Method and device for constructing medical training samples, and medical text retrieval method

Abstract

The present application relates to the field of artificial intelligence technology, and specifically provides a method and apparatus for constructing medical training samples, and a method for retrieving medical texts. The method for constructing medical training samples includes: obtaining a question-and-answer set; calculating the authority score of the reply text based on the institutional authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text; screening the question-and-answer corpus in the question-and-answer set based on the authority score of the reply text, and adding the screened question-and-answer corpus to a candidate question-and-answer set; combining the reply texts in the candidate question-and-answer set to obtain a target popular science article corresponding to a target medical topic; associating the target popular science article with the target question-and-answer corpus to obtain a medical training sample, where the target question-and-answer corpus is a question-and-answer corpus obtained from the candidate question-and-answer set; this solution realizes the automatic construction of medical training samples.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and more specifically, to a method and device for constructing medical training samples, and a method for retrieving medical texts. Background Art

[0002] In order to improve retrieval efficiency, a medical question-answering model built based on artificial intelligence technology is applied to the retrieval of medical-related questions. By inputting the medical question text into the medical question-answering model, the medical question-answering model automatically outputs the medical answer text for the medical question text based on its understanding of the medical question text, thereby realizing automatic retrieval of the medical answer text for the medical question text.

[0003] To ensure the accuracy of the medical answers generated by the medical question-answering model, it must be trained using a number of medical training samples. These training samples consist of question-answer-article triples, where the question represents the medical question, the answer represents the medical answer to the question, and the article represents the popular science article from which the answer originated. In the prior art, these question-answer-article triples are manually constructed, which is labor-intensive and inefficient. Summary of the Invention

[0004] The embodiments of the present application provide a method and apparatus for constructing medical training samples, and a method for retrieving medical texts, so as to realize automatic construction of medical training samples.

[0005] Other features and advantages of the present application will become apparent from the following detailed description, or may be learned in part by practice of the present application.

[0006] According to one aspect of an embodiment of the present application, a method for constructing a medical training sample is provided, comprising:

[0007] Obtaining a question and answer set, wherein the question and answer set includes a plurality of question and answer corpora for a target medical topic, wherein the question and answer corpora include question texts and answer texts;

[0008] Calculating the authority score of the reply text based on the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text;

[0009] Screening the question and answer corpus in the question and answer set according to the authority score of the reply text, and adding the screened question and answer corpus to the candidate question and answer set;

[0010] Combining the reply texts in the candidate question and answer set to obtain a target popular science article corresponding to the target medical topic;

[0011] The target popular science article is associated with a target question and answer corpus to obtain a medical training sample, where the target question and answer corpus is a question and answer corpus obtained from the candidate question and answer set; the medical training sample is used to train a medical question and answer model, wherein the medical question and answer model is used to obtain medical answer text for an input medical question text from a number of medical popular science articles.

[0012] According to one aspect of an embodiment of the present application, a method for retrieving medical texts is provided, comprising:

[0013] Obtaining a search request, wherein the search request indicates a medical question text;

[0014] Obtaining medical answer texts for the medical question texts from a number of medical popular science articles using a medical question-answering model; the medical question-answering model is trained using the medical training samples constructed using the above-mentioned method for constructing medical training samples;

[0015] Using the medical answer text as a text summary of the medical popular science article corresponding to the medical answer text;

[0016] The address information and text summary of the medical popular science article where the medical answer text is located are returned to the initiator of the search request.

[0017] According to one aspect of an embodiment of the present application, a device for constructing a medical training sample is provided, comprising:

[0018] A question and answer set acquisition module is used to acquire a question and answer set, wherein the question and answer set includes a plurality of question and answer corpora for a target medical topic, and the question and answer corpora include question texts and answer texts;

[0019] An authority score calculation module, configured to calculate the authority score of the reply text based on the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text;

[0020] A screening module, configured to screen question and answer corpora in the question and answer set according to the authority score of the reply text, and add the screened question and answer corpora to the candidate question and answer set;

[0021] A combining module, configured to combine the reply texts in the candidate question and answer set to obtain a target popular science article corresponding to the target medical topic;

[0022] An association module is used to associate the target popular science article with a target question and answer corpus to obtain a medical training sample, where the target question and answer corpus is a question and answer corpus obtained from the candidate question and answer set; the medical training sample is used to train a medical question and answer model, wherein the medical question and answer model is used to obtain medical answer text for an input medical question text from a number of medical popular science articles.

[0023] According to one aspect of an embodiment of the present application, a medical text retrieval device is provided, comprising:

[0024] A search request acquisition module, configured to acquire a search request, wherein the search request indicates a medical question text;

[0025] The medical answer text acquisition module is used to obtain the medical answer text for the medical question text from a number of medical popular science articles through the medical question answering model; the medical question answering model is trained using the medical training samples constructed by the above-mentioned method for constructing medical training samples.

[0026] a text summary determination module, configured to use the medical answer text as a text summary of the medical popular science article corresponding to the medical answer text;

[0027] The return module is used to return the address information and text summary of the medical popular science article where the medical answer text is located to the initiator of the search request.

[0028] According to one aspect of an embodiment of the present application, an electronic device is provided, comprising: a processor; and a memory, wherein the memory stores computer-readable instructions, and when the computer-readable instructions are executed by the processor, the method for constructing medical training samples or the method for retrieving medical texts as described above is implemented.

[0029] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, on which computer-readable instructions are stored. When the computer-readable instructions are executed by a processor, the method for constructing medical training samples or the method for retrieving medical texts as described above is implemented.

[0030] In the solution of this application, with the help of the large amount of medical-related question and answer corpora existing in the Internet medical content platform, the target popular science article is automatically constructed in reverse based on multiple question and answer corpora under the same medical theme. That is, the reply texts of the question and answer corpora in the candidate question and answer set are combined. On this basis, based on the question and answer relationship between the question text and the reply text in the target question and answer corpus selected from the candidate question and answer set, the target popular science article is associated with the target question and answer corpus, forming a medical training sample of article-question-answer triples, realizing the automatic construction of medical training samples, which no longer requires manual construction, and improving the construction efficiency of medical training samples.

[0031] Moreover, in the scheme of the present application, the authority score of the reply text in the question and answer corpus is calculated based on the institutional authority level characteristics of the medical institution to which the author of the source of the reply text in the question and answer corpus belongs and the medical title characteristics corresponding to the source author of the reply text, and the question and answer corpus in the question and answer set is screened according to the authority score of the reply text to obtain a candidate question and answer set, which can ensure that the authority of the reply text in the candidate question and answer corpus is high, and ensure the scientific nature and reliability of the content of the reply text and target popular science articles in the candidate question and answer set, thereby ensuring the reliability of the obtained medical training samples.

