Entity recommendation method, device, storage medium and equipment

Abstract

The embodiment of the present application provides an entity recommendation method, device, storage medium and equipment. After obtaining the input text, the input text is first semantically understood to determine the search intent and seed entity words. When the search intent matches the seed entity words and the search intent belongs to the precise recommendation intent, the entity relationship is determined based on the search intent, and the seed entity words are mapped to the standard entity words in the medical knowledge graph; the entity words in the medical knowledge graph that have an entity relationship with the standard entity words are recommended as recommended entities. In other words, in the embodiment of the present application, entity recommendations are made based on the medical knowledge graph only when the search intent matches the seed entity words and the search intent belongs to the precise recommendation intent, thereby ensuring the accuracy of medical entity recommendations.

Description

Technical Field

[0001] The present application relates to the field of information processing technology, and in particular to an entity recommendation method, device, storage medium and equipment. Background Art

[0002] Entity recommendation is a common function of search engines. It can use the Named Entities Recognition (NER) algorithm to extract entity words from the user input text, and then recommend other entity words related to the extracted entity words that are most likely to meet the user's needs and their information links, eliminating the trouble of users looking for answers in the search results (i.e., search results obtained based on the user input text), greatly improving the user experience.

[0003] However, the inventors found in the process of implementing this application that the existing entity recommendation method only considers the relevance between the entity word to be recommended and the search entity (i.e., the entity word extracted from the search content input by the user). In the medical field, which requires very high accuracy for entity recommendation, the recommendation results of this recommendation method are too vague and the risk of recommendation errors is very high. Therefore, how to improve the accuracy of medical entity recommendation has become a technical problem that needs to be solved urgently. Summary of the invention

[0004] In view of this, embodiments of the present application provide an entity recommendation method, apparatus, storage medium, and device to improve the accuracy of medical entity recommendations.

[0005] To achieve the above objectives, the present application provides the following technical solutions:

[0006] In a first aspect, a method for recommending an entity is provided, comprising:

[0007] Get input text;

[0008] Performing semantic understanding on the input text to determine search intent and extract seed entity words;

[0009] If the search intent matches the seed entity word, and the search intent belongs to the precise recommendation intent, then determining the entity relationship based on the search intent, and mapping the seed entity word to a standard entity word in the medical knowledge graph;

[0010] The entity words in the medical knowledge graph that have the entity relationship with the standard entity words are recommended as recommended entities.

[0011] In a second aspect, an entity recommendation device is provided, comprising:

[0012] The acquisition module is used to obtain the input text;

[0013] An understanding module, used for performing semantic understanding on the input text to determine the search intent and extract seed entity words;

[0014] A determination module, configured to determine an entity relationship based on the search intent and map the seed entity word to a standard entity word in a medical knowledge graph if the search intent matches the seed entity word and the search intent belongs to an accurate recommendation intent;

[0015] A recommendation module is used to recommend entity words in the medical knowledge graph that have the entity relationship with the standard entity words as recommended entities.

[0016] In a third aspect, an electronic device is provided, including a memory and a processor;

[0017] The memory is used to store programs;

[0018] The processor is used to execute the program to implement each step of the entity recommendation method as described in any one of the above items.

[0019] In a fourth aspect, a readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps of the entity recommendation method as described in any one of the above items are implemented.

[0020] The entity recommendation method, device, storage medium and equipment provided in the embodiment of the present application, after obtaining the input text, first perform semantic understanding on the input text, determine the search intent and seed entity words, and only when the search intent matches the seed entity words and the search intent belongs to the precise recommendation intent, determine the entity relationship based on the search intent, and map the seed entity words to the standard entity words in the medical knowledge graph; recommend the entity words in the medical knowledge graph that have an entity relationship with the standard entity words as recommended entities. That is to say, in the embodiment of the present application, only when the search intent matches the seed entity words and the search intent belongs to the precise recommendation intent, entity recommendations are made based on the medical knowledge graph, ensuring the accuracy of medical entity recommendations. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0022] Figure 1 A flowchart for implementing the entity recommendation method provided in an embodiment of the present application;

