Method and apparatus for normalizing biomedical entity mentions and storage medium

Abstract

The present disclosure relates to a method, apparatus and storage medium for normalizing biomedical entity mentions. According to one embodiment of the present disclosure, the method comprises: receiving a biomedical entity mention as a mention to be mapped; generating a candidate concept set of the mention to be mapped by retrieving a biomedical lexicon; determining whether the same concept as the mention to be mapped is included in the candidate concept set; in case of a negative determination, updating the candidate concept set by expanding each candidate concept in the candidate concept set based on a related concept set obtained from the biomedical lexicon; obtaining a semantic similarity set by determining semantic similarity of each candidate concept in the updated candidate concept set to the mention to be mapped; and mapping the mention to be mapped to a candidate concept corresponding to a maximum semantic similarity in the semantic similarity set. The method, apparatus and storage medium of the present disclosure have at least the beneficial effect of improving accuracy of determining a canonical concept.

Description

Technical Field

[0001] This disclosure generally relates to knowledge discovery, and more specifically to methods, apparatus, and storage media for the standardization of biomedical entity references. Background Technology

[0002] With the rapid development of biomedical technology, the amount of biomedical literature, such as scientific papers and patent documents, is increasing daily. This has promoted the development of text mining techniques in the biomedical field. Biomedical terms mentioned in the literature are called biomedical entity mentions. Text mining techniques include the normalization of biomedical entity mentions. The purpose of the normalization task is to determine the corresponding unique identifier of entity mentions in biomedical literature within a knowledge graph, thereby establishing a connection between the entity mention and the knowledge graph. Establishing this connection is significant for technological research in the biomedical field.

[0003] Conventional entity mention normalization methods typically consist of two modules: candidate generation and candidate ranking. While existing conventional entity mention normalization methods have achieved good results in the normalization of biomedical entities, they still have certain limitations. First, the generated candidate names are similar, making it difficult to determine the correct candidate based solely on the name. Second, since candidates and mentions are entities within the same domain, their contextual information is also similar, making it difficult to correctly rank the candidates even using contextual information. Summary of the Invention

[0004] A brief overview of this disclosure is provided below to offer a basic understanding of certain aspects of it. It should be understood that this overview is not an exhaustive summary of the disclosure. It is not intended to identify key or essential parts of the disclosure, nor is it intended to limit its scope. Its purpose is merely to present certain concepts in a simplified form as a prelude to the more detailed description that follows.

[0005] According to one aspect of this disclosure, a computer-implemented method for normalizing biomedical entity mentions is provided. The method includes: receiving a biomedical entity mention as a mention to be mapped; generating a candidate concept set of the mention to be mapped by retrieving a biomedical dictionary; determining whether the candidate concept set includes a concept identical to the mention to be mapped; if the determination result is "no", updating the candidate concept set by expanding the candidate concept based on a related concept set obtained from a biomedical dictionary for each candidate concept in the candidate concept set; obtaining a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the mention to be mapped; and mapping the mention to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set; wherein the related concept set contains synonyms of the corresponding candidate concept, and if the corresponding candidate concept has a corresponding parent concept, the related concept set also contains the corresponding parent concept and its synonyms.

[0006] According to one aspect of this disclosure, an apparatus for normalizing biomedical entity references is provided. The apparatus includes: a memory storing instructions; and one or more processors, the processors being able to communicate with the memory to execute instructions retrieved from the memory, and the instructions causing the processors to: receive a biomedical entity reference as a reference to be mapped; generate a candidate concept set of the reference to be mapped by retrieving from a biomedical dictionary; determine whether the candidate concept set includes a concept identical to the reference to be mapped; if the determination result is "no", update the candidate concept set by expanding the candidate concept set based on a related concept set obtained from a biomedical dictionary for each candidate concept in the candidate concept set; obtain a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped; and map the reference to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set; wherein the related concept set includes synonyms of the corresponding candidate concept, and if the corresponding candidate concept has a corresponding parent concept, the related concept set also includes the corresponding parent concept and its synonyms.

[0007] According to another aspect of this disclosure, a computer-readable storage medium having a program stored thereon is provided. The program causes a computer running the program to: receive a biomedical entity mention as a mention to be mapped; generate a candidate concept set of the mention to be mapped by searching a biomedical dictionary; determine whether the candidate concept set includes a concept identical to the mention to be mapped; if the determination result is "no", update the candidate concept set by expanding the candidate concept based on a related concept set obtained from a biomedical dictionary for each candidate concept in the candidate concept set; obtain a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the mention to be mapped; and map the mention to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set; wherein the related concept set contains synonyms of the corresponding candidate concept, and if the corresponding candidate concept has a corresponding parent concept, the related concept set also contains the corresponding parent concept and its synonyms.