[0032] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0033] The accompanying drawings are incorporated into and constitute a part of the specification, illustrating embodiments consistent with the present application and, together with the specification, explaining the principles of the present application. Obviously, the drawings described below are only some embodiments of the present application, and those skilled in the art can derive other drawings based on these drawings without inventive effort. In the drawings:

[0034] Figure 1 A schematic diagram shows an exemplary system architecture to which the technical solutions of the embodiments of the present application can be applied.

[0035] Figure 2 FIG. 1 is a flowchart of a method for constructing medical training samples according to an embodiment of the present application.

[0036] Figure 3 This is a flowchart illustrating an embodiment of the present application for obtaining the authority level characteristics of the medical institution to which the source author of the reply text belongs.

[0037] Figure 4 Schematic diagram of constructing medical training samples according to an embodiment of the present application.

[0038] Figure 5 This is a flowchart of the steps before step 240 according to an embodiment of the present application.

[0039] Figure 6 FIG. 5 is a flowchart of step 510 according to an embodiment of the present application.

[0040] Figure 7 FIG. 6 is a flowchart of step 610 according to an embodiment of the present application.

[0041] Figure 8 This is a schematic diagram of the model structure of the word2vec model.

[0042] Figure 9 FIG. 6 is a flowchart illustrating step 620 according to an embodiment of the present application.

[0043] Figure 10 This is a flowchart of a medical text retrieval method according to an embodiment of the present application.

[0044] Figure 11 4 is a block diagram of an apparatus for constructing medical training samples according to an embodiment of the present application.

[0045] Figure 12 It is a block diagram of a medical text retrieval device according to an embodiment of the present application.

[0046] Figure 13 A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown. DETAILED DESCRIPTION

[0047] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be thorough and complete and will fully convey the concepts of the example embodiments to those skilled in the art.

[0048] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner.In the following description, many specific details are provided so as to provide a full understanding of the embodiments of the present application. However, it will be appreciated by those skilled in the art that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps etc. can be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring the various aspects of the application.

[0049] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically separate entities. That is, these functional entities may be implemented in software, in one or more hardware modules or integrated circuits, or in different networks and / or processor devices and / or microcontroller devices.

[0050] The flowcharts shown in the accompanying drawings are for illustrative purposes only and do not necessarily include all contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may be decomposed, while others may be combined or partially combined. Therefore, the actual execution order may vary depending on the actual situation.

[0051] It should be noted that the term "plurality" used in this document refers to two or more. "And / or" describes a relationship between associated objects, indicating that three possible relationships exist. For example, "A and / or B" can represent: A alone, A and B together, or B alone. The character " / " generally indicates an "or" relationship between the associated objects.

[0052] Artificial Intelligence (AI) refers to the theories, methods, techniques, and application systems that use digital computers or machines controlled by digital computers to simulate, extend, and expand human intelligence, to 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 seeks 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 possess the capabilities of perception, reasoning, and decision-making.

[0053] Artificial intelligence (AI) technology is a comprehensive discipline encompassing a wide range of fields, encompassing both hardware and software technologies. Foundational AI technologies generally include sensors, specialized AI chips, cloud computing, distributed storage, big data processing, operating / interaction systems, and mechatronics. AI software technologies primarily encompass computer vision, speech processing, natural language processing, and machine learning / deep learning.

[0054] As users pay more attention to medical and health issues, more and more popular science articles on medical and health are available. When users need to know more about a disease, they usually enter a question to search for relevant information on the Internet and then read the retrieved results to find the answer.

[0055] In order to improve the retrieval efficiency of medical-related information, a medical question-answering model constructed based on artificial intelligence technology is applied to the retrieval of medical-related questions. By inputting the user's medical question text into the medical question-answering model, the medical question-answering model can understand the medical question text and automatically output the medical answer text for the medical question text based on the understanding. The medical question-answering model realizes automatic retrieval of the medical answer text for the medical question text.

[0056] To ensure the accuracy of the medical answers generated by the medical question-answering model, it must be trained using a number of medical training samples. These training samples consist of question-answer-article triples, where the question represents the medical question, the answer represents the medical answer to the question, and the article represents the source article. In the prior art, medical editors typically compile this question-answer-article triple corpus by understanding the medical question and various medical articles. They then determine whether each article contains a medical answer to the question. If so, they use this data as a basis for constructing medical training samples.

[0057] To ensure the accuracy of the medical question-answering model, a large number of medical training samples are required. However, if the manual construction method in related technologies is used, the construction of medical training samples is labor-intensive and inefficient. Based on this, the solution of this application is proposed to achieve automatic construction of medical training samples.

[0058] Figure 1 A schematic diagram shows an exemplary system architecture to which the technical solutions of the embodiments of the present application can be applied.

[0059] like Figure 1 As shown, the system architecture may include terminal devices (such as Figure 1 101, tablet computer 102, and portable computer 103, which may also be a desktop computer, etc.), network 104, and server 105. Network 104 is a medium for providing a communication link between the terminal device and server 105. Network 104 can include various connection types, such as wired communication links, wireless communication links, etc.

[0060] It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is merely illustrative. Any number of terminal devices, networks and servers may be provided as needed. For example, the server 105 may be a server cluster consisting of multiple servers.

[0061] In some embodiments of the present application, the method for constructing medical training samples can be executed by the server 105 to construct medical training samples based on the question and answer corpus in the question and answer set according to the method for constructing medical training samples of the present application.

[0062] In some embodiments of the present application, a medical question-answering model may be constructed in server 105 and then trained using the constructed medical training samples. After training is complete, server 105 may automatically retrieve medical answers to medical questions from popular medical science articles based on the trained medical question-answering model.

[0063] In some embodiments of the present application, a user can initiate a retrieval request to the server 105 based on a terminal device, and the retrieval request indicates a medical question text. The server 105 then automatically searches through a medical question-and-answer model to obtain the medical answer text for the medical question text, and obtain the address information of the medical popular science article from which the medical answer text originates, such as a link to the medical popular science article.

[0064] In some embodiments of the present application, the server 105 uses the obtained medical answer text as a text summary of the medical popular science article from which the medical answer text originates, and returns the text summary and address information of the medical popular science article from which the medical answer text originates to the terminal device, and displays the text summary in the terminal device, so that the user can quickly understand the medical answer text for the medical question text contained in the medical popular science article, without the user having to search for the medical answer text for the medical question text by reading the medical popular science article.

