An entity linking method, device, electronic device, and computer-readable medium

Abstract

The application discloses an entity linking method and device, electronic equipment and computer readable medium. The method comprises the following steps: after determining a to-be-linked entity and context information of the to-be-linked entity from to-be-processed text, determining a first linking result by using at least one candidate object corresponding to the to-be-linked entity recorded in a linking relationship library, and determining a target linking object of the to-be-linked entity according to the first linking result and the context information of the to-be-linked entity. Thus, the entity linking effect is improved.

Description

Technical Field

[0001] This application relates to the field of data processing technology, and in particular to a method, apparatus, electronic device, and computer-readable medium for linking entities. Background Technology

[0002] Entity linking is one of the fundamental tasks in natural language processing and knowledge graphs. Generally, entity linking can be, for example, associating an entity in a text with an object in a database, so that other business processes can be completed based on this association (e.g., knowledge question answering, information flow recommendation, text semantic understanding, web page recommendation, and semantic search).

[0003] However, how to establish physical links is a technical problem that urgently needs to be solved. Summary of the Invention

[0004] To address the aforementioned technical problems, this application provides a method, apparatus, electronic device, and computer-readable medium for linking entities.

[0005] To achieve the above objectives, the technical solution provided in this application is as follows:

[0006] This application provides an entity linking method, the method comprising:

[0007] Determine the entity to be linked and its context information from the text to be processed;

[0008] Using a pre-built link relationship database, a first link result corresponding to the entity to be linked is determined; the link relationship database includes the correspondence between the entity to be linked and at least one candidate object; the first link result is determined based on the at least one candidate object.

[0009] Based on the first linking result and the context information of the entity to be linked, the target linking object of the entity to be linked is determined.

[0010] In one possible implementation, the link relationship database further includes recommended representation data for each of the candidate objects; the recommended representation data of the candidate objects is used to indicate the probability of associating the candidate object with the entity to be linked;

[0011] The first link result is determined based on the at least one candidate object and the recommended representation data of the at least one candidate object.

[0012] In one possible implementation, the process of determining the first linking result includes:

[0013] The recommended representation data of the at least one candidate object are compared to obtain the comparison result;

[0014] If the comparison result indicates that the recommended representation data of the first object among the at least one candidate object is the largest, then the first link result is generated based on the first object.

[0015] In one possible implementation, the recommended representation data of the candidate object is determined based on the historical feedback operations corresponding to the candidate object.

[0016] In one possible implementation, after determining the target link object of the entity to be linked based on the first linking result and the context information of the entity to be linked, the method further includes:

[0017] Obtain the actual feedback operation corresponding to the target link object;

[0018] Update the recommendation representation data corresponding to the target link object based on the actual feedback operation corresponding to the target link object;

[0019] The link relationship database is updated based on the recommended representation data corresponding to the target link object.

[0020] In one possible implementation, determining the target link object of the entity to be linked based on the first linking result and the context information of the entity to be linked includes:

[0021] If the context information of the entity to be linked meets the preset information quantity condition, then the second link result corresponding to the entity to be linked is determined based on the entity to be linked, the context information of the entity to be linked, and the pre-built entity linking model.

[0022] If the entity to be linked is an abbreviation, and the text to be processed contains the full name information corresponding to the entity to be linked, then the third link result corresponding to the entity to be linked is determined based on the full name information.

[0023] Based on the first linking result, the second linking result, and the third linking result, the target linking object of the entity to be linked is determined.

[0024] In one possible implementation, the process of determining the second linking result includes:

[0025] The entity to be linked and its context information are input into the entity linking model to obtain the entity linking description result output by the entity linking model; wherein, the entity linking description result includes the confidence scores of at least one candidate object corresponding to the entity to be linked; the at least one candidate object includes a second object and a third object; the confidence score of the second object is the maximum value among the confidence scores of the at least one candidate object; the confidence score of the third object is the second largest value among the confidence scores of the at least one candidate object;

[0026] If the confidence level of the second object is greater than the first threshold, and the difference between the confidence level of the second object and the confidence level of the third object is greater than the second threshold, then the second link result is generated based on the second object.

[0027] In one possible implementation, determining the third link result corresponding to the entity to be linked based on the full name information includes:

[0028] The full name information is matched with at least one candidate object in the entity word lexicon to obtain the matching result;

[0029] If the matching result indicates that there is a fourth object among the at least one candidate object that matches the full name information, then the third link result is determined based on the fourth object.

[0030] In one possible implementation, the method further includes:

[0031] If the entity to be linked is an abbreviation, then the abbreviation feature of the entity to be linked is determined;

[0032] Generate the template to be used for the entity to be linked according to the full name template generation rules corresponding to the abbreviation feature;

[0033] Using the template to be used, a template retrieval process is performed on the text to be processed to obtain the retrieval results;

[0034] Based on the search results, determine the full name information corresponding to the entity to be linked.

[0035] In one possible implementation, the method further includes:

[0036] If the entity to be linked meets the preset abbreviation conditions, then the entity to be linked is determined to be an abbreviation.

[0037] In one possible implementation, before determining the first link result corresponding to the entity to be linked using a pre-built link relationship library, the method further includes:

[0038] Obtain the text scene identifier of the text to be processed;

[0039] The link relationship library corresponding to the text scene identifier is found from the pre-built mapping relationship; wherein the mapping relationship includes the correspondence between the text scene identifier and the link relationship library.

[0040] This application provides a physical linking device, including:

[0041] The first determining unit is used to determine the entity to be linked and the context information of the entity to be linked from the text to be processed;

[0042] The second determining unit is used to determine the first linking result corresponding to the entity to be linked by using a pre-built linking relationship library; the linking relationship library includes the correspondence between the entity to be linked and at least one candidate object; the first linking result is determined based on the at least one candidate object;

[0043] The third determining unit is used to determine the target linking object of the entity to be linked based on the first linking result and the context information of the entity to be linked.

[0044] This application provides an electronic device, the device comprising: a processor and a memory;

[0045] The memory is used to store instructions or computer programs;

[0046] The processor is configured to execute the instructions or computer program in the memory, so that the electronic device performs the entity linking method provided in this application.

[0047] This application provides a computer-readable medium storing instructions or a computer program that, when executed on a device, causes the device to perform the entity linking method provided in this application.

[0048] This application provides a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the entity linking method provided in this application.

[0049] Compared with the prior art, this application has at least the following advantages:

[0050] In the technical solution provided in this application, after determining the entity to be linked and its context information from the text to be processed, at least one candidate object corresponding to the entity to be linked is recorded in a pre-built link relationship library to determine the first link result corresponding to the entity to be linked. Based on the first link result and the context information of the entity to be linked, the target link object of the entity to be linked is determined, thereby achieving the purpose of entity linking. Attached Figure Description

[0051] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0052] Figure 1 A flowchart illustrating an entity linking method provided in an embodiment of this application;

[0053] Figure 2 A schematic diagram illustrating an entity linking process provided in an embodiment of this application;

[0054] Figure 3 This is a schematic diagram of the structure of a physical linking device provided in an embodiment of this application;

[0055] Figure 4 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0056] Research has found that for entity linking tasks, in some application scenarios, entity words appearing in the same text in different contexts are linked to different objects in the information database (e.g., linked to different definitions), which increases the difficulty of entity linking tasks.

[0057] The study also found that some implementations of entity linking tasks can utilize pre-built machine learning models for entity link disambiguation; however, these implementations have at least the following drawbacks:

[0058] ① The processing effect on short texts is poor. The specific reasons are as follows: For the entity words involved in short texts, the information carried by the short text is limited, which makes it impossible for the short text to provide sufficient context information for the entity words, resulting in insufficient context. In addition, the machine learning model mentioned above usually relies on sufficient context for entity link disambiguation. When the context is insufficient or non-existent, the processing performance of the machine learning model will drop significantly, resulting in poor processing effect for short texts.

[0059] ② The processing effect of abbreviations is not good. The specific reason is that abbreviations may be associated with multiple entity interpretations, making it impossible for the machine learning model mentioned above to locate their accurate definition.

[0060] Based on the above research, in order to improve the entity linking effect, this application provides an entity linking method. The method includes: after determining the entity to be linked and its context information from the text to be processed, using at least one candidate object corresponding to the entity to be linked recorded in a pre-built link relationship library, determining the first linking result corresponding to the entity to be linked, and determining the target linking object of the entity to be linked based on the first linking result and the context information of the entity to be linked, thereby achieving the purpose of entity linking.

