

[0170] Calling a status classification model to classify the status information to obtain a first classification result of the user;

[0171] The first inquiry information is determined based on the first classification result, the classification decision path information, and the state frequency distribution information.

[0172] Optionally, when the processing module 902 is used to determine the first query information based on the first classification result, the classification decision path information, and the state frequency distribution information, it is specifically used to:

[0173] Querying classification decision path information according to the first classification result to obtain a first data set, wherein the first data set includes one or more candidate state description data;

[0174] Querying state frequency distribution information according to the first classification result to obtain a second data set, wherein the second data set includes one or more candidate state description data;

[0175] The first query information is obtained by fusing the first data set and the second data set.

[0176] Optionally, when the classification module 903 is used to determine the target classification result of the user according to the status information and the query result, it is specifically used to:

[0177] Calling a state classification model to perform feature extraction processing on the state information and the query result to obtain feature representation information of the user;

[0178] The state classification model is called to perform classification processing on the feature representation information to obtain the target classification result of the user.

[0179] Optionally, when the classification module 903 is used to call the state classification model to perform feature extraction processing on the state information and the query result to obtain the feature representation information of the user, it is specifically used to:

[0180] Calling the representation layer of the state classification model to perform feature extraction processing on each initial state description data included in the state information to obtain a feature vector of each initial state description data;

[0181] Calling the representation layer of the state classification model to perform feature extraction processing on each recommended state description data included in the query result to obtain a feature vector of each recommended state description data;

[0182] The representation layer of the state classification model is called to fuse the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain feature representation information of the user.

[0183] Optionally, when the classification module 903 is used to call the state classification model to classify the feature representation information and obtain the target classification result of the user, it is specifically used to:

[0184] Calling the state classification model to classify the feature representation information to obtain a second classification result for the user and the number of classifications, wherein the second classification result includes a target classification probability of the user and a confidence level of the target classification;

[0185] Determining whether the second classification result or the number of classifications meets a preset condition;

[0186] If the above-mentioned preset conditions are met, the above-mentioned second classification result will be used as the target classification result of the above-mentioned user. The above-mentioned preset conditions include that the above-mentioned target classification probability is greater than the probability threshold, and the above-mentioned confidence is greater than the confidence threshold; or, the above-mentioned preset conditions include that the above-mentioned classification times are greater than the times threshold.

[0187] Optionally, the classification module 903 is further configured to:

[0188] If the above preset conditions are not met, the second inquiry information is determined based on the above second classification result, classification decision path information and state frequency distribution information;

[0189] updating the query result based on the user's response operation to the second query information to obtain an updated query result;

[0190] The target classification result of the user is determined based on the status information and the updated query result.

[0191] Optionally, when the classification module 903 is used to call the state classification model to classify the feature representation information and obtain the second classification result of the user, it is specifically used to:

[0192] Calling multiple neuron combinations in the state classification model to classify the feature representation information to obtain multiple prediction results, each of the multiple prediction results including a probability distribution of the user in multiple categories;

[0193] A second classification result of the user is determined based on the multiple prediction results.

[0194] Optionally, when the classification module 903 is used to determine the second classification result of the user based on the multiple prediction results, it is specifically used to:

[0195] Determine a probability mean for each of the multiple classifications based on the multiple prediction results;

[0196] Obtain the target classification with the largest corresponding probability mean among the above multiple classifications, and use the probability mean of the above target classifications as the target classification probability;

[0197] The probability variance corresponding to the target classification is determined based on the probability mean of the target classification and the multiple prediction results, and the probability variance is used as the confidence of the target classification.

[0198] Optionally, the state information further includes state definition information of each initial state description data, and the query result includes each recommended state description data in the one or more recommended state description data and the state definition information of each recommended state description data, wherein the state definition information is used to indicate whether the state exists;

[0199] The classification module 903 is used to call the state classification model to perform feature extraction processing on the state information and the query result to obtain the feature representation information of the user, specifically for:

[0200] Calling the representation layer of the state classification model to perform feature extraction processing on each initial state description and state limitation information data included in the state information to obtain a feature vector of each initial state description data;

[0201] Calling the representation layer of the state classification model to perform feature extraction processing on each recommended state description data and state qualification information included in the query result to obtain a feature vector of each recommended state description data;

[0202] The representation layer of the state classification model is called to fuse the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain feature representation information of the user.

[0203] Optionally, the classification module 903 is further configured to:

[0204] Acquire training data, the training data including multiple training samples and classification labels for each training sample, each training sample including one or more sample state description data;

[0205] Using the initial neural network model to process each of the above training samples, to obtain a prediction classification result for each of the above training samples;

[0206] The model parameters of the initial neural network model are adjusted according to the predicted classification results and classification labels of each training sample to obtain a state classification model.