[0065] The following is a detailed description of the implementation details of the technical solution of the embodiment of the present application:

[0066] Figure 2 The flowchart of the method for constructing medical training samples according to one embodiment of the present application is shown. The method can be executed by a computer device with processing capabilities, such as a server, etc., which is not specifically limited here. Figure 2 As shown, the method includes at least steps 210 to 250, which are described in detail as follows:

[0067] Step 210: Obtain a question and answer set, where the question and answer set includes a number of question and answer corpora for a target medical topic, and the question and answer corpora include question texts and answer texts.

[0068] The target medical topic refers to the medical topic that is commonly related to each question and answer corpus in the question and answer set. In a specific embodiment, the granularity of the medical topic can be set according to actual needs.

[0069] In some embodiments of the present application, the granularity of the medical topic can be based on disease, for example, the medical topic can be hypertension, diabetes, facial paralysis, rheumatism, etc. In some embodiments of the present application, a smaller granularity can be selected to define the medical topic, such as "disease + symptoms", "disease + classification", or "disease + diet". For example, "hypertension + dizziness" is a medical topic, and "hypertension + cause" is a medical topic.

[0070] The question text in the question-answering corpus refers to the text used to indicate the question asked, and the reply text refers to the text used to indicate the answer to the question corresponding to the question text.

[0071] In some embodiments of the present application, in order to ensure the number of constructed medical training samples, question and answer corpora can be collected for different medical topics, and a corresponding question and answer set for each medical topic can be obtained to further construct medical training samples.

[0072] On some internet platforms, users can ask questions about healthcare. That is, users can post questions about medical topics on the internet platform. Other users on the platform, such as those certified as medical professionals, can answer the questions posted by the users and post their responses on the platform. Therefore, we collect question-and-answer corpora on these internet platforms for different medical topics.

[0073] Step 220 , calculating the authority score of the reply text based on the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical professional title characteristics corresponding to the source author of the reply text.

[0074] In this solution, in order to ensure the scientificity and accuracy of the reply text corresponding to the question indicated by the question text, the collected reply texts are from users who are certified as medical workers. At the same time, the medical institution to which the source author of the reply text belongs and the medical title of the source author are recorded in the platform from which the reply text comes.

[0075] Therefore, the user information of the source author of the reply text (i.e., the user who posted the reply text) can be collected from the page where the reply text is located, and the medical title of the source author of the reply text and the medical institution to which the source author of the reply text belongs can be extracted from the user information.

[0076] The medical title feature is used to represent the user's medical title, which includes junior titles (medical assistant, physician / resident), intermediate titles (attending physician), deputy senior titles (associate chief physician), and senior titles (chief physician, chief Chinese medicine physician, chief pharmacist, chief Chinese medicine pharmacist, chief nurse, and chief laboratory technician).

[0077] The institutional authority level feature is used to characterize the authority of a medical institution. Its authority can be reflected by its ranking in the medical institution rankings, or by the level of the medical institution.

[0078] In the scenario where the authority of a medical institution is reflected by its grade, the grade of the medical institution reflects the comprehensive situation of the medical institution, so the grade of the medical institution can be used as the institutional authority grade feature of the medical institution. In China, hospitals are divided into three levels and ten grades (first-level hospitals, second-level hospitals, and third-level hospitals, each level of hospitals is divided into three grades: A, B, and C, among which third-level hospitals have additional specific grades). Of course, the classification of medical institutions may vary in different countries and regions. Therefore, the different classifications of the grades of medical institutions listed in China are considered to be limitations on the scope of use of this application.

[0079] In some embodiments of the present application, the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text may be weighted to obtain the authority score of the reply text.

[0080] Specifically, step 220 may further include: obtaining the institutional authority level characteristics of the medical institution to which the source author of the reply text belongs, and obtaining the medical title characteristics corresponding to the source author of the reply text; weighting the numerical value mapped by the institutional authority level characteristics and the numerical value mapped by the medical title characteristics to obtain the authority score corresponding to the reply text.

[0081] In this embodiment, a first numerical mapping table for the authority level characteristics of the institution and a second numerical mapping table for the medical title characteristics are pre-constructed. In the first numerical mapping table, each authority level characteristic of the institution corresponds to a numerical value; similarly, in the second numerical mapping table, each medical title characteristic corresponds to a numerical value. On this basis, after determining the authority level characteristics of the institution and the corresponding medical title characteristics corresponding to the reply text, the numerical value mapped to the authority level characteristics of the medical institution to which the source author of the reply text belongs is determined according to the first numerical mapping table, and the numerical value mapped to the medical title characteristics corresponding to the source author of the reply text is determined according to the second numerical mapping table.

[0082] In some embodiments of the present application, Figure 3 The process shown is used to obtain the institutional authority level characteristics of the medical institution to which the source author of the reply text belongs, such as Figure 3 Shown, including:

[0083] Step 310: Obtain the institution information corresponding to the reply text, where the institution information indicates the target medical institution to which the source author of the reply text belongs.

[0084] The organization information corresponding to the reply text can be obtained from the user information of the source author of the reply text.

[0085] Step 320: Determine the ranking of the target medical institution in the target medical institution ranking list.

[0086] The target medical institution ranking list refers to the medical ranking list that includes the target medical institution.

[0087] In some embodiments of the present application, step 320 further includes: querying a medical institution ranking list set to obtain a target medical institution ranking list including the target medical institution; and obtaining the ranking of the target medical institution in the target medical institution ranking list.

[0088] In this embodiment, various medical institution rankings are pre-collected and a medical institution ranking set is constructed. Examples of medical institution rankings include the annual XX Medical Rankings and the XX Top 100 Hospitals rankings. Of course, in other embodiments, more medical institution rankings may be included, not limited to the ones listed above, and may be selected based on actual needs.

[0089] In some embodiments of the present application, since the medical institution rankings are updated at a certain period, for example, once a year, in order to ensure the timeliness of the medical institution rankings in the medical institution ranking set, the medical institution rankings in the medical institution ranking set can be the medical institution rankings published within the current set time period, such as the medical institution rankings published in the last 5 years, the last 3 years, etc.

[0090] After determining the target medical institution to which the source author of the reply text belongs, the target medical institution can be queried in the medical institution ranking list set to determine the target medical institution ranking list including the target medical institution and the ranking of the target medical institution in the target medical institution ranking list.

[0091] Step 330 : Determine the authority level feature corresponding to the target medical institution according to the target medical institution ranking list and the ranking of the target medical institution in the target medical institution ranking list.

