Symptom information determination method and apparatus, electronic device, and storage medium

Abstract

The application relates to a symptom information determination method and device, electronic equipment and a storage medium. The method comprises the following steps: acquiring object description information to be processed; using a target symptom recognition network to obtain corresponding target symptom information by taking the object description information as input, wherein the target symptom recognition network is obtained by machine learning training of multiple sample pairs and adjusting parameters of a preset network during the training process, each sample pair indicates a pair of object description information samples and symptom information samples, the training process comprises learning a first type correlation degree and a second type correlation degree, the first type correlation degree represents a correlation degree between two heterogeneous samples in the sample pair, and the second type correlation degree represents a correlation degree between two heterogeneous samples from different sample pairs. The application improves the accuracy and efficiency of symptom recognition. The embodiments of the application can be applied to various scenes such as cloud technology, artificial intelligence, intelligent transportation and auxiliary driving.

Description

Technical Field

[0001] This application relates to the field of Internet communication technology, and in particular to a method, apparatus, electronic device and storage medium for determining symptom information. Background Technology

[0002] With the development of internet communication technology, it has been applied in fields such as smart healthcare, providing users with convenience in medical care, such as symptom identification based on user-provided information. In related technologies, symptom identification based on user-provided information often involves first performing entity recognition on the provided information, and then standardizing the entity recognition results to obtain symptom information. However, the accuracy of the symptom information obtained in this way is relatively low. Therefore, there is a need to provide an accurate and effective symptom information determination scheme. Summary of the Invention

[0003] To address at least one of the aforementioned technical problems, this application provides a method, apparatus, electronic device, and storage medium for determining symptom information:

[0004] According to a first aspect of this application, a method for determining symptom information is provided, the method comprising:

[0005] Obtain the description information of the object to be processed;

[0006] Using the object description information as input, the corresponding target symptom information is obtained by using a target symptom recognition network. The target symptom recognition network is obtained by machine learning training on multiple sample pairs and adjusting the parameters of a preset network during the training process. Each sample pair indicates a pair of object description information samples and symptom information samples. The training process includes learning a first type of correlation and a second type of correlation. The first type of correlation represents the correlation between two dissimilar samples in the sample pair, and the second type of correlation represents the correlation between two dissimilar samples from different sample pairs.

[0007] According to a second aspect of this application, a symptom information determining device is provided, the device comprising:

[0008] Acquisition module: Used to obtain description information of the object to be processed;

[0009] The determination module is used to obtain the corresponding target symptom information by taking the object description information as input and using a target symptom recognition network. The target symptom recognition network is obtained by machine learning training on multiple sample pairs and adjusting the parameters of a preset network during the training process. Each sample pair indicates a pair of object description information samples and symptom information samples. The training process includes learning a first type of correlation and a second type of correlation. The first type of correlation represents the correlation between two dissimilar samples in the sample pair, and the second type of correlation represents the correlation between two dissimilar samples from different sample pairs.

[0010] According to a third aspect of this application, an electronic device is provided, the electronic device including at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores at least one instruction or at least one program, the at least one instruction or at least one program being loaded and executed by the at least one processor to implement the symptom information determination method as described in the first aspect.

[0011] According to a fourth aspect of this application, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or at least one program being loaded and executed by a processor to implement the symptom information determination method as described in the first aspect.

[0012] According to a fifth aspect of this application, a computer program product is provided, the computer program product comprising at least one instruction or at least one program segment, the at least one instruction or at least one program segment being loaded and executed by a processor to implement the symptom information determination method as described in the first aspect.

[0013] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this application.

[0014] Implementing this application will have the following beneficial effects:

[0015] This application utilizes a target symptom recognition network to obtain target symptom information corresponding to object description information, improving the accuracy and efficiency of symptom recognition, and thus enhancing the effectiveness of business processing based on target symptom information. For training the target symptom recognition network, the training process involves learning two types of correlations: the correlation between two dissimilar samples in the same sample pair can be used as "positive samples" to be learned, and the correlation between two dissimilar samples in different sample pairs can be used as "negative samples" to be learned. This application provides learnable positive and negative samples for the training of the target symptom recognition network, which is beneficial to improving training effectiveness and network generalization ability.

[0016] Other features and aspects of this application will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0017] To more clearly illustrate the technical solutions and advantages in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This diagram illustrates an application environment according to an embodiment of the present application.

[0019] Figure 2 A flowchart illustrating a method for determining symptom information according to an embodiment of this application is shown.

[0020] Figure 3 A schematic diagram illustrating the process of training a target symptom recognition network according to an embodiment of this application is shown.

[0021] Figure 4 A schematic diagram illustrating the process of determining target negative sample pairs according to an embodiment of this application is shown;

[0022] Figure 5 This diagram illustrates a process for obtaining target negative sample pairs based on target candidate outlier samples and specified samples according to an embodiment of this application.

[0023] Figure 6 A schematic diagram of the architecture of a preset network according to an embodiment of this application is shown;

[0024] Figure 7 A schematic diagram of an interface for a pre-consultation scenario according to an embodiment of this application is shown;

[0025] Figure 8 This diagram illustrates a device block diagram according to an embodiment of the present application;

[0026] Figure 9 A schematic diagram of an electronic device according to an embodiment of this application is shown. Detailed Implementation

[0027] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0028] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or server that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or devices.

[0029] Various exemplary embodiments, features, and aspects of this application will now be described in detail with reference to the accompanying drawings. The same reference numerals in the drawings denote elements that have the same or similar functions. Although various aspects of the embodiments are shown in the drawings, they are not necessarily drawn to scale unless specifically indicated otherwise.

[0030] The term “exemplary” as used herein means “serving as an example, embodiment, or illustration.” Any embodiment illustrated herein as “exemplary” is not necessarily to be construed as superior to or better than other embodiments.

[0031] In this document, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent three cases: A alone, A and B simultaneously, and B alone. Furthermore, the term "at least one" in this document means any combination of at least two of any one or more elements. For example, including at least one of A, B, and C can mean including any one or more elements selected from the set consisting of A, B, and C.

[0032] Furthermore, to better illustrate this application, numerous specific details are provided in the following detailed description. Those skilled in the art should understand that this application can be implemented without certain specific details. In some instances, methods, means, components, and circuits well-known to those skilled in the art have not been described in detail in order to highlight the main points of this application.

[0033] Before providing a further detailed description of the embodiments of this application, the nouns and terms involved in the embodiments of this application will be explained, and the nouns and terms involved in the embodiments of this application shall be interpreted as follows.

