Information source retrieval methods, systems, and storage media including alias identification and same-name entity disambiguation

Abstract

This invention discloses an information source retrieval method, system, and storage medium that includes alias recognition and homonym disambiguation. The method includes: inputting the entity name to be retrieved; obtaining a set of aliases for the entity name using alias recognition technology; traversing the information source database to be retrieved based on the alias set to construct an information source network; initializing the information source nodes in the information source network as embeddings; constructing a loss function for real target entities based on the embeddings of the information source nodes; constructing a target loss function for positive and negative sample sets based on representation learning technology of the information source network; and implementing a joint homonym disambiguation technique based on the loss function for real target entities and the target loss function for positive and negative sample sets to obtain information sources corresponding to different real target entities for the user to choose from. This invention is more adaptable to real-world environments and can retrieve more comprehensive relevant information sources.

Description

Technical Field

[0001] This invention relates to the field of data query technology, and in particular to an information source retrieval method, system, terminal device, and computer-readable storage medium that includes alias identification and homonym disambiguation. Background Technology

[0002] In the information age, the aggregation and dissemination of massive amounts of data has given rise to the phenomenon of "entity name explosion." In the digital wave, the names of various entities, such as personal names, corporate brands, and geographical identifiers, are growing exponentially. Simultaneously, the proliferation of aliases profoundly impacts social cognition and data governance. The same entity can have multiple names due to cultural differences, usage scenarios, or dissemination needs. For example, the chemical substance "hydrogen peroxide" is often called "hydrogen peroxide water," and the academic institution "Massachusetts Institute of Technology" is simplified to "MIT" in Chinese online media. Even after Twitter was renamed "X," it still faces identity fragmentation due to user cognitive inertia. Similarly, when users search for information sources (such as web pages, PDF documents, and papers) of entities they are interested in in databases, aliasing issues can easily lead to incomplete searches, while name ambiguity issues can cause the searched information sources to not match the entities the user is truly interested in. Furthermore, the exact number of real target entities included in the searched material (e.g., does searching for "apple" only include the two real target entities of edible apples and Apple Inc., or are there more?) is a problem that existing name disambiguation methods struggle to solve.

[0003] Existing name-disambiguation methods, such as those in the literature (Baichuan Zhang, Mohammad Al Hasan: Name Disambiguation in Anonymized Graphs using Network Embedding. CIKM 2017: 1239-1248) and the patent with publication number CN118378107A, can be effective in entity disambiguation. However, they may have the following problems in practical retrieval applications: 1) They cannot handle entity aliases, which may result in incomplete data processing and unsatisfactory search results; 2) They require prior knowledge of the number of actual target entities to be identified, or the clustering disambiguation process and representation learning optimization process are performed separately, making unified optimization impossible and limiting the final disambiguation performance. Therefore, in information source retrieval, a name-disambiguation and alias identification method that can automatically confirm the number of actual target entities is crucial. Summary of the Invention

[0004] To address at least one of the technical problems in the prior art, the present invention provides an information source retrieval method, system, terminal device, and computer-readable storage medium that includes alias identification and same-name entity disambiguation.

[0005] The first objective of this invention is to provide an information source retrieval method that includes alias identification and same-name entity disambiguation.

[0006] The second objective of this invention is to provide an information source retrieval system that includes alias identification and same-name entity disambiguation.

[0007] The third objective of this invention is to provide a terminal device.

[0008] A fourth objective of this invention is to provide a computer-readable storage medium.

[0009] The first objective of this invention can be achieved by adopting the following technical solution:

[0010] A method for retrieving information sources that includes alias recognition and same-name entity disambiguation, the method comprising:

[0011] Input the entity name to be searched; based on the entity name, use alias recognition technology to obtain a set of aliases for the entity name;

[0012] Based on the alias set, traverse the database of information sources to be retrieved and construct an information source network;

[0013] Initialize the information source nodes in the information source network as embeddings; construct the loss function for the real target entity based on the embeddings of the information source nodes;

[0014] Based on representation learning techniques of information source networks, a target loss function for positive and negative sample sets is constructed.