[0023] Figure 2 A schematic diagram of a structure of a semantic understanding model provided in an embodiment of the present application;

[0024] Figure 3 A partial example diagram of the medical knowledge graph provided in the embodiment of the present application;

[0025] Figure 4 A flowchart for implementing the mapping of seed entity words to standard entity words in a medical knowledge graph provided in an embodiment of the present application;

[0026] Figure 5 A program logic diagram of the entity recommendation method provided in an embodiment of the present application;

[0027] Figure 6 A schematic diagram of a structure of an entity recommendation device provided in an embodiment of the present application;

[0028] Figure 7 This is an example diagram of a hardware structure block diagram of an electronic device provided in an embodiment of the present application. DETAILED DESCRIPTION

[0029] The following will be combined with the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0030] See also Figure 1 , Figure 1 A flowchart for implementing the entity recommendation method provided in the embodiment of the present application may include:

[0031] Step S11: Get input text.

[0032] When a user wants to know relevant information in the medical field, he can enter a search string in the medical field in the search engine. For example, when the user or his friend is not feeling well, he may ask the search engine for information related to the symptoms, such as what is wrong with a bump on the head. Based on this, the above input text can be the search string entered by the user into the search engine.

[0033] The search engine may be a dedicated search engine in the medical field or a general search engine.

[0034] Step S12: Perform semantic understanding on the input text to determine the search intent and extract seed entity words.

[0035] In an embodiment of the present application, in addition to determining the user's search intent through input text, entity words are also extracted from the input text. Since the input text is medical-related input text, the search intent is the search intent in the medical field, and the extracted entity words are also entity words in the medical field.

[0036] Since the entity words extracted from the input text are a basis for obtaining the entity words to be recommended, for the sake of distinction, the entity words extracted from the input text are recorded as seed entity words.

[0037] The pre-trained semantic understanding model can be used to perform semantic understanding on the input text to obtain the search intent and seed entity words.

[0038] Step S13: If the search intent matches the seed entity word and the search intent belongs to the precise recommendation intent, the entity relationship is determined based on the search intent, and the seed entity word is mapped to the standard entity word in the medical knowledge graph.

[0039] Based on the search intent, one entity relationship or multiple entity relationships can be determined. The specific number is determined by the entity relationships defined in the medical knowledge graph. If one entity relationship is defined in the medical knowledge graph for the search intent, one entity relationship is determined based on the search intent. If multiple entity relationships are defined in the medical knowledge graph for the search intent, multiple entity relationships are determined based on the search intent.

[0040] Optionally, the corresponding relationship between mutually matching search intents and entity words, as well as the precise recommendation intent, may be predefined, that is, which search intent and which entity word are matched, and which search intent belongs to the precise recommendation intent, may be predefined.

[0041] For example, if the search intent is to ask about symptoms and the entity word is a hospital name, it is obvious that asking about symptoms and the hospital name are logically conflicting. Therefore, asking about symptoms and the hospital name do not match, and no corresponding relationship between asking about symptoms and the hospital name can be established. If the search intent is to ask about symptoms and the entity word is a disease name, since there are usually certain symptoms when you are sick, asking about symptoms and the disease name do match, and a corresponding relationship between asking about symptoms and the disease name can be established. For another example, "find a hospital" is an accurate recommendation intent, and the hospital name needs to be directly and accurately returned. However, "breast cancer" does not have a clear intent, so "breast cancer" does not belong to the accurate recommendation intent.

[0042] If a corresponding relationship is defined between the search intent and the seed entity (i.e., the seed entity word is an entity word with a predefined corresponding relationship with the search intent), it can be determined that the search intent matches the seed entity word, otherwise, it is determined that the search intent does not match the seed entity word. If the search intent belongs to the predefined precise recommendation intent, the search intent belongs to the precise recommendation intent.

[0043] If the search intent does not match the seed entity word, error handling is required, that is, no entity recommendation is performed, but the traditional search method is executed to serve the user.