[0092] It is understandable that different medical institution rankings correspond to different levels of social recognition, and different rankings within the same medical institution ranking also reflect the social recognition of different medical institutions. Therefore, the authority level characteristics corresponding to the target medical institution can be comprehensively determined based on the target medical institution ranking and the target medical institution's ranking within the target medical institution ranking.

[0093] In some embodiments of the present application, step 330 includes: obtaining a weight coefficient corresponding to the target medical institution ranking list; multiplying the weight coefficient corresponding to the target medical institution ranking list by the ranking of the target medical institution in the target medical institution ranking list to obtain a target value; and using the obtained target value as the institutional authority level feature corresponding to the target medical institution.

[0094] In this embodiment, a weight coefficient is pre-configured for each medical institution ranking in the medical institution ranking set. The weight coefficient is used to characterize the social recognition of the medical institution ranking. For example, for the XX medical ranking, the XX top 100 hospital ranking, and the YY top 300 hospital ranking, the weight coefficients of the three rankings can be configured, such as in descending order of weight: XX medical ranking, XX top 100 hospital ranking, and YY top 300 hospital ranking.

[0095] In this embodiment, a weight coefficient is configured for the medical institution ranking list. Thus, after determining the target medical institution to which the source author of the reply text belongs, the weight coefficient corresponding to the target medical institution ranking list is multiplied by the ranking of the target medical institution in the target medical institution ranking list to obtain a target value, and the obtained target value is used as the institutional authority level feature corresponding to the target medical institution, thereby realizing the numerical quantification of the institutional authority level feature corresponding to the target medical institution.

[0096] In some embodiments of the present application, the target medical institution ranking list where the target medical institution is located and the ranking of the target medical institution in the target medical institution ranking list can also be directly used as the institutional authority level features corresponding to the target medical institution.

[0097] In some embodiments of the present application, the question and answer corpus in the question and answer set can also be screened based on the medical institutions included in the medical institution rankings in the medical institution rankings set, and the question and answer corpus that meets the preset conditions can be screened out. For example, the preset condition can be that the target medical institution to which the source author of the reply text belongs is a medical institution included in the medical institution rankings in the medical institution rankings set.

[0098] Please continue reading Figure 2 , step 230, screening the question and answer corpus in the question and answer set according to the authority score of the reply text, and adding the screened question and answer corpus to the candidate question and answer set.

[0099] In one embodiment, a score threshold may be set so that, based on the authority score of each reply text, the question and answer corpus containing reply texts with authority scores higher than the score threshold is filtered out from the question and answer set to construct a candidate question and answer set.

[0100] In one embodiment, the reply texts can be sorted in descending order according to their authority scores, and then a set number of reply texts located before the order are filtered out from the obtained order, and the question and answer corpus containing the filtered reply texts is added to the candidate question and answer set.

[0101] By screening the question and answer corpus in the question and answer set according to the authority score of the reply text, it is possible to ensure that the question and answer corpus containing reply texts with higher authority scores is screened out from the question and answer set, thereby ensuring the scientificity and reliability of the constructed medical training samples.

[0102] Step 240 : Combining the reply texts in the candidate question and answer set to obtain a target popular science article corresponding to the target medical topic.

[0103] Target popular science articles refer to medical popular science articles corresponding to target medical topics.

[0104] In some embodiments of the present application, multiple reply texts in the candidate question and answer set may be combined and spliced ​​in segments to generate a target popular science article corresponding to the target medical topic.

[0105] Step 250: Associating the target popular science article with the target question and answer corpus to obtain a medical training sample, where the target question and answer corpus is a question and answer corpus obtained from the candidate question and answer set; the medical training sample is used to train a medical question and answer model, wherein the medical question and answer model is used to obtain medical answer text for the input medical question text from several medical popular science articles.

[0106] Among them, a question and answer corpus can be randomly selected from the candidate question and answer set as the target question and answer corpus.

[0107] The target popular science article is associated with the target corpus to form a medical training sample of triples, that is, a training sample of article-question-answer.

[0108] Figure 4 is a schematic diagram of constructing a medical training sample according to an embodiment of the present application, such as Figure 4As shown, if after the question and answer corpus is screened through step 230, the candidate question and answer set for the target medical topic includes three groups of question and answer corpora, namely, question and answer corpus 1 (including question text 1 and reply text 1), question and answer corpus 2 (including question text 2 and reply text 2) and question and answer corpus 3 (including question text 3 and reply text 3), on this basis, reply text 1, reply text 2 and reply text 3 in the candidate question and answer set are spliced ​​to obtain the target medical popular science article; and question and answer corpus 2 is selected from the candidate question and answer set as the target question and answer corpus, and the target medical popular science article is associated with question text 2 and reply text 2 in the target question and answer corpus, so as to obtain a triplet of medical training samples.

[0109] In the process of training a medical question-answering model using medical training samples, the question text in the medical training samples is input into the medical question-answering model, and the medical question-answering model outputs a reply text for the question text based on the semantic features of the question text, as well as outputs article information for indicating the medical popular science article from which the output reply text comes; then, the function value of the loss function of the medical question-answering model is calculated based on the reply text and article information output by the medical question-answering model and the reply text and target medical popular science article in the medical training samples; and then, the parameters of the medical question-answering model are adjusted based on the calculated function value of the loss function until the loss function converges.

[0110] In the solution of this application, with the help of a large amount of medical-related question and answer corpora on the Internet medical content platform, the reply texts in the question and answer corpora are generally short. Therefore, multiple question and answer corpora under the same medical topic are used to reversely and automatically construct the target popular science article, that is, the reply texts of the question and answer corpora in the selected candidate question and answer set are combined. On this basis, based on the question and answer relationship between the question text and the reply text in the target question and answer corpus selected from the candidate question and answer set, the target popular science article is associated with the target question and answer corpus to form a medical training sample of the article-question-answer triple, realizing the automatic construction of medical training samples, which does not require manual construction, thereby improving the construction efficiency of medical training samples.

[0111] Moreover, in the scheme of the present application, the authority score of the reply text in the question and answer corpus is calculated based on the institutional authority level characteristics of the medical institution to which the author of the source of the reply text in the question and answer corpus belongs and the medical title characteristics corresponding to the source author of the reply text, and the question and answer corpus in the question and answer set is screened according to the authority score of the reply text to obtain a candidate question and answer set, which can ensure that the authority of the reply text in the candidate question and answer corpus is high, and ensure the scientific nature and reliability of the content of the reply text and target popular science articles in the candidate question and answer set, thereby ensuring the reliability of the obtained medical training samples.