[0015] A joint name-based disambiguation technique is implemented based on the loss function of real target entities and the target loss function of positive and negative sample sets to obtain information sources corresponding to different real target entities for users to choose from.

[0016] Furthermore, the joint name-based disambiguation technique, which achieves collaborative optimization based on the loss function of the real target entity and the target loss function of the positive and negative sample sets, obtains information sources corresponding to different real target entities for the user to choose from, including:

[0017] The objective function is constructed based on the loss function of the real target entity and the target loss function of the positive and negative sample sets as follows:

[0018] ;

[0019] In the formula, Let be the target loss function for the positive and negative sample sets. The loss function is the actual target entity. for Norm regularization term to prevent model overfitting; parameters and These are used to control the target entity loss function and penalty term, respectively.

[0020] Based on the learning rate, the Adam optimization method is used to optimize the objective function. The information source node set is obtained when the model converges overall or reaches the maximum number of iterations. The final embedding and the set of real target entities ;

[0021] Based on the final embedding and the set of real target entities Calculate the embedding of each information source To each real target entity distance The minimum distance indicates that the information source belongs to the real target entity; return all information sources corresponding to the real target entities to the user for the user to select the final search results.

[0022] Furthermore, the process of constructing the target loss function for the positive and negative sample sets is as follows:

[0023] Calculate the information source network using community detection algorithms. Community structure If node i and node j are in the same community, then ;

[0024] In information source network Choose any node ; and These are the node set and edge set of the information source network, respectively;

[0025] Random walk sequences are generated based on a community-oriented random walk strategy, and the calculation is performed. Node reaches The probability of a node is:

[0026] ;

[0027] In the formula, w is the edge weight. Represents a node The community; according to probability A random walk sequence is obtained by performing L random walks. ;

[0028] Traversing the information source network Each node serves as a training node. Take node random walk sequence As a training node Positive samples ), and randomly sample Not in the random walk sequence The nodes on the list are used as negative samples. );

[0029] Calculate the target loss function based on positive and negative sample sets:

[0030] ;

[0031] In the formula, The minimum value greater than 0.

[0032] Furthermore, the construction process of the real target entity loss function is as follows:

[0033] Initialize the information sources in the information source network as candidate real target entities; the initial set of candidate real target entities is denoted as... The number of candidate real target entities k is initialized to the number of information sources n;

[0034] For each candidate real target entity Calculate its density ;

[0035] set Candidate real target entities in the data are ranked by their density Sort in descending order to prioritize processing high-density real target entities in subsequent steps;

[0036] The process generates new candidate real target entities and splits and merges existing candidate real target entities, then updates the set. At the same time, update the number of candidate real target entities. ;

[0037] Based on the embedding of each information source and the embedding of the candidate real target entity, in order to make each information source closer to the candidate real target entity and to make each candidate real target entity farther away from each other, the real target entity loss function is designed as follows:

[0038] ;

[0039] In the formula, Embedded as an information source.

[0040] Furthermore, the process involves generating new candidate real target entities, splitting and merging existing candidate real target entities, and updating the set. ,include:

[0041] Calculate the probability p of generating a new candidate real target entity and the probability (1-p) of splitting and merging existing candidate real target entities; where the probability p of the new candidate real target entity is calculated as follows:

[0042] ;

[0043] In the formula, This indicates the number of times the process generates new candidate real target entities. Indicates the total number of runs;

[0044] Based on the current number and Embedding of candidate real target entities and The following candidate new real target entities are generated:

[0045] ;

[0046] In the formula, Set value;

[0047] Will Add to collection And update the number of candidate real target entities. ;

[0048] set Candidate real target entities in the data are ranked by their density Sort in descending order; targeting current high-density real-world target entities. ,Will optimal radius The average value of all real target entities within the range is taken as the current high-density real target entity. The representation of:

[0049] ;

[0050] for Real target entities within the optimal radius optimal radius Unable to cover In order to ensure and Overlay each other, perform a split operation to update the real target entity. :

[0051] ;

[0052] It has already been integrated into From the candidate real target entities in Remove from the list and update the actual target entity count. .