[0034] Noise Contrastive Estimation (NCE) is a statistical model estimation method and a type of contrastive learning method. It learns a similarity function by distinguishing between positive samples and noise (usually negative samples). NCE is often used to solve complex computational problems in neural networks and is widely used in natural language processing tasks such as information retrieval and automatic question answering.

[0035] In-batch negative sampling is a method for sampling negative samples during model training. When the training dataset lacks negative samples, the conventional approach is to obtain them through random sampling. In-batch negative sampling, however, uses other samples within the same batch as negative samples for the current sample. For example, a common example in information retrieval or question answering is matching a question to a correct answer. Information retrieval or question answering models are typically based on matching or similarity models, and these types of models require negative samples for training. In-batch negative sampling utilizes the answers to other questions within the same batch as incorrect answers and negative samples for training.

[0036] Please see Figure 1 , Figure 1 The diagram illustrates an application environment according to an embodiment of this application. The application environment may include a client 10 and a server 20. The client 10 and server 20 can be directly or indirectly connected via wired or wireless communication. Related objects (such as users or simulators) can send corpus processing instructions to the server 20 through the client 10. The server 20 determines the object description information to be processed based on the received corpus processing instructions, and then uses a target symptom recognition network to obtain the target symptom information corresponding to the object description information. Alternatively, in response to a corpus processing instruction triggered by a related object, the client 10 determines the object description information to be processed based on the corpus processing instructions, and then uses a target symptom recognition network to obtain the target symptom information corresponding to the object description information. It should be noted that... Figure 1 This is just one example.

[0037] Client 10 can be a physical device such as a smartphone, computer (e.g., desktop computer, tablet, laptop), augmented reality (AR) / virtual reality (VR) device, digital assistant, smart voice interaction device (e.g., smart speaker), smart wearable device, smart home appliance, in-vehicle terminal, etc., or it can be software running on the physical device, such as a computer program. The operating system corresponding to the client can be Android, iOS (a mobile operating system developed by Apple), Linux (an operating system), Microsoft Windows, etc.

[0038] The server-side component 20 can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server may include network communication units, processors, and memory, etc. The server-side component can provide backend services to the corresponding clients.

[0039] The symptom information determination scheme provided in this application embodiment can utilize technologies related to Artificial Intelligence (AI). Artificial intelligence is the theory, 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 obtain optimal results. Artificial intelligence technology is a comprehensive discipline involving a wide range of fields, encompassing both hardware and software technologies. Fundamental AI technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, big data processing technology, operating / interactive systems, and mechatronics. AI software technologies mainly include computer vision (CV), speech processing, natural language processing (NLP), as well as machine learning / deep learning, autonomous driving, and intelligent transportation.

[0040] Natural Language Processing (NLP) is an important area within computer science and artificial intelligence. It studies the theories and methods for enabling effective communication between humans and computers using natural language. NLP is a science that integrates linguistics, computer science, and mathematics. Therefore, research in this field involves natural language—the language people use in daily life—and thus it has a close connection with linguistic research. NLP technologies typically include text processing, semantic understanding, machine translation, question answering, and knowledge graphs.

[0041] It should be noted that when object description information, object description information samples, target symptom information, symptom information samples, etc., which are related to user information, are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0042] Figure 2 This diagram illustrates a flowchart of a method for determining symptom information according to an embodiment of this application. Figure 2 As shown, the method includes:

[0043] S201: Obtain description information of the object to be processed;

[0044] In this embodiment, the client or server obtains the object description information to be processed. The object description information is used to describe the target object's physical discomfort, medical condition, etc. It is understood that the object description information to be processed needs to undergo symptom identification.

[0045] Regardless of whether it's the client or the server, the object description information to be processed can be determined based on the received corpus processing instructions to achieve acquisition. This acquisition can be done from local storage, via the internet, or from relevant business systems. These relevant business systems can be external systems or internal systems unrelated to the symptom recognition system involved in this application's embodiments. It should be noted that: 1) the object providing the object description information is not necessarily the target object. For example, the first object provides object description information indicating the second object based on the interactive interface provided by the client. 2) the form of the object description information provided by the providing object is not limited to text, voice, images, etc.

[0046] S202: Using the object description information as input, obtain the corresponding target symptom information using the target symptom recognition network.

[0047] In this embodiment, the object description information to be processed is used as input, and the client or server uses a target symptom recognition network to obtain the corresponding target symptom information. It can be understood that the object description information to be processed is input into a locally deployed target symptom recognition network, and the output of the target symptom recognition network is the target symptom information. Considering the aforementioned format of the object description information, if the object description information to be processed does not meet the format requirements of the target symptom recognition network for input data, the object description information to be processed can be format-processed first, and then the format-processed object description information can be input into the target symptom recognition network.

[0048] The target symptom recognition network is obtained by machine learning training on multiple sample pairs and adjusting the parameters of the preset network during the training process. Each sample pair indicates a pair of object description information samples and symptom information samples. The training process includes learning a first type of relevance and a second type of relevance. The first type of relevance represents the relevance between two dissimilar samples in the sample pair, and the second type of relevance represents the relevance between two dissimilar samples from different sample pairs.

[0049] For example, a sample pair consists of an object description information sample and a symptom information sample. The object description information sample and the symptom information sample belong to different categories; they are dissimilar samples. Whether an object description information sample and a symptom information sample are paired can be determined based on whether they are related. The object description information sample and the symptom information sample in the same sample pair are related. To a certain extent, the symptom information sample in the same sample pair can serve as a symptom representation of the object description information sample, and correspondingly, the object description information sample in the same sample pair can serve as a descriptive representation of the symptom information sample.

[0050] Taking multiple sample pairs (sample pairs 1-10) as an example, sample pair i consists of object description information sample i and symptom information sample i, where i ranges from 1 to 10. The correlation between object description information sample i and symptom information sample i in sample pair i is used as a "positive sample" to be learned, and the correlation between object description information sample i and symptom information sample j (from sample pair j) can be used as a "negative sample" to be learned, where i ≠ j and j ranges from 1 to 10. To obtain the target symptom recognition network, multiple batches of sample pair sets are needed, with each batch involving multiple sample pairs. During the training process, each batch of sample pair sets provides positive and negative samples for learning, thus improving training effectiveness and efficiency.

[0051] In one exemplary implementation, such as Figure 3 As shown, the target symptom recognition network is obtained through the following training method:

[0052] S301: For each of the object description information samples involved in the plurality of sample pairs, obtain the heterogeneous sample prediction result corresponding to each of the object description information samples based on the preset network;

[0053] S302: For each of the symptom information samples involved in the plurality of sample pairs, obtain the outlier sample prediction result corresponding to each of the symptom information samples based on the preset network;

[0054] S303: Calculate the loss function value based on the outlier prediction results corresponding to each object description information sample and the outlier prediction results corresponding to each symptom information sample;

[0055] S304: Adjust the parameters of the preset network based on the loss function value until the convergence condition is met to obtain the target symptom recognition network.