[0008] The beneficial effects of this disclosure on the methods, apparatus, and storage media for standardizing references to biomedical entities include at least the following: improving the accuracy of defining standard concepts. Attached Figure Description

[0009] The embodiments of this disclosure are described below with reference to the accompanying drawings, which will help to more easily understand the above and other objects, features, and advantages of this disclosure. The drawings are only for illustrating the principles of this disclosure. The dimensions and relative positions of the elements are not necessarily drawn to scale in the drawings. The same reference numerals may denote the same features. In the drawings:

[0010] Figure 1 An exemplary flowchart of a computer-implemented method for normalizing biomedical entity references according to an embodiment of the present disclosure is shown;

[0011] Figure 2 An exemplary flowchart of a diffusion candidate concept method according to an embodiment of the present disclosure is shown;

[0012] Figure 3 An exemplary flowchart of a method for determining semantic similarity according to an embodiment of the present disclosure is shown;

[0013] Figure 4 An exemplary block diagram of an apparatus for standardizing biomedical entity references according to one embodiment of this disclosure is shown;

[0014] Figure 5 An exemplary block diagram of a device for standardizing references to biomedical entities according to an embodiment of this disclosure is shown; and

[0015] Figure 6An exemplary block diagram of an information processing apparatus according to one embodiment of the present disclosure is shown. Detailed Implementation

[0016] Exemplary embodiments of this disclosure will be described below with reference to the accompanying drawings. For clarity and brevity, not all features of actual embodiments are described in the specification. However, it should be understood that many embodiment-specific decisions can be made in the development of any such actual embodiment to achieve the developer’s specific objectives, and these decisions may vary as the embodiments differ.

[0017] It should also be noted that, in order to avoid obscuring the contents of this disclosure with unnecessary details, only the device structure closely related to the solution according to this disclosure is shown in the accompanying drawings, while other details that are not closely related to this disclosure are omitted.

[0018] It should be understood that this disclosure is not limited to the described embodiments by virtue of the following description with reference to the accompanying drawings. In this document, embodiments may be combined with each other, features may be substituted or borrowed between different embodiments, and one or more features may be omitted in one embodiment, where feasible.

[0019] Computer program code used to perform the operations of various aspects of embodiments of this disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, Smalltalk, C++, and conventional procedural programming languages ​​such as the "C" programming language or similar programming languages.

[0020] The method disclosed herein can be implemented using circuitry with corresponding functional configurations. The circuitry includes circuitry for a processor.

[0021] One aspect of this disclosure provides, for example, a computer-implemented method for standardizing biomedical entities. Reference is made below. Figure 1 The method is described exemplarily.

[0022] Figure 1 An exemplary flowchart of a computer-implemented method 100 for normalizing biomedical entities, according to one embodiment of the present disclosure, is shown. Method 100, for example, is capable of outputting a normalization result for a biomedical entity reference in relation to a biomedical dictionary based on an input biomedical entity reference, the normalization result indicating the corresponding normalized representation of the biomedical entity reference shown in the biomedical dictionary.

[0023] In step S101, a biomedical entity mention is received as a mention mm to be mapped. mm is, for example, a biomedical term mentioned in biomedical literature. A term extraction application can be used to extract the biomedical terms appearing in the images or text of biomedical literature. An example of a mention mm to be mapped is “sporadic t-cell leukemia”. The mention mm to be mapped is, for example, an English term or another Indo-European term.

[0024] In step S103, a candidate concept set Sc for the reference mm to be mapped is generated by retrieving the biomedical dictionary D. Sc = {mc[1], mc[2], ..., mc[i], ..., mc[i_max]}, where i_max is the number of elements in the candidate concept set. For example, the full-text search engine Lucene is used to index each concept and its identifier in the biomedical dictionary D (biomedical knowledge base); then, Lucene is used to retrieve from the index, and the top 20 search results are used as candidate concepts to form the candidate concept set. The upper limit of the number of elements in the candidate concept set can be limited, for example, to 20. When the retrieved candidate concepts do not reach the upper limit, the candidate concept set is constructed using all retrieved candidate concepts. Each candidate concept mc[i] has an index on the biomedical dictionary D, denoted as mc[i].ind. The concepts in the biomedical dictionary D are standardized biomedical terms that have been included. The biomedical dictionary D is, for example, an English dictionary or other Indo-European dictionary.

[0025] In step S105, it is determined whether the candidate concept set Sc includes the same concept as the reference to be mapped. For example, for the reference to be mapped mm = "sporadic t-cell leukemia", if Sc contains the candidate concept mc[i] = "sporadic t-cell leukemia", then it is determined that the candidate concept set Sc includes the same concept as the reference to be mapped.

[0026] If the result is determined to be "No", proceed to step S107. In step S107, the candidate concept set Sc is updated by expanding the candidate concept mc[i] based on the related concept set Sr[i] obtained from the biomedical dictionary D for each candidate concept mc[i]. The related concept set contains synonyms of the corresponding candidate concept, and if the corresponding candidate concept has a corresponding parent concept, the related concept set also contains the corresponding parent concept and its synonyms. For example, the related concept set Sr[i] of the candidate concept mc[i] contains all the synonyms of the candidate concept mc[i] shown in the biomedical dictionary D, where j can be 1 to the number of synonyms j_max; and if the candidate concept mc[i] has a parent concept, the related concept set Sr[i] also contains the parent concept mp[i] of the candidate concept mc[i] and all the synonyms of the parent concept mp[i] shown in the biomedical dictionary D, where k can be 1 to the number of synonyms k_max. In biomedical dictionaries, concepts have hierarchical relationships. This means that a concept has a parent concept or a child concept, or both; where the child concept contains information about the parent concept. For example, the concept “lymphoma, non-hodgkins” is a type of “lymphoma”, where “lymphoma, non-hodgkins” is a child concept of “lymphoma”, and “lymphoma” is the parent concept of “lymphoma, non-hodgkins”. Each concept entry in the biomedical dictionary D records the synonyms of that concept and their parent concepts (if any). It should be noted that if mc[i] has multiple parent concepts, the preferred related concept set Sr[i] contains these multiple parent concepts, and also includes the synonyms of each of these multiple parent concepts. Preferably, when the candidate concept mc[i] has at least one parent concept, the related concept set Sr[i] also contains the at least one parent concept and the synonyms of each of the at least one parent concept.