[0053] Furthermore, the step of traversing the information source database to be retrieved based on the alias set and constructing the information source network includes:

[0054] Traverse the information source database. If any alias in the alias set appears in the name or content of an information source in the information source database, then that information source is related to the entity name A to be retrieved, and add that information source to the node set. middle;

[0055] For node sets Entity identification is performed on the name and content of each information source to obtain the entity set of each information source; for each entity name B in the entity set, the alias table is retrieved to obtain the alias set B_all; and the entity name B and related aliases in the entity set are uniformly replaced with the entity name B.

[0056] Building an information source network , For the edge set of the information source network, This indicates that two information sources have an intersection, and the number of intersections is used as the weight of the edge.

[0057] Furthermore, based on the entity name, an alias recognition technology is used to obtain a set of aliases for the entity name, including:

[0058] Based on entity name A, call the alias generation interface to generate a set of aliases for the entity name;

[0059] If an entity name A is retrieved from the alias table, and no alias exists, entity name A is inserted into the entity table, and the alias set generated by the alias generation interface is added to the alias table accordingly. At the same time, the relationship between entity name A in the entity table and the corresponding alias in the alias table is constructed in the relationship mapping table. If an alias exists, all existing aliases for each entity corresponding to that alias are updated, and the missing aliases are added to the alias table.

[0060] Retrieve the alias table again using entity name A, and return the set of all corresponding aliases.

[0061] The second objective of this invention can be achieved by adopting the following technical solution:

[0062] An information source retrieval system that includes alias recognition and same-name entity disambiguation, the system comprising:

[0063] The identification module is used to input the entity name to be retrieved; based on the entity name, alias identification technology is used to obtain a set of aliases for the entity name;

[0064] The first construction module is used to traverse the information source database to be retrieved based on the alias set and construct the information source network.

[0065] The second construction module is used to initialize the information source nodes in the information source network as embeddings; based on the embeddings of the information source nodes, a loss function for the real target entity is constructed.

[0066] The third building module is used to construct the target loss function for positive and negative sample sets based on representation learning techniques from information source networks.

[0067] The collaborative optimization module is used to implement a joint same-name disambiguation technique based on the loss function of the real target entity and the target loss function of the positive and negative sample sets, and to obtain information sources corresponding to different real target entities for users to choose from.

[0068] The third objective of this invention can be achieved by adopting the following technical solution:

[0069] A terminal device includes a processor and a memory for storing a processor-executable program. When the processor executes the program stored in the memory, it implements the above-described information source retrieval method that includes alias identification and same-name entity disambiguation.

[0070] The fourth objective of this invention can be achieved by adopting the following technical solution:

[0071] A computer-readable storage medium storing a program that, when executed by a processor, implements the above-described information source retrieval method including alias identification and same-name entity disambiguation.

[0072] The present invention has the following advantages over the prior art:

[0073] (1) A key issue in name-based disambiguation is determining the number of real target entities, which is difficult to ascertain in real-world environments. To address this, common methods include density-based clustering (DBSCAN), which does not require prior knowledge of the number of real target entities. However, the clustering disambiguation process and the representation learning process are often performed separately, limiting model performance. This invention proposes a joint disambiguation method that achieves coordinated optimization of the loss function for real target entities and the target loss function for positive and negative sample sets. This solves the problem that the number of real target entities is unknown in real-world environments and that the clustering disambiguation process and representation learning cannot be coordinated for optimization, making it more adaptable to real-world environments.

[0074] (2) This invention incorporates alias recognition technology during the retrieval process, enabling the retrieval of more comprehensive relevant information sources. Furthermore, the larger number of data samples also ensures sufficient training for the subsequent same-name disambiguation model.