[0056] Taking multiple sample pairs (sample pairs 1-10) as an example, sample pair i consists of object description information sample i and symptom information sample i, where i ranges from 1 to 10. During training, all samples (i.e., 20 samples) are input into a pre-defined network to obtain the prediction result of the out-of-class sample corresponding to each sample. That is, using object description information sample i as input, the pre-defined network obtains the candidate symptom information corresponding to object description information sample i. Similarly, using symptom information sample i as input, the pre-defined network obtains the candidate description information corresponding to symptom information sample i. For calculating the loss function value, the ideal prediction result can be obtained based on out-of-class samples belonging to the same sample pair as the sample. The loss function value can be obtained based on the difference between the prediction result of the out-of-class sample corresponding to each sample and the ideal prediction result. The parameters of the pre-defined network are adjusted according to the loss function value until the convergence condition is met, resulting in the target symptom recognition network.

[0057] The target symptom recognition network is used to determine relevant symptom information based on related object description information. Training with object description information samples as input data can be viewed as forward training, which aligns with the target data flow involved in the target symptom recognition network. Correspondingly, training with symptom information samples as input data can be viewed as backward training. Through training in both directions, the network can not only learn to predict symptom information based on description information, but also learn to predict description information based on symptom information. This allows for a more thorough learning of the correlation between dissimilar samples, which is beneficial for improving the prediction accuracy and adaptability of the target symptom recognition network.

[0058] Furthermore, to obtain the target symptom recognition network, multiple batches of sample pairs are needed, each batch involving multiple sample pairs. The training of the current batch of sample pairs can be used to adjust the parameters of the first network to obtain the second network. The adjustment process is as follows: 1) For each object description information sample involved in the current batch of sample pairs, obtain the prediction result of the outlier sample corresponding to each object description information sample based on the first network; 2) For each symptom information sample involved in the current batch of sample pairs, obtain the prediction result of the outlier sample corresponding to each symptom information sample based on the first network; 3) Calculate the loss function value based on the prediction results of the outlier samples corresponding to each object description information sample and the prediction results of the outlier samples corresponding to each symptom information sample; 4) Adjust the parameters of the first network based on the loss function value until the convergence condition is met to obtain the second network.

[0059] The first network can be the aforementioned preset network, or it can be an intermediate network involved in the process of obtaining the target symptom recognition network based on the preset network. Taking the current batch of sample pairs as sample pairs 1-10 as an example, sample pair i consists of object description information sample i and symptom information sample i, where i ranges from 1 to 10. The outlier prediction result corresponding to each object description information sample can include the matching probability corresponding to all outlier samples in the current batch of sample pairs. That is, the outlier prediction result of object description information sample i includes its matching probability with each of the 10 symptom information samples. The outlier prediction result corresponding to each symptom information sample is similar and will not be elaborated further.

[0060] In practical applications, such as Figure 6 As shown, the architecture of the preset network can include an encoding part and a noise contrast estimation part, specifically: Bi-Encoder, forward noise contrast estimation (e.g., Figure 6 NCE-FORWARD in the middle) and inverse noise contrast estimation (e.g. Figure 6 (NCE-BACKWARD in the context of a batch of sample pairs). Given a set of sample pairs in a batch, an encoder is used for all patient inputs (e.g., ...). Figure 6 Q in 1-m One encoder encodes the sample of the object description information mentioned above; another encoder encodes all symptoms ( Figure 6 S in 1-m The symptom information samples mentioned above are encoded. Then, both forward and reverse noise contrast estimation are performed simultaneously. Forward noise contrast estimation uses the loss function in Formula 1 below, while reverse noise contrast estimation uses the loss function in Formula 2 below. The encoding part aims to transform both patient input and symptoms into a high-dimensional space. This can preserve the original semantics of the text to the greatest extent. After obtaining the high-dimensional vectors of patient input and symptoms, semantic matching can more accurately identify the symptoms.

[0061] For positive noise contrast estimation: Assume there are X sample pairs (patient input, symptom), where the symptom in the same sample pair is the corresponding symptom contained in the patient input of the same sample pair. For each batch of sample pairs, N sample pairs are drawn from the X sample pairs, and these N sample pairs are input into a Bi-encoder (one encoder encodes the patient input, and the other encodes the symptom) to obtain the corresponding vector. The resulting vector includes two matrices: one is the encoded matrix of the patient input (Q...). emb ), a matrix that encodes the symptoms (S) emb Both matrices are of size N*M, where N is the number of sample pairs in the current batch set, and M is the vector dimension. This is achieved using scores = Q. emb *S emb T This yields a similarity score matrix, where each row corresponds to the similarity between a patient's input and all symptoms within the batch. For each row, the loss function value is obtained using Formula 1 below.

[0062]

[0063] Where qi is the input of the i-th patient, s i + For the symptoms corresponding to qi (qi and s) i + (belonging to the same sample pair), while s j - Let sim be the j-th (j≠i) other symptom within the batch, where sim is the similarity function and τ is a hyperparameter. i s i + ) represents the similarity between the input of the i-th patient and its corresponding symptoms. It is the sum of similarities between the i-th patient's input and other symptoms within the batch. During the process of using gradient descent to reduce the loss function value, the distance between the patient's input and its corresponding symptom continuously decreases, while the distance between it and other symptoms continuously increases.

[0064] For inverse noise contrast estimation: the difference from the above forward noise contrast estimation is that: by using scores = S emb *Q emb TA similarity score matrix is ​​obtained, where each row corresponds to the similarity between a symptom and the inputs of all patients within the batch. For each row, the loss function value is obtained using Formula 2 below. The parameter involved in Formula 2 differs from that in Formula 1 in that: q j - Enter the value for the j-th (j≠i) other patient in the batch. It is the sum of similarity between the i-th symptom and the inputs of other patients in the batch. During the process of using gradient descent to reduce the loss function value, the distance between the symptom and its corresponding patient input is continuously decreasing, while the distance between it and the inputs of other patients is continuously increasing.

[0065]

[0066] The combination of forward noise contrast estimation and reverse noise contrast estimation can not only make full use of the information in the training dataset, but also increase the difficulty of the training task, thereby obtaining a more efficient and robust symptom recognition network.