[0044] If the search intent matches the seed entity word, but the seed entity is not an exact recommendation intent, fuzzy processing is performed, that is, it is assumed that there are multiple intents related to the seed entity word at the same time. For example, if the input "breast cancer" does not have a clear intent, entity recommendations are made based on multiple common intents such as "find a doctor" and "ask about a disease".

[0045] Optionally, determining the entity relationship based on the search intent may include: determining the entity relationship corresponding to the search intent according to the pre-established correspondence between the search intent and the entity relationship. For example, assuming that the intent is "find a doctor" and the entity word is "breast cancer", the entity relationship may be "good at doctor", and the entity word with "good at doctor" in common with the entity word "breast cancer" may be "doctor's name", that is, the doctor with "doctor's name" is a doctor who is good at treating "breast cancer".

[0046] Optionally, mapping the seed entity words to standard entity words in the medical knowledge graph may include: mapping the seed entity words to standard entity words in the medical knowledge graph according to a preset mapping relationship (eg, a mapping table).

[0047] In order to correctly match the seed entity to the node in the medical knowledge graph, a vocabulary alias mapping module is pre-established, which is used to map the colloquial entity words to the standard entity words in the knowledge graph. The specific method is to use the colloquial entity words mined from the medical corpus spoken by the user as the initial vocabulary through entity linking (i.e. the mapping relationship between colloquial entities and standardized entities), and then ensure the correctness of the mapping through professional manual review.

[0048] In addition, if the search engine is not a dedicated search engine in the medical field, but a general search engine, then before determining whether the search intent matches the seed entity word and whether the search intent belongs to the precise recommendation intention, you can first determine whether the search intent or the seed entity word is the search intent or the seed entity word in the medical field. If the search intent is the search intent in the medical field, and / or the seed entity word is an entity word in the medical field, then determine whether the search intent matches the seed entity word and whether the search intent belongs to the precise recommendation intention.

[0049] Step S14: recommending entity words in the medical knowledge graph that have the above-mentioned entity relationship with the standard entity words as recommended entities.

[0050] Of course, in addition to entity recommendations, search results (usually web page links) obtained based on traditional search methods can also be returned to users.

[0051] The entity recommendation method provided in the embodiment of the present application, after obtaining the input text, first performs semantic understanding on the input text to determine the search intent and seed entity words. Only when the search intent matches the seed entity words and the search intent belongs to the precise recommendation intent, entity recommendation is performed based on the medical knowledge graph, thereby ensuring the accuracy of medical entity recommendation.

[0052] In an optional embodiment, a structural diagram of the above semantic understanding model is as follows: Figure 2 As shown, it may include:

[0053] Bidirectional Encoder Representation from Transformers (BERT) network;

[0054] A feed-forward neural network connected to the first output unit of the BERT network, the feed-forward neural network is used to predict search intent;

[0055] A conditional random field (CRF) model connected to multiple second output units of the BERT network is used to extract seed entity words in the input text.

[0056] Among them, the input of the BERT network is also the input of the semantic understanding model. During the training phase, the input of the semantic understanding model is sample data labeled with intent labels and slot labels. Figure 2 In the example, the query is the input sample data, Tok i (i=1, 2, ..., N) represents the i-th word in the input query, CLS and SEP are preset identifiers. Every time the BERT network receives an input query, it will add the identifier CLS before the input query and the identifier SEP at the end of the input query, and process the text after adding the identifiers to obtain the hidden vector expression C corresponding to the identifier CLS, the hidden vector expression S corresponding to the identifier SEP, and the hidden vector expression T corresponding to each word i . The BERT network has an output unit for each word, an output unit for the CLS marker, and an output unit for the SEP marker. Optionally, the triage model can be trained using a fine-tuning method based on Google's open source BERT Chinese pre-trained model. Of course, other methods can also be used to train the triage model, which is not specifically limited here.