[0061] Furthermore, since the entity linking scheme provided in this application can be implemented based on the three linking methods shown below, the entity linking scheme can effectively overcome the defects shown in ①-② above, thereby effectively improving the entity linking effect.

[0062] Furthermore, this application does not limit the executing entity of the above entity linking method. For example, the entity linking method provided in the embodiments of this application can be applied to devices with data processing capabilities, such as terminal devices or servers. Alternatively, the entity linking method provided in the embodiments of this application can also be implemented through data communication between different devices (e.g., a terminal device and a server, two terminal devices, or two servers). The terminal device can be a smartphone, computer, personal digital assistant (PDA), or tablet computer, etc. The server can be a standalone server, a cluster server, or a cloud server.

[0063] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0064] To better understand the technical solution provided in this application, the entity linking method provided in this application will be explained below with reference to some accompanying drawings. For example... Figure 1 As shown, the entity linking method provided in this application includes S1-S3 as described below. Wherein, the Figure 1 A flowchart illustrating an entity linking method provided in an embodiment of this application.

[0065] S1: Determine the entity to be linked and its context information from the text to be processed.

[0066] The text to be processed refers to any text data that needs to undergo entity recognition and entity linking processing.

[0067] The entity to be linked refers to the entity word identified from the text to be processed above; moreover, this application does not limit the process of determining the entity to be linked. For example, it can be implemented using any existing or future entity recognition method.

[0068] The phrase "context information of the entity to be linked" above describes the context of the entity to be linked within the text to be processed above. Furthermore, this application does not limit the process of determining the "context information of the entity to be linked." For example, it could specifically involve deleting the entity to be linked from the text to be processed to obtain its context information. Alternatively, it could involve first deleting the entity to be linked from the text to obtain the remaining content; then deleting some preset invalid characters from the remaining content to obtain the context information of the entity to be linked. These preset invalid characters can be pre-defined; moreover, this application does not limit the preset invalid characters, which could include punctuation marks, numbers, and other characters.

[0069] S2: Using a pre-built link relationship library, determine the first link result corresponding to the entity to be linked; the link relationship library includes the correspondence between the entity to be linked and at least one candidate object; the first link result is determined based on the at least one candidate object.

[0070] The link relationship database describes the usage of link objects by a target user for certain entity words within a historical time period, enabling the database to represent the link usage status of the target user during that historical time period. The target user refers to the user of the device displaying the "target link object of the entity to be linked" below. The historical time period refers to the time period prior to the execution time of S1 above. The link object refers to the object that needs to establish an association with the entity word; moreover, this application does not limit the link object. For example, in some application scenarios, the link object can refer to a definition existing in a thesaurus (e.g., "entity word thesaurus" below).

[0071] In fact, the "link relationship database" mentioned above can include at least the correspondence between the m-th entity word and some historical link objects corresponding to that m-th entity word. Here, the historical link objects refer to the link objects that were provided to the target user for the m-th entity word within a historical time period, so that these historical link objects were previously presented to the target user as link objects associated with the m-th entity word. m is a positive integer, m≤M, where M is a positive integer and M represents the number of entities involved in the link relationship database.

[0072] In fact, to further improve the effectiveness of entity linking, this application also provides a possible implementation of the link relationship database mentioned above. In this implementation, the link relationship database includes not only the correspondence between the m-th entity word and some historical link objects corresponding to the m-th entity word, but also recommendation representation data for each historical link object. The recommendation representation data of the k-th historical link object is used to represent the probability of associating the k-th historical link object with the m-th entity word (that is, the probability of recommending the k-th historical link object as a link object associated with the m-th entity word to the target user). Furthermore, the larger the value of the "recommendation representation data of the k-th historical link object," the greater the probability of associating the k-th historical link object with the m-th entity word. k is a positive integer, k≤K, where K is a positive integer and represents the number of objects in the "some historical link objects corresponding to the m-th entity word".

[0073] Furthermore, regarding the "recommendation representation data of the k-th historical link object" mentioned above, this "recommendation representation data of the k-th historical link object" can be determined based on the historical feedback operations corresponding to the k-th historical link object. Here, the "historical feedback operation corresponding to the k-th historical link object" refers to the feedback operation triggered by the target user when the k-th historical link object is recommended to the target user as a link object associated with the m-th entity word within a historical time period. This feedback operation indicates whether the k-th historical link object is the correct object expected by the target user.

[0074] It should be noted that this application does not limit the implementation method of the "feedback operation" in the above paragraph. For example, it can be implemented using the two feedback methods shown in (I)-(II) below.

[0075] (i) Explicit feedback method, specifically: the target user explicitly expresses whether the "kth historical link object" mentioned above is the correct object they expected through certain actions. It should be noted that this application does not limit these actions; for example, it can be a like or a dislike. The like action explicitly expresses that the "kth historical link object" is the correct object they expected (i.e., explicit correctness); however, the dislike action explicitly expresses that the "kth historical link object" is not the correct object they expected (i.e., explicit error).

[0076] (ii) Implicit feedback method, which is as follows: After the target user receives the feedback of "kth historical link object" for "mth entity word" above, if the target user actively views the "kth historical link object" or opens the related document of the "kth historical link object", it is considered that the implicit feedback is correct; however, if the target user switches to view other link objects and their related content, it is considered that the implicit feedback is incorrect.

[0077] It should also be noted that this application does not limit the process of determining the "recommendation representation data of the kth historical link object" mentioned above. For example, it can be implemented using the formula below.

[0078] Recommended characterization data k = Cumulative number of links k +Number of correct feedbacks k ×weight 正确 -Number of error reports k ×weight 错误

[0079] In the formula, the above "recommended representation data" k "" indicates the "recommendation representation data of the k-th historical link object" mentioned above; "cumulative link count" mentioned above. k"Indicates the number of times the k-th historical link object was recommended to the target user as the link object associated with the m-th entity word in the preceding text within a historical time period; the preceding text "number of correct feedbacks" k "" indicates the number of times the target user provided correct feedback for the k-th historical link object within the historical time period; the "number of incorrect feedback" mentioned above. k "" indicates the number of times the target user provided error feedback for the k-th historical link object within the historical time period; the "cumulative link count" indicates the number of times the target user provided error feedback for the k-th historical link object within the historical time period. k = The number of times the feedback was correct k "+Number of error reports" k "; The above "weight" 正确 "This refers to pre-planning the number of correct feedback responses." k The set weighting values; the "weights" mentioned above. 错误 "This refers to pre-planning for the number of feedback errors" k The weighted values ​​set.

[0080] Based on the above content regarding the "link relationship database," it is known that this database can record a large number of historical link objects corresponding to entity words, as well as the recommendation representation data of these historical link objects under the entity words. This allows the database to represent the user feedback results of target users on these historical link objects within a historical time period. Consequently, the database can better represent the user's link usage status for these entity words, and subsequently provide entity link results that match the user's link usage, thus improving the effectiveness of entity links.

[0081] In fact, in different text scenarios (e.g., document writing, instant messaging, etc.), the link usage of the target user for the same entity word may be different. Therefore, in order to better improve the effect of entity linking, this application also provides the process of determining the "link relationship library" mentioned above, which may specifically include steps 11-12 below.

[0082] Step 11: Obtain the text scene identifier of the text to be processed.

[0083] The text scene identifier is used to indicate the text scene to which the preceding "text to be processed" belongs. Furthermore, this application does not limit the process of determining the text scene identifier. For example, if the text to be processed is a string extracted from a document, then the text scene identifier "document" can be determined as the text scene identifier of the text to be processed. Similarly, if the text to be processed is a string extracted from a chat page of instant messaging software, then the text scene identifier "instant messaging" can be determined as the text scene identifier of the text to be processed.

[0084] It should be noted that this application does not limit the execution time of step 11 above. For example, step 11 can be executed after the "text to be processed" above is obtained.

[0085] Step 12: Find the link relationship library corresponding to the above text scene identifier from the pre-built mapping relationship; wherein, the mapping relationship includes the correspondence between the text scene identifier and the link relationship library.

[0086] The mapping relationship is used to record the link relationship library corresponding to multiple text scene identifiers.

[0087] Furthermore, this application does not limit the mapping relationship described above. For example, it may include the correspondence between the y-th text scene identifier and the y-th link relationship library. The y-th text scene identifier is used to uniquely identify the y-th text scene; the y-th link relationship library is determined based on some entity words in the y-th text scene, the link objects provided to the target user for these entity words, and the historical feedback operations of these link objects, so that the y-th link relationship library can represent the link usage presented by the target user in the y-th text scene. y is a positive integer, y≤Y, where Y is a positive integer and represents the number of text scene identifiers involved in the mapping relationship.

