Entity recommendation method and apparatus, storage medium, and electronic device

Abstract

Embodiments of the present application disclose an entity recommendation method and device, a storage medium and an electronic device. The method comprises: in response to an input query text, determining a link entity corresponding to the query text; based on the link entity, obtaining a corresponding entity set from each of at least two entity embedding representations, the at least two entity embedding representations comprising at least a preset granularity label embedding representation and / or a network topology embedding representation; determining a recommendation weight corresponding to each entity in the entity set; and performing entity recommendation based on the recommendation weight. According to the technical solution of the embodiments of the present application, the features of an entity can be described from different angles by fusing entity fine-grained embedding and network topology embedding, thereby improving the accuracy of entity recommendation.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to an entity recommendation method, apparatus, storage medium, and electronic device. Background Technology

[0002] With the development of internet technology, more and more people are obtaining various online content through online searches, and physical search and recommendation are important scenarios in search services.

[0003] One technical solution employs vectorization methods such as word2vec to vectorize both the user's query text and the entity's summary information, and then calculates the similarity between the two. However, since user query texts are generally short, while entity summary information is usually long text, accurate entity recommendations are difficult to achieve. Therefore, improving the accuracy of entity recommendations has become a pressing technical challenge. Summary of the Invention

[0004] This application provides an entity recommendation method, apparatus, storage medium, and electronic device, which can improve the accuracy of entity recommendation. The technical solution is as follows:

[0005] In a first aspect, embodiments of this application provide an entity recommendation method, the method comprising:

[0006] In response to the input query text, determine the link entity corresponding to the query text;

[0007] Based on the link entity, obtain the corresponding entity set from each of the at least two entity embedding representations, wherein the at least two entity embedding representations include at least: a preset granularity label embedding representation and / or a network topology embedding representation;

[0008] Determine the recommendation weight for each entity in the entity set, and make entity recommendations based on the recommendation weight.

[0009] Secondly, embodiments of this application provide an entity recommendation device, the device comprising:

[0010] The link entity determination module is used to determine the link entity corresponding to the input query text in response to the input query text;

[0011] The entity set determination module is used to obtain the corresponding entity set from each of at least two entity embedding representations based on the linked entity, wherein the at least two entity embedding representations include: a preset granularity label embedding representation and a network topology embedding representation;

[0012] The entity recommendation module is used to determine the recommendation weight of each entity in the entity set and to make entity recommendations based on the recommendation weight.

[0013] Thirdly, embodiments of this application provide a computer storage medium storing a plurality of instructions adapted for loading by a processor and executing the steps of the method described above.

[0014] Fourthly, embodiments of this application provide an electronic device, including: a processor and a memory; wherein the memory stores a computer program adapted to be loaded by the processor and to execute the steps of the above-described method.

[0015] The beneficial effects of the technical solutions provided in some embodiments of this application include at least the following:

[0016] On the one hand, in response to the input query text, the system identifies the link entities corresponding to the query text, accurately determining the target entities corresponding to the query text through entity links. On the other hand, based on the link entities, the system obtains the corresponding entity set from at least two entity embedding representations, including at least a preset granularity label embedding representation and a network topology embedding representation. By fusing fine-grained entity embedding and network topology embedding, the system can characterize the features of entities from different perspectives, thereby improving the accuracy of entity recommendation. Furthermore, the system determines the recommendation weight corresponding to each entity in the entity set, and performs entity recommendation based on the recommendation weight. This allows for fine-grained ranking of the recommended content according to the recommendation weight, further improving the accuracy of entity recommendation. Attached Figure Description

[0017] 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 of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 A schematic diagram illustrating an application scenario of the entity recommendation method provided according to an embodiment of this application is shown;

[0019] Figure 2 A flowchart illustrating an entity recommendation method provided according to some embodiments of this application is shown;

[0020] Figure 3 A schematic flowchart illustrating the generation of a preset granularity tag embedding representation is shown according to some other embodiments of this application;

[0021] Figure 4A flowchart illustrating an entity recommendation method provided according to further embodiments of this application is shown;

[0022] Figure 5 This paper shows a schematic diagram of the structure of a physical recommendation device provided in an embodiment of this application;

[0023] Figure 6 A schematic diagram of the structure of an electronic device provided in an embodiment of this application is shown. Detailed Implementation

[0024] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0025] First, the terms used in the embodiments of this application will be explained and described.