[0057] In an embodiment of the present application, the feedforward neural network is connected to the output unit (i.e., the first output unit) corresponding to the identifier CLS of the BERT network, that is, the feedforward neural network processes the hidden layer vector expression corresponding to the identifier CLS, and uses a classification function (e.g., a softmax function) to process the processing result of the feedforward neural network to obtain and output the intent label.

[0058] The CRF model is connected to the output unit corresponding to each character in the input query (the output unit corresponding to each character is a second output unit), that is, the CRF model processes the hidden vector expression corresponding to each character in the input query to obtain the word slot label corresponding to each character. The word slot label corresponding to a character indicates which entity word a character belongs to.

[0059] After determining the word slot labels of each word, adjacent words with the same word slot labels are extracted as seed entity words. In some cases, different entity words may overlap, for example, one entity word is part of another entity word, such as "breast" is part of "breast cancer", then the longest word is selected as the seed entity word, that is, "breast cancer" is selected as the seed entity word.

[0060] Of course, in addition to the above-mentioned parts, the semantic understanding model also includes a loss function (not shown in the figure). The loss function calculates the difference between the prediction result and the label of the labeled data, so as to iteratively optimize the model according to the difference.

[0061] In the embodiment of the present application, a powerful BERT intent-slot joint model is used for query understanding, which has very high accuracy and recall rate, further improving the accuracy of entity recommendation.

[0062] In an optional embodiment, the search intention may include at least any one of the following: an intention to ask about related diseases, an intention to ask about symptoms, and an intention to find a doctor.

[0063] The intention to ask about related diseases usually means that the user knows a specific disease and wants to query information related to the disease. For example, if the user inputs the text: What should I pay attention to for skin allergies? or how to avoid skin allergies, etc., the search intention is: the intention to ask about related diseases, and the related disease here refers to skin allergies.

[0064] Symptoms are usually when users do not know what the disease is, but only know the symptoms. They may use the symptom information to find the cause or the disease. For example, if the user inputs the text: What is the cause of a red pimple on the head?, the search intent is symptoms, and the symptom here refers to a red pimple on the head.

[0065] The intention of finding a doctor usually means that the user wants to go to the hospital for treatment. At this time, the user may search for a suitable doctor by entering the disease name and / or symptom information. For example, the user enters the text: the most professional doctor for treating glaucoma.

[0066] In the embodiment of the present application, in order to adapt to the entity recommendation in the medical field, the recommendation strategy is deeply customized in the medical field. Specifically, it can be reflected as follows:

[0067] Optionally, when the search intent is to ask about related diseases, determining entity relationships based on the search intent may include:

[0068] Determine the pathological correlation and site correlation based on search intent.

[0069] like Figure 3 As shown, it is a partial example diagram of the medical knowledge graph provided in the embodiment of the present application. In the embodiment of the present application, for the question-related disease intent, a dual strategy of site-related and pathology-related is adopted. For example, for the disease entity word "infectious rash", the entity word "skin allergy" can be recommended based on site-relatedness, and the entity word "mycoplasma pneumonia" can be recommended based on pathology-relatedness. In the medical knowledge graph, site-relatedness can be converted into a two-degree search strategy, i.e., seed entity->site, then site->recommended entity. The recommendation strategy for pathology-relatedness is the same, i.e., seed entity->pathology, then pathology->recommended entity.

[0070] Optionally, when the search intent is to ask about symptoms and the seed entity is an entity word of the symptom, the above-mentioned flowchart for mapping the seed entity word to a standard entity word in the medical knowledge graph is as follows: Figure 4 The shown may include:

[0071] Step S41: Select a candidate entity word from the standard entity words in the medical knowledge graph, and the candidate entity word has the greatest similarity with the seed entity word.

[0072] Step S42: Match the part entity words in the candidate entity words with the part entity words in the seed entity words. If the matching result meets the matching condition, map the seed entity words to the candidate entity words.