[0112] In the related art, medical training samples constructed based on the manual editing and production of reading comprehension corpus have a single syntactic structure for question and reply texts, which results in the medical question-answering model trained with such artificially constructed medical training samples being unable to adapt to scenarios with diverse medical questions and answers, leading to shortcomings such as insufficient recall and generalization in the trained medical question-answering model. Since different users ask and answer the same question in different ways, the medical training samples constructed in this solution are derived from the question-answering corpus of many users on the internet platform, so that the multiple questions and answers collected are expected to cover a variety of question-asking and reply methods for the same target medical topic, ensuring the syntactic diversity of the question and reply texts in the multiple medical training samples obtained. After training the medical question-answering model with the obtained medical training samples, the medical question-answering model can effectively cope with question texts with diverse syntactic structures.

[0113] In some embodiments of the present application, Figure 5 As shown, before step 240, the method further includes:

[0114] Step 510: Calculate the semantic similarity between any two question and answer corpora in the candidate question and answer set.

[0115] The semantic feature vector of each question-answer corpus may be constructed first, and then the semantic similarity between the two question-answer corpora may be calculated based on the semantic feature vectors of the two question-answer corpora.

[0116] In some embodiments of the present application, since the question and answer corpus includes question text and reply text, the semantic feature vector of the question text and the semantic feature vector of the reply text can be constructed separately first, and then the semantic feature vector of the question text and the semantic feature vector of the reply text in a question and answer corpus are spliced, and the spliced ​​vector is used as the semantic feature vector of the question and answer corpus.

[0117] Step 520 : Filter the question and answer corpus in the candidate question and answer set according to the semantic similarity, and use the reply texts in the filtered question and answer corpus in the candidate question and answer set as objects to be combined.

[0118] The semantic similarity between any two question-and-answer corpora is used to filter out question-and-answer corpora with high semantic similarity from the candidate question-and-answer set. Specifically, a semantic similarity threshold can be set to filter out question-and-answer corpora with semantic similarity exceeding the threshold. This allows the target medical popular science article to be assembled based on the response text from question-and-answer corpora with low semantic similarity, ensuring that the content duplication rate of the target medical popular science article is low.

[0119] In some embodiments of the present application, Figure 6As shown, step 510 includes:

[0120] Step 610 : For the first question and answer corpus and the second question and answer corpus in the candidate question and answer set that require similarity calculation, calculate the first semantic similarity between the question text in the first question and answer corpus and the question text in the second question and answer corpus.

[0121] Step 620: Calculate a second semantic similarity between the reply text in the first question-and-answer corpus and the reply text in the second question-and-answer corpus.

[0122] Step 630: Weight the first semantic similarity and the second semantic similarity to obtain the semantic similarity between the first question-and-answer corpus and the second question-and-answer corpus.

[0123] The first question-answer corpus refers to one of the two question-answer corpora for which semantic similarity calculation is to be performed, and the other question-answer corpus is referred to as the second question-answer corpus.

[0124] The first semantic similarity refers to the semantic similarity between the question text in the first question and answer corpus and the question text in the second question and answer corpus.

[0125] The second semantic similarity refers to the semantic similarity between the reply text in the first question-and-answer corpus and the reply text in the second question-and-answer corpus.

[0126] The weighting coefficients set for the first semantic similarity and the second semantic similarity can be set according to actual needs and are not specifically limited here.

[0127] In this embodiment, the semantic similarity between the two question and answer corpora is obtained by calculating the semantic similarity of the question text and the reply text in the first question and answer corpus and the second question and answer corpus respectively, and weighting the first semantic similarity between the question texts and the second semantic similarity between the reply texts.

[0128] In some embodiments of the present application, Figure 7 As shown, step 610 includes:

[0129] Step 710: Obtain a first word vector corresponding to each word segment in the question text of the first question and answer corpus; and obtain a second word vector corresponding to each word segment in the question text of the second question and answer corpus.

[0130] In step 720, the first word vectors corresponding to the first question and answer corpus are accumulated bit by bit to obtain the question semantic vector corresponding to the question text in the first question and answer corpus; and the second word vectors corresponding to the second question and answer corpus are accumulated bit by bit to obtain the question semantic vector corresponding to the question text in the second question and answer corpus.

[0131] The first word vector refers to the word vector of the word segmentation in the question text of the first question and answer corpus; the second word vector refers to the word vector of the word segmentation in the question text of the second question and answer corpus; and the question semantic vector refers to the semantic vector of the question text.

[0132] In some embodiments of the present application, since the question text in the question-answer corpus is shorter than the reply text, a word vector can be constructed for each word segment in the question text, and a semantic vector of the question text can be constructed based on this.

[0133] In some embodiments of the present application, a word vector for each word segment in the question text can be constructed using a word2vec model, which is a word vector calculation model proposed by Mikolov et al. of Google.

[0134] Figure 8 A schematic diagram of the model structure of the word2vec model is shown, as Figure 8 As shown, the word2vec model includes an input layer, a hidden layer, and an output layer. The input layer is used to receive the text after word segmentation, and the output layer is used to output the word vector of each word segmentation. In this embodiment, before step 710, the word2vec model needs to be trained with medical-related text. After training, the word2vec model can be used to map each word to a vector, namely the word vector of the word.

[0135] In some embodiments of the present application, in step 710, the word2vec model can output a 200-dimensional word vector for each word segment in the question text, and then accumulate the 200-dimensional word vectors of each word segment in the question text bit by bit to obtain a 200-dimensional word vector (i.e., question semantic vector) representation of the question text.

[0136] Step 730: Calculate the distance between the question semantic vector corresponding to the question text in the first question and answer corpus and the question semantic vector corresponding to the question text in the first question and answer corpus to obtain a first semantic similarity between the question text in the first question and answer corpus and the question text in the second question and answer corpus.

[0137] In step 730, the cosine similarity between the question semantic vectors can be calculated based on the question semantic vector corresponding to the question text in the first question-and-answer corpus and the question semantic vector corresponding to the question text in the first question-and-answer corpus. The cosine similarity ranges from 0 to 1, with a value closer to 1 indicating greater similarity and a value closer to 0 indicating less similarity.

[0138] In some embodiments of the present application, Figure 9 As shown, step 620 includes:

[0139] Step 910: Calculate the binary signature corresponding to the reply text in the first question and answer corpus and the binary signature corresponding to the reply text in the second question and answer corpus according to the SimHash algorithm.

[0140] In some embodiments of the present application, instead of calculating the binary signature of the entire reply text, a preset number of sentences (eg, the three longest sentences) may be selected from the reply text to calculate the corresponding binary signature.