[0027] In step S109, a semantic similarity set Ss is obtained by determining the semantic similarity sm[i] between each candidate concept mc[i] in the updated candidate concept set and the mention mm to be mapped.

[0028] In step S111, the mention to be mapped, mm, is mapped to the candidate concept mc[i_smax] corresponding to the maximum semantic similarity in the semantic similarity set Ss. Specifically, the index mc[i_smax].ind of the candidate concept mc[i_smax] corresponding to the maximum semantic similarity in the semantic similarity set Ss with respect to the biomedical dictionary D is used to map the mention to be mapped, mm, to the corresponding concept mc[i_smax] in the biomedical dictionary D, so as to indicate the normalized concept of the mention to be mapped, mm.ind of the mention to be mapped, mm, to be mapped, is set to mc[i_smax].ind, thereby mapping mm to the candidate concept mc[i_smax] corresponding to the maximum semantic similarity, and the normalized concept of the mention to be mapped, mm, is set to the candidate concept mc[i_smax] corresponding to the maximum semantic similarity.

[0029] If the determination result in step S105 is "yes", then step S113 is executed. In step S113, the mention to be mapped, mm, is mapped to the same candidate concept mc[i_s] in the candidate concept set Sc. For example, the index mm.ind of the biomedical dictionary D of the mention to be mapped, mm, is set to mc[i_s].ind, thereby mapping mm to the same candidate concept mc[i_s].

[0030] In one embodiment, generating a candidate concept set for the reference to be mapped by searching the biomedical dictionary D includes preprocessing the reference mm. The preprocessing may include at least one of the following: converting abbreviations in the reference to be mapped to full names; replacing non-Arabic numerals in the reference to be mapped with Arabic numerals; and converting plural references to singular entities. Optionally, the preprocessing may further include converting the first letter of the reference mm to a lowercase letter if the first letter is capitalized.

[0031] In one embodiment of this disclosure, candidate concepts may be expanded based on word frequency and mentions to be mapped. Figure 2 An exemplary flowchart of a candidate concept diffusion method 200 according to one embodiment of this disclosure is shown. For each candidate concept in the candidate concept set (i.e., selected candidate concept), method 200 can be executed to expand the candidate concept based on a set of relevant concepts obtained from a biomedical dictionary, thereby updating the candidate concept set.

[0032] In step S203, the word sequence Sw[i], arranged in descending order of word frequency, is determined based on the word frequencies of the words in the related concept set Sr[i] of the selected candidate concept mc[i] in the candidate concept set. When the selected candidate concept mc[i] has a parent concept mp[i], the same word may appear in multiple concepts in the related concept set Sr, which consists of all synonyms of the selected candidate concept (sys[i][1], ..., sys[i][j], ..., sys[i][j_max]), its parent concept mp[i], and all synonyms of the parent concept (syp[i][1], ..., sys[i][k], ..., sys[i][k_max]). When the selected candidate concept mc[i] does not have a parent concept, the same word may also appear in multiple concepts in the related concept set Sr, which consists of all synonyms of the selected candidate concept (sys[i][1], ..., sys[i][j], ..., sys[i][j_max]). Note that the number of parent concepts of mc[i] can be multiple. Sr can be constructed by selecting the first parent concept from the multiple parent concepts shown in D. The relevant concept set Sr preferably includes all parent concepts except mc[i], as well as synonyms of other parent concepts. The frequency of each word is determined by counting the number of occurrences of each word in the relevant concept set (e.g., a total of 30 different words). This allows us to determine the original sequence So[i] (e.g., an original sequence containing 30 words) composed of words arranged in descending order of frequency. The word sequence Sw[i] can be set to be the same as the original sequence So[i]. Alternatively, if the number of words in the original sequence is more than N (N is a natural number, e.g., N=5), the word sequence Sw[i] can be set to be the sequence composed of the first N words of the original sequence So[i]. Or, the word sequence Sw[i] can contain only words in the original sequence whose frequency is greater than a frequency threshold. In other words, the length (i.e., the number of words) of the word sequence Sw[i] can be limited based on a frequency threshold or a sequence length threshold.