[0073] Since the colloquial symptom entity words are very complicated, it is unrealistic to use a pre-established mapping relationship (such as a mapping table) to process all symptom entity words, that is, it is impossible to exhaustively list all symptom entity words, which easily leads to the situation that the mapping cannot be performed, resulting in a low entity word recall rate, thereby reducing the accuracy of entity recommendation. In order to further improve the accuracy of entity recommendation, in the embodiment of the present application, a hybrid strategy of similarity strategy and strong knowledge constraint is adopted to perform entity linking. Specifically,

[0074] Fasttext average word vector can be used to calculate the cosine similarity between the standard entity word and the seed entity word, and a standard entity word that is most similar to the seed entity word is selected as the candidate entity word according to the cosine similarity. Then, the part entity word is extracted from the candidate entity word (for the convenience of distinction, it is recorded as the first part entity word), and the part entity word is extracted from the seed entity word (for the convenience of distinction, it is recorded as the second part entity word), and the extracted first part entity word and the second part entity word are matched. If the matching result meets the matching condition (for example, the matching degree is greater than the threshold), it means that the candidate entity word is correct, and the seed entity word is mapped to the candidate entity word. If the matching result does not meet the matching condition, the candidate entity is considered to be wrong, and no entity word recommendation is performed, and only the traditional search method is executed. For example, assuming that the seed entity word is "a lump on the head" and the candidate entity word is "a lump on the foot", since the part entity word in "a lump on the head" is "head", and the part entity word in "a lump on the foot" is "foot", the part entity words of the two are different, so the candidate entity word "a lump on the foot" is wrong, and "a lump on the head" cannot be mapped to "a lump on the foot". If the candidate entity word is "a bump on the head", since the part entity word in "a bump on the head" is "head", the part entity word in "a bump on the head" is also "head", and the two part entity words are the same, the candidate entity word "a bump on the head" is correct, and "a bump on the head" can be mapped to "a bump on the head".

[0075] That is to say, in the embodiment of the present application, if the search intent is to ask about symptoms, a hybrid strategy of similarity strategy and strong knowledge constraint can be used to map seed entity words to standard entity words in the medical knowledge graph; if the search intent is not to ask about symptoms, a preset mapping relationship can be used to map seed entity words to standard entity words in the medical knowledge graph. This improves the accuracy of entity recommendation while ensuring data processing efficiency.

[0076] In an optional embodiment, when the search intent is to find a doctor, and the seed entity words include symptom entity words but do not include disease entity words, an implementation method for determining entity relationships based on the search intent may be:

[0077] Based on the symptom-asking intention, the first entity relationship is determined, and the disease entity word having the first entity relationship with the seed entity (specifically the symptom entity word in the seed entity word) is determined in the medical knowledge graph. The first entity relationship is the entity relationship corresponding to the symptom-asking intention. The first entity relationship represents which disease the symptom described by the symptom entity word belongs to.

[0078] Based on the intention of finding a doctor, a second entity relationship is determined. The second entity relationship may be a disease treatment-related relationship, which may refer to the doctor's experience, expertise, popularity, title, etc. in treating the disease.

[0079] Accordingly, entity words in the medical knowledge graph that have entity relationships with standard entity words are recommended as recommended entities, including:

[0080] The doctor name entity word that has a second entity relationship with the disease entity word in the medical knowledge graph is recommended as a recommended entity.

[0081] In the embodiment of the present application, for the case where the search intent is to find a doctor, and the seed entity words only have symptom entity words but no disease entity words, it is decomposed into two stages: "ask about symptoms" and "find a doctor". First, the disease entity words are determined according to the "ask about symptoms" intent strategy, and then, suitable doctors are recommended based on the disease entity words.

[0082] In summary, in the embodiments of the present application, in order to ensure the accuracy of entity recommendations, a method of professional vocabulary mapping is selected in the entity linking stage, and a deep customization strategy is adopted in the entity recommendation stage. This can handle multiple search intentions more effectively and accurately, improve the accuracy and usefulness of medical entity recommendations, enhance explainability, and reduce risks.

[0083] Of course, the above only exemplifies three types of search intents that are deeply customized in the medical field. In the embodiments of the present application, the search intent is not limited to the above three types, but may also include other search intents, such as finding a hospital, asking about treatment drugs, etc., which will not be illustrated one by one here.