[0026] Entities: Entities are language objects in natural language processing. For example, entities can be organizations, quantities, currencies, personal names, company names, geographical locations, etc.

[0027] Linked entities: refer to the target entities in the preset entity library that correspond to the query text, and the target entities in the preset entity library that correspond to the entity objects (e.g., person names, place names, organization names, etc.) that have been identified in the query text.

[0028] Preset granularity tags: These represent tags with a preset granularity for entities in natural language processing. For example, preset granularity tags can be fine-grained tags such as short sentences or phrases.

[0029] Embedded representation: refers to representing an entity object using a low-dimensional dense vector. The entity object can be a word, a product, a news article, a movie, etc.

[0030] Network topology embedding representation is a method for embedding entity nodes in a graph network. The resulting entity node embedding representation generally includes global structural information of the graph and local similarity information of neighboring nodes.

[0031] Currently, entity search and recommendation is a crucial scenario in search services, with the core challenge being calculating the semantic similarity between the user's query text and the entities. One technical solution employs vectorization methods, such as word2vec, to vectorize both the user's query text and the entity's summary information. Then, the similarity between the two is calculated, and the title corresponding to the entity is displayed as the recommended content to the user. However, this approach suffers from the following problems:

[0032] (1) User query text is usually very short, while entity summary is usually long text. Because the summary contains other information (such as other entities), it cannot accurately express the semantics of the entity, resulting in semantic calculation distortion of both. Therefore, more accurate entity description information is needed.

[0033] (2) Coarse entity labels cannot represent the specificity of an entity. For example, internet companies can be further subdivided into security companies, retail companies, etc. Therefore, more granular labels are needed.

[0034] (3) Although methods such as fastext can capture information in a literal sense, they do not take into account the network graph structure between entities. Therefore, it is necessary to introduce another entity representation method that can characterize the network information of entities.

[0035] Based on the above, this application provides an entity recommendation method that integrates fine-grained entity tags and entity internal link networks. By integrating fine-grained entity embedding and network topology embedding, the entity recommendation method can characterize the features of entities from different perspectives, thereby improving the accuracy of entity recommendation.

[0036] The technical solution of the entity recommendation method according to the embodiments of this application will now be described in detail with reference to the accompanying drawings.

[0037] Figure 1 A schematic diagram illustrating an application scenario of the entity recommendation method provided according to an embodiment of this application is shown.

[0038] Reference Figure 1 As shown, when a user enters a query text such as "apple" on the client's search application interface and clicks the search control (i.e., the "Search" control), the server uses the entered query text "apple" and employs an entity recommendation method that integrates fine-grained entity tags and entity internal link networks to obtain the recommended entities below, including Apple mobile phones, Apple watches, Apple tablets, etc.

[0039] In an example embodiment, a search application is installed on the client. In response to the query text entered on the client's search application, the server determines the link entity corresponding to the query text. Based on the link entity, the server obtains the corresponding entity set from each of at least two entity embedding representations, wherein the at least two entity embedding representations include at least: a preset granularity label embedding representation and / or a network topology embedding representation. The server determines the weight of each entity in the entity set and makes entity recommendations based on the weights.

[0040] It should be noted that the aforementioned client can be an application terminal device configured with a search application. This application terminal device can include, but is not limited to, at least one of the following: smartphones, laptops, tablets, PDAs, MIDs (Mobile Internet Devices), desktop computers, smart TVs, etc., without limitation. The target client can be a video application client, an instant messaging client, a browser client, an educational application client, etc. The aforementioned server can be a single server, a server cluster consisting of multiple servers, or a cloud server. The above is merely an example, and no special limitations are made in this embodiment.

[0041] Figure 2 A flowchart illustrating an entity recommendation method according to some embodiments of this application is shown. The entity recommendation method can be executed by a computing device with computational processing capabilities, such as a server. The entity recommendation method includes steps S210 to S230. The entity recommendation method in the example embodiments will now be described in detail with reference to the accompanying drawings.

[0042] Reference Figure 2 As shown, in step S210, in response to the input query text, the link entity corresponding to the query text is determined.