[0088] Based on the relevant content of steps 11 to 12 above, if entity recognition and entity linking processing are required for the text to be processed, after obtaining the text to be processed, first determine the text scene identifier of the text to be processed so that the text scene identifier can indicate which text scene the text to be processed comes from; then determine whether the text scene identifier is the same as the y-th text scene identifier in the mapping relationship above. If they are the same, the y-th link relationship library corresponding to the y-th text scene identifier is directly determined as the link relationship library corresponding to the text to be processed, so that the link result of any entity word in the text to be processed can be determined with the help of the corresponding link relationship library in the future; if they are different, they can be discarded. Where y is a positive integer, y≤Y.

[0089] The “candidate object” mentioned above refers to the link object that exists in the link relationship database above and has a corresponding relationship with the entity to be linked above.

[0090] Furthermore, this application does not limit the determination process of "at least one candidate object" mentioned above. For example, when the link relationship database includes the correspondence between the m-th entity word and some historical link objects corresponding to the m-th entity word, the determination process can specifically be as follows: after obtaining the entity to be linked, the entity to be linked is compared with the m-th entity word in the link relationship database. If the comparison result indicates that the entity to be linked is the same as the m-th entity word, then all historical link objects corresponding to the m-th entity word can be determined as candidate objects corresponding to the entity to be linked, so that the link result of the entity to be linked can be determined from these candidate objects in the future. Where m is a positive integer, m≤M.

[0091] The "first link result" mentioned above refers to the link result determined using the link relationship database for the entity to be linked above. For example, the "first link result" could refer to... Figure 2 The image shows "Link results determined based on historical behavior".

[0092] In addition, this application does not limit the determination process of the "first link result" mentioned above. For example, when the link relationship database includes the correspondence between the entity to be linked and at least one candidate object, the determination process may specifically be: randomly select a candidate object from these candidate objects and determine it as the first link result.

[0093] For example, when the link relationship database includes the correspondence between the entity to be linked and at least one candidate object, as well as the recommended representation data of each candidate object, the determination process of the "first link result" mentioned above can specifically be as follows: based on the at least one candidate object and the recommended representation data of the at least one candidate object, determine the first link result corresponding to the entity to be linked. Here, the "recommended representation data of the candidate object" is used to indicate the probability of associating the candidate object with the entity to be linked; moreover, the "recommended representation data of the candidate object" can be determined based on the historical feedback operations corresponding to the candidate object. Furthermore, the relevant content of the "recommended representation data of the candidate object" is similar to the "recommended representation data of the kth historical linked object" mentioned above, and for the sake of brevity, it will not be elaborated further here.

[0094] To facilitate understanding of the above content, the following explanation will be provided with examples.

[0095] As an example, when the above link relationship database includes the correspondence between the above entity to be linked and at least one candidate object, as well as the recommendation representation data of each candidate object, the determination process of the above "first link result" may specifically include steps 21-22 below.

[0096] Step 21: Compare the recommended representation data of at least one candidate object mentioned above to obtain a comparison result, so that the comparison result can represent the relative size between the recommended representation data of these candidate objects.

[0097] It should be noted that this application does not limit the implementation of step 21 above. For example, it can specifically be a comparison result of the size between the recommendation representation data of any two objects among the "at least one candidate object" mentioned above. Furthermore, in some application scenarios, step 21 can specifically be: sorting the recommendation representation data of the at least one candidate object in descending order to obtain a comparison result, so that the comparison result can represent the relative size between the recommendation representation data of these candidate objects.

[0098] Step 22: If the comparison results above indicate that the first object has the largest recommended representation data among at least one candidate object above, then generate the first link result based on the first object.

[0099] The first object refers to the candidate object with the largest recommendation representation data among the "at least one candidate object" mentioned above; and this application does not limit the first object, for example, it may refer to the explanatory content.

[0100] Furthermore, this application does not limit the implementation of step 22, "generating a first link result based on the first object". For example, it can be specifically: directly determining the first object as the first link result, so that the first link result can indicate that the object recommended to the target user based on the link relationship database determined above, which can be associated with the link object to be linked above, is the first object.

[0101] Based on the content related to S2 above, after identifying the entity to be linked from a text, several candidate objects corresponding to the entity to be linked can be searched from a pre-built link relationship database. Then, using the recommendation representation data of these candidate objects recorded in the link relationship database, these candidate objects are sorted to obtain a sorting result, so that the sorting result can represent the arrangement of these candidate objects according to the recommendation representation data from largest to smallest. Finally, the candidate object with the highest sorting order is determined as the first link result corresponding to the entity to be linked (e.g., ...). Figure 2 The "link results determined based on historical behavior" shown above enable the first link result to represent the objects that can be associated with the linked entities in the preceding text, as determined by the link relationship database, and recommended to the target user. This achieves the purpose of entity linking by leveraging the target user's historical behavior.

[0102] S3: Based on the first link result and the context information of the entity to be linked, determine the target link object of the entity to be linked.

[0103] To facilitate understanding of S3, an example will be provided below.

[0104] As an example, S3 above may specifically include steps one through three below.

[0105] Step 1: If the context information of the entity to be linked meets the preset information quantity condition, then the second link result corresponding to the entity to be linked is determined based on the entity to be linked, the context information of the entity to be linked, and the pre-built entity linking model.

[0106] The preset information quantity condition can be pre-set. For example, it can be that the number of valid characters in the "context information of the entity to be linked" is greater than a preset threshold. Valid characters refer to semantically meaningful characters present in the "context information of the entity to be linked." Furthermore, this application does not limit the process for determining these valid characters. For example, it can be that after obtaining the "context information of the entity to be linked," some preset invalid characters (e.g., punctuation marks, numbers, etc.) are deleted from the "context information of the entity to be linked" to obtain the valid characters in the "context information of the entity to be linked."

[0107] The "entity linking model" mentioned above refers to a pre-built machine learning model with entity linking disambiguation function; moreover, this application does not limit the entity linking model, for example, it can be implemented using any existing or future machine learning model with entity linking disambiguation function.

[0108] The "second link result" mentioned above refers to the link result determined for the entity to be linked using the entity linking model described above. For example, this "second link result" refers to... Figure 2 The results shown are "links determined by the model".

[0109] Furthermore, this application does not limit the process of determining the "second link result" mentioned above. For example, it can be implemented using any existing or future method that can perform entity linking based on the model.

[0110] In fact, in order to better improve the effect of entity linking, this application also provides a process for determining the "second linking result" mentioned above, which may specifically include steps 31-32 below.

[0111] Step 31: If the context information of the entity to be linked meets the preset information content condition, then input the entity to be linked and its context information into the entity linking model to obtain the entity linking description result output by the entity linking model. The entity linking description result includes the confidence level of at least one candidate object corresponding to the entity to be linked; the at least one candidate object includes a second object and a third object; the confidence level of the second object is the maximum value among the confidence levels of the at least one candidate object; the confidence level of the third object is the second largest value among the confidence levels of the at least one candidate object.

[0112] The entity link description result refers to the entity link result given by the entity link model above for the entity to be linked; and the embodiments of this application do not limit the entity link description result. For example, it may include at least one candidate object and the confidence level of each candidate object.

[0113] The t-th candidate object refers to a link object determined by the entity linking model above for the entity to be linked; moreover, this application does not limit the t-th candidate object, for example, it can be explanatory content. Where t is a positive integer, t≤T, T is a positive integer, and T represents the number of objects in "at least one candidate object" above.

[0114] The confidence level of the t-th candidate object is used to represent the probability that the t-th candidate object is the link object associated with the entity to be linked above; and the higher the confidence level of the t-th candidate object, the greater the probability that the t-th candidate object is the link object associated with the entity to be linked. Where t is a positive integer, t≤T.

[0115] The "second object" mentioned above refers to the candidate with the highest confidence among the "at least one alternative object" mentioned above.

[0116] The "third object" mentioned above refers to the candidate object with the second highest confidence level among the "at least one candidate object" mentioned above. That is, the confidence level of the third object is lower than that of the second object mentioned above, and the confidence level of the third object is not lower than that of the other entities in the "at least one candidate object" mentioned above besides the second object.