[0075] (3) The representation learning technique used in this invention only requires the information source network constructed from the information source entity set, without the need for other information, and has a wide range of applications. In order to ensure model performance and prevent network edge sparsity, and taking into account the idea of ​​clustering similar things in real life, community-oriented random walks are added to the model learning to fully explore the community relationships between entities, further expanding the application scope of representation learning technique. Attached Figure Description

[0076] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the structures shown in these drawings without creative effort.

[0077] Figure 1 This is a flowchart of an information source retrieval method that includes alias identification and same-name entity disambiguation according to Embodiment 1 of the present invention.

[0078] Figure 2 This is an example of entity alias storage in Embodiment 1 of the present invention.

[0079] Figure 3 This is a structural block diagram of the information source retrieval system including alias recognition and same-name entity disambiguation according to Embodiment 2 of the present invention.

[0080] Figure 4 This is a structural block diagram of the terminal device according to Embodiment 3 of the present invention. Detailed Implementation

[0081] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention. It should be understood that the specific embodiments described are merely used to explain this application and are not intended to limit this application.

[0082] Example 1:

[0083] like Figure 1 As shown, this embodiment provides an information source retrieval method that includes alias recognition and same-name entity disambiguation, comprising the following steps:

[0084] S101. Input the name of the entity to be searched and access the information source database to be searched.

[0085] In this embodiment, the entity to be retrieved is named A, and the information source database to be retrieved is an existing one. It mainly uses an existing relational database (such as MySQL) to store the information source metadata (such as the information source name and information source storage location).

[0086] S102. Based on the entity name, use alias recognition technology to obtain the set of aliases for the entity name.

[0087] The alias system used in this embodiment includes an alias table, an entity table, and a relationship mapping table, as shown in Tables 1 to 3. When the system initializes and runs, common entity names and their aliases are pre-imported from public data such as Wikidata as basic data.

[0088] (1) Based on the entity name, call the alias generation interface to generate a set of aliases for the entity name.

[0089] The alias generation interface consists of a series of customizable and extensible generation methods. Use the simplified or traditional Chinese generation method (such as calling the zhconv package) to generate the corresponding simplified or traditional Chinese aliases or , use the pinyin generation method (such as calling the pypinyin package) to generate pinyin aliases , use the cross-language generation method (such as calling the googletrans package) to generate aliases in various languages, such as English aliases ; You can also manually add known aliases using the manual management interface .

[0090] For example, when entering "苹果" (Apple), the alias generation interface automatically generates "蘋果", "ping guo", and the English alias "Apple". Users can also manually add other aliases for "苹果" (Apple).

[0091] (2) Based on the set of aliases for the entity name, update the alias table, the entity table, and the relationship mapping table.

[0092] Retrieve the alias table using the entity name to check if there are relevant aliases. If not, insert the entity name A entered by the user into the entity table, and correspondingly add the set of aliases generated by the alias generation interface to the alias table, and build the relationship between the entity name and the corresponding set of aliases in the relationship mapping table. If it exists, update all the existing aliases for each entity corresponding to this alias, and supplement the aliases that may be missing in the system. For example, when retrieving "苹果" (Apple), both "苹果公司" (Apple Inc.) and "苹果" (Apple) in the system entity table contain the alias "苹果" (Apple), so the system needs to add the aliases "蘋果" and "ping guo" that do not exist in the alias table for "苹果公司" (Apple Inc.) and "苹果" (Apple) respectively.

[0093] (3) Retrieve the alias table using the entity name and return all the sets of aliases for the entity name.

[0094] Retrieve the alias table again using the entity name A and return all the sets of aliases A_all for the entity name A. For example, when entering "苹果" (Apple), the returned set of aliases A_all should contain all the sets of aliases for "苹果公司" (Apple Inc.) and "苹果" (Apple), and A_all is: "苹果", "Apple", "Apple Inc.", "The Fruit Company", "蘋果", and "ping guo".

[0095] For examples related to entity aliases, please refer to Figure 2 .

[0096] Table 1. Alias ​​Table

[0097] ;

[0098] Table 2 Relationship Mapping Table

[0099] ;

[0100] Table 3 Entity Table

[0101] ;

[0102] S103. Traverse the information source database based on the alias set to construct an information source network; initialize the information source nodes in the information source network as embedded nodes.