[0084] See also Figure 5 , Figure 5 The program logic diagram of the entity recommendation method provided in the embodiment of the present application may include:

[0085] After obtaining the search intent and seed entity words, first determine whether the search intent and the seed entity words match. If the search intent and the seed entity words match, then determine whether the search intent is an accurate recommendation intent. If the search intent is an accurate recommendation intent, then select the entity words to be recommended in the medical knowledge graph based on the search intent and the standard entities corresponding to the seed entities in the medical knowledge graph (corresponding Figure 5 ); if the search intent is not an exact recommendation intent, fuzzy processing is performed, that is, it is assumed that there are multiple intents related to the seed entity word at the same time, and then based on the multiple intents and the standard entities corresponding to the seed entity in the medical knowledge graph, the entity words to be recommended are selected in the medical knowledge graph (corresponding to Figure 5 In addition, before executing the entity word selection strategy, the seed entity needs to be converted into a graph node through alias mapping. The entity word selection strategy recommends entities based on the entity words at the graph nodes. If the search intent does not match the seed entity word, an error is handled.

[0086] Corresponding to the embodiment of the method, the present application also provides an entity recommendation device. A structural schematic diagram of the entity recommendation device provided in the embodiment of the present application is as follows: Figure 6 As shown, it may include:

[0087] Acquisition module 61, understanding module 62, determination module 63 and recommendation module 64; wherein,

[0088] The acquisition module 61 is used to acquire input text;

[0089] The understanding module 62 is used to perform semantic understanding on the input text to determine the search intent and extract seed entity words;

[0090] The determination module 63 is used to determine the entity relationship based on the search intent and map the seed entity word to a standard entity word in the medical knowledge graph if the search intent matches the seed entity word and the search intent belongs to the precise recommendation intent;

[0091] The recommendation module 64 is used to recommend the entity words in the medical knowledge graph that have the entity relationship with the standard entity words as recommended entities.

[0092] The entity recommendation device provided in the embodiment of the present application, after obtaining the input text, first performs semantic understanding on the input text, determines the search intent and the seed entity words, and only when the search intent matches the seed entity words and the search intent belongs to the precise recommendation intent, performs entity recommendation based on the medical knowledge graph, thereby ensuring the accuracy of the medical entity recommendation.

[0093] In an optional embodiment, the understanding module 62 may be specifically used to: input the input text into a preset semantic understanding model to obtain the search intent and seed entity words; wherein the semantic understanding model includes:

[0094] Transformer-based bidirectional encoding representation BERT network;

[0095] a feed-forward neural network connected to the first output unit of the BERT network, the feed-forward neural network being used to predict search intent;

[0096] A conditional random field model connected to the plurality of second output units of the BERT network, wherein the conditional random field model is used to extract seed entity words from the input text.

[0097] In an optional embodiment, matching the search intent with the seed entity word may include:

[0098] The seed entity word is an entity word that has a predefined corresponding relationship with the search intent.

[0099] In an optional embodiment, the search intention includes at least any one of the following: an intention to ask about related diseases, an intention to ask about symptoms, and an intention to find a doctor.

[0100] In an optional embodiment, when the search intent is to ask about related diseases, when the determination module 63 determines the entity relationship based on the search intent, it can be specifically used to:

[0101] Based on the search intent, the pathological correlation and the site of disease correlation are determined.

[0102] In an optional embodiment, when the search intent is to ask about symptoms and the seed entity word is a symptom entity word, when the determination module 63 maps the seed entity word to a standard entity word in the medical knowledge graph, it can be specifically used for:

[0103] Selecting a candidate entity word from the standard entity words in the medical knowledge graph, wherein the candidate entity word has the greatest similarity to the seed entity word;

[0104] The part entity words in the candidate entity words are matched with the part entity words in the seed entity words, and if the matching result meets the matching condition, the seed entity word is mapped to the candidate entity word.