[0117] Based on the content of step 31 above, after obtaining the entity to be linked and its context information, the entity to be linked and its context information can be input into a pre-built entity linking model. This allows the entity linking model to perform entity link disambiguation processing on the entity to be linked, obtaining and outputting entity link description results. These results represent the disambiguation order of links corresponding to the entity to be linked and the confidence level of at least one candidate object corresponding to the entity to be linked. This allows for the subsequent determination of the linked objects associated with the entity to be linked based on these two outputs. Specifically, the disambiguation order describes the arrangement of all candidate objects according to their confidence levels from highest to lowest.

[0118] It should be noted that the "second object" mentioned above refers to the object that is first in the "disambiguation sorting order" in the previous paragraph; the "third object" mentioned above refers to the object that is second in the "disambiguation sorting order" in the previous paragraph.

[0119] Step 32: If the confidence level of the second object mentioned above is greater than the first threshold, and the difference between the confidence level of the second object and the confidence level of the third object is greater than the second threshold, then generate the second link result based on the second object.

[0120] The first threshold can be preset, and the second threshold can also be preset.

[0121] Furthermore, this application does not limit the implementation of step 32 above, "generating a second link result based on the second object". For example, it can specifically be: determining the second object as the second link result so that the second link result can represent the object that can be recommended to the target user as a link object associated with the entity to be linked above, based on the entity link model above. In this way, the purpose of entity link processing with the help of machine learning model can be achieved.

[0122] Based on the relevant content of steps 31 to 32 above, for the entity to be linked, the pre-built entity linking model can be used to perform entity link disambiguation processing on the entity to be linked, to obtain the disambiguation order of the links corresponding to the entity to be linked and the confidence level of at least one candidate object corresponding to the entity to be linked; then, based on the disambiguation order of the links, the second object with the highest confidence and the third object with the second highest confidence are determined. When the confidence level of the second object is greater than the first threshold, and the difference between the confidence level of the second object and the confidence level of the third object is greater than the second threshold, it can be determined that the second object is more reliable and less prone to ambiguity as the link object associated with the entity to be linked above. Therefore, the second object can be directly determined as the second link result, so that the second link result can represent the object that can be recommended to the target user as the link object associated with the entity to be linked based on the entity linking model above. In this way, the purpose of entity linking processing with the help of machine learning model can be achieved.

[0123] In fact, in order to better improve the effect of entity linking, this application also provides another process for determining the "second linking result" mentioned above, which may specifically include steps 41-44 below.

[0124] Step 41: After obtaining the context information of the entity to be linked, determine whether the context information meets the preset information quantity condition. If yes, proceed with steps 42-44 below; otherwise, determine the first preset result (e.g., the string "no result") as the second link result corresponding to the entity to be linked.

[0125] The first preset result is used to indicate the link result of an entity word that cannot be accurately determined by the above entity link model; and this application does not limit the implementation of the first preset result. For example, it can be implemented using the string "no result".

[0126] Based on the relevant content of step 41 above, for the entity to be linked, after obtaining the context information of the entity to be linked, if it is determined that the context information does not meet the preset information quantity condition, it can be determined that the context information cannot provide sufficient context for the entity to be linked. Thus, it can be determined that the entity linking model cannot obtain sufficient context, resulting in the entity linking model giving an accurate linking result for the entity to be linked. Therefore, the first preset result can be determined as the second linking result corresponding to the entity to be linked, so that the second linking result can represent the linking result that cannot be accurately determined by the entity linking model above.

[0127] Step 42: When it is determined that the context information of the entity to be linked meets the preset information content condition, the entity to be linked and its context information are input into the entity linking model to obtain the entity linking description result output by the entity linking model. The entity linking description result includes the confidence scores of at least one candidate object corresponding to the entity to be linked; the at least one candidate object includes a second object and a third object; the confidence score of the second object is the maximum value among the confidence scores of the at least one candidate object; the confidence score of the third object is the second largest value among the confidence scores of the at least one candidate object.

[0128] It should be noted that the relevant content of step 42 can be found in step 31 above, and will not be repeated here for the sake of brevity.

[0129] Step 43: If the confidence level of the second object mentioned above is greater than the first threshold, and the difference between the confidence level of the second object and the confidence level of the third object is greater than the second threshold, then generate the second link result based on the second object.

[0130] It should be noted that the relevant content of step 43 can be found in step 32 above, and will not be repeated here for the sake of brevity.

[0131] Step 44: If the confidence level of the second object mentioned above is not greater than the first threshold, or the difference between the confidence level of the second object and the confidence level of the third object is not greater than the second threshold, then the first preset result mentioned above is determined as the second linking result corresponding to the entity to be linked.

[0132] In this application, if it is determined that the confidence level of the second object is not greater than the first threshold, it can be determined that it is unreliable to associate the second object with the linked entity in the preceding text. Therefore, the first preset result can be directly determined as the second link result corresponding to the linked entity, so that the second link result can represent the link result that cannot be accurately determined by the linked entity model in the preceding text. In addition, if it is determined that the difference between the confidence level of the second object and the confidence level of the third object is not greater than the second threshold, it can be determined that ambiguity is likely to occur when determining the linked object associated with the linked entity based on the linked entity description result in the preceding text. Therefore, the first preset result can be directly determined as the second link result corresponding to the linked entity, so that the second link result can represent the link result that cannot be accurately determined by the linked entity model in the preceding text.

[0133] Based on the relevant content of steps 41 to 44 above, it can be seen that for the entity to be linked, if the linking object associated with the entity cannot be accurately given by the pre-built entity linking model, the first preset result can be directly determined as the second linking result corresponding to the entity to be linked, so that the second linking result can represent the linking result of the entity to be linked that cannot be accurately determined by the entity linking model above.

[0134] Based on the relevant content in S3 above, after obtaining the entity to be linked and its context information, the second link result corresponding to the entity to be linked can be determined by using a pre-built entity linking model (e.g., Figure 2 The "linking results determined by the model" shown above enable the second linking result to represent the linking result given by the entity linking model for the entity to be linked, thereby achieving the purpose of entity linking processing with the help of a machine learning model.

[0135] Step 2: If the entity to be linked is an abbreviation, and the text to be processed contains the full name information corresponding to the entity to be linked, then determine the third link result corresponding to the entity to be linked based on the full name information.

[0136] Abbreviations can be used to represent a phrase (e.g., the phrases shown in Table 1 below).

[0137]

[0138] Table 1. Some phrases and their corresponding abbreviations

[0139] In fact, since abbreviations usually conform to a certain format, in order to better improve the recognition effect of abbreviations, this application also provides an abbreviation recognition process, which may include steps 51-53 below.

[0140] Step 51: After obtaining the entity to be linked, it can be determined whether the entity to be linked meets the preset abbreviation conditions. If yes, proceed to step 52 below; if no, proceed to step 53 below.

[0141] The preset abbreviation conditions are used to represent the common format characteristics of all abbreviations. Furthermore, this application does not limit the preset abbreviation conditions. For example, in some application scenarios, the preset abbreviation conditions may specifically be: the abbreviation should consist only of English letters, numbers, and specific symbols, and the first character should not be a symbol. Similarly, in other application scenarios, the preset abbreviation conditions may be: the abbreviation should consist only of English letters, numbers, and specific symbols, with no more than one consecutive digit, and the first character should not be a symbol. Therefore, in one possible implementation, the preset abbreviation conditions can be set in advance according to the specific application scenario.

[0142] Step 52: If the entity to be linked above meets the preset abbreviation conditions, then it can be determined that the entity to be linked belongs to an abbreviation.

[0143] Step 53: If the entity to be linked above does not meet the preset abbreviation conditions, then it can be determined that the entity to be linked does not belong to the abbreviation.

[0144] Based on the relevant content of steps 51 to 53 above, after obtaining the entity to be linked, it can be determined whether the entity to be linked meets the preset abbreviation conditions. If it does, it can be determined that the entity to be linked belongs to an abbreviation. Therefore, it can be queried from the text to be processed above to see if there is full name information corresponding to the entity to be linked.

[0145] The "fully qualified name information corresponding to the entity to be linked" mentioned above refers to the phrase represented by that entity. For example, if the entity to be linked is the abbreviation of "URL" mentioned above, then the "fully qualified name information corresponding to the entity to be linked" would be the phrase "uniform resource locator" mentioned above.

[0146] Research has revealed a correspondence between abbreviations and their corresponding full names, as shown in (1)-(3) below.