[0103] Understandably, embedding is also called vector representation.

[0104] (1) Traverse the information source database. If any alias in the alias set A_all appears in the name or content of the currently traversed information source, it indicates that the information source is related to the entity name A to be retrieved. Then add the information source to the node set. middle.

[0105] (2) Traverse the set of information source nodes For each information source name and content, entity recognition is performed using tools such as BiLSTM or BiLSTM+CRF to obtain the entity set for each information source. For each entity name B in the entity set, an alias table is retrieved to obtain an alias set B_all. Entity name B and related aliases in the entity set are then uniformly replaced with entity name B.

[0106] (3) Constructing an information source network , For the edge set of the information source network, This indicates that there is an edge between two information sources (the entity sets of the two information sources intersect, and the number of intersections represents the edge weight). The information source nodes are then initialized respectively. Represented as an f-dimensional vector The vector matrix composed of all information source nodes (n) is as follows: .

[0107] S104. Based on the embedding of information source nodes, construct a loss function for the real target entity.

[0108] (1) Initialize the information source as a candidate real target entity.

[0109] Embed each information source Considered as the initial candidate real target entity, i.e. The initial set of candidate real target entities is denoted as And initialize the number of real target entities to the number of information sources, that is .

[0110] (2) For each candidate real target entity Calculate its density as follows:

[0111] ;

[0112] in, This is an indicator function that takes the value 1 when the condition is met, and 0 otherwise. Let be the distance function, representing arrive The distance; Based on preset quantiles The calculation radius, ;in and They represent Calculate the maximum and minimum distances to other real-world targets. Calculate different quantiles for each. The real target entity below The density, take the radius when the density is maximum. As the real target entity optimal radius .

[0113] (3) Set the candidate real target entities According to density Sort in descending order to facilitate subsequent steps starting with high-density real target entities. Simultaneously calculate the probability p of generating a new real target entity and the probability (1-p) of splitting and merging existing real target entities. The probability value p is calculated as follows:

[0114] ;

[0115] in This indicates the number of times the process generates new real-world target entities. This indicates the total number of runs.

[0116] (4) When generating a new real target entity, based on the current number of times And last time Real target entity embedding and The new real target entity is generated as follows:

[0117] ;

[0118] in, .

[0119] Will Add to the real target entity set and update the actual number of target entities. And use step (2) to calculate. density and optimal radius .

[0120] (5) When splitting and merging existing real target entities, sort the results in descending order of density according to step (3). For the current high-density real target entities, optimal radius The average value of all real target entities within the range is taken as the current high-density real target entity. The representation of:

[0121] ;

[0122] At the same time, the merger resulted in Real target entities within the optimal radius optimal radius Unable to cover The situation is split to ensure and They can overwrite each other, and the split operation updates the real target entity. The representation is as follows:

[0123] ;

[0124] Based on the results of the merging and splitting, those that have already been integrated will be... From the candidate real target entities in Remove from the list and update the actual target entity count. .

[0125] (6) Based on the embedding of each information source and the embedding of the candidate real target entity, in order to make each information source close to the real target entity and make each real target entity far away from each other, the real target entity loss function is designed as follows:

[0126] ;

[0127] S105. Based on representation learning techniques of information source networks, construct the target loss function for positive and negative sample sets.

[0128] (1) Use the Louvain community detection algorithm to calculate the information source network. Community structure If node i and node j are in the same community, then .

[0129] (2) In the information source network Choose any node Random walk sequences are generated based on a community-oriented random walk strategy, where the calculation... Node reaches The probability is:

[0130] ;

[0131] Where w is the edge weight. Represents a node The community; according to probability The random walk sequence is obtained by performing L random walks: .

[0132] (3) Traverse the information source network Each node serves as a training node. Take node random walk sequence As a training node Positive samples ), and randomly sample Not in the random walk sequence The nodes on the list are used as negative samples. ).