[0207] It should be noted that the functions of each functional module of the data processing device of the embodiment of the present application can be specifically implemented according to the method in the above method embodiment. The specific implementation process can refer to the relevant description of the above method embodiment, which will not be repeated here.

[0208] See also Figure 10 , Figure 10This is a schematic block diagram of a computer device provided in one embodiment of the present application. As shown in the figure, the intelligent terminal in this embodiment may include: a processor 1001, a storage device 1002, and a network interface 1003. The processor 1001, storage device 1002, and network interface 1003 may exchange data.

[0209] The above-mentioned storage device 1002 may include a volatile memory (volatile memory), such as a random-access memory (RAM); the storage device 1002 may also include a non-volatile memory (non-volatile memory), such as a flash memory, a solid-state drive (SSD), etc.; the above-mentioned storage device 1002 may also include a combination of the above-mentioned types of memory.

[0210] The processor 1001 may be a central processing unit (CPU). In one embodiment, the processor 1001 may also be a graphics processing unit (GPU). The processor 1001 may also be a combination of a CPU and a GPU. In one embodiment, the storage device 1002 is used to store program instructions, and the processor 1001 may call the program instructions to perform the following operations:

[0211] Obtaining user status information, the status information including one or more initial status description data;

[0212] Determining first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data;

[0213] Obtaining a response operation of the user to the first inquiry information, and determining an inquiry result of the first inquiry information according to the response operation;

[0214] The target classification result of the user is determined based on the status information and the query result.

[0215] Optionally, when the processor 1001 is used to determine the first query information based on the state information, the classification decision path information, and the state frequency distribution information, it is specifically used to:

[0216] Calling a status classification model to classify the status information to obtain a first classification result of the user;

[0217] The first inquiry information is determined based on the first classification result, the classification decision path information, and the state frequency distribution information.

[0218] Optionally, when the processor 1001 is used to determine the first query information based on the first classification result, the classification decision path information, and the state frequency distribution information, it is specifically used to:

[0219] Querying classification decision path information according to the first classification result to obtain a first data set, wherein the first data set includes one or more candidate state description data;

[0220] Querying state frequency distribution information according to the first classification result to obtain a second data set, wherein the second data set includes one or more candidate state description data;

[0221] The first query information is obtained by fusing the first data set and the second data set.

[0222] Optionally, when the processor 1001 is used to determine the target classification result of the user according to the state information and the query result, it is specifically used to:

[0223] Calling a state classification model to perform feature extraction processing on the state information and the query result to obtain feature representation information of the user;

[0224] The state classification model is called to perform classification processing on the feature representation information to obtain the target classification result of the user.

[0225] Optionally, when the processor 1001 is used to call the state classification model to perform feature extraction processing on the state information and the query result to obtain the feature representation information of the user, it is specifically used to:

[0226] Calling the representation layer of the state classification model to perform feature extraction processing on each initial state description data included in the state information to obtain a feature vector of each initial state description data;

[0227] Calling the representation layer of the state classification model to perform feature extraction processing on each recommended state description data included in the query result to obtain a feature vector of each recommended state description data;

[0228] The representation layer of the state classification model is called to fuse the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain feature representation information of the user.

[0229] Optionally, when the processor 1001 is used to call the state classification model to perform classification processing on the feature representation information to obtain the target classification result of the user, it is specifically used to:

[0230] Calling the state classification model to classify the feature representation information to obtain a second classification result for the user and the number of classifications, wherein the second classification result includes a target classification probability of the user and a confidence level of the target classification;

[0231] Determining whether the second classification result or the number of classifications meets a preset condition;

[0232] If the above-mentioned preset conditions are met, the above-mentioned second classification result will be used as the target classification result of the above-mentioned user. The above-mentioned preset conditions include that the above-mentioned target classification probability is greater than the probability threshold, and the above-mentioned confidence is greater than the confidence threshold; or, the above-mentioned preset conditions include that the above-mentioned classification times are greater than the times threshold.

[0233] Optionally, the processor 1001 is further configured to:

[0234] If the above preset conditions are not met, the second inquiry information is determined based on the above second classification result, classification decision path information and state frequency distribution information;

[0235] updating the query result based on the user's response operation to the second query information to obtain an updated query result;

[0236] The target classification result of the user is determined based on the status information and the updated query result.