[0147] (1) When an abbreviation has the characteristic of being “pure English”, there is a correspondence between the abbreviation and its corresponding full name information as shown in (1.1)-(1.2) below.

[0148] (1.1) If there is no special lowercase form in the middle of the abbreviation, the number of words in the full name of the abbreviation is the same as the number of letters in the abbreviation, and any connector can be used between the words in the full name (e.g., spaces, underscores _, hyphens -, and the use of multiple connectors). For example, when the abbreviation is "URL" as mentioned above, the full name information corresponding to the abbreviation conforms to the template shown in Table 2 below.

[0149] abbreviations URL Full Name Template 1 Uxxx Rxxx Lxxx Full Name Template 2 Uxxx-Rxxx-Lxxx Full Name Template 3 Uxxx_Rxxx_Lxxx Full Name Template 4 Uxxx-Rxxx Lxxx

[0150] Table 2 shows the correspondence between abbreviations and full name templates.

[0151] It should be noted that "xxx" mentioned in Table 2 above can be used to represent strings of any length.

[0152] (1.2) If there is a special lowercase form in the middle of the abbreviation, it can be distinguished from the uppercase at both ends. The lowercase part is considered to be a combination of a noun and an abbreviation to form a new word. As for the uppercase part, it can be processed in the manner shown in (1.1) above. For example, when the abbreviation is "WebRTC" as mentioned above, the full name information corresponding to the abbreviation conforms to the template shown in Table 3 below.

[0153] abbreviations WebRTC Full Name Template 1 Web Rxxx Txxx Txxx Full Name Template 2 Web Rxxx-Txxx-Txxx Full Name Template 3 Web Rxxx_Txxx_Txxx Full Name Template 4 Web Rxxx-Txxx Txxx

[0154] Table 3 shows another correspondence between abbreviations and full name templates.

[0155] It should be noted that "xxx" mentioned in Table 3 above can be used to represent strings of any length.

[0156] (2) When an abbreviation has the feature of "English + symbol", the number of words in the full name corresponding to the English part of the abbreviation is the same as the number of letters in the abbreviation. The symbols in the abbreviation are retained in the full name, and the link symbols in the English part should be distinguished from the symbols in the abbreviation. For example, when the abbreviation is "TFT-LCD" as mentioned above, the full name information corresponding to the abbreviation conforms to the template shown in Table 4 below.

[0157] abbreviations TFT-LCD Full Name Template 1 Txxx Fxxx Txxx-Lxxx Cxxx Dxxx Full Name Template 2 Txxx_Fxxx_Txxx-Lxxx_Cxxx_Dxxx

[0158] Table 4 shows another type of correspondence between abbreviations and full name templates.

[0159] It should be noted that "xxx" mentioned in Table 4 above can be used to represent strings of any length.

[0160] (3) When an abbreviation has the feature of “English + number”, there is a correspondence between the abbreviation and its corresponding full name information as shown in (3.1)-(3.2) below.

[0161] (3.1) If an abbreviation has the characteristic of "single number + letter", then the number in the abbreviation is a combination of the first English words that are repeated a certain number of times to form the full name, and the abbreviation is also connected by a connector. For example, when the abbreviation is "4S" or "W3C" as mentioned above, the full name information corresponding to the abbreviation conforms to the template shown in Table 5 below.

[0162]

[0163] Table 5 shows another type of correspondence between abbreviations and full name templates.

[0164] It should be noted that "xxx" mentioned in Table 5 above can be used to represent strings of any length.

[0165] (3.2) If an abbreviation has the characteristic of "multiple consecutive numbers + letters", then the numbers in the abbreviation represent multiple strings of English letters, which are unrelated to other parts of the abbreviation and only indicate the length of the string. For example, when the abbreviation is "i18n" as mentioned above, the full name information corresponding to the abbreviation conforms to the template shown in Table 6 below.

[0166] Table 6 also shows the correspondence between abbreviations and full name templates.

[0167] It should be noted that "xxx" in Table 6 above refers to a string consisting of 18 letters.

[0168] Based on the relevant content in (1) to (3) above, it can be seen that different types of abbreviations have different abbreviation characteristics, and different types of abbreviations correspond to different full name structures (that is, full name templates). Based on this, this application also provides a process for determining the "full name information corresponding to the entity to be linked" mentioned above, which may specifically include steps 61-64 below.

[0169] Step 61: If the entity to be linked is an abbreviation, then determine the abbreviation feature of the entity to be linked so that the abbreviation feature can represent the abbreviation state presented by the entity to be linked (for example, any abbreviation state shown in (1) to (3) above).

[0170] It should be noted that this application does not limit the process of determining the "abbreviation features of the entity to be linked" mentioned above. For example, it can be implemented using a pre-built machine learning model with abbreviation feature extraction function. Or, it can be implemented using pre-defined abbreviation feature determination rules (e.g., regular expressions).

[0171] Step 62: Generate the template to be used for the entity to be linked, according to the full name template generation rules corresponding to the abbreviation features mentioned above.

[0172] Among them, the full name template generation rule refers to the rule pre-set for generating the full name template based on the "abbreviation features of the entity to be linked" mentioned above; and this application does not limit the full name template generation rule. For example, when the "abbreviation features of the entity to be linked" are the abbreviation features shown in (1.1) above (that is, the two features of "pure English" and "no special lowercase form in the middle of the abbreviation"), the full name template generation rule can be the rule shown in (1.1) above (that is, the rule that "the number of words in the full name corresponding to the abbreviation is the same as the number of letters in the abbreviation, and any connector can be used between the words in the full name").

[0173] The “template to be used” mentioned above refers to the full name template generated for the entity to be linked above; moreover, this application does not limit the “template to be used”. For example, when the “entity to be linked” above is the abbreviation of “URL” above, the “template to be used” may include the four full name templates shown in Table 2 above.

[0174] Step 63: Using the template to be used above, perform template retrieval processing on the text to be processed to obtain the retrieval results.

[0175] The template retrieval process is used to determine the string that matches a certain template from text data; and this application does not limit the implementation of the template retrieval process. For example, it can be implemented by string matching, regular expression matching, dictionary matching, etc.

[0176] The search results are used to indicate whether there is a string in the text to be processed that matches the template to be used.

[0177] Step 64: Based on the search results above, determine the full name information of the entity to be linked.

[0178] It should be noted that this application does not limit the implementation of step 64. For example, it may include steps 641-643 as described below.

[0179] Step 641: If the above search results indicate that there is no string in the text to be processed that matches the template to be used, then it can be determined that there is no full name information corresponding to the entity to be linked in the text to be processed. Therefore, the pre-set second preset result (e.g., the string "no result") can be determined as the third link result corresponding to the entity to be linked, so that the third link result can indicate that the link object associated with the entity to be linked cannot be determined by strong related information.

[0180] The second preset result is used to indicate the link result where an entity word cannot be determined by using strong related information; and this application does not limit the implementation of the second preset result. For example, it can be implemented using the string "no result".

[0181] Step 642: If the above search results indicate that there is only one string in the text to be processed that matches the template to be used, then use that string as the full name information of the entity to be linked, and determine the third link result corresponding to the entity to be linked based on the full name information.

[0182] Step 643: If the above search results indicate that there are multiple strings in the text to be processed that match the template to be used, then according to the preset selection rules, filter out the full name information corresponding to the entity to be linked from these strings, and determine the third link result corresponding to the entity to be linked based on the full name information.

[0183] Among them, the preset selection rules can be set in advance according to the application scenario. For ease of understanding, the following explanation is based on steps 71-73.

[0184] As an example, when the search results above indicate that there are E strings in the text to be processed that match the template to be used, the process of determining the "full name information corresponding to the entity to be linked" above may include steps 71-73 below.

[0185] Step 71: Determine whether there are preset indicator words before and after the position of the e-th string in the text to be processed, and obtain the judgment result corresponding to the e-th string. Where e is a positive integer, e≤E, and E is a positive integer.

[0186] The preset indicator words can be pre-defined, for example, they can be words such as "full name", "abbreviation", "represent", or "point".

[0187] Step 72: Calculate the distance between the position of the e-th string in the text to be processed and the position of the preceding entity to be linked in the text to be processed, thus obtaining the distance corresponding to the e-th string. Here, e is a positive integer, e≤E, and E is a positive integer.

[0188] Step 73: Based on the judgment results and distances of the E strings, select the optimal string from the E strings as the full name information of the entity to be linked.