[0105] In an optional embodiment, when the search intent is to find a doctor, and the seed entity words include symptom entity words but do not include disease entity words, determining the entity relationship based on the search intent includes:

[0106] Based on the symptom question intention, a first entity relationship is determined, and a disease entity word having a first entity relationship with the seed entity is determined in the medical knowledge graph.

[0107] Based on the intention of finding a doctor, a second entity relationship is determined to determine the doctor's name entity word that has the second entity relationship with the disease entity word.

[0108] The present application also provides an electronic device, which may be a terminal device (such as a computer) or a server or a server cluster, and the electronic device may be configured with the above-mentioned entity recommendation device. Figure 7 As shown, it may include:

[0109] Processor 1, communication interface 2, memory 3 and communication bus 4;

[0110] The processor 1, the communication interface 2, and the memory 3 communicate with each other via the communication bus 4;

[0111] Optionally, the communication interface 2 may be an interface of a communication module, such as an interface of a GSM module;

[0112] The processor 1 may be a central processing unit CPU, or an application-specific integrated circuit ASIC (Application Specific Integrated Circuit), or one or more integrated circuits configured to implement the embodiments of the present application.

[0113] The memory 3 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0114] The processor 1 is specifically used to execute the program stored in the memory, and performs the following steps:

[0115] Get input text;

[0116] Performing semantic understanding on the input text to determine search intent and extract seed entity words;

[0117] If the search intent matches the seed entity word, and the search intent belongs to the precise recommendation intent, then determining the entity relationship based on the search intent, and mapping the seed entity word to a standard entity word in the medical knowledge graph;

[0118] The entity words in the medical knowledge graph that have the entity relationship with the standard entity words are recommended as recommended entities.

[0119] Optionally, the detailed functions and extended functions of the program may refer to the above description.

[0120] The present application also provides a readable storage medium, which may store a program suitable for execution by a processor, wherein the program is used to:

[0121] Get input text;

[0122] Performing semantic understanding on the input text to determine search intent and extract seed entity words;

[0123] If the search intent matches the seed entity word, and the search intent belongs to the precise recommendation intent, then determining the entity relationship based on the search intent, and mapping the seed entity word to a standard entity word in the medical knowledge graph;

[0124] The entity words in the medical knowledge graph that have the entity relationship with the standard entity words are recommended as recommended entities.

[0125] Optionally, the detailed functions and extended functions of the program may refer to the above description.

[0126] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0127] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0128] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0129] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0130] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit.

[0131] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0132] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the core idea or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An entity recommendation method, characterized in that: include: Get input text; Performing semantic understanding on the input text to determine search intent and extract seed entity words; If the search intent matches the seed entity word, and the search intent belongs to the precise recommendation intent, then the entity relationship is determined based on the search intent, and the seed entity word is mapped to a standard entity word in the medical knowledge graph, wherein the search intent matches the seed entity word including: the seed entity word is an entity word having a predefined corresponding relationship with the search intent; Recommending the entity words in the medical knowledge graph that have the entity relationship with the standard entity words as recommended entities; If the search intent matches the seed entity word and the search intent does not belong to the precise recommendation intention, the entity word to be recommended is selected in the medical knowledge graph based on the search intent having multiple intentions related to the seed entity word and the standard entity in the medical knowledge graph corresponding to the seed entity.

2. The method according to claim 1, characterized in that The semantic understanding of the input text to determine the search intent and extract seed entity words includes: The input text is input into a preset semantic understanding model to obtain the search intent and seed entity words; wherein the semantic understanding model includes: Transformer-based bidirectional encoding representation BERT network; a feed-forward neural network connected to the first output unit of the BERT network, the feed-forward neural network being used to predict search intent; A conditional random field model connected to the plurality of second output units of the BERT network, wherein the conditional random field model is used to extract seed entity words from the input text.

3. The method according to claim 1 or 2, characterized in that: The search intention includes at least any one of the following: an intention to ask about related diseases, an intention to ask about symptoms, and an intention to find a doctor.

4. The method according to claim 3, characterized in that When the search intent is to ask about related diseases, determining the entity relationship based on the search intent includes: Based on the search intent, the pathological correlation and the site of disease correlation are determined.