[0043] In the example embodiment, the linked entity refers to the target entity in a preset entity library that corresponds to the query text. A search application or browser is installed on the client, and the query text is entered on the search page of the search application or browser. In response to the entered query text, the server determines the linked entity corresponding to the query text from the preset entity library.

[0044] For example, in response to the input query text, the server uses a vectorization method such as word2vec to determine the embedding vector corresponding to the query text, calculates the similarity between the embedding vector corresponding to the query text and the embedding vectors of entities in a preset entity library, and identifies entities with a similarity greater than a predetermined threshold as the link entities corresponding to the query text.

[0045] In step S220, based on the linked entity, a corresponding entity set is obtained from each of at least two entity embedding representations, and the at least two entity embedding representations at least include: a preset granularity label embedding representation and / or a network topology embedding representation.

[0046] In an exemplary embodiment, an embedding representation refers to representing an entity object by a low-dimensional dense vector. The entity object can be a word, a commodity, a news article, a movie, etc. The at least two entity embedding representations at least include: a preset granularity label embedding representation and / or a network topology embedding representation. For example, the at least two entity embedding representations at least include a preset granularity label embedding representation and a network topology embedding representation; or, the at least two entity embedding representations at least include a preset granularity label embedding representation and a knowledge triple embedding representation; or, the at least two entity embedding representations at least include a network topology embedding representation and a knowledge triple embedding representation, etc. The preset granularity label of an entity can be a fine-grained label such as a short sentence or phrase corresponding to the entity. For example, if the entity is "Qi Li Xiang", the preset granularity label can be "a song sung by Jay Chou". The preset granularity label embedding representation is the embedding representation of the fine-grained label of the entity point.

[0047] The network topology embedding of an entity is a method of embedding the entity nodes in a graph network. The finally generated entity node embedding representation generally contains the global structure information of the graph and the local similarity information of the neighbor nodes. The network topology embedding representation of an entity can be determined by DeepWalk, EGES, Node2vec, etc. Taking DeepWalk as an example, by parsing the encyclopedia page of the entity, an edge is constructed between the current page entity and each internal link entity to form an undirected graph network. By performing random walks on the graph network, an entity sequence is generated, and the entity sequence is used as a training sample to be input into a context prediction neural network such as Skip-gram for training to obtain the network topology embedding of the entity.

[0048] Further, the similarity between the linked entity and the embedding representations of each entity in the at least two entity embedding representations is calculated to obtain an entity set with a similarity greater than a predetermined threshold. For example, the similarity between the linked entity and the embedding representations of each entity in the predetermined granularity label embedding representation is calculated to obtain an entity set composed of multiple preset granularity label embedding representations with a similarity greater than the predetermined threshold.

[0049] In step S230, the recommendation weights corresponding to each entity in the entity set are determined, and entity recommendations are made based on the recommendation weights.

[0050] In the example embodiment, each entity embedding represents a corresponding entity set. That is, the preset granularity label corresponds to the first entity set, and the network topology embedding represents the second entity set. The first entity set and the second entity set are merged and deduplicated to generate a merged entity set. The recommendation weight of each entity in the merged entity set is determined, and entity recommendations are made based on the recommendation weight of each entity. For example, the title corresponding to at least one entity with a recommendation weight greater than a predetermined threshold is recommended to the client.

[0051] For example, in some exemplary embodiments, the recommendation weight includes entity weights. The frequency of occurrence of each entity in the merged entity set within the first entity set, the second entity set, and the third entity set is determined. Based on the frequency of occurrence of each entity in the merged entity set, the entity weight corresponding to each entity in the merged entity set is determined, and entity recommendations are performed based on the entity weights corresponding to each entity. For instance, titles corresponding to multiple entities with entity weights greater than a predetermined threshold are recommended to the client.

[0052] In other example embodiments, the recommendation weight includes entity weight and tag weight. The preset granularity tag corresponding to each entity in the merged entity set in the <entity, preset granularity tag, tag weight> triple is determined, resulting in a preset granularity tag set corresponding to the merged entity set. Based on each preset granularity tag in the preset granularity tag set, the entities in the merged entity set are classified, resulting in an entity set corresponding to each preset granularity tag. Based on the entity weight of each entity in the entity set corresponding to each preset granularity tag, the total entity weight of each preset granularity tag is determined. Based on the total entity weight and tag weight corresponding to each preset granularity tag, the recommendation weight corresponding to each preset granularity tag is determined, and entity recommendations are performed based on the recommendation weight corresponding to each preset granularity tag.