[0133] (4) Calculate the target loss function based on the positive and negative sample sets:

[0134] ;

[0135] in It prevents the value from being a minimum of 0.

[0136] S106. A joint name-based disambiguation technique that achieves collaborative optimization based on the loss function of the real target entity and the target loss function of the positive and negative sample sets, obtains information sources corresponding to different real target entities for users to choose from.

[0137] (1) The objective function is constructed as follows:

[0138] ;

[0139] in, It is Norm regularization term to prevent model overfitting, parameters and These are used to control the effects of the target entity loss function and the penalty term, respectively.

[0140] Based on learning rate The Adam optimization method is used to optimize the overall objective function. When the model converges or reaches the maximum number of iterations, the set of information source nodes is obtained. Final Embedding Result and Real Target Entity Set Number of actual target entities .

[0141] (2) Embedding based on optimization results and Calculate the embedding of each information source To each real target entity distance The information source with the smallest distance belongs to the real target entity. All information sources corresponding to the real target entities are returned to the user for selection of the final search results.

[0142] Those skilled in the art will understand that all or part of the steps in the methods of the above embodiments can be implemented by a program instructing related hardware, and the corresponding program can be stored in a computer-readable storage medium.

[0143] It should be noted that although the method operations of the above embodiments are described in a specific order in the accompanying drawings, this does not require or imply that these operations must be performed in that specific order, or that all the operations shown must be performed to achieve the desired result. On the contrary, the order of execution of the described steps may be changed. Additionally or alternatively, certain steps may be omitted, multiple steps may be combined into one step, and / or one step may be broken down into multiple steps.

[0144] Example 2:

[0145] like Figure 3 As shown, this embodiment provides an information source retrieval system that includes alias recognition and same-name entity disambiguation. The system includes an identification module 301, a first construction module 302, a second construction module 303, a third construction module 304, and a collaborative optimization module 305, wherein:

[0146] The identification module 301 is used to input the entity name to be retrieved; based on the entity name, it uses alias identification technology to obtain a set of aliases for the entity name;

[0147] The first construction module 302 is used to traverse the information source database to be retrieved based on the alias set and construct the information source network.

[0148] The second construction module 303 is used to initialize the information source nodes in the information source network as embeddings; and to construct the loss function of the real target entity based on the embeddings of the information source nodes.

[0149] The third construction module 304 is used to construct the target loss function for positive and negative sample sets based on representation learning techniques from information source networks.

[0150] The collaborative optimization module 305 is used to implement a joint same-name disambiguation technique based on the loss function of the real target entity and the target loss function of the positive and negative sample sets, and to obtain information sources corresponding to different real target entities for the user to choose from.

[0151] The specific implementation of each module in this embodiment can be found in Embodiment 1 above, and will not be repeated here. It should be noted that the system provided in this embodiment is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure can be divided into different functional modules to complete all or part of the functions described above.

[0152] Example 3:

[0153] This embodiment provides a terminal device, which can be a computer, such as... Figure 4 As shown, the processor 402, memory, input device 403, display 404, and network interface 405 are connected via system bus 401. The processor provides computing and control capabilities. The memory includes a non-volatile storage medium 406 and internal memory 407. The non-volatile storage medium 406 stores the operating system, computer programs, and database. The internal memory 407 provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. When the processor 402 executes the computer programs stored in the memory, it implements the information source retrieval method of Embodiment 1 described above, which includes alias recognition and same-name entity disambiguation, as follows:

[0154] Input the entity name to be searched; based on the entity name, use alias recognition technology to obtain a set of aliases for the entity name;

[0155] Based on the alias set, traverse the database of information sources to be retrieved and construct an information source network;

[0156] Initialize the information source nodes in the information source network as embeddings; construct the loss function for the real target entity based on the embeddings of the information source nodes;

[0157] Based on representation learning techniques of information source networks, a target loss function for positive and negative sample sets is constructed.

[0158] A joint name-based disambiguation technique is implemented based on the loss function of real target entities and the target loss function of positive and negative sample sets to obtain information sources corresponding to different real target entities for users to choose from.