[0141] Step 920: Calculate the Hamming distance between the binary signature corresponding to the reply text in the first question-and-answer corpus and the binary signature corresponding to the reply text in the second question-and-answer corpus.

[0142] After obtaining the binary signatures corresponding to the response texts in the first Q&A corpus and the binary signatures corresponding to the response texts in the second Q&A corpus, the two binary signatures are compared. The number of different bits is the Hamming distance between the two binary signatures. A larger Hamming distance indicates a less similar response text, while a smaller Hamming distance indicates a more similar response text.

[0143] Step 930: Determine a second semantic similarity between the reply text in the first question-and-answer corpus and the reply text in the second question-and-answer corpus based on the calculated Hamming distance.

[0144] Figure 10 The flowchart of the medical text retrieval method according to one embodiment of the present application is shown. The method can be executed by a computer device with processing capabilities, such as a server. Figure 10 As shown, the method includes at least steps 1010 to 1040, which are described in detail as follows:

[0145] Step 1010: Obtain a search request, wherein the search request indicates a medical question text.

[0146] Step 1020: Obtain medical answer text for the medical question text from several medical popular science articles through a medical question answering model, wherein the medical question answering model is trained using the medical training samples constructed by the method described in steps 210 to 250.

[0147] In step 1020, the medical question-answering model needs to locate medical popular science articles that contain answers to the questions indicated by the medical question text, and on the other hand, it needs to obtain the content representing the answers to the questions indicated by the medical question text (i.e., medical answer text) in the located medical popular science articles.

[0148] Step 1030: Use the medical answer text as a text summary of the medical popular science article where the medical answer text is located.

[0149] Step 1040: Return the address information and text summary of the medical popular science article where the medical answer text is located to the initiator of the search request.

[0150] The address information of the medical popular science article may be a link to the medical popular science article. The user may enter the page of the medical popular science article based on the address information of the medical popular science article to read the medical popular science article.

[0151] Because the medical answer text serves as a text summary of the medical popular science article containing the medical answer text, once the search results are returned to the initiator of the search request, the text summary of the medical popular science article containing the medical answer text, i.e., the medical answer text, can be directly displayed on the search results page displayed on the terminal of the initiator of the search request. This allows users to directly obtain the answer to the question indicated by the medical question text on the search results page, without having to read the entire medical popular science article containing the medical answer text to find the answer themselves.

[0152] Therefore, through the retrieval method of the present application, it is possible to automatically locate and obtain the medical answer text representing the answer to the question indicated by the medical question text in the medical popular science article, without the user having to find the answer by reading the popular science article.

[0153] The following describes an embodiment of the device of the present application, which can be used to perform the method described in the above embodiment of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the above method embodiment of the present application.

[0154] Figure 11 is a block diagram of an apparatus for constructing medical training samples according to an embodiment. Figure 11 As shown, the device for constructing medical training samples includes:

[0155] A question and answer set acquisition module 1110 is configured to acquire a question and answer set, wherein the question and answer set includes a plurality of question and answer corpora for a target medical topic, wherein the question and answer corpora include question texts and answer texts;

[0156] The authority score calculation module 1120 is used to calculate the authority score of the reply text based on the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text;

[0157] A screening module 1130 is configured to screen question and answer materials in the question and answer set according to the authority score of the reply text, and add the screened question and answer materials to the candidate question and answer set;

[0158] A combining module 1140 is configured to combine the reply texts in the candidate question and answer set to obtain a target popular science article corresponding to the target medical topic;

[0159] The association module 1150 is used to associate the target popular science article with the target question and answer corpus to obtain a medical training sample, where the target question and answer corpus is a question and answer corpus obtained from the candidate question and answer set; the medical training sample is used to train a medical question and answer model, wherein the medical question and answer model is used to obtain medical answer text for the input medical question text from a number of medical popular science articles.

[0160] In some embodiments of the present application, the authority score calculation module 1120 includes:

[0161] An institution authority level feature acquisition unit, configured to acquire the institution authority level feature of the medical institution to which the source author of the reply text belongs;

[0162] and a medical title feature acquisition unit, configured to acquire the medical title feature corresponding to the source author of the reply text;

[0163] A weighting unit is used to weight the numerical value mapped by the institutional authority level feature and the numerical value mapped by the medical title feature to obtain the authority score corresponding to the reply text.

[0164] In some embodiments of the present application, the institution authority level feature acquisition unit includes:

[0165] An institution information acquisition unit, configured to acquire the institution information corresponding to the reply text, wherein the institution information indicates a target medical institution to which the source author of the reply text belongs;

[0166] a ranking determination unit, configured to determine the ranking of the target medical institution in a target medical institution ranking list;

[0167] The institution authority level characteristic determination unit is used to determine the institution authority level characteristic corresponding to the target medical institution according to the target medical institution ranking list where the target medical institution is located and the ranking of the target medical institution in the target medical institution ranking list.

[0168] In some embodiments of the present application, the institution authority level characteristic determination unit includes:

[0169] A weight coefficient obtaining unit, configured to obtain a weight coefficient corresponding to the target medical institution ranking list;

[0170] a multiplication unit, configured to multiply the weight coefficient corresponding to the target medical institution ranking list by the ranking of the target medical institution in the target medical institution ranking list to obtain a target value;

[0171] The first determining unit is configured to use the obtained target value as an institutional authority level feature corresponding to the target medical institution.

[0172] In some embodiments of the present application, the apparatus for constructing medical training samples further includes:

[0173] A semantic similarity calculation module, used to calculate the semantic similarity between any two question and answer corpora in the candidate question and answer set;

[0174] A filtering module is used to filter the question and answer corpus in the candidate question and answer set according to the semantic similarity, and use the reply texts in the question and answer corpus in the candidate question and answer set after filtering as objects to be combined.

[0175] In some embodiments of the present application, the semantic similarity calculation module includes:

[0176] A first semantic similarity calculation unit is configured to calculate, for a first question and answer corpus and a second question and answer corpus in the candidate question and answer set, a first semantic similarity between a question text in the first question and answer corpus and a question text in the second question and answer corpus for which similarity calculation is required;

[0177] A second semantic similarity calculation unit, configured to calculate a second semantic similarity between a reply text in the first question-and-answer corpus and a reply text in the second question-and-answer corpus;

[0178] A semantic similarity calculation unit is used to weight the first semantic similarity and the second semantic similarity to obtain the semantic similarity between the first question and answer corpus and the second question and answer corpus.