[0053] according to Figure 2 The technical solution in the example embodiment, on the one hand, responds to the input query text and determines the link entity corresponding to the query text, which can accurately determine the target entity corresponding to the query text through entity linking; on the other hand, based on the link entity, obtains the corresponding entity set from at least two entity embedding representations, which include at least: a preset granularity label embedding representation and a network topology embedding representation. By fusing the entity recommendation method of fine-grained entity embedding and network topology embedding, the characteristics of the entity can be characterized from different perspectives, thereby improving the accuracy of entity recommendation; furthermore, it determines the recommendation weight corresponding to each entity in the entity set, and performs entity recommendation based on the recommendation weight. It can sort the content to be recommended according to the recommendation weight, thereby further improving the accuracy of entity recommendation.

[0054] Furthermore, in order to determine the network topology embedding representation of an entity, in an example embodiment, by parsing the encyclopedia page of the entity to be processed, an edge is constructed between the entity to be processed and its corresponding internal link entity to form an undirected graph network; by traversing the graph network, the entity sequence corresponding to the entity to be processed is extracted; and the network topology embedding representation corresponding to the entity to be processed is determined based on the entity sequence.

[0055] For example, entity network topology embedding representation includes: (1) network graph construction; and (2) entity embedding learning based on the network graph. These two parts are explained in detail below:

[0056] (1) Network graph construction

[0057] By parsing the encyclopedia page, an edge is constructed between the current page entity and each internal link entity, forming an undirected graph.

[0058] (2) Entity embedding learning based on network graph

[0059] By using network embedding methods such as DeepWalk and Node2vec, entity sequences are generated by walking through the network. These entity sequences are then used as training samples and input into a context prediction neural network such as Skip-gram for processing, resulting in the network topology embedding of the entities.

[0060] According to the technical solution of this example embodiment, by generating a network topology embedding representation of entities, it is possible to learn the network characteristics between entities using the internal chain information of entities, thereby simulating the user's click-to-jump behavior between entities.

[0061] Furthermore, in some example embodiments, at least two entity embedding representations also include triplet embedding representations. For example, triplet embedding representations of entities can be trained using knowledge models such as TransE. Based on linked entities, the corresponding entity set is obtained from each of the at least two entity embedding representations, including: obtaining a first entity set based on linked entities from a preset granularity label embedding representation; obtaining a second entity set based on linked entities from a network topology embedding representation; and obtaining a third entity set based on linked entities from a triplet embedding representation.

[0062] According to the technical solution in this example embodiment, by integrating entity fine-grained embedding, knowledge triple embedding, and network topology embedding, the characteristics of entities can be characterized from different perspectives to the greatest extent, thereby improving the accuracy of entity recommendation.

[0063] Furthermore, the first entity set, the second entity set, and the third entity set are merged and deduplicated to generate a merged entity set; the frequency of occurrence of each entity in the merged entity set in the first entity set, the second entity set, and the third entity set is determined; based on the frequency of occurrence of each entity in the merged entity set, the entity weight corresponding to each entity in the merged entity set is determined; and entity recommendation is performed based on the entity weight.

[0064] According to the technical solution in this example embodiment, entity recommendation is performed by integrating fine-grained entity embedding, knowledge triple embedding, and network topology embedding, thereby further improving the accuracy of entity recommendation.

[0065] Figure 3 A schematic diagram of a process for generating a preset granularity label embedding representation is shown according to some other embodiments of this application.

[0066] Reference Figure 3 As shown, in step S310, preset granularity tags of each entity are extracted from the unstructured text using a predetermined entity extraction method, generating a tuple of <entity, preset granularity tag>.

[0067] In the example embodiment, the preset granularity label is a fine-grained label such as a short sentence, phrase or sentence corresponding to the entity. The preset granularity label of each entity is extracted from the unstructured text using a predetermined entity extraction method, such as sequence labeling.