[0033] In step S205, the word sequence Sw[i] is updated. For example, the word sequence Sw[i] is updated based on the mention to be mapped mm. Specifically, each word in mm can be checked to determine whether the word has appeared in the word sequence Sw[i]. For words that have appeared, the word sequence Sw[i] is updated by moving the word forward in the word sequence Sw[i]. For example, the word in the word sequence Sw[i] that appears in the mention to be mapped mm is moved forward before the first word in the word sequence Sw[i], that is, the word that has appeared is set as the first word of the word sequence. When there are multiple words in the word sequence Sw[i] that appear in the mention to be mapped mm, these words are all moved forward to a earlier position in the word sequence Sw[i]. For example, the forward movement is performed so that these words occupy the first, second, and third positions of the word sequence (the case where there are three words in the word sequence Sw[i] that appear in the mention to be mapped mm). In one example, when the word currently being checked in mm is already the first word in the current word sequence Sw[i], the process jumps to checking the next word in mm (that is, no forward movement is required). As an example, if Sw[i] is w1, w2, w3, w4, w5, where w4 has appeared in mm, then after performing the shift, Sw[i] will be updated to w4, w1, w2, w3, w5.

[0034] In one example, the word sequence Sw[i] determined in step S203 can be the same as the original sequence So[i]. In step S203, after updating the word sequence Sw[i] based on the mentions to be mapped mm, the word sequence is updated by extracting only the first N words of the word sequence Sw[i]. That is, updating the word sequence can also include updating the word sequence based on a sequence length threshold.

[0035] The word sequence Sw[i] output in step S205 can be denoted as: w[i][1], w[i][2], ..., w[i][n], ..., w[i][n_max].

[0036] In step S207, an attempt is made to expand the selected candidate concept by adding words from the word sequence Sw[i] to the end of the selected candidate concept mc[i]. The sequence position pointer P can be used to indicate the index of the word to be checked in the word sequence Sw[i] (initially set to 1). The word indicated by the sequence position pointer P in the word sequence Sw[i] is checked to see if it is in the selected candidate concept mc[i]. If the check result is "no", the word is added to the end of the selected candidate concept mc[i], and P is updated to P+1. For example, if mc[i] = "b-cell lymphoma" and the word to be added is "non-hodgkins", then the expanded mc[i] is "b-cell lymphoma non-hodgkins". In one example, step S207 includes adding the first word in the word sequence Sw[i] that is not in the selected candidate concept mc[i] to the end of the selected candidate concept mc[i] to expand and update the selected candidate concept. For example, if w[i][1] is not included in mc[i], then mc[i] is updated to mc[i] + space + w[i][1], where space represents a space. If the check result is "yes", then P is updated to P+1; and step S209 is performed (i.e., the word not indicated by P is expanded).

[0037] In step S209, it is determined whether the number of words in the selected candidate concept mc[i] is greater than the length threshold Lth. If the result is "yes", the expansion of the selected candidate concept ends. The length threshold is determined, for example, as follows: Lth = min{2 * number of words to be mapped, M}, where M is the maximum number of words in the concept in D, and min{} represents the minimum parameter in the selection parameters. Here, "2" is only an example and can be adjusted based on experience.

[0038] In step S211, it is determined whether the selected candidate concept is the same as the mention to be mapped, ignoring word order. If the result is "yes", the expansion of the selected candidate concept ends. It should be noted that, in a variant example, the sameness judgment (S211) can be performed first, followed by the word count judgment (S209).

[0039] In step S213, it is determined whether the word sequence has reached the end. For example, this can be achieved by checking whether the current sequence position pointer P has reached the end of the word sequence (whether P equals n_max+1). If P = n_max+1, it means that all words in the word sequence Sw[i] have been checked, and the check has reached the end of the word sequence.

[0040] In step S215, it is determined whether the selected candidate concept mc[i] has a higher-level parent concept. This can be determined based on the indication in dictionary D: check whether dictionary D shows a parent concept for the parent concept. If it is determined that the selected candidate concept mc[i] has no higher-level parent concept, then proceed to step S219, outputting a prompt message to inform the user, so that the user can, for example, manually determine the processing method. Optionally, it can also be set so that if it is determined that the selected candidate concept mc[i] has no higher-level parent concept, then the expansion of the selected candidate concept ends.

[0041] In step S217, the word sequence is updated based on a higher-level parent concept. For example, Sw[i] is updated by constructing a word sequence ordered in descending order of word frequency using the higher-level parent concept and its synonyms. Similarly, the upper limit of the number of words contained in the word sequence Sw[i] can be limited based on a word frequency threshold or a sequence length threshold. For example, when the original word sequence contains 30 words, Sw[i] is constructed using its first 5 words. For example, Sw[i] can be constructed using only words with a word frequency greater than 1. Then, the process returns to step S205.

[0042] In other words, method 200 may include: conditionally expanding the selected candidate concept based on a higher-level parent concept of the selected candidate concept.

[0043] The method for obtaining semantic similarity sets in this disclosure is described below.

[0044] The semantic similarity set can be represented as Ss={sm[1],……,sm[i],……sm[i_max]}, where sm[i] represents the semantic similarity between the candidate concept mc[i] and the mention mm to be mapped.

[0045] In one example, sm[i] can be determined using conventional semantic similarity calculation methods. For instance, the feature vectors of the two corresponding word strings can be determined, and the similarity between these two feature vectors can be directly calculated as the corresponding semantic similarity.

[0046] In one embodiment of this disclosure, attention matrices and convolutional neural network models are used to determine the semantic similarities in the semantic similarity set Ss. Below, refer to... Figure 3 This example illustrates the method for determining semantic similarity.