5. The method according to claim 3, characterized in that: When the search intent is to ask about symptoms, and the seed entity word is a symptom entity word, mapping the seed entity word to a standard entity word in the medical knowledge graph includes: Selecting a candidate entity word from the standard entity words in the medical knowledge graph, wherein the candidate entity word has the greatest similarity to the seed entity word; The part entity words in the candidate entity words are matched with the part entity words in the seed entity words, and if the matching result meets the matching condition, the seed entity word is mapped to the candidate entity word.

6. The method according to claim 3, characterized in that When the search intent is to find a doctor, and the seed entity words include symptom entity words but do not include disease entity words, determining the entity relationship based on the search intent includes: Based on the symptom-asking intention, determining a first entity relationship, and determining a disease entity word having the first entity relationship with the seed entity in the medical knowledge graph; Based on the intention of finding a doctor, a second entity relationship is determined to determine the doctor's name entity word that has the second entity relationship with the disease entity word.

7. An entity recommendation device, characterized in that: include: The acquisition module is used to obtain the input text; An understanding module, used for performing semantic understanding on the input text to determine the search intent and extract seed entity words; A determination module, for determining an entity relationship based on the search intent and mapping the seed entity word to a standard entity word in a medical knowledge graph if the search intent matches the seed entity word and the search intent belongs to an accurate recommendation intent, wherein the search intent matches the seed entity word including: the seed entity word is an entity word having a predefined corresponding relationship with the search intent; A recommendation module, used for recommending entity words in the medical knowledge graph that have the entity relationship with the standard entity words as recommended entities; The determination module is also used to select the entity word to be recommended in the medical knowledge graph based on the search intent having multiple intentions related to the seed entity word and the standard entity corresponding to the seed entity in the medical knowledge graph if the search intent matches the seed entity word and the search intent does not belong to the precise recommendation intention.

8. The device according to claim 7, characterized in that The understanding module is specifically used for: The input text is input into a preset semantic understanding model to obtain the search intent and seed entity words; wherein the semantic understanding model includes: Transformer-based bidirectional encoding representation BERT network; a feed-forward neural network connected to the first output unit of the BERT network, the feed-forward neural network being used to predict search intent; A conditional random field model connected to the plurality of second output units of the BERT network, wherein the conditional random field model is used to extract seed entity words from the input text.

9. The device according to claim 7 or 8, characterized in that The search intention includes at least any one of the following: an intention to ask about related diseases, an intention to ask about symptoms, and an intention to find a doctor.

10. The device according to claim 9, characterized in that When the search intent is to ask about related diseases, the determination module determines the entity relationship based on the search intent, specifically for: Based on the search intent, the pathological correlation and the site of disease correlation are determined.

11. The device according to claim 9, characterized in that When the search intent is to ask about symptoms and the seed entity word is a symptom entity word, the determination module maps the seed entity word to a standard entity word in the medical knowledge graph, specifically for: Selecting a candidate entity word from the standard entity words in the medical knowledge graph, wherein the candidate entity word has the greatest similarity to the seed entity word; The part entity words in the candidate entity words are matched with the part entity words in the seed entity words, and if the matching result meets the matching condition, the seed entity word is mapped to the candidate entity word.

12. The device according to claim 9, characterized in that When the search intent is to find a doctor, and the seed entity words include symptom entity words but do not include disease entity words, the determination module determines the entity relationship based on the search intent, specifically for: Based on the symptom-asking intention, determining a first entity relationship, and determining a disease entity word having the first entity relationship with the seed entity in the medical knowledge graph; Based on the intention of finding a doctor, a second entity relationship is determined to determine the doctor's name entity word that has the second entity relationship with the disease entity word.

13. An electronic device, characterized in that: including memory and processor; The memory is used to store programs; The processor is used to execute the program to implement each step of the entity recommendation method as described in any one of claims 1-6.

14. A readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, each step of the entity recommendation method according to any one of claims 1 to 6 is implemented.

Images