[0068] For example, let unstructured text be: "Seven Mile Fragrance is a Chinese..." Taiwan The song, performed by singer Jay Chou, is included in his self-titled album "Seven Mile Fragrance," released in July 2004. In 2004, this song won a Chinese award. Hongkong TVB Top Ten Golden Songs Awards: Best Composer, Best Producer, Best Arranger (3 awards); 2005: Best Song of the Year at the 11th Global Chinese Music Awards, and many other awards. The following <entities, preset granularity labels> were extracted from the above unstructured text using sequence labeling: <Seven Mile Fragrance, a song performed by Jay Chou>; <Seven Mile Fragrance, 2004 China...> Hongkong TVB Top Ten Golden Songs Best Composer and other binary groups.

[0069] It should be noted that although the example of extracting tuples using sequence labeling is given, those skilled in the art should understand that other methods can also be used for tuple extraction, such as rule-based sampling or MRC (Machine Reading Comprehension) methods, which are also within the scope of this disclosure.

[0070] In step S320, clustering operations are performed on the <entity, preset granularity label> tuples of each entity to form a set of preset granularity labels corresponding to different meanings under the same entity.

[0071] In the example embodiment, since entities with the same name may have different meanings—for example, the entity "apple" can refer to fruit, mobile phone, or company—fine-grained tags of entities are aggregated through clustering operations such as DBSCAN to form a preset set of tags corresponding to different meanings under the same entity name. For example, the tag set corresponding to the mobile phone meaning of the entity "apple" is {IPHONE7, IPHONE6S, IPHONE10}.

[0072] In step S330, the preset granularity tag set corresponding to different meanings is taken as text, and the preset granularity tag embedding representation is obtained by using a preset vector representation method.

[0073] In the example embodiment, the set of preset granularity tags corresponding to different meanings is taken as a text, and the preset granularity tag embedding representation of the entity is obtained by using a preset vector representation method such as FastText.

[0074] It should be noted that although FastText is used as an example for illustration, those skilled in the art should understand that other appropriate methods, such as word2vec or Item2vec vector representation methods, can also be used to obtain the preset granularity tag embedding representation of entities.

[0075] according to Figure 3 The technical solution in the example embodiment, on the one hand, extracts preset granularity tags of each entity from unstructured text through a predetermined entity extraction method, which can mine a large number of specific fine-grained entity tags from large-scale unstructured text, thereby providing more accurate entity description information and making entity links more precise; on the other hand, by using finer-grained tags, the distinctiveness between entities can be improved, and the recommendation results can be more diverse.

[0076] Furthermore, in the example embodiment, the tag weight of each sense of the same entity is determined based on the tag frequency of the preset granularity tags corresponding to each sense of the same entity; and the triple of <entity, preset granularity tag, tag weight> corresponding to each sense of the same entity is generated based on the <entity, preset granularity tag> tuple and the tag weight.

[0077] For example, to identify the popularity of different senses of the same entity, we can select the number of fine-grained labels and the label frequency corresponding to different senses of the same entity, and perform normalization processing. The result of the normalization processing is used as the label weight of each sense, such as <apple, mobile phone, 0.8>, <apple, fruit, 0.7>, etc.

[0078] According to the technical solution of this example embodiment, the tag weight is determined by the tag frequency of the preset granularity tags corresponding to each sense of the same entity, which enables more accurate entity recommendation based on the tag weight.

[0079] Furthermore, in the example embodiment, since semantically consistent but differently expressed preset granular tags (i.e., fine-grained tags) are extracted from large-scale unstructured text, it is necessary to align and merge these different preset granular tags into a single preset granular tag. For example, "China's largest cybersecurity company" and "China's largest cybersecurity company" are both aligned to the unified expression "China's largest cybersecurity company".

[0080] By aligning and fusing pre-defined granularity tags that are semantically consistent but express different ideas, the pre-defined granularity tags can express entities more accurately, thereby enabling more accurate entity recommendations.

[0081] Figure 4 A flowchart illustrating an entity recommendation method provided according to further embodiments of this application is shown.

[0082] Reference Figure 4 As shown, in step S405, a query text vector corresponding to the query text is generated, and the query text vector is matched with the preset granularity label embedding representation for similarity to obtain the first candidate entity set.

[0083] In the example embodiment, for the query text input by the user, the query text is vectorized using methods such as FastText to obtain the query text vector, and the query text vector is matched with the entities in the preset granularity embedding representation to obtain the first candidate entity set.

[0084] In step S410, the query text vector is matched with the entity vectors in the preset entity library to obtain the second candidate entity set.