[0067] In addition to using Bi-Encoder, Cross-Encoder can also be used for the encoding part. Cross-Encoder has only one encoder. For training tasks like symptom recognition, which take a pair of texts as input, the common practice is to concatenate the two texts into one, but add a special character [SEP] in the middle for differentiation. The choice between Bi-Encoder and Cross-Encoder for the encoding part can be determined based on the application scenario of the target symptom recognition network. If there are no time constraints, Cross-Encoder can be chosen because it has a word-level attention mechanism and good encoding performance. If there are time constraints (such as in online applications), Bi-Encoder can be chosen because it can better balance encoding performance and encoding efficiency. In practical online applications, taking symptom recognition as an example, all symptoms can be encoded offline first, and only the patient input needs to be encoded and similarity calculated online, which can greatly improve matching efficiency.

[0068] In one exemplary implementation, to further improve training effectiveness, in addition to learning the first and second types of relevance during the training process, a third type of relevance between target negative sample pairs is added. For target negative sample pairs, it is understood that the relevance of the dissimilar samples corresponding to the sample pair is generally positive and has the highest relevance score, while the relevance between two dissimilar samples from different sample pairs may indicate a negative or positive correlation. Target negative sample pairs consist of two dissimilar samples from different sample pairs, and the relevance of the dissimilar samples corresponding to the target negative sample pairs is generally positive and has a high relevance score. The introduction of target negative sample pairs helps to improve the fine-grainedness of relevance learning, enhances the network's ability to distinguish between at least two highly relevant dissimilar samples for the same sample, and thus improves the predictive performance of the target symptom recognition network.

[0069] Obtaining a target symptom recognition network requires multiple sample pairs. During training, first-class and second-class relevance can be learned, followed by third-class relevance. Although the target symptom recognition network is used to determine relevant symptom information based on relevant object description information, the construction of target negative sample pairs can either use object description information samples as baseline samples to find highly relevant symptom information samples other than those belonging to the same sample pair, or use symptom information samples as baseline samples to find highly relevant object description information samples other than those belonging to the same sample pair. To accelerate network fitting and simultaneously improve prediction accuracy, high-quality negative sample pairs are added to both the positive and negative sides in addition to batch negative sampling to increase the difficulty of contrastive learning.

[0070] To obtain a target symptom recognition network, multiple batches of sample pairs are needed, with each batch containing multiple sample pairs. During training, each batch of sample pairs can first learn the first and second types of relevance, and then learn the third type of relevance. Similarly, multiple rounds of sample pairs are needed, with each round involving multiple batches of sample pairs, and each batch containing multiple sample pairs. During training, each round of sample pairs can first learn the first and second types of relevance, and then learn the third type of relevance. It should be noted that in this batch or multi-round training scenario, the highly relevant outlier samples used to construct the target negative sample pairs along with the baseline samples can come not only from the current batch or round of sample pairs, but also from previous batches or rounds of sample pairs.

[0071] like Figure 4As shown, the target symptom recognition network is obtained through multiple rounds of training. Each round of training includes a first task and a second task. The first task is used to learn the first type of relevance and the second type of relevance, and the second task is used to learn the third type of relevance between target negative sample pairs. The target negative sample pairs are determined by the following method:

[0072] S401: Input the specified sample into the candidate symptom recognition network to obtain the matching probability corresponding to multiple candidate outlier samples respectively. The specified sample refers to any sample involved in the sample pair set in this round. The candidate symptom recognition network refers to the network that completes the first task. The multiple candidate outlier samples are not outlier samples in the sample pair to which the specified sample belongs.

[0073] S402: Based on the matching probabilities corresponding to the multiple candidate outlier samples, the sample with a matching probability greater than the first probability threshold is determined as the target candidate outlier sample;

[0074] S403: Obtain the target negative sample pair based on the target candidate outlier sample and the specified sample.

[0075] For example, the current sample pair set includes multiple sample pairs, which can come from at least one batch of sample pairs. Taking these multiple sample pairs as sample pairs 1-100 as an example, sample pair i consists of object description information sample i and symptom information sample i, where i ranges from 1 to 100. The specified sample is any one of the description information samples 1-100 and the symptom information samples 1-100. The training involving this sample pair set can be used to adjust the parameters of the third network based on the first task to obtain a candidate symptom recognition network, and then adjust the parameters of the candidate symptom recognition network based on the second task to obtain a fourth network. The third network can be the aforementioned preset network, or it can be an intermediate network involved in the process of obtaining the target symptom recognition network based on the preset network.

[0076] For the task "Inputting a specified sample into a candidate symptom recognition network to obtain the matching probabilities corresponding to multiple candidate outlier samples", the process involves using object description information sample i as input and utilizing the candidate symptom recognition network to obtain the matching probabilities between object description information sample i and multiple candidate outlier samples. These multiple candidate outlier samples may not only be all samples from symptom information samples 1-100 excluding symptom information sample i, but may also be all or part of the symptom information samples involved in the previous round of sample pair sets. (The process is repeated three times in the original text.)

[0077] If the first probability threshold is 50%, samples with a matching probability greater than 50% among multiple candidate outlier samples are identified as target candidate outlier samples. Of course, the first probability threshold can be flexibly set according to needs, such as considering historical feedback and training efficiency. When object description information sample 1 is the designated sample, if the target candidate outlier sample is symptom information sample 5, then a target negative sample pair is formed based on object description information sample 1 and symptom information sample 5. For the object description information sample as the designated sample, the number of relevant target candidate outlier samples can be 0, 1, or greater than 1. Correspondingly, the number of relevant target negative sample pairs can be 0, 1, or greater than 1. When symptom information sample 3 is the designated sample, if the target candidate outlier sample is object description information sample 100, then a target negative sample pair is formed based on object description information sample 100 and symptom information sample 3. For the symptom information sample as the designated sample, the number of relevant target candidate outlier samples can be 0, 1, or greater than 1. Correspondingly, the number of relevant target negative sample pairs can be 0, 1, or greater than 1.

[0078] Regarding the process of "obtaining target negative sample pairs based on target candidate outlier samples and specified samples," it's worth noting that if the number of target negative sample pairs is consistent with the number of target candidate outlier samples, there might be an excessive number of target negative sample pairs, affecting training efficiency. The following section details the process of selecting and obtaining target negative sample pairs:

[0079] (i) First, when there are at least two target candidate outlier samples, the top few samples with the highest matching probability among the at least two target candidate outlier samples are determined as target outlier samples; then, the target negative sample pair is obtained based on the target outlier samples and the specified sample.

[0080] For example, when object description information sample 1 is used as the specified sample, if there are 50 target candidate outlier samples that are greater than the target number (e.g., 30), these 50 target candidate outlier samples can be sorted in descending order of matching probability to obtain a sequence; then, the top 30 target candidate outlier samples in this sequence are extracted to obtain 30 target outlier samples; furthermore, target negative sample pairs are constructed based on these 30 target outlier samples and object description information sample 1 to obtain 30 target negative sample pairs.