[0159] Example 4:

[0160] This embodiment provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the information source retrieval method of Embodiment 1 described above, which includes alias identification and same-name entity disambiguation, as follows:

[0161] Input the entity name to be searched; based on the entity name, use alias recognition technology to obtain a set of aliases for the entity name;

[0162] Based on the alias set, traverse the database of information sources to be retrieved and construct an information source network;

[0163] Initialize the information source nodes in the information source network as embeddings; construct the loss function for the real target entity based on the embeddings of the information source nodes;

[0164] Based on representation learning techniques of information source networks, a target loss function for positive and negative sample sets is constructed.

[0165] A joint name-based disambiguation technique is implemented based on the loss function of real target entities and the target loss function of positive and negative sample sets to obtain information sources corresponding to different real target entities for users to choose from.

[0166] It should be noted that the computer-readable storage medium in this embodiment can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. The 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.

[0167] The above description is merely a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope disclosed in the present invention, based on the technical solution and inventive concept of the present invention, shall fall within the scope of protection of the present invention.

Claims

1. An information source retrieval method including alias recognition and homonym entity disambiguation, characterized by, The method comprises: input the entity name to be searched; based on the entity name, an alias recognition technology is used to obtain an alias set of the entity name; based on the alias set, an information source database to be searched is traversed to construct an information source network; an information source node in the information source network is initialized as an embedding; based on the embedding of the information source node, a real target entity loss function is constructed; based on a representation learning technology of the information source network, a target loss function of a positive and negative sample set is constructed; a collaborative alias disambiguation technology is realized based on collaborative optimization of the real target entity loss function and the target loss function of the positive and negative sample set, so as to obtain information sources corresponding to different real target entities for user selection; the construction process of the real target entity loss function is as follows: initializing information sources in an information source network as candidate real target entities; an initial candidate real target entity set is denoted as and initializing the number of candidate real target entities k as the number of information sources n; for each candidate real target entity , compute its density ; Sort the candidate real target entities in the set by their densities in descending order to facilitate subsequent processing starting from high-density real target entities first; running generates new candidate real target entities and running splits and merges existing candidate real target entities and updates the set while updating the number of candidate real target entities ; based on the embedding of each information source and the embedding of a candidate real target entity, in order to make each information source close to the candidate real target entity and make each candidate real target entity far away from each other, the real target entity loss function is designed as follows: ； In the formula, is an information source embedding; The running generates new candidate real target entities and runs split and merge of existing candidate real target entities, and updates the set , comprising: the probability value p of generating a new candidate real target entity and the probability (1-p) of running splitting and merging of the existing candidate real target entity are calculated; wherein the probability value p of the new candidate real target entity is calculated as follows: ； In the formula, denotes the number of runs generating new candidate real target entities, denotes the total number of runs; based on the current number of times and candidate real target entity embeddings of the number of times and candidate new real target entities are generated as follows: ； In the formulae, is a set value; Add to the set Add to the set , and update the number of candidate real target entities ; Sort the candidate real target entities in the set by their density in descending order; for the current high-density real target entity , average all the real target entities within the optimal radius of to get the representation of the current high-density real target entity : ； For real target entities within the optimal radius of the optimal radius cannot be covered In order to ensure and mutual coverage, a splitting operation is performed to update the real target entities : ； remove from the set of candidate real target entities the candidate real target entities that have been fused into the set of real target entities, while updating the number of real target entities .

2. The information source retrieval method according to claim 1, characterized by, the collaborative alias disambiguation technology realized based on collaborative optimization of the real target entity loss function and the target loss function of the positive and negative sample set, so as to obtain information sources corresponding to different real target entities for user selection, comprises: the target function is constructed based on the real target entity loss function and the target loss function of the positive and negative sample set as follows: ； In the formula, is a target loss function of positive and negative sample sets, is a real target entity loss function, is a norm regularization term to prevent model overfitting; the parameter and are used to control the target entity loss function and the penalty term, respectively; Based on the learning rate, the objective function is optimized using an Adam optimization method, and the final embedding of the information source node set is obtained when the model converges as a whole or reaches the maximum number of iterations as a whole and the real target entity set ;​ based on the final embedding and the set of real target entities , calculate the distance of each information source embedding to each real target entity ; the smallest distance indicates that the information source belongs to the real target entity; return the information sources corresponding to all real target entities to the user for the user to select the final search result.