[0189] It should be noted that the filtering rule for the "optimal string" mentioned above is: prioritize strings that have preset indicator words and are relatively close; and this filtering rule can be preset.

[0190] Based on the relevant content of steps 61 to 64 above, for the entity to be linked, after determining that the entity to be linked is an abbreviation, the full name template corresponding to the entity to be linked can be determined first based on the abbreviation characteristics of the entity to be linked; then, based on the full name template, template retrieval processing is performed on the text to be processed above to obtain the retrieval results; finally, based on the retrieval results, the full name information corresponding to the entity to be linked is determined so that the third link result corresponding to the entity to be linked can be determined based on the full name information in the future.

[0191] The "third-party link result" mentioned above refers to the link result determined for the entity to be linked, using strongly related information (e.g., full name information). For example, this "third-party link result" could be... Figure 2 The image shows "Link results determined based on strong relevant information".

[0192] Furthermore, this application does not limit the process of determining the "third link result" mentioned above. For example, when the entity linking method provided in this application is used to associate the entity to be linked with an object (e.g., definition content) in the entity thesaurus, the process of determining the "third link result" may specifically include steps 81-83 below.

[0193] Step 81: Match the full name information of the entity to be linked with at least one candidate object in the entity word lexicon to obtain the matching result.

[0194] The entity thesaurus refers to the database to which the linked objects associated with the "entity to be linked" mentioned above belong; moreover, this application is not limited to this entity thesaurus, for example, it can be implemented using any existing or future database that requires entity linking. It should be noted that all objects involved in this application come from this entity thesaurus.

[0195] The “candidate object” mentioned above refers to any object that can be associated with entity words in the “entity word lexicon” mentioned above; and this application does not limit the “candidate object”, for example, it can refer to the definition content.

[0196] The term "matching result" above indicates whether there exists a candidate object in the "entity term lexicon" above that matches the "full name information corresponding to the entity to be linked" above. Furthermore, this application does not limit the specific meaning of this match. For example, when all objects involved in this application refer to explanatory content, the specific meaning of this match can be: if the "full name information corresponding to the entity to be linked" appears in the q-th candidate object, or if the "full name information corresponding to the entity to be linked" is an alias of the q-th candidate object, then it can be determined that the q-th candidate object matches the "full name information corresponding to the entity to be linked"; otherwise, there is no match. Here, q is a positive integer, q≤Q, where Q is a positive integer and represents the number of candidate objects in the "entity term lexicon" above.

[0197] Step 82: If the above matching result indicates that there is a fourth object among the candidate objects that matches the full name information, then determine the third link result corresponding to the entity to be linked based on the fourth object.

[0198] The fourth object refers to the candidate object that exists in the "entity word thesaurus" mentioned above and matches the "full name information corresponding to the entity to be linked" mentioned above.

[0199] Furthermore, this application does not limit the implementation of step 82 above, "determine the third link result based on the fourth object". For example, it can specifically be: determine the fourth object as the third link result corresponding to the entity to be linked above.

[0200] Step 83: If the above matching result indicates that there is no candidate object matching the full name information among at least one candidate object in the above, then the second preset result in the above is determined as the third link result corresponding to the entity to be linked in the above, so that the third link result can indicate that the link object associated with the entity to be linked cannot be determined by strong related information.

[0201] Based on the relevant content of steps 81 to 83 above, if the entity to be linked is an abbreviation, and the full name information corresponding to the entity to be linked exists in the text to be processed above, then the third link result corresponding to the entity to be linked can be determined based on the full name information.

[0202] In fact, in order to better improve the effect of entity linking, this application also provides another possible implementation of the above-mentioned "third link result", which may specifically include steps 91-93 below.

[0203] Step 91: If the entity to be linked is not an abbreviation, then the second preset result above is determined as the third link result corresponding to the entity to be linked above.

[0204] Step 92: If the entity to be linked is an abbreviation, but the full name information corresponding to the entity to be linked does not exist in the text to be processed above, then the second preset result above will be determined as the third link result corresponding to the entity to be linked above.

[0205] Step 93: If the entity to be linked is an abbreviation, and the full name of the entity to be linked exists in the preceding text, then determine the third link result corresponding to the entity to be linked based on the full name information.

[0206] Based on the relevant content of steps 91 to 93 above, for the entity to be linked, if it is determined that the entity to be linked is not an abbreviation, or if it is determined that the entity to be linked is an abbreviation, but the full name information corresponding to the entity to be linked does not exist in the text to be processed above, then it can be determined that there is no strong relevant information about the entity to be linked in the text to be processed. Therefore, it can be determined that the linked object associated with the entity to be linked cannot be determined by using strong relevant information. Thus, the second preset result above can be determined as the third link result corresponding to the entity to be linked, so that the third link result can indicate that the linked object associated with the entity to be linked cannot be determined by using strong relevant information. However, if it is determined that the entity to be linked is an abbreviation, and the full name information corresponding to the entity to be linked exists in the text to be processed, then it can be determined that there is strong relevant information about the entity to be linked in the text to be processed. Therefore, the third link result corresponding to the entity to be linked can be determined by using the strong relevant information (that is, the full name information). Figure 2The "link results determined based on strong relevant information" shown above enable the third link result to represent the objects that can be recommended to the target user as linked objects associated with the entities to be linked above, as determined based on strong relevant information. This achieves the purpose of entity linking processing by leveraging the strong relevant information present in the text to be processed.

[0207] Step 3: Based on the first link result, the second link result, and the third link result, determine the target link object of the entity to be linked.

[0208] The target link object refers to the link object that is finally determined for the entity to be linked above, so that the target link object can be provided to the target user as the link object associated with the entity to be linked.

[0209] In addition, this application does not limit the process of determining the "target link object" mentioned above. For example, it may specifically include steps 101-103 below.

[0210] Step 101: If the "third link result" above is the fourth object above, then it can be determined that there is a link object determined based on strong related information. Therefore, the fourth object can be determined as the target link object of the entity to be linked above.

[0211] Step 102: If the above "third link result" is the second preset result, and the above "second link result" is the above second object, then it can be determined that there is no link object determined based on strong relevant information, but there is a link object determined based on the model. Therefore, the second object can be determined as the target link object of the above entity to be linked.

[0212] Step 103: If the above "third link result" is the second preset result and the above "second link result" is the first preset result, then it can be determined that there is neither a link object determined based on strong relevant information nor a link object determined based on the model. Therefore, the above first link result (that is, the above "first object") can be determined as the target link object of the above entity to be linked.

[0213] Based on the relevant content of steps 101 to 103 above, for the entity to be linked, after obtaining multiple link results determined for the entity to be linked, the link object determined based on strong relevant information is selected first as the link object associated with the entity to be linked; the link object determined based on the model is selected second as the link object associated with the entity to be linked; and the link object determined based on historical behavior is selected as the last resort as the link object associated with the entity to be linked.

[0214] Based on the relevant content of steps one to three above, in one possible implementation, the entity linking method may include: after determining the entity to be linked and its context information from the text to be processed, using at least one candidate object corresponding to the entity to be linked recorded in a pre-built link relationship library, determining a first linking result corresponding to the entity to be linked; furthermore, if the context information of the entity to be linked meets a preset information content condition, then determining a second linking result corresponding to the entity to be linked based on the entity to be linked, its context information, and a pre-built entity linking model; furthermore, if the entity to be linked is an abbreviation, and the text to be processed contains the full name information corresponding to the entity to be linked, then determining a third linking result corresponding to the entity to be linked based on the full name information; finally, determining the target linking object of the entity to be linked based on the first linking result, the second linking result, and the third linking result, thus achieving the purpose of entity linking. Since the target linking object is determined based on three linking methods, it can better represent which object the entity to be linked to, thereby improving the entity linking effect.

[0215] Based on the relevant content of S1 to S3 above, the entity linking method provided in this application embodiment may include: after determining the entity to be linked and its context information from the text to be processed, using at least one candidate object corresponding to the entity to be linked recorded in a pre-built link relationship library, determining the first linking result corresponding to the entity to be linked, and determining the target linking object of the entity to be linked based on the first linking result and the context information of the entity to be linked, thereby achieving the purpose of entity linking.

[0216] In fact, to further improve the accuracy of the "linked objects determined based on historical behavior" mentioned above, this application also provides a possible implementation of the entity linking method mentioned above. In this implementation, the entity linking method may include not only S1-S3 mentioned above, but also steps 111-113 mentioned below. Among them, the execution time of step 111 is later than the execution time of S3 mentioned above.