[0047] Figure 3 An exemplary flowchart of a method 300 for determining semantic similarity according to one embodiment of the present disclosure is shown.

[0048] In step S301, attention matrix A is determined based on the concept vector Fcv of the target candidate concept mc[i] and the mention vector Fmv of the mention mm to be mapped. The elements a of attention matrix A are... uv=match_score(Vwm[u],Vwc[v]), where Vwm[u] is the word vector of the u-th word in mc[i], and Vwc[v] is the word vector of the v-th word in mc[i]. match_score(wm[u],wc[v]) represents the degree of matching between the word wc[v] in the target candidate concept mc[i] and the word wm[u] in the mention to be mapped in mc[i]. Vwm[u] is a component of Fmv; Vwc[v] is a component of Fcv. The degree of matching between two words can be defined based on predetermined rules. The word vectors generated from the given words are generated using conventional techniques (e.g., word vectors generated through a language model such as Word2Vec), which will not be elaborated here.

[0049] In step S302, based on the attention matrix A, the attention feature vector Fma of the mention mm to be mapped and the attention feature vector Fca of the target candidate concept mc[i] are determined, where Fma is the attention feature vector of the mention mm to be mapped and Fca is the attention feature vector of the target candidate concept mc[i]. ma =W0·A T F ca =W1·A, where T represents the transpose transformation. W0 and W1 are the parameters that need to be learned during model training.

[0050] In step S303, a convolutional neural network (CNN) layer is used to generate a mention feature Fm for the mention to be mapped, based on the mention vector Fmv of the mention to be mapped mm and the attention feature vector Fma. In one example, the feature Fm' can be formed by concatenating Fma and Fmv (i.e., Fm' = Fma + Fmv). The feature Fm' is input into the CNN layer, and the CNN processes Fm' and outputs Fm. This can be represented as: Fm = CNN(Fm').

[0051] In step S304, a convolutional neural network (CNN) layer is used to generate a concept feature Fc for the target candidate concept mc[i] based on the concept vector Fcv and the attention feature vector Fca. In one example, the connection between Fca and Fcv can be used to form a feature Fc' (i.e., Fc' = Fca + Fcv). The feature Fc' is then input into the CNN layer, and the CNN processes Fc' before outputting Fc. This can be represented as: Fc = CNN(Fc').

[0052] In step S305, a deep feature Fd is generated using at least one hidden layer based on the mention feature Fm and the concept feature Fc. In one example, the connection between Fm and Fc can be used to form feature Fd' (i.e., Fd' = Fm + Fc), feature Fd' is input into the hidden layer, and after the hidden layer processes Fd', it outputs Fd. This can be represented as: Fd = Hidden(Fd').

[0053] In step S306, the Softmax layer is used to determine the semantic similarity sm[i] between the target candidate concept mc[i] and the mention mm to be mapped, based on the deep feature Fd. In one example, the output of the Softmax layer is a two-dimensional vector, and one component of this two-dimensional vector is the semantic similarity between the target candidate concept and the mention to be mapped.

[0054] Those skilled in the art will understand that before using a convolutional neural network (CNN) model to determine semantic similarity, the CNN model needs to be trained with samples to determine its parameters. During the training phase of the CNN model, for each entity mention in the training corpus that already has normalized results, a candidate concept set is generated using the candidate concept generation method described in this disclosure. If the normalized result of a candidate concept matches the tagging of the entity mention, then the mention-candidate pair is labeled.<mm,mc[i]> The semantic similarity (a component of the output of the Softmax layer; corresponding to a similarity score) is set to 1, otherwise it is marked as 0. During the testing phase, the similarity scores are used for ranking, and the candidate concept with the highest similarity score is selected as the normalization result.

[0055] This disclosure also provides a device for standardizing references to biomedical entities. See below for reference. Figure 4 An exemplary description is provided. Figure 4 An exemplary block diagram of an apparatus 400 for normalizing biomedical entity references according to an embodiment of the present disclosure is shown. Apparatus 400 includes: a retrieval unit 403, a determination unit 405, an update unit 407, an acquisition unit 409, and a mapping unit 411. The apparatus includes: a retrieval unit 403 for generating a candidate concept set of references to be mapped as biomedical entity references by retrieving from a biomedical dictionary; a determination unit 405 for determining whether the candidate concept set includes the same concept as the reference to be mapped; an update unit 407 for updating the candidate concept set by, if the determination result is "no", expanding the candidate concept set based on a related concept set obtained from a biomedical dictionary for each candidate concept, wherein the related concept set contains synonyms of the corresponding candidate concept, and, if the corresponding candidate concept has a corresponding parent concept, the related concept set also contains the corresponding parent concept and its synonyms; and an acquisition unit 409 for obtaining a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped. Mapping unit 411 is used to map the mention to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set. Apparatus 400 corresponds to method 100. Further configuration of apparatus 400 can be found in the description of method 100 in this disclosure. For example, mapping unit 411 is also configured to map the mention to be mapped to the same candidate concept in the candidate concept set if the result is "no".