[0085] In the example embodiment, the query text input by the user is vectorized using methods such as FastText to obtain a query text vector. The entity database is then matched with the query text vector to obtain a second candidate entity set.

[0086] In step S415, the entities in the first candidate entity set and the second candidate entity set are scored and sorted, and the link entity corresponding to the query text is determined based on the sorting result.

[0087] In the example embodiment, the entities in the first candidate entity set and the second candidate entity set are scored and ranked, and the candidate entity ranked first is taken as the final link entity.

[0088] In step S420, based on the linked entity, the corresponding first entity set is obtained from the preset granularity tag embedding representation.

[0089] In an example embodiment, the link entity obtained in step S415 is used to find a first set of entities similar to the link entity in the preset granularity tag embedding representation, i.e., fine-grained tag embedding.

[0090] In step S425, based on the link entities, the corresponding second entity set is obtained from the network topology embedding representation.

[0091] In the example embodiment, the link entity obtained in step S415 is used to find a second set of entities similar to the link entity in the network topology embedding.

[0092] In step S430, based on the linked entity, the corresponding third entity set is obtained from the triplet embedding representation.

[0093] In an example embodiment, the linked entity obtained in step S415 is used to find a set of third entities similar to the linked entity in the triple embedding. For example, a knowledge model such as TransE can be used to obtain the triple embedding representation of the entity through training. The similarity between the embedding representation of the linked entity and the embedding representations of the three groups is calculated to determine the set of third entities with a similarity greater than a predetermined threshold.

[0094] In step S435, the first entity set, the second entity set, and the third entity set are merged and deduplicated to generate a merged entity set, and the entity weights of the entities are determined based on their frequency of occurrence.

[0095] In the example embodiment, the first entity set, the second entity set, and the third entity set are merged and deduplicated, and the entity weight of the entity is determined based on the frequency of occurrence. That is, the frequency of occurrence of each entity in the merged entity set in the first entity set, the second entity set, and the third entity set is determined, and the entity weight corresponding to each entity in the merged entity set is determined based on the frequency of occurrence of each entity in the merged entity set.

[0096] In step S440, the preset granularity label corresponding to each entity in the merged entity set in the triplet <entity, preset granularity label, label weight> is determined, and the preset granularity label set corresponding to each entity in the merged entity set is obtained.

[0097] In the example embodiment, the merged entity set in step S435 is traversed, and for each entity, the preset granularity label, i.e. fine granularity label, corresponding to the entity is found from the triple <entity, preset granularity label, label weight>, to obtain the preset granularity label set corresponding to each entity in the merged entity set.

[0098] In step S445, based on each preset granularity label of the preset granularity label set, the entities in the merged entity set are classified to obtain the entity set corresponding to each preset granularity label.

[0099] In the example embodiment, based on the preset granularity label set corresponding to each entity in the merged entity set, the entities in the merged entity set are classified using the preset granularity label, i.e., the fine granularity label, as the key value, to obtain the entity set under the preset granularity label. The total entity weight of the preset granularity label is the sum of the weights of all entities in the entity set under the preset granularity label.

[0100] In step S450, the recommendation weight corresponding to the preset granularity label is determined.

[0101] In some example embodiments, the recommended weight includes entity weight. The total entity weight of a preset granularity label is the sum of the entity weights in the entity set under that preset granularity label. The total entity weight of each entity in the entity set corresponding to each preset granularity label is determined by summing the entity weights.

[0102] In other example embodiments, the recommendation weight includes entity weight and tag weight. The recommendation weight for each preset granularity tag is determined based on the total entity weight corresponding to each preset granularity tag and the tag weight in the triplet. For example, the recommendation weight for each preset granularity tag is determined by performing a weighted calculation on the total entity weight and tag weight corresponding to each preset granularity tag.

[0103] In step S455, entity recommendations are made based on recommendation weights.

[0104] In the example embodiment, based on the recommendation weight corresponding to each preset granularity tag in the above-mentioned preset granularity tag set, the preset granularity tags are displayed as recommendation results. For example, multiple preset granularity tags with recommendation weights greater than a predetermined threshold are determined, recommended, and displayed to the client.