[0217] Step 111: Obtain the actual feedback operation corresponding to the target link object mentioned above.

[0218] In this context, "actual feedback operation" refers to the feedback actions triggered by the target user after the target link object is provided as the link object associated with the entity to be linked. For example, this "actual feedback operation" could refer to... Figure 2 The "link feedback operation" shown.

[0219] In addition, the relevant content of "actual feedback operation" above is similar to the relevant content of "feedback operation" above, so for the sake of brevity, it will not be repeated here.

[0220] Furthermore, this application does not limit the process of obtaining the "actual feedback operation" mentioned above. For example, it can be implemented using any existing or future operation acquisition method.

[0221] Step 112: Update the recommendation representation data corresponding to the target link object based on the actual feedback operation corresponding to the target link object mentioned above.

[0222] It should be noted that this application does not limit the implementation of step 112 above. For example, it can be implemented using the formula above.

[0223] Step 113: Update the link relationship database based on the recommended representation data corresponding to the target link object mentioned above.

[0224] It should be noted that this application does not limit the implementation of step 113. For example, it may specifically include steps 1131-1132 below.

[0225] Step 1131: If there is no correspondence between the entity to be linked and the target linked object in the link relationship database, the correspondence and the recommendation representation data corresponding to the target linked object can be added to the link relationship database.

[0226] Step 1132: If there is a correspondence between the entity to be linked and the target linked object in the link relationship database, then the "recommendation representation data corresponding to the target linked object" recorded in the link relationship database can be updated directly using the recommendation representation data corresponding to the target linked object.

[0227] Based on the relevant content of steps 111 to 113 above, after obtaining the target link object of the entity to be linked, the target link object can be provided to the target user as the link object associated with the entity to be linked. After receiving the actual feedback operation triggered by the target user on the target link object, the link relationship database above is updated based on the actual feedback operation, so that the link relationship database can represent the link usage habits of the target user as realistically as possible. This helps to improve the accuracy of the "link object determined based on historical behavior" above, thereby improving the entity link determination effect.

[0228] Based on the entity linking method provided in the embodiments of this application, the embodiments of this application also provide an entity linking device, which is described below in conjunction with... Figure 3 Explanation and clarification will be provided. Among them, Figure 3This is a schematic diagram of a physical linking device provided in an embodiment of this application. It should be noted that for technical details of the physical linking device provided in this embodiment, please refer to the relevant content of the physical linking method described above.

[0229] like Figure 3 As shown, the entity linking device 300 provided in this application embodiment includes:

[0230] The first determining unit 301 is used to determine the entity to be linked and the context information of the entity to be linked from the text to be processed;

[0231] The second determining unit 302 is used to determine the first linking result corresponding to the entity to be linked by using a pre-built linking relationship library; the linking relationship library includes the correspondence between the entity to be linked and at least one candidate object; the first linking result is determined based on the at least one candidate object;

[0232] The third determining unit 303 is used to determine the target linking object of the entity to be linked based on the first linking result and the context information of the entity to be linked.

[0233] In one possible implementation, the link relationship database further includes recommended representation data for each of the candidate objects; the recommended representation data of the candidate objects is used to indicate the probability of associating the candidate object with the entity to be linked;

[0234] The first link result is determined based on the at least one candidate object and the recommended representation data of the at least one candidate object.

[0235] In one possible implementation, the second determining unit 302 is specifically used to: compare the recommended representation data of the at least one candidate object to obtain a comparison result; if the comparison result indicates that the recommended representation data of the first object among the at least one candidate object is the largest, then generate the first link result based on the first object.

[0236] In one possible implementation, the recommended representation data of the candidate object is determined based on the historical feedback operations corresponding to the candidate object.

[0237] In one possible implementation, the physical linking device 300 further includes:

[0238] The feedback acquisition unit is used to acquire the actual feedback operation corresponding to the target link object;

[0239] The first update unit is used to update the recommended representation data corresponding to the target link object according to the actual feedback operation corresponding to the target link object;

[0240] The second update unit is used to update the link relationship database based on the recommended representation data corresponding to the target link object.

[0241] In one possible implementation, the third determining unit 303 includes:

[0242] The first determining subunit is used to determine the second linking result corresponding to the entity to be linked based on the entity to be linked, the context information of the entity to be linked, and the pre-constructed entity linking model if the context information of the entity to be linked meets the preset information quantity condition.

[0243] The second determining subunit is used to determine the third linking result corresponding to the entity to be linked based on the full name information if the entity to be linked is an abbreviation and the text to be processed contains the full name information corresponding to the entity to be linked.

[0244] The third determining subunit is used to determine the target link object of the entity to be linked based on the first linking result, the second linking result, and the third linking result.

[0245] In one possible implementation, the first determining subunit is specifically configured to: input the entity to be linked and its context information into the entity linking model to obtain an entity linking description result output by the entity linking model; wherein the entity linking description result includes the confidence level of at least one candidate object corresponding to the entity to be linked; the at least one candidate object includes a second object and a third object; the confidence level of the second object is the maximum value among the confidence levels of the at least one candidate object; the confidence level of the third object is the second largest value among the confidence levels of the at least one candidate object; if the confidence level of the second object is greater than a first threshold, and the difference between the confidence level of the second object and the confidence level of the third object is greater than a second threshold, then the second linking result is generated based on the second object.

[0246] In one possible implementation, the second determining subunit is specifically used to: match the full name information with at least one candidate object in the entity word lexicon to obtain a matching result; if the matching result indicates that there is a fourth object among the at least one candidate object that matches the full name information, then determine the third link result based on the fourth object.

[0247] In one possible implementation, the physical linking device 300 further includes:

[0248] The fourth determining unit is used to determine the abbreviation feature of the entity to be linked if the entity to be linked is an abbreviation.

[0249] The template generation unit is used to generate the template to be used corresponding to the entity to be linked, according to the full name template generation rules corresponding to the abbreviation feature.

[0250] The template retrieval unit is used to perform template retrieval processing on the text to be processed using the template to be used, and to obtain retrieval results.

[0251] The fifth determining unit is used to determine the full name information corresponding to the entity to be linked based on the search results.

[0252] In one possible implementation, the physical linking device 300 further includes:

[0253] The sixth determining unit is used to determine that the entity to be linked belongs to an abbreviation if the entity to be linked meets the preset abbreviation conditions.

[0254] In one possible implementation, the physical linking device 300 further includes:

[0255] A scene acquisition unit is used to acquire the text scene identifier of the text to be processed;

[0256] An information lookup unit is used to search for the link relationship library corresponding to the text scene identifier from a pre-built mapping relationship; wherein the mapping relationship includes the correspondence between the text scene identifier and the link relationship library.

[0257] Based on the above-mentioned content of the entity linking device 300, it can be seen that, for the entity linking device 300 provided in this application embodiment, after determining the entity to be linked and the context information of the entity to be linked from the text to be processed, the first linking result corresponding to the entity to be linked is determined by using at least one candidate object corresponding to the entity to be linked recorded in the pre-built linking relationship library, and the target linking object of the entity to be linked is determined according to the first linking result and the context information of the entity to be linked, which is conducive to improving the entity linking effect.

[0258] In addition, this application also provides an electronic device, the device including a processor and a memory: the memory is used to store instructions or computer programs; the processor is used to execute the instructions or computer programs in the memory, so that the electronic device performs any implementation of the entity linking method provided in this application.

[0259] See Figure 4This diagram illustrates a structural schematic of an electronic device 400 suitable for implementing embodiments of the present disclosure. The terminal devices in these embodiments may include, but are not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (personal digital assistants), PADs (tablet computers), PMPs (portable multimedia players), in-vehicle terminals (e.g., in-vehicle navigation terminals), and fixed terminals such as digital TVs and desktop computers. Figure 4 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0260] like Figure 4 As shown, electronic device 400 may include a processing device (e.g., a central processing unit, a graphics processing unit, etc.) 401, which can perform various appropriate actions and processes according to a program stored in read-only memory (ROM) 402 or a program loaded from storage device 408 into random access memory (RAM) 403. RAM 403 also stores various programs and data required for the operation of electronic device 400. The processing device 401, ROM 402, and RAM 403 are interconnected via bus 404. Input / output (I / O) interface 405 is also connected to bus 404.