[0237] Optionally, when the processor 1001 is used to call the state classification model to perform classification processing on the feature representation information to obtain the second classification result of the user, it is specifically used to:

[0238] Calling multiple neuron combinations in the state classification model to classify the feature representation information to obtain multiple prediction results, each of the multiple prediction results including a probability distribution of the user in multiple categories;

[0239] A second classification result of the user is determined based on the multiple prediction results.

[0240] Optionally, when the processor 1001 is used to determine the second classification result of the user according to the multiple prediction results, it is specifically used to:

[0241] Determine a probability mean for each of the multiple classifications based on the multiple prediction results;

[0242] Obtain the target classification with the largest corresponding probability mean among the above multiple classifications, and use the probability mean of the above target classifications as the target classification probability;

[0243] The probability variance corresponding to the target classification is determined based on the probability mean of the target classification and the multiple prediction results, and the probability variance is used as the confidence of the target classification.

[0244] Optionally, the state information further includes state definition information of each initial state description data, and the query result includes each recommended state description data in the one or more recommended state description data and the state definition information of each recommended state description data, wherein the state definition information is used to indicate whether the state exists;

[0245] When the processor 1001 is used to call the state classification model to perform feature extraction processing on the state information and the query result to obtain the feature representation information of the user, it is specifically used to:

[0246] Calling the representation layer of the state classification model to perform feature extraction processing on each initial state description and state limitation information data included in the state information to obtain a feature vector of each initial state description data;

[0247] Calling the representation layer of the state classification model to perform feature extraction processing on each recommended state description data and state qualification information included in the query result to obtain a feature vector of each recommended state description data;

[0248] The representation layer of the state classification model is called to fuse the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain feature representation information of the user.

[0249] Optionally, the processor 1001 is further configured to:

[0250] Acquire training data, the training data including multiple training samples and classification labels for each training sample, each training sample including one or more sample state description data;

[0251] Using the initial neural network model to process each of the above training samples, to obtain a prediction classification result for each of the above training samples;

[0252] The model parameters of the initial neural network model are adjusted according to the predicted classification results and classification labels of each training sample to obtain a state classification model.

[0253] In a specific implementation, the processor 1001, the storage device 1002, and the network interface 1003 described in the embodiment of the present application can execute the embodiment of the present application. Figure 2 or Figure 6 The implementation described in the relevant embodiments of the data processing method provided can also be performed in the embodiments of this application Figure 9 The implementation methods described in the relevant embodiments of the provided data processing device will not be repeated here.

[0254] In the several embodiments provided in this application, it should be understood that the disclosed methods, devices, and systems can be implemented in other ways. For example, the device embodiments described above are merely schematic; for example, the division of the units is merely a logical function division, and there may be other division methods in actual implementation; for example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed may be through some interfaces, indirect coupling or communication connection of devices or units, which may be electrical, mechanical, or other forms.

[0255] In addition, it should be noted that the embodiment of the present application further provides a computer-readable storage medium, and the computer-readable storage medium stores a computer program executed by the aforementioned data processing device, and the computer program includes program instructions. When the processor executes the above program instructions, it can execute the above Figure 2 、 Figure 6 The method in the corresponding embodiment will therefore not be described in detail here. In addition, the description of the beneficial effects of adopting the same method will not be repeated. For technical details not disclosed in the computer-readable storage medium embodiment involved in this application, please refer to the description of the method embodiment of this application. As an example, the program instructions can be deployed on a computer device, or executed on multiple computer devices located in one location, or executed on multiple computer devices distributed in multiple locations and interconnected by a communication network. Multiple computer devices distributed in multiple locations and interconnected by a communication network can constitute a blockchain system.

[0256] According to one aspect of the present application, a computer program product or computer program is provided, the computer program product or computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device can perform the aforementioned Figure 2 、 Figure 6 The method in the corresponding embodiment will therefore not be described in detail here.

[0257] Those skilled in the art will appreciate that all or part of the processes in the above-described method embodiments can be implemented by instructing related hardware through a computer program. The program can be stored in a computer-readable storage medium, and when executed, the program can include the processes in the above-described method embodiments. The storage medium can be a magnetic disk, an optical disk, a read-only memory (ROM), or a random access memory (RAM).

[0258] The above disclosure is only part of the embodiments of the present application, and it is certainly not intended to limit the scope of the rights of the present application. A person skilled in the art can understand that all or part of the processes of the above embodiments and equivalent changes made in accordance with the claims of the present application are still within the scope of the invention.