[0179] In some embodiments of the present application, the first semantic similarity calculation unit includes:

[0180] A first word vector acquisition unit is used to obtain a first word vector corresponding to each word segment in the question text of the first question and answer corpus;

[0181] A second word vector acquisition unit is used to obtain a second word vector corresponding to each word in the question text of the second question and answer corpus;

[0182] a first accumulating unit, configured to accumulate the first word vectors corresponding to the first question-answer corpus bit by bit to obtain a question semantic vector corresponding to the question text in the first question-answer corpus;

[0183] A second accumulating unit, configured to accumulate the second word vectors corresponding to the second question-answer corpus bit by bit to obtain a question semantic vector corresponding to the question text in the second question-answer corpus;

[0184] A distance calculation unit is used to calculate the distance between the question semantic vector corresponding to the question text in the first question and answer corpus and the question semantic vector corresponding to the question text in the first question and answer corpus, and obtain the first semantic similarity between the question text in the first question and answer corpus and the question text in the second question and answer corpus.

[0185] In some embodiments of the present application, the second semantic similarity calculation unit includes:

[0186] a binary signature calculation unit, configured to calculate the binary signature corresponding to the reply text in the first question and answer corpus and the binary signature corresponding to the reply text in the second question and answer corpus according to the SimHash algorithm;

[0187] A Hamming distance calculation unit, configured to calculate a Hamming distance between a binary signature corresponding to a reply text in the first question-and-answer corpus and a binary signature corresponding to a reply text in the second question-and-answer corpus;

[0188] A second semantic similarity determination unit is configured to determine a second semantic similarity between the reply text in the first question-and-answer corpus and the reply text in the second question-and-answer corpus according to the calculated Hamming distance.

[0189] Figure 12 is a block diagram of a medical text retrieval device according to an embodiment. Figure 12 As shown, the medical text retrieval device includes:

[0190] A search request acquisition module 1210 is configured to acquire a search request, wherein the search request indicates a medical question text;

[0191] The medical answer text acquisition module 1220 is used to acquire the medical answer text for the medical question text from a plurality of medical popular science articles using the medical question-answering model;

[0192] A text summary determination module 1230 is configured to use the medical answer text as a text summary of the medical popular science article corresponding to the medical answer text;

[0193] The return module 1240 is used to return the address information and text summary of the medical popular science article where the medical answer text is located to the initiator of the search request.

[0194] Figure 13 A schematic diagram of the structure of a computer system suitable for implementing an electronic device according to an embodiment of the present application is shown.

[0195] It should be noted that Figure 13 The computer system 1300 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present application.

[0196] like Figure 13 As shown, computer system 1300 includes a central processing unit (CPU) 1301, which can perform various appropriate actions and processes, such as executing the methods described in the above embodiments, based on programs stored in read-only memory (ROM) 1302 or programs loaded from storage 1308 into random access memory (RAM) 1303. RAM 1303 also stores various programs and data required for system operation. CPU 1301, ROM 1302, and RAM 1303 are interconnected via bus 1304. An input / output (I / O) interface 1305 is also connected to bus 1304.

[0197] The following components are connected to the I / O interface 1305: an input section 1306 including a keyboard, mouse, and the like; an output section 1307 including devices such as a cathode ray tube (CRT), a liquid crystal display (LCD), and speakers; a storage section 1308 including a hard disk; and a communication section 1309 including a network interface card such as a LAN (Local Area Network) card or a modem. The communication section 1309 performs communication processing via a network such as the Internet. A drive 1310 is also connected to the I / O interface 1305 as needed. Removable media 1311, such as a magnetic disk, optical disk, magneto-optical disk, or semiconductor memory, is installed in the drive 1310 as needed, allowing computer programs read from the media to be installed in the storage section 1308 as needed.

[0198] In particular, according to embodiments of the present application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for executing the methods illustrated in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via the communication section 1309 and / or installed from removable media 1311. When executed by the central processing unit (CPU) 1301, the computer program performs the various functions defined in the system of the present application.

[0199] It should be noted that the computer-readable medium described in the embodiments of this application may be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal transmitted in baseband or as part of a carrier wave, which carries computer-readable program code. Such propagated data signals may take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium may also be any computer-readable medium other than a computer-readable storage medium that can transmit, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer-readable medium may be transmitted using any suitable medium, including but not limited to wireless, wired, or any suitable combination thereof.

[0200] The flowcharts and block diagrams in the accompanying drawings illustrate the possible implementation architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present application. Among them, each box in the flowchart or block diagram can represent a module, program segment, or part of the code, and the above-mentioned module, program segment, or part of the code contains one or more executable instructions for implementing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in an order different from that marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flowchart, and the combination of boxes in the block diagram or flowchart, can be implemented with a dedicated hardware-based system that performs the specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0201] The units involved in the embodiments described in this application may be implemented by software or hardware, and the units described may also be set in a processor. In some cases, the names of these units do not constitute limitations on the units themselves.

[0202] As another aspect, the present application further provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments, or may exist independently without being incorporated into the electronic device. The computer-readable storage medium carries computer-readable instructions, and when the computer-readable instructions are executed by a processor, the method of any of the above embodiments is implemented.

[0203] According to one aspect of the present application, an electronic device is also provided, which includes: a processor; a memory, wherein computer-readable instructions are stored in the memory, and when the computer-readable instructions are executed by the processor, the method in any of the above embodiments is implemented.

[0204] According to one aspect of an embodiment of the present application, a computer program product or computer program is provided, the computer program product or computer program including 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, causing the computer device to perform the method of any of the above embodiments.

[0205] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiment of the application, the features and functions of two or more modules or units described above can be concretized in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be concretized.

[0206] Through the description of the above embodiments, it is easy for those skilled in the art to understand that the example embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solution according to the embodiments of the present application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, and includes a number of instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present application.

[0207] Those skilled in the art will readily conceive of other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present application that follow the general principles of this application and include common knowledge or customary techniques in the art that are not disclosed herein.

[0208] It should be understood that the present application is not limited to the exact structures described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.

Claims

1. A method for constructing medical training samples, characterized in that: include: Obtaining a question and answer set, wherein the question and answer set includes a plurality of question and answer corpora for a target medical topic, wherein the question and answer corpora include question texts and answer texts; Calculating the authority score of the reply text based on the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text; Screening the question and answer corpus in the question and answer set according to the authority score of the reply text, and adding the screened question and answer corpus to the candidate question and answer set; Calculating the semantic similarity between any two question-answer corpora in the candidate question-answer set; According to the semantic similarity, the question and answer corpus with a semantic similarity higher than a semantic similarity threshold in the candidate question and answer set is filtered out; Combining the response texts in the filtered candidate question and answer set to obtain a target popular science article corresponding to the target medical topic; The target popular science article is associated with a target question and answer corpus to obtain a medical training sample, where the target question and answer corpus is a question and answer corpus obtained from the candidate question and answer set; the medical training sample is used to train a medical question and answer model, wherein the medical question and answer model is used to obtain medical answer text for an input medical question text from a number of medical popular science articles.