[0261] Typically, the following devices can be connected to I / O interface 405: input devices 406 including, for example, touchscreens, touchpads, keyboards, mice, cameras, microphones, accelerometers, gyroscopes, etc.; output devices 407 including, for example, liquid crystal displays (LCDs), speakers, vibrators, etc.; storage devices 408 including, for example, magnetic tapes, hard disks, etc.; and communication devices 409. Communication device 409 allows electronic device 400 to communicate wirelessly or wiredly with other devices to exchange data. Although Figure 4 An electronic device 400 with various devices is shown; however, it should be understood that it is not required to implement or possess all of the devices shown. More or fewer devices may be implemented or possessed alternatively.

[0262] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device 409, or installed from a storage device 408, or installed from a ROM 402. When the computer program is executed by the processing device 401, it performs the functions defined in the methods of embodiments of this disclosure.

[0263] The electronic device provided in this embodiment belongs to the same inventive concept as the method provided in the above embodiments. Technical details not described in detail in this embodiment can be found in the above embodiments, and this embodiment has the same beneficial effects as the above embodiments.

[0264] This application also provides a computer-readable medium storing instructions or a computer program that, when executed on a device, causes the device to perform any implementation of the entity linking method provided in this application.

[0265] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0266] In some implementations, clients and servers can communicate using any currently known or future-developed network protocol such as HTTP (Hypertext Transfer Protocol) and can interconnect with digital data communication (e.g., communication networks) of any form or medium. Examples of communication networks include local area networks (“LANs”), wide area networks (“WANs”), the Internet (e.g., the Internet of Things), and end-to-end networks (e.g., ad hoc end-to-end networks), as well as any currently known or future-developed networks.

[0267] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0268] The aforementioned computer-readable medium carries one or more programs, which, when executed by the electronic device, enable the electronic device to perform the aforementioned methods.

[0269] Computer program code for performing the operations of this disclosure can be written in one or more programming languages ​​or a combination thereof, including but not limited to object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0270] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this disclosure. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0271] The units described in the embodiments of this disclosure can be implemented in software or hardware. The names of the units / modules do not necessarily limit the specific unit itself.

[0272] The functions described above in this document can be performed, at least in part, by one or more hardware logic components. For example, exemplary types of hardware logic components that can be used, without limitation, include: Field Programmable Gate Arrays (FPGAs), Application-Specific Integrated Circuits (ASICs), Application Standard Products (ASSPs), System-on-Chip (SoCs), Complex Programmable Logic Devices (CPLDs), and so on.

[0273] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0274] It should be noted that the various embodiments in this specification are described in a progressive manner, with each embodiment focusing on the differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the systems or apparatus disclosed in the embodiments, since they correspond to the methods disclosed in the embodiments, the descriptions are relatively simple, and relevant parts can be referred to the method section.

[0275] It should be understood that in this application, "at least one (item)" means one or more, and "more than" means two or more. "And / or" is used to describe the relationship between related objects, indicating that three relationships can exist. For example, "A and / or B" can represent three cases: only A exists, only B exists, and both A and B exist simultaneously, where A and B can be singular or plural. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship. "At least one (item) of the following" or similar expressions refer to any combination of these items, including any combination of single or plural items. For example, at least one (item) of a, b, or c can represent: a, b, c, "a and b", "a and c", "b and c", or "a and b and c", where a, b, and c can be single or multiple.

[0276] It should also be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0277] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0278] The above description of the disclosed embodiments enables those skilled in the art to make or use this application. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of this application. Therefore, this application is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for linking entities, characterized in that, The method includes: Determine the entity to be linked and its context information from the text to be processed; Using a pre-built link relationship database, a first link result corresponding to the entity to be linked is determined; the link relationship database includes the correspondence between the entity to be linked and at least one candidate object; the first link result is determined based on the at least one candidate object. Based on the first linking result and the context information of the entity to be linked, the target linking object of the entity to be linked is determined. The target linking object is determined based on the first linking result and the third linking result corresponding to the entity to be linked. The process of determining the third linking result includes: in response to the entity to be linked being an abbreviation, determining the abbreviation feature of the entity to be linked; generating the full name template corresponding to the entity to be linked according to the full name template generation rule corresponding to the abbreviation feature; performing template retrieval processing on the text to be processed based on the full name template to obtain retrieval results; determining the full name information corresponding to the entity to be linked based on the retrieval results; and determining the third linking result corresponding to the entity to be linked based on the full name information.

2. The method according to claim 1, characterized in that, The link relationship database also includes recommended representation data for each candidate object; the recommended representation data of the candidate object is used to indicate the probability of associating the candidate object with the entity to be linked. The first link result is determined based on the at least one candidate object and the recommended representation data of the at least one candidate object.

3. The method according to claim 2, characterized in that, The process of determining the first link result includes: The recommended representation data of the at least one candidate object are compared to obtain the comparison result; If the comparison result indicates that the recommended representation data of the first object among the at least one candidate object is the largest, then the first link result is generated based on the first object.

4. The method according to claim 2, characterized in that, The recommended representation data of the candidate object is determined based on the historical feedback operations corresponding to the candidate object.

5. The method according to claim 1, characterized in that, After determining the target link object of the entity to be linked based on the first linking result and the context information of the entity to be linked, the method further includes: Obtain the actual feedback operation corresponding to the target link object; Update the recommendation representation data corresponding to the target link object based on the actual feedback operation corresponding to the target link object; The link relationship database is updated based on the recommended representation data corresponding to the target link object.

6. The method according to claim 1, characterized in that, The target link object is also determined based on the second link result corresponding to the entity to be linked; The process of determining the second link result includes: if the context information of the entity to be linked meets the preset information quantity condition, then the second link result corresponding to the entity to be linked is determined based on the entity to be linked, the context information of the entity to be linked, and the pre-constructed entity link model.

7. The method according to claim 6, characterized in that, The process of determining the second link result includes: The entity to be linked and its context information are input into the entity linking model to obtain the entity linking description result output by the entity linking model; wherein, the entity linking description result includes the confidence scores of at least one candidate object corresponding to the entity to be linked; the at least one candidate object includes a second object and a third object; the confidence score of the second object is the maximum value among the confidence scores of the at least one candidate object; the confidence score of the third object is the second largest value among the confidence scores of the at least one candidate object; If the confidence level of the second object is greater than the first threshold, and the difference between the confidence level of the second object and the confidence level of the third object is greater than the second threshold, then the second link result is generated based on the second object.

8. The method according to claim 1, characterized in that, The step of determining the third link result corresponding to the entity to be linked based on the full name information includes: The full name information is matched with at least one candidate object in the entity word lexicon to obtain the matching result; If the matching result indicates that there is a fourth object among the at least one candidate object that matches the full name information, then the third link result is determined based on the fourth object.

9. The method according to claim 1, characterized in that, The method further includes: If the entity to be linked meets the preset abbreviation conditions, then the entity to be linked is determined to be an abbreviation.

10. The method according to claim 1, characterized in that, Before determining the first link result corresponding to the entity to be linked using a pre-built link relationship library, the method further includes: Obtain the text scene identifier of the text to be processed; The link relationship library corresponding to the text scene identifier is found from the pre-built mapping relationship; wherein the mapping relationship includes the correspondence between the text scene identifier and the link relationship library.

11. A physical linking device, characterized in that, include: The first determining unit is used to determine the entity to be linked and the context information of the entity to be linked from the text to be processed; The second determining unit is used to determine the first linking result corresponding to the entity to be linked by using a pre-built linking relationship library; the linking relationship library includes the correspondence between the entity to be linked and at least one candidate object; The first linking result is determined based on the at least one candidate object; The third determining unit is used to determine the target linking object of the entity to be linked based on the first linking result and the context information of the entity to be linked. The target linking object is determined based on the first linking result and the third linking result corresponding to the entity to be linked. The process of determining the third link result includes: in response to the fact that the entity to be linked is an abbreviation, determining the abbreviation feature of the entity to be linked, generating a full name template corresponding to the entity to be linked according to the full name template generation rule corresponding to the abbreviation feature; performing template retrieval processing on the text to be processed based on the full name template to obtain retrieval results; determining the full name information corresponding to the entity to be linked based on the retrieval results; and determining the third link result corresponding to the entity to be linked based on the full name information.

12. An electronic device, characterized in that, The device includes: a processor and a memory; The memory is used to store instructions or computer programs; The processor is configured to execute the instructions or computer program in the memory to cause the electronic device to perform the method according to any one of claims 1-10.

13. A computer-readable medium, characterized in that, The computer-readable medium stores instructions or computer programs that, when executed on the device, cause the device to perform the method according to any one of claims 1-10.

Images