[0056] This disclosure also provides a device for standardizing references to biomedical entities. See below for reference. Figure 5 An exemplary description is provided. Figure 5 An exemplary block diagram of an apparatus 500 for standardizing references to biomedical entities, according to one embodiment of this disclosure, is shown. The apparatus 500 includes a memory 501 and one or more processors 503. Instructions are stored on the memory 501. One or more processors 503 are capable of communicating with memory to execute instructions fetched from memory 501, and the instructions cause one or more processors 503 to: receive a biomedical entity mention as a mention to be mapped; generate a candidate concept set of the mention to be mapped by searching a biomedical dictionary; determine whether the candidate concept set includes a concept identical to the mention to be mapped; if the determination result is "no", update the candidate concept set by expanding the candidate concept based on a related concept set obtained from a biomedical dictionary for each candidate concept in the candidate concept set; obtain a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the mention to be mapped; and map the mention to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set; wherein the related concept set contains synonyms of the corresponding candidate concept, and if the corresponding candidate concept has a corresponding parent concept, the related concept set also contains the corresponding parent concept and the corresponding synonym of the parent concept. This apparatus corresponds to the method for normalizing biomedical entity mentions of this disclosure. Further configuration of the apparatus can be referred to the description of method 100 in this disclosure.

[0057] One aspect of this disclosure provides a computer-readable storage medium having a program stored thereon. The program causes a computer running the program to: receive a biomedical entity mention as a mention to be mapped; generate a candidate concept set of the mention to be mapped by searching a biomedical dictionary; determine whether the candidate concept set includes a concept identical to the mention to be mapped; if the determination is "no", update the candidate concept set by expanding the candidate concept based on a related concept set obtained from a biomedical dictionary for each candidate concept in the candidate concept set; obtain a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the mention to be mapped; and map the mention to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set; wherein the related concept set contains synonyms of the corresponding candidate concept, and if the corresponding candidate concept has a corresponding parent concept, the related concept set also contains the corresponding parent concept and its synonyms. Further configuration details of the program can be found in the description of the method for normalizing biomedical entity mentions in this disclosure.

[0058] According to one aspect of this disclosure, an information processing device is also provided.

[0059] Figure 6 This is an exemplary block diagram of an information processing device 600 according to one embodiment of the present disclosure. Figure 6 In this system, the central processing unit (CPU) 601 performs various processes based on programs stored in the read-only memory (ROM) 602 or programs loaded from the storage section 608 into the random access memory (RAM) 603. The RAM 603 also stores, as needed, data required by the CPU 601 when performing various processes.

[0060] CPU 601, ROM 602 and RAM 603 are interconnected via bus 604. Input / output interface 605 is also connected to bus 604.

[0061] The following components are connected to the input / output interface 606: an input section 605 including a soft keyboard, etc.; an output section 607 including a display such as a liquid crystal display (LCD) and speakers, etc.; a storage section 608 such as a hard disk; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet, a local area network, a mobile network, or a combination thereof.

[0062] The driver 610 is also connected to the input / output interface 605 as needed. A removable medium 611, such as a semiconductor memory, is mounted on the driver 610 as needed, so that programs read from it can be installed into the storage section 608 as needed.

[0063] CPU 601 can run a program for standardizing biomedical entity references. This program is capable of performing the functions of method 100.

[0064] The scheme disclosed herein includes determining the normative concepts mentioned by biomedical entities based on extended sibling and superior concepts. This is beneficial for improving the accuracy of determining normative concepts. The scheme disclosed herein also includes determining the normative concepts mentioned by biomedical entities based on an attention matrix, which is beneficial for further improving the accuracy of determining normative concepts.

[0065] As described above, this disclosure provides the principles mentioned in the standardized biomedical entity specification. It should be noted that the effects of the solutions described in this disclosure are not necessarily limited to those described above, and any effect shown in this specification or other effects that can be understood from this specification may be achieved in addition to or in place of the effects described in the preceding paragraphs.

[0066] Although the invention has been disclosed above through a description of specific embodiments, it should be understood that those skilled in the art can design various modifications (including combinations or substitutions of features between embodiments, where applicable), improvements, or equivalents to the invention within the spirit and scope of the appended claims. These modifications, improvements, or equivalents should also be considered to be included within the scope of protection of the invention. For example, in method 200, the following exemplary modifications can be made: step S207 is adjusted so that, in the current case of P, after determining that the expansion condition is not met, if P+1 does not exceed the word sequence range, P is updated, and expansion is attempted again; if the expansion attempt is successful, step S209 is performed; if P+1 exceeds the word sequence range, step S215 is executed.

[0067] The above examples illustrate how to standardize terminology in the biomedical field. It is understood that the above methods, with simple adaptive adjustments (e.g., selecting a dictionary relevant to the field), can be used to standardize terminology in other fields (e.g., chemistry).

[0068] It should be emphasized that the term "including / comprises" as used herein refers to the presence of a feature, element, step, or component, but does not exclude the presence or addition of one or more other features, elements, steps, or components.

[0069] Furthermore, the methods of the various embodiments of the present invention are not limited to being performed in the chronological order described in the specification or shown in the drawings, but may also be performed in other chronological orders, in parallel, or independently. Therefore, the execution order of the methods described in this specification does not constitute a limitation on the technical scope of the present invention.

[0070] Postscript

[0071] This disclosure includes, but is not limited to, the following schemes.