2. The method according to claim 1, characterized in that Calculating the authority score of the reply text according to the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text includes: Obtaining the authority level characteristics of the medical institution to which the source author of the reply text belongs, and obtaining the medical title characteristics corresponding to the source author of the reply text; The numerical value mapped by the institutional authority level feature and the numerical value mapped by the medical professional title feature are weighted to obtain the authority score corresponding to the reply text.

3. The method according to claim 2, characterized in that The authority level characteristics of the medical institution to which the source author of the reply text belongs are obtained, including: Obtaining the institution information corresponding to the reply text, wherein the institution information indicates the target medical institution to which the source author of the reply text belongs; Determining the ranking of the target medical institution in a list of target medical institutions; According to the target medical institution ranking list where the target medical institution is located and the ranking of the target medical institution in the target medical institution ranking list, the institutional authority level characteristics corresponding to the target medical institution are determined.

4. The method according to claim 3, characterized in that The determining of the authority level feature corresponding to the target medical institution according to the target medical institution ranking list and the ranking of the target medical institution in the target medical institution ranking list includes: Obtaining a weight coefficient corresponding to the target medical institution ranking list; Multiplying the weight coefficient corresponding to the target medical institution ranking list by the ranking of the target medical institution in the target medical institution ranking list to obtain a target value; The obtained target value is used as the institutional authority level feature corresponding to the target medical institution.

5. The method according to claim 1, wherein Calculating the semantic similarity between any two question-answer corpora in the candidate question-answer set includes: For a first question and answer corpus and a second question and answer corpus in the candidate question and answer set that require similarity calculation, calculate a first semantic similarity between the question text in the first question and answer corpus and the question text in the second question and answer corpus, and calculate a second semantic similarity between the reply text in the first question and answer corpus and the reply text in the second question and answer corpus; The first semantic similarity and the second semantic similarity are weighted to obtain the semantic similarity between the first question and answer corpus and the second question and answer corpus.

6. The method according to claim 5, characterized in that The calculating a first semantic similarity between the question text in the first question-answer corpus and the question text in the second question-answer corpus includes: Obtain a first word vector corresponding to each word segment in the question text of the first question and answer corpus; and obtain a second word vector corresponding to each word segment in the question text of the second question and answer corpus; Accumulate the first word vectors corresponding to the first question and answer corpus bit by bit to obtain the question semantic vector corresponding to the question text in the first question and answer corpus; and accumulate the second word vectors corresponding to the second question and answer corpus bit by bit to obtain the question semantic vector corresponding to the question text in the second question and answer corpus; Calculate the distance between the question semantic vector corresponding to the question text in the first question and answer corpus and the question semantic vector corresponding to the question text in the first question and answer corpus to obtain a first semantic similarity between the question text in the first question and answer corpus and the question text in the second question and answer corpus.

7. The method according to claim 5, characterized in that The calculating the second semantic similarity between the reply text in the first question-and-answer corpus and the reply text in the second question-and-answer corpus includes: Calculate the binary signature corresponding to the reply text in the first question and answer corpus and the binary signature corresponding to the reply text in the second question and answer corpus according to the SimHash algorithm; Calculating the Hamming distance between the binary signature corresponding to the reply text in the first question-and-answer corpus and the binary signature corresponding to the reply text in the second question-and-answer corpus; A second semantic similarity between the reply text in the first question-and-answer corpus and the reply text in the second question-and-answer corpus is determined based on the calculated Hamming distance.

8. A medical text retrieval method, characterized in that: include: Obtaining a search request, wherein the search request indicates a medical question text; Obtaining medical answer texts for the medical question texts from a number of medical popular science articles using a medical question-answering model; the medical question-answering model is trained using medical training samples constructed using the method according to any one of claims 1 to 7; Using the medical answer text as a text summary of the medical popular science article corresponding to the medical answer text; The address information and text summary of the medical popular science article where the medical answer text is located are returned to the initiator of the search request.

9. A device for constructing medical training samples, characterized in that: include: A question and answer set acquisition module is used to acquire a question and answer set, wherein the question and answer set includes a plurality of question and answer corpora for a target medical topic, and the question and answer corpora include question texts and answer texts; An authority score calculation module, configured to calculate the authority score of the reply text based on the authority level characteristics of the medical institution to which the source author of the reply text belongs and the medical title characteristics corresponding to the source author of the reply text; A screening module, configured to screen question and answer corpora in the question and answer set according to the authority score of the reply text, and add the screened question and answer corpora to the candidate question and answer set; Calculating the semantic similarity between any two question-answer corpora in the candidate question-answer set; According to the semantic similarity, the question and answer corpus with a semantic similarity higher than a semantic similarity threshold in the candidate question and answer set is filtered out; A combining module, configured to combine the reply texts in the filtered candidate question and answer set to obtain a target popular science article corresponding to the target medical topic; An association module is used to associate the target popular science article with a target question and answer corpus to obtain a medical training sample, where the target question and answer corpus is a question and answer corpus obtained from the candidate question and answer set; the medical training sample is used to train a medical question and answer model, wherein the medical question and answer model is used to obtain medical answer text for an input medical question text from a number of medical popular science articles.

10. A medical text retrieval device, characterized in that: include: A retrieval request post-module is used to obtain a retrieval request, wherein the retrieval request indicates a medical question text; a medical answer text acquisition module, configured to acquire medical answer texts for the medical question texts from a plurality of medical popular science articles using a medical question-answering model; the medical question-answering model is trained using medical training samples constructed using the method according to any one of claims 1 to 7; a text summary determination module, configured to use the medical answer text as a text summary of the medical popular science article corresponding to the medical answer text; The return module is used to return the address information and text summary of the medical popular science article where the medical answer text is located to the initiator of the search request.

11. An electronic device, characterized in that: include: processor; A memory having computer-readable instructions stored thereon, wherein when the computer-readable instructions are executed by the processor, the method for constructing a medical training sample according to any one of claims 1 to 7 is implemented; or the method for retrieving medical text according to claim 8 is implemented.

12. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer-readable instructions. When the computer-readable instructions are executed by a processor, the method for constructing medical training samples according to any one of claims 1 to 7 is implemented; or the method for retrieving medical texts according to claim 8 is implemented.

Images