[0081] By analyzing the relationship between the number of targets and the matching probability, target candidate outlier samples are filtered. The resulting target outlier samples construct target negative sample pairs of higher quality, which is more conducive to improving the fine-grainedness of relevance learning and enhancing the relevance network's ability to learn the differences between at least two highly relevant outlier samples for the same sample.

[0082] (ii) First, when there are at least two target candidate outlier samples, obtain the target number; then, randomly select the target number of samples from the at least two target candidate outlier samples as target outlier samples; furthermore, obtain the target negative sample pair based on the target outlier samples and the specified sample.

[0083] For example, when symptom information sample 3 is the designated sample, if there are 100 target candidate outlier samples that is greater than the target number (e.g., 20), 20 can be randomly selected from these 100 target candidate outlier samples as target outlier samples; then, target negative sample pairs are constructed based on these 20 target outlier samples and symptom information sample 3 respectively to obtain 20 target negative sample pairs.

[0084] Randomly selecting outlier samples based on the target number can speed up the determination of outlier samples, thereby improving the efficiency of constructing target negative sample pairs and training efficiency. Meanwhile, considering that the reliance on matching probability stems from the assumption that the network's prediction results are accurate, and that the network may also have problems learning partial relevance, randomly selecting outlier samples and then using the constructed target negative sample pairs as input for training can improve the network's generalization ability to a certain extent, making the target symptom recognition network more adaptable and reliable. It should be noted that the target number here can be the same as or different from the target number in (a) above.

[0085] (iii) such as Figure 5 As shown, obtaining the target negative sample pair based on the target candidate outlier sample and the specified sample includes:

[0086] S501: Obtain the associated heterogeneous sample corresponding to the specified sample;

[0087] S502: When there are at least two target candidate outlier samples, filter out the samples that match the associated outlier samples from the at least two target candidate outlier samples to obtain the target outlier samples;

[0088] S503: Obtain the target negative sample pair based on the target outlier sample and the specified sample.

[0089] Filtering out samples that match associated outlier samples from at least two target candidate outlier samples can prevent these samples from forming target negative sample pairs with the specified sample, thus avoiding interference from such target negative sample pairs in training. A sample pair consists of an object description information sample and a symptom information sample. The object description information sample and the symptom information sample in the same sample pair are related. Associated outlier samples are also related to the specified sample, but the associated outlier samples and the specified sample do not form a sample pair. This may stem from omissions in the sample pair acquisition process. For example, the sample pair acquisition process involves manual annotation. Patient input 'a' contains symptoms 1-5, but the staff only annotated symptoms 1-3, while symptoms 4-5 are also related to patient input 'a'. If target negative sample pairs are constructed based on symptoms 4-5 and patient input 'a' respectively to obtain two target negative sample pairs, the relevant network will learn a third relevance based on these two target negative sample pairs, which will affect the network training effect, and thus affect the prediction accuracy and effectiveness of the target symptom recognition network. In practical applications, sample pairs formed by associated outlier samples and the specified sample can be regarded as pseudo-negative samples.

[0090] For example, when symptom information sample 25 is used as the designated sample, the associated outlier samples 1-10 corresponding to symptom information sample 25 are obtained. If there are 50 target candidate outlier samples, samples that match associated outlier samples 1-10 can be identified from these 50 target candidate outlier samples (such as target candidate outlier sample 1 that matches associated outlier sample 1, and target candidate outlier sample 10 that matches associated outlier sample 5), and these are filtered out to obtain the target outlier samples (such as target candidate outlier samples 2-9, target candidate outlier samples 11-50).

[0091] The target candidate outlier sample is not an outlier sample in the sample pair to which the specified sample belongs. Generally, associated outlier samples are not outlier samples in the sample pair to which the specified sample belongs. The target candidate outlier sample to be filtered matches the associated outlier sample, which can be determined based on their similarity. For example, if the similarity is greater than a similarity threshold, they are considered a match.

[0092] Furthermore, obtaining the associated outlier sample corresponding to the specified sample may include the following steps: based on the matching probabilities corresponding to the multiple candidate outlier samples respectively, determining the sample with a matching probability greater than a second probability threshold as the associated outlier sample, wherein the second probability threshold is greater than the first probability threshold, the second probability threshold indicates the baseline probability when the candidate symptom recognition network processes the validation set and the prediction accuracy for the validation set is greater than the accuracy threshold, and the baseline probability is the basis for the candidate symptom recognition network to output the prediction result.

[0093] If the second probability threshold is 70%, samples with a matching probability greater than 70% among multiple candidate outlier samples are identified as associated outlier samples. Samples with a matching probability greater than 70% among multiple candidate outlier samples can be filtered out as associated outlier samples, and can also be used as a reference for further filtering of target candidate outlier samples.

[0094] The second probability threshold is greater than the first probability threshold, and it is related to the validation set processed by the candidate symptom recognition network. For the latter, it can be understood that the validation set includes multiple sample pairs, where the object description information samples and symptom information samples within the same sample pair are correlated. Taking sample pairs 201-300 as an example, sample pair k consists of object description information sample k and symptom information sample k, where k ranges from 201 to 300. Sample pairs 201-300 are input into the candidate symptom recognition network for outlier sample prediction. For each sample involved in sample pairs 201-300, the matching probabilities corresponding to multiple candidate outlier samples are obtained. Then, the prediction result is determined based on samples whose matching probabilities are greater than the prediction probability threshold k. Multiple candidate outlier samples can be not only all outlier samples in sample pairs 201-300, but also all or part of the object description information samples involved in the current round of sample pairs or the previous round of sample pairs. The prediction accuracy is determined based on the prediction results of all samples in sample pairs 201-300. If the prediction accuracy is greater than the accuracy threshold, then the current prediction probability threshold k is determined as the second probability threshold. Otherwise, the prediction probability threshold k is increased, and the prediction result and prediction accuracy are redefined accordingly. This process continues, checking whether the redefined prediction accuracy is greater than the accuracy threshold. If it is, the current prediction probability threshold k is determined as the second probability threshold; otherwise, the process of "increasing the prediction probability threshold k and redefined the prediction result and prediction accuracy" is repeated until "the prediction accuracy is greater than the accuracy threshold," at which point the current prediction probability threshold k is determined as the second probability threshold. In practical applications, the initial value of the probability threshold k can be 50%, and the increase in the probability threshold k is 5% each time. Each training round has a corresponding accuracy threshold, the initial value of which can be 80%, and the increase in the accuracy threshold is 1% each round.