[0072] 1. A computer-implemented method for standardizing biomedical entity references, characterized in that it comprises:

[0073] Receive the biomedical entity mention as a mention to be mapped;

[0074] The candidate concept set to be mapped is generated by searching a biomedical dictionary;

[0075] Determine whether the candidate concept set includes the same concept as the reference to be mapped;

[0076] If the result is determined to be "no",

[0077] The candidate concept set is updated by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set.

[0078] A semantic similarity set is obtained by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped; and

[0079] Map the reference to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set;

[0080] The related concept set includes synonyms of the corresponding candidate concept, and when the corresponding candidate concept has a corresponding parent concept, the related concept set also includes the corresponding parent concept and its synonyms.

[0081] 2. The method according to Appendix 1, wherein generating the candidate concept set of the references to be mapped by searching a biomedical dictionary includes preprocessing the references to be mapped; and

[0082] The preprocessing includes at least one of the following:

[0083] Convert the abbreviations in the references to be mapped to their full names;

[0084] Replace the non-Arabic numerals in the reference to be mapped with Arabic numerals; and

[0085] Convert the plural form of the reference to be mapped into the singular form of the entity.

[0086] 3. The method according to Appendix 1, wherein updating the candidate concept set by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set comprises:

[0087] Based on the word frequencies of words in the related concept sets of the selected candidate concepts in the candidate concept set, a word sequence arranged in descending order of word frequency is determined;

[0088] Update the word sequence; and

[0089] An attempt is made to expand the candidate concept by selecting words from the word sequence based on the selected candidate concept and adding them to the end of the selected candidate concept.

[0090] 4. The method according to Appendix 3, wherein updating the word sequence comprises:

[0091] The word in the word sequence that appears in the reference to be mapped is set as the first word of the word sequence.

[0092] 5. According to the method described in Appendix 3, the attempt to expand the candidate concept by selecting words from the word sequence and adding them to the end of the selected candidate concept includes:

[0093] The first word in the word sequence that does not appear in the selected candidate concept is added to the end of the selected candidate concept to update the selected candidate concept.

[0094] 6. The method according to Appendix 3, wherein updating the candidate concept set by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set comprises:

[0095] Determine whether the number of words in the selected candidate concept is greater than a length threshold; and

[0096] If the number of words in the selected candidate concept exceeds the length threshold, the expansion of the selected candidate concept is terminated.

[0097] 7. The method according to Appendix 3, wherein updating the candidate concept set by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set comprises:

[0098] Determine whether the selected candidate concept is the same as the mention to be mapped, ignoring word order; and

[0099] When it is determined that the selected candidate concept is the same as the reference to be mapped, ignoring word order, the expansion of the selected candidate concept ends.

[0100] 8. The method according to Appendix 3, wherein updating the candidate concept set by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set comprises:

[0101] The selected candidate concept is conditionally extended based on a higher-level parent concept.

[0102] 9. The method according to Appendix 1, wherein a convolutional neural network model is used to determine the similarity between each candidate concept in the updated candidate concept set and the reference to be mapped.

[0103] 10. The method according to Appendix 9, wherein the convolutional neural network model is configured as follows:

[0104] The attention matrix is ​​determined based on the concept vector of the target candidate concept and the mention vector of the mention to be mapped;

[0105] Based on the attention matrix, determine the attention feature vector of the mention to be mapped and the attention feature vector of the target candidate concept;

[0106] The mention features of the mention to be mapped are generated using a convolutional neural network layer based on the mention vector and attention feature vector of the mention to be mapped;

[0107] The convolutional neural network layer is used to generate conceptual features of the target candidate concept based on the concept vector and attention feature vector of the target candidate concept;

[0108] Deep features are generated based on the mention features and the concept features using at least one hidden layer; and

[0109] The Softmax layer is used to determine the semantic similarity between the target candidate concept and the reference to be mapped based on the deep features;

[0110] Each element of the attention matrix represents the degree of matching between words in the target candidate concept and words in the mention to be mapped.

[0111] 11. The method according to Appendix 10, wherein the output of the Softmax layer is a two-dimensional vector, and one component of the two-dimensional vector is the semantic similarity between the target candidate concept and the reference to be mapped.

[0112] 12. A device for standardizing references to biomedical entities, characterized in that it comprises:

[0113] The retrieval unit is configured to generate a set of candidate concepts to be mapped as biomedical entity references by searching a biomedical dictionary;

[0114] The determining unit is configured to determine whether the candidate concept set includes the same concept as the reference to be mapped;

[0115] The updating unit is configured to, when the result is determined to be "no", update the candidate concept set by expanding the candidate concept based on the relevant concept set obtained from the biomedical dictionary for each candidate concept in the candidate concept set;

[0116] The obtaining unit is configured to obtain a semantic similarity set by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped; and

[0117] The mapping unit is configured to map the reference to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set;

[0118] The related concept set includes synonyms of the corresponding candidate concepts, and when the corresponding candidate concept has a corresponding parent concept, the related concept set also includes the corresponding parent concept and the synonym of the corresponding parent concept.

[0119] 13. A device for standardizing references to biomedical entities, characterized in that the device comprises:

[0120] A memory, on which instructions are stored; and

[0121] One or more processors, said one or more processors being able to communicate with said memory to execute said instructions fetched from said memory, and said instructions causing said one or more processors to:

[0122] Receive the biomedical entity mention as a mention to be mapped;

[0123] The candidate concept set to be mapped is generated by searching a biomedical dictionary;

[0124] Determine whether the candidate concept set includes the same concept as the reference to be mapped; if the result is "no",

[0125] The candidate concept set is updated by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set.