3. The information source retrieval method according to any one of claims 1 to 2, characterized in that, the construction process of the target loss function of the positive and negative sample set is as follows: Computing community structure of information source networks using community finding algorithms ;​ If node i is in the same community as node j, then ; In the information source network optionally one node ; and are the set of nodes and edges of the information source network, respectively; According to the community-oriented random walk strategy, a random walk sequence is generated, and a probability that a node walks to another node is calculated. another node is calculated. ； In the formula, w is an edge weight, representing a community of nodes ; according to a probability , a random walk sequence is obtained by a random walk L steps ; Traversing an information source network Each node as a training node , take the random walk sequence of the node As a positive sample of the training node ), and randomly sample Nodes not on the random walk sequence As a negative sample )​ the target loss function is calculated based on the positive and negative sample set: ； In the formula, a minimum value greater than 0.

4. The information source retrieval method of claim 1, wherein, the information source network is constructed based on the alias set traversing the information source database to be searched, comprising: traversing the information source database, if any of the aliases in the alias set appears in the name or content of an information source of the information source database, the information source is related to the entity name A to be searched, and the information source is added to the node set in the middle; For each information source entity recognition is performed on the name and content of each information source to obtain an entity set of each information source; for each entity name B in the entity set, an alias table is searched to obtain an alias set B_all; and the entity name B and the related aliases in the entity set are uniformly replaced by the entity name B; Constructing information source networks , For the edge set of information source networks, indicates that there is an intersection between two information sources, and the number of intersections serves as the weight of the edge.

5. The information source retrieval method according to any one of claims 1, 4, characterized by, the alias set of the entity name is obtained based on the entity name by using the alias recognition technology, comprising: based on the entity name A, an alias generation interface is called to generate an alias set of the entity name; the alias table is searched with the entity name A, if there is no alias, the entity name A is inserted into the entity table, and the alias set generated by the alias generation interface is added to the alias table, and the relationship between the entity name A in the entity table and the corresponding alias in the alias table is constructed in the relationship mapping table; if there is an alias, all existing aliases of each entity corresponding to the alias are updated, and the missing alias is supplemented in the alias table; the alias table is re-searched with the entity name A, and all corresponding alias sets are returned.

6. An information source retrieval system comprising alias recognition and homonym entity disambiguation, characterized in that, The system comprises: an identification module for inputting an entity name to be searched; based on the entity name, an alias recognition technology is used to obtain an alias set of the entity name; a first construction module for traversing an information source database to be searched based on the alias set to construct an information source network; a second construction module for initializing an information source node in the information source network as an embedding; based on the embedding of the information source node, a real target entity loss function is constructed; a third construction module for constructing a target loss function of a positive and negative sample set based on a representation learning technology of the information source network; a fourth construction module for realizing a collaborative alias disambiguation technology based on collaborative optimization of the real target entity loss function and the target loss function of the positive and negative sample set, so as to obtain information sources corresponding to different real target entities for user selection. The synergistic optimization module is used for realizing the synergistic optimization of the homonym disambiguation technology based on the real target entity loss function and the target loss function of the positive and negative sample sets, and obtaining information sources corresponding to different real target entities for user selection. The information source retrieval system realizes the information source retrieval method in any one of claims 1-5 when running.

7. A terminal device comprising a processor and a memory for storing a processor-executable program, characterized in that The processor realizes the information source retrieval method in any one of claims 1-5 when executing the program stored in the memory.

8. A computer-readable storage medium having stored thereon a computer program, characterized in that, The computer program realizes the information source retrieval method in any one of claims 1-5 when executed by the processor.

Images