[0095] The three methods described above for obtaining target negative sample pairs can also be combined in twos or all three. For example, after obtaining multiple target candidate outlier samples, we can first use method (iii) above to remove associated outlier samples to obtain a sample set; then, we can use method (ii) above to randomly select a number of target samples from the sample set as target outlier samples. When randomly selecting a number of target outlier samples, we can use multinomial distribution sampling based on the model's predicted probability (the predicted probability here can correspond to the matching probability mentioned above). The specific Python sampling function is as follows: torch.multinomial(probs, num_negatives, replacement=False). Here, probs is the probability set of the candidate set (corresponding to the multiple target candidate outlier samples mentioned above). num_negatives is the number of negative samples to be selected (corresponding to the target outlier samples mentioned above), which can be determined by the parameter tuning of the candidate symptom recognition network during the processing of the validation set. replacement means that the selected negative samples will not have duplicate samples. The function will return the index of the selected sample.

[0096] Furthermore, based on the examples in steps S301-S304 above, the target negative sample pairs can be used as training data input into the relevant network, and the resulting loss function values ​​are as follows: Formula 3 and Formula 4:

[0097]

[0098]

[0099] Formula 3 is obtained by updating Formula 1, and Formula 4 is obtained by updating Formula 2. qi is the input for the i-th patient, s... i + For the symptoms corresponding to qi (qi and s) i + (belonging to the same sample pair), while s j - For the j-th (j≠i) other symptom in the batch, q j - For the j-th (j≠i) other patient in the batch, sim is the similarity function, and τ is a hyperparameter. sim(q i s i + ) represents the similarity between the input of the i-th patient and its corresponding symptoms, and N is the similarity between the input and the symptoms. hard This represents the number of target negative sample pairs. Finally, the loss function values ​​involved in Equations 3 and 4 can be combined, and the combined loss function value can be used to adjust the network parameters. The combined loss function value is Loss = Loss forward +Lossbackward .

[0100] The symptom information determination scheme provided in this application can be used in pre-diagnosis scenarios. See also... Figure 7 , Figure 7 The "Feels like there's a lump on my neck" option in the interface diagram on the left provides a description of the object to be processed. Figure 6 The "Neck lump", "Neck rash", "Neck numbness", "Neck lymph node enlargement" and "rash" in the interface diagram on the right can indicate the target symptom information.

[0101] The intelligent pre-consultation system can utilize the target symptom recognition network provided in this application embodiment. Symptom recognition is a core functional module of the intelligent pre-consultation system during the patient information collection stage. The purpose of intelligent pre-consultation is to collect relevant information about the patient's condition in advance to assist doctors in analyzing the patient's condition and making risk assessments more effectively and accurately. The system can recommend several high-frequency symptoms for the patient to choose from based on the department selected by the patient. If the patient cannot find a corresponding symptom among the symptom options provided by the system, the system will provide an input box for the patient to input relevant information. The system can utilize natural language processing technology (involving the application of the target symptom recognition network provided in this application embodiment) to process the patient's input and extract relevant information. The information processing steps consist of two steps. The first step is to analyze the patient's input intent, such as physical examination, follow-up visit, consultation, or description of the condition. If the patient's input intent is identified as a description of the condition, the target symptom recognition network will identify relevant symptoms from the patient's input, such as... Figure 7 As shown. Because patient input is often colloquial, the entity recognition and standardization processes used in related technologies often struggle to identify the symptoms. If information cannot be collected effectively and accurately, the final system-generated report may affect the doctor's consultation. Applying the target symptom recognition network provided in this application can improve the hit rate of the pre-consultation system in symptom recognition, thereby enhancing the system's ability to collect patient symptom information. For more colloquial patient input (such as "small bumps on the eyelids") and patient input primarily describing the patient's symptoms and their characteristics (such as "I can feel my intestines are hard and bulging when I wake up on an empty stomach in the morning, and I always hear my stomach rumbling"), the target symptom recognition network provided in this application can effectively improve the accuracy of symptom recognition. When the description of the patient's symptoms and their characteristics is vague, the target symptom recognition network provided in this application also has strong adaptability in symptom recognition.

[0102] To verify the effectiveness of the target symptom recognition network provided in this application embodiment, it was tested using a target set. The target set contained 22,000 patient inputs, including 429 related symptoms. 2,000 patient inputs were randomly selected as the test set, and the remaining data was used for training and validation. Performance metrics used were NDCG (Normalized Discounted Cumulative Gain) and RECALL (Recall), commonly used metrics for ranking tasks. The test results demonstrate that the target symptom recognition network provided in this application embodiment possesses excellent recognition performance and efficiency.

[0103] As can be seen from the technical solutions provided in the embodiments of this application above, the embodiments of this application utilize a target symptom recognition network to obtain target symptom information corresponding to object description information, thereby improving the accuracy and efficiency of symptom recognition and thus enhancing the effectiveness of business processing based on target symptom information. For the training of the target symptom recognition network, the training process involves learning two types of correlations: the correlation between two dissimilar samples in the same sample pair can be used as a "positive sample" to be learned, and the correlation between two dissimilar samples in different sample pairs can be used as a "negative sample" to be learned. The embodiments of this application provide positive and negative samples for training the target symptom recognition network, which is beneficial for improving training effectiveness and network generalization ability.

[0104] This application also provides a symptom information determination device, such as... Figure 8 As shown, the symptom information determining device 80 includes:

[0105] Acquisition module 801: Used to acquire description information of the object to be processed;

[0106] Determination module 802: used to obtain corresponding target symptom information by taking the object description information as input and using a target symptom recognition network. The target symptom recognition network is obtained by machine learning training on multiple sample pairs and adjusting the parameters of a preset network during the training process. Each sample pair indicates a pair of object description information samples and symptom information samples. The training process includes learning a first type of correlation and a second type of correlation. The first type of correlation represents the correlation between two dissimilar samples in the sample pair, and the second type of correlation represents the correlation between two dissimilar samples from different sample pairs.

[0107] It should be noted that the apparatus and method embodiments described in the device embodiments are based on the same inventive concept.

[0108] In some embodiments, the functions or modules of the apparatus provided in this application can be used to perform the methods described in the above method embodiments. The specific implementation can be referred to the description of the above method embodiments, and for the sake of brevity, it will not be repeated here.

[0109] This application also provides a computer-readable storage medium storing at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the above-described method. The computer-readable storage medium may be a non-volatile computer-readable storage medium.

[0110] This application also provides an electronic device, which includes at least one processor and a memory communicatively connected to the at least one processor; wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the at least one processor to implement the above method.

[0111] Electronic devices can be provided as terminals, servers, or other forms of devices.