[0126] A semantic similarity set is obtained by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped; and

[0127] Map the reference to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set;

[0128] The related concept set includes synonyms of the corresponding candidate concept, and when the corresponding candidate concept has a corresponding parent concept, the related concept set also includes the corresponding parent concept and its synonyms.

[0129] 14. A computer-readable storage medium having a program stored thereon, characterized in that the program causes a computer executing the program to:

[0130] Receive the biomedical entity mention as a mention to be mapped;

[0131] The candidate concept set to be mapped is generated by searching a biomedical dictionary;

[0132] Determine whether the candidate concept set includes the same concept as the reference to be mapped;

[0133] If the result is determined to be "no",

[0134] The candidate concept set is updated by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set.

[0135] A semantic similarity set is obtained by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped; and

[0136] Map the reference to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set;

[0137] The related concept set includes synonyms of the corresponding candidate concept, and when the corresponding candidate concept has a corresponding parent concept, the related concept set also includes the corresponding parent concept and its synonyms.

Claims

1. A computer-implemented method for standardizing biomedical entity references, characterized in that, include: The biomedical entity mention is received as a reference to be mapped, wherein the reference to be mapped is an English term or other Indo-European term; The candidate concept set to be mapped is generated by searching a biomedical dictionary, which is an English dictionary or other Indo-European dictionary; Determine whether the candidate concept set includes the same concept as the reference to be mapped; If the result is determined to be "no", The candidate concept set is updated by expanding the candidate concept based on a set of related concepts obtained from the biomedical dictionary for each candidate concept in the candidate concept set. A semantic similarity set is obtained by determining the semantic similarity between each candidate concept in the updated candidate concept set and the reference to be mapped; as well as Map the reference to be mapped to the candidate concept corresponding to the maximum semantic similarity in the semantic similarity set; The related concept set includes synonyms of the corresponding candidate concepts, and when a candidate concept has a corresponding parent concept, the related concept set also includes the corresponding parent concept and its synonyms; and The process of updating the candidate concept set by expanding the candidate concept based on a relevant concept set obtained from the biomedical dictionary for each candidate concept in the candidate concept set includes: Based on the word frequencies of words in the related concept sets of the selected candidate concepts in the candidate concept set, a word sequence arranged in descending order of word frequency is determined; Update the word sequence by setting the word appearing in the mentioned to be mapped as the first word of the word sequence; and An attempt is made to expand the candidate concept by selecting words from the word sequence based on the selected candidate concept and adding them to the end of the selected candidate concept.

2. The method according to claim 1, wherein, Generating a candidate concept set for the references to be mapped by searching a biomedical dictionary includes preprocessing the references to be mapped; and The preprocessing includes at least one of the following: Convert the abbreviations in the references to be mapped to their full names; Replace the non-Arabic numerals in the reference to be mapped with Arabic numerals; and Convert the plural form of the reference to be mapped into the singular form of the entity.

3. The method according to claim 1, wherein, Expanding a candidate concept by selecting words from the word sequence and adding them to the end of the selected candidate concept includes: The first word in the word sequence that does not appear in the selected candidate concept is added to the end of the selected candidate concept to update the selected candidate concept.

4. The method according to claim 1, wherein, Updating the candidate concept set by expanding the candidate concept based on a relevant concept set obtained from the biomedical dictionary for each candidate concept in the candidate concept set includes: Determine whether the number of words in the selected candidate concept is greater than a length threshold; and If the number of words in the selected candidate concept exceeds the length threshold, the expansion of the selected candidate concept is terminated.

5. The method according to claim 1, wherein, A convolutional neural network model is used to determine the similarity between each candidate concept in the updated candidate concept set and the reference to be mapped.

6. The method according to claim 5, wherein, The convolutional neural network model is configured as follows: The attention matrix is ​​determined based on the concept vector of the target candidate concept and the mention vector of the mention to be mapped; Based on the attention matrix, determine the attention feature vector of the mention to be mapped and the attention feature vector of the target candidate concept; The mention features of the mention to be mapped are generated using a convolutional neural network layer based on the mention vector and attention feature vector of the mention to be mapped; The convolutional neural network layer is used to generate conceptual features of the target candidate concept based on the concept vector and attention feature vector of the target candidate concept; Deep features are generated based on the mention features and the concept features using at least one hidden layer; as well as The Softmax layer is used to determine the semantic similarity between the target candidate concept and the reference to be mapped based on the deep features; Each element of the attention matrix represents the degree of matching between words in the target candidate concept and words in the mention to be mapped.

7. A device for standardizing references to biomedical entities, characterized in that, The device includes: A memory, on which instructions are stored; and One or more processors, the one or more processors being able to communicate with the memory to execute the instructions fetched from the memory, and the instructions causing the one or more processors to implement the method according to any one of claims 1 to 6.

8. A computer-readable storage medium having a program stored thereon, characterized in that, The program causes the computer running the program to implement the method according to any one of claims 1 to 6.

Images