Claims

1. A data processing method, characterized in that: The method comprises: Acquire user status information, wherein the status information includes one or more initial status description data; determining first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data; Obtaining a response operation of the user to the first inquiry information, and determining an inquiry result of the first inquiry information according to the response operation; Determine the target classification result of the user according to the status information and the query result; The determining of the first query information based on the state information, the classification decision path information, and the state frequency distribution information includes: Calling a status classification model to classify the status information to obtain a first classification result of the user; querying classification decision path information according to the first classification result to obtain a first data set, wherein the first data set includes one or more candidate state description data; querying state frequency distribution information based on the first classification result to obtain a second data set, wherein the second data set includes one or more candidate state description data; wherein the state frequency distribution information includes ranking information of the state description data corresponding to each category in the multiple categories; Performing an intersection extraction operation on the first data set and the second data set to obtain common data; The shared data is sorted according to the state frequency distribution information, and first query information is determined according to the sorted shared data.

2. The method according to claim 1, characterized in that The determining the target classification result of the user according to the status information and the query result includes: Calling a state classification model to perform feature extraction processing on the state information and the query result to obtain feature representation information of the user; The state classification model is called to perform classification processing on the feature representation information to obtain a target classification result of the user.

3. The method according to claim 2, characterized in that The calling of the state classification model to classify the feature representation information to obtain the target classification result of the user includes: Calling the state classification model to perform classification processing on the feature representation information to obtain a second classification result of the user and a classification count, wherein the second classification result includes a target classification probability of the user and a confidence level of the target classification; Determining whether the second classification result or the number of classifications meets a preset condition; If the preset condition is met, the second classification result is used as the target classification result of the user, and the preset condition includes that the target classification probability is greater than the probability threshold and the confidence is greater than the confidence threshold; or, the preset condition includes that the number of classifications is greater than the number threshold.

4. The method according to claim 3, characterized in that The calling of the state classification model to classify the feature representation information to obtain a second classification result of the user includes: Calling multiple neuron combinations in the state classification model to classify the feature representation information respectively to obtain multiple prediction results, each of the multiple prediction results including a probability distribution of the user in multiple categories; A second classification result for the user is determined based on the multiple prediction results.

5. The method according to claim 4, characterized in that Determining a second classification result of the user according to the multiple prediction results includes: Determining a probability mean of each of the multiple categories based on the multiple prediction results; Obtaining a target classification having the largest corresponding probability mean among the multiple classifications, and taking the probability mean of the target classification as the target classification probability; The probability variance corresponding to the target classification is determined according to the probability mean of the target classification and the multiple prediction results, and the probability variance is used as the confidence of the target classification.

6. The method according to claim 2, characterized in that The state information also includes state definition information for each initial state description data; the query result includes each recommended state description data in the one or more recommended state description data and state definition information for each recommended state description data; the state definition information is used to indicate whether a state exists; and calling the state classification model to perform feature extraction processing on the state information and the query result to obtain feature representation information of the user includes: Calling the representation layer of the state classification model to perform feature extraction processing on each initial state description data and state limitation information included in the state information to obtain a feature vector of each initial state description data; Calling the representation layer of the state classification model to perform feature extraction processing on each recommended state description data and state limitation information included in the query result to obtain a feature vector of each recommended state description data; The representation layer of the state classification model is called to perform fusion processing on the feature vector of each initial state description data and the feature vector of each recommended state description data to obtain feature representation information of the user.

7. A data processing device, characterized in that: The device comprises: An acquisition module, configured to acquire user status information, wherein the status information includes one or more initial status description data; a processing module, configured to determine first query information based on the state information, the classification decision path information, and the state frequency distribution information, wherein the first query information includes one or more recommended state description data; The processing module is further configured to obtain a response operation of the user to the first inquiry information, and determine an inquiry result of the first inquiry information according to the response operation; A classification module, configured to determine a target classification result of the user based on the status information and the query result; Wherein, the processing module is specifically used to: Calling a status classification model to classify the status information to obtain a first classification result of the user; querying classification decision path information according to the first classification result to obtain a first data set, wherein the first data set includes one or more candidate state description data; querying state frequency distribution information based on the first classification result to obtain a second data set, wherein the second data set includes one or more candidate state description data; wherein the state frequency distribution information includes ranking information of the state description data corresponding to each category in the multiple categories; Performing an intersection extraction operation on the first data set and the second data set to obtain common data; The shared data is sorted according to the state frequency distribution information, and first query information is determined according to the sorted shared data.

8. A computer-readable storage medium, characterized in that The computer storage medium stores a computer program, wherein the computer program includes program instructions. The program instructions are executed by a processor to perform the data processing method according to any one of claims 1 to 6.

9. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the data processing method according to any one of claims 1 to 6.

Images