[0112] Figure 9 A block diagram of an electronic device according to an embodiment of this application is shown. For example, electronic device 1900 may be provided as a server. (Refer to...) Figure 9 The electronic device 1900 includes a processing component 1922, which further includes one or more processors, and memory resources represented by memory 1932 for storing instructions, such as application programs, that can be executed by the processing component 1922. The application programs stored in memory 1932 may include one or more modules, each corresponding to a set of instructions. Furthermore, the processing component 1922 is configured to execute instructions to perform the methods described above.

[0113] Electronic device 1900 may also include a power supply component 1926 configured to perform power management of electronic device 1900, a wired or wireless network interface 1950 configured to connect electronic device 1900 to a network, and an input / output (I / O) interface 1958. Electronic device 1900 can operate on an operating system stored in memory 1932, such as Windows Server™, Mac OS X™, Unix™, Linux™, FreeBSD™, or similar.

[0114] In an exemplary embodiment, a non-volatile computer-readable storage medium is also provided, such as a memory 1932 including computer program instructions that can be executed by a processing component 1922 of an electronic device 1900 to perform the above-described method.

[0115] This application may be a system, method, and / or computer program product. A computer program product may include a computer-readable storage medium having computer-readable program instructions loaded thereon for causing a processor to implement various aspects of this application.

[0116] Computer-readable storage media can be tangible devices capable of holding and storing instructions for use by an instruction execution device. Computer-readable storage media can be, for example—but not limited to—electrical storage devices, magnetic storage devices, optical storage devices, electromagnetic storage devices, semiconductor storage devices, or any suitable combination thereof. More specific examples (a non-exhaustive list) of computer-readable storage media include: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), static random access memory (SRAM), portable compact disc read-only memory (CD-ROM), digital multifunction disc (DVD), memory sticks, floppy disks, mechanical encoding devices, such as punch cards or recessed protrusions storing instructions thereon, and any suitable combination thereof. The computer-readable storage media used herein are not to be construed as transient signals themselves, such as radio waves or other freely propagating electromagnetic waves, electromagnetic waves propagating through waveguides or other transmission media (e.g., light pulses through fiber optic cables), or electrical signals transmitted through wires.

[0117] The computer-readable program instructions described herein can be downloaded from computer-readable storage media to various computing / processing devices, or downloaded via a network, such as the Internet, local area network, wide area network, and / or wireless network, to an external computer or external storage device. The network may include copper transmission cables, fiber optic transmission, wireless transmission, routers, firewalls, switches, gateway computers, and / or edge servers. A network adapter card or network interface in each computing / processing device receives the computer-readable program instructions from the network and forwards them to the computer-readable storage media in the respective computing / processing device.

[0118] The computer program instructions used to perform the operations of this application may be assembly instructions, instruction set architecture (ISA) instructions, machine instructions, machine-dependent instructions, microcode, firmware instructions, status setting data, or source code or object code written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Smalltalk, C+, etc., and conventional procedural programming languages ​​such as the "C" language or similar programming languages. The computer-readable program instructions may be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving a remote computer, the remote computer may be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or may be connected to an external computer (e.g., via the Internet using an Internet service provider). In some embodiments, electronic circuits, such as programmable logic circuits, field-programmable gate arrays (FPGAs), or programmable logic arrays (PLAs), are personalized by utilizing the status information of the computer-readable program instructions. These electronic circuits can execute the computer-readable program instructions to implement various aspects of this application.

[0119] Various aspects of this application are described herein with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It should be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer-readable program instructions.

[0120] These computer-readable program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, or other programmable data processing apparatus to produce a machine such that, when executed by the processor of the computer or other programmable data processing apparatus, they create means for implementing the functions / actions specified in one or more blocks of the flowchart and / or block diagram. These computer-readable program instructions can also be stored in a computer-readable storage medium that causes a computer, programmable data processing apparatus, and / or other device to operate in a particular manner; thus, the computer-readable medium storing the instructions comprises an article of manufacture that includes instructions for implementing aspects of the functions / actions specified in one or more blocks of the flowchart and / or block diagram.

[0121] Computer-readable program instructions may also be loaded onto a computer, other programmable data processing apparatus, or other device to cause a series of operational steps to be performed on the computer, other programmable data processing apparatus, or other device to produce a computer-implemented process, thereby causing the instructions executed on the computer, other programmable data processing apparatus, or other device to perform the functions / actions specified in one or more boxes of a flowchart and / or block diagram.

[0122] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of an instruction, which includes one or more executable instructions for implementing a specified logical function. In some alternative implementations, the functions specified in the blocks may occur in a different order than those specified in the drawings. For example, two consecutive blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or action, or using a combination of dedicated hardware and computer instructions.

[0123] The various embodiments of this application have been described above. These descriptions are exemplary and not exhaustive, nor are they limited to the disclosed embodiments. Many modifications and variations will be apparent to those skilled in the art without departing from the scope and spirit of the described embodiments. The terminology used herein is chosen to best explain the principles, practical applications, or technological improvements to the embodiments in the market, or to enable others skilled in the art to understand the embodiments disclosed herein.

Claims

1. A symptom information determination method characterized by comprising: The method comprises: acquiring object description information to be processed; using a target symptom recognition network to obtain corresponding target symptom information, taking the object description information as input, the target symptom recognition network being obtained by machine learning training of a plurality of sample pairs and adjusting parameters of a preset network during the training process, each sample pair indicating a pair of object description information samples and symptom information samples, the training process comprising learning a first type of correlation and a second type of correlation, the first type of correlation representing the correlation between two heterogeneous samples in the sample pair, and the second type of correlation representing the correlation between two heterogeneous samples from different sample pairs; wherein the target symptom recognition network is obtained through multiple rounds of training, each round of training task comprising a first task and a second task, the first task being a task for learning the first type of correlation and the second type of correlation, and the second task being a task for learning a third type of correlation between a target negative sample pair, the target negative sample pair being determined by the following method: inputting a specified sample into a candidate symptom recognition network to obtain a plurality of candidate heterogeneous samples corresponding to matching probabilities, the specified sample indicating any one sample involved in the current sample pair set, the candidate symptom recognition network indicating a network that has completed the first task, and the plurality of candidate heterogeneous samples not being heterogeneous samples in the sample pair to which the specified sample belongs; based on the matching probabilities corresponding to the plurality of candidate heterogeneous samples, determining a sample with a matching probability greater than a first probability threshold as a target candidate heterogeneous sample; and obtaining the target negative sample pair based on the target candidate heterogeneous sample and the specified sample.

2. The method of claim 1, wherein, The target symptom recognition network is obtained by the following training method: for each object description information sample involved in the plurality of sample pairs, obtaining a heterogeneous sample prediction result corresponding to each object description information sample based on the preset network; for each symptom information sample involved in the plurality of sample pairs, obtaining a heterogeneous sample prediction result corresponding to each symptom information sample based on the preset network; based on the heterogeneous sample prediction result corresponding to each object description information sample and the heterogeneous sample prediction result corresponding to each symptom information sample, calculating a loss function value; based on the loss function value, adjusting the parameters of the preset network to satisfy a convergence condition to obtain the target symptom recognition network.

3. The method of claim 1, wherein, The target negative sample pair is obtained based on the target candidate heterogeneous sample and the specified sample, comprising: when there are at least two target candidate heterogeneous samples, determining the first target number of samples with the largest matching probability among the at least two target candidate heterogeneous samples as target heterogeneous samples; obtaining the target negative sample pair based on the target heterogeneous samples and the specified sample.

4. The method of claim 1, wherein, The target negative sample pair is obtained based on the target candidate heterogeneous sample and the specified sample, comprising: when there are at least two target candidate heterogeneous samples, obtaining a target number; randomly selecting the target number of samples from the at least two target candidate heterogeneous samples as target heterogeneous samples; The target negative sample pair is obtained based on the target heterogeneous sample and the specified sample.

5. The method of claim 1, wherein, The target negative sample pair is obtained based on the target heterogeneous sample and the specified sample. An associated heterogeneous sample corresponding to the specified sample is obtained. When there are at least two target candidate heterogeneous samples, the target candidate heterogeneous samples that match the associated heterogeneous sample are filtered out to obtain a target heterogeneous sample. The target negative sample pair is obtained based on the target heterogeneous sample and the specified sample.

6. The method of claim 5, wherein, The associated heterogeneous sample corresponding to the specified sample is obtained by: Based on the matching probabilities corresponding to the plurality of candidate heterogeneous samples, a sample with a matching probability greater than a second probability threshold is determined as the associated heterogeneous sample, the second probability threshold is greater than the first probability threshold, the second probability threshold indicates a benchmark probability when the prediction accuracy of the candidate symptom recognition network processing a verification set and for the verification set is greater than an accuracy threshold, and the benchmark probability is a basis for the candidate symptom recognition network outputting a prediction result.

7. A symptom information determination apparatus characterized by comprising: The device comprises: An acquisition module configured to acquire object description information to be processed; A determination module configured to obtain corresponding target symptom information by taking the object description information as input and using a target symptom recognition network, the target symptom recognition network being obtained by machine learning training of a plurality of sample pairs and adjustment of parameters of a preset network during the training process, each sample pair indicating a pair of object description information sample and symptom information sample, the training process comprising learning a first type of correlation and a second type of correlation, the first type of correlation representing the correlation between two heterogeneous samples in the sample pair, and the second type of correlation representing the correlation between two heterogeneous samples from different sample pairs; The target symptom recognition network is obtained through multiple rounds of training, each round of training task comprising a first task and a second task, the first task being a task for learning the first type of correlation and the second type of correlation, and the second task being a task for learning a third type of correlation between a target negative sample pair, the target negative sample pair being determined by the following method: inputting a specified sample into a candidate symptom recognition network to obtain matching probabilities corresponding to a plurality of candidate heterogeneous samples, the specified sample indicating any one sample involved in the current sample pair set, the candidate symptom recognition network indicating a network that has completed the first task, and the plurality of candidate heterogeneous samples not being heterogeneous samples in the sample pair to which the specified sample belongs; based on the matching probabilities corresponding to the plurality of candidate heterogeneous samples, a sample with a matching probability greater than a first probability threshold is determined as a target candidate heterogeneous sample; and the target negative sample pair is obtained based on the target candidate heterogeneous sample and the specified sample.

8. The apparatus of claim 7, wherein, The target symptom recognition network is obtained by the following training method: For each object description information sample involved in the plurality of sample pairs, a heterogeneous sample prediction result corresponding to each object description information sample is obtained based on the preset network. For each of the symptom information samples involved in the plurality of sample pairs, a heterogeneous sample prediction result corresponding to each of the symptom information samples is obtained based on the preset network; A loss function value is calculated based on the heterogeneous sample prediction result corresponding to each of the object description information samples and the heterogeneous sample prediction result corresponding to each of the symptom information samples; Parameters of the preset network are adjusted based on the loss function value to meet a convergence condition, and the target symptom recognition network is obtained.

9. The apparatus of claim 7, wherein, The target negative sample pair is obtained based on the target candidate heterogeneous sample and the specified sample, and the target negative sample pair includes: When there are at least two target candidate heterogeneous samples, a target number of samples with the largest matching probability among the at least two target candidate heterogeneous samples are determined as target heterogeneous samples; The target negative sample pair is obtained based on the target heterogeneous sample and the specified sample.

10. The apparatus of claim 7, wherein, The target negative sample pair is obtained based on the target candidate heterogeneous sample and the specified sample, and the target negative sample pair includes: When there are at least two target candidate heterogeneous samples, a target number of samples with the largest matching probability among the at least two target candidate heterogeneous samples are determined as target heterogeneous samples; The target negative sample pair is obtained based on the target heterogeneous sample and the specified sample. The target negative sample pair is obtained based on the target candidate heterogeneous sample and the specified sample, and the target negative sample pair includes:

11. The apparatus of claim 7, wherein, An associated heterogeneous sample corresponding to the specified sample is obtained; When there are at least two target candidate heterogeneous samples, samples matching the associated heterogeneous sample among the at least two target candidate heterogeneous samples are filtered out to obtain target heterogeneous samples; The target negative sample pair is obtained based on the target heterogeneous sample and the specified sample. The associated heterogeneous sample corresponding to the specified sample is obtained, and the associated heterogeneous sample includes:

12. The apparatus of claim 11, wherein, Based on the matching probabilities of the plurality of candidate heterogeneous samples, samples with a matching probability greater than a second probability threshold value are determined as the associated heterogeneous sample, the second probability threshold value is greater than the first probability threshold value, the second probability threshold value indicates a benchmark probability when a prediction accuracy of the candidate symptom recognition network processing a verification set and for the verification set is greater than an accuracy threshold value, and the benchmark probability is a basis for the candidate symptom recognition network to output a prediction result. The electronic device includes at least one processor and a memory in communication with the at least one processor; wherein the memory stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the at least one processor to implement the symptom information determination method of any one of claims 1-6.

13. An electronic device, comprising: The computer readable storage medium stores at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the symptom information determination method of any one of claims 1-6.

14. A computer-readable storage medium, characterized in that, The computer program product includes at least one instruction or at least one program, and the at least one instruction or at least one program is loaded and executed by the processor to implement the symptom information determination method of any one of claims 1-6.

15. A computer program product, characterised in that, ​

Images