[0105] according to Figure 4The technical solution in the example embodiment has several advantages. First, the preset granularity is fine-grained labeling, which can provide more accurate entity description information, making the obtained linked entities more precise. Second, the network characteristics between entities can be learned by utilizing the internal link information of entities, which can simulate the user's click-to-jump behavior between entities. Third, the entity recommendation method that integrates entity fine-grained embedding, knowledge triple embedding and network topology embedding can characterize the features of entities from different perspectives to the greatest extent, thereby improving the accuracy of entity recommendation.

[0106] The following are embodiments of the apparatus described in this application, which can be used to execute the embodiments of the method described in this application. For details not disclosed in the apparatus embodiments of this application, please refer to the embodiments of the method described in this application.

[0107] Figure 5 A schematic diagram of the structure of an entity recommendation device provided in an exemplary embodiment of this application is shown.

[0108] Reference Figure 5 As shown, the entity recommendation device 500 can be implemented as all or part of a device through software, hardware, or a combination of both, and is applied to electronic devices. The entity recommendation device 500 includes a linked entity determination module 510, an entity set determination module 520, and an entity recommendation module 530. Wherein:

[0109] The link entity determination module 510 is used to determine the link entity corresponding to the input query text in response to the input query text.

[0110] The entity set determination module 520 is used to obtain the corresponding entity set from each of at least two entity embedding representations based on the linked entity, wherein the at least two entity embedding representations include at least: a preset granularity label embedding representation and / or a network topology embedding representation.

[0111] The entity recommendation module 530 is used to determine the recommendation weight of each entity in the entity set and to make entity recommendations based on the recommendation weight.

[0112] In some example embodiments, based on the above solution, the apparatus further includes:

[0113] The binary tuple generation module is used to extract preset granularity tags of each entity from unstructured text through a predetermined entity extraction method, and generate <entity, preset granularity tag> binary tuples.

[0114] The tag set generation module is used to perform clustering processing on the preset granularity tags of the binary tuples of each entity through clustering operations to form a preset granularity tag set corresponding to different meanings under the same entity.

[0115] The tag embedding representation module is used to take the preset granularity tag set corresponding to different meanings as text and obtain the preset granularity tag embedding representation using a preset vector representation method.

[0116] In some example embodiments, based on the above solution, the apparatus further includes:

[0117] The tag frequency determination module is used to determine the tag frequency of preset granularity tags corresponding to each definition of the same entity;

[0118] The tag weight determination module is used to normalize the frequency of the tags and determine the tag weight of each sense of the same entity based on the result of the normalization process.

[0119] The triple generation module is used to generate triples of <entity, preset granularity label, label weight> corresponding to each sense of the same entity based on the <entity, preset granularity label> binary and the label weight.

[0120] In some example embodiments, based on the above scheme, the at least two entity embedding representations further include triplet embedding representations, and the entity set determination module 520 is further configured to:

[0121] Based on the link entity, obtain the corresponding first entity set from the preset granularity tag embedding representation;

[0122] Based on the link entity, obtain the corresponding second entity set from the network topology embedding representation;

[0123] Based on the linked entity, the corresponding third entity set is obtained from the triple embedding representation.

[0124] In some example embodiments, based on the above scheme, the entity recommendation module 530 is further used for:

[0125] The first entity set, the second entity set, and the third entity set are merged and deduplicated to generate a merged entity set.

[0126] Determine the frequency of occurrence of each entity in the merged entity set within the first entity set, the second entity set, and the third entity set.

[0127] The entity weights corresponding to each entity in the merged entity set are determined based on the frequency of occurrence of each entity in the merged entity set.

[0128] In some example embodiments, based on the above scheme, the recommendation weight includes entity weight and tag weight, and the entity recommendation module 530 is further used for:

[0129] Determine the preset granularity label corresponding to each entity in the merged entity set in the <entity, preset granularity label, label weight> triple, and obtain the preset granularity label set corresponding to each entity in the merged entity set;

[0130] Based on each preset granularity label of the preset granularity label set, the entities in the merged entity set are classified to obtain the entity set corresponding to each preset granularity label;

[0131] Based on the entity weights of each entity in the entity set corresponding to each preset granularity label, the total entity weight of each preset granularity label is determined.

[0132] Based on the total entity weight and the tag weight corresponding to each preset granularity tag, the recommendation weight corresponding to each preset granularity tag is determined.

[0133] In some example embodiments, based on the above solution, the apparatus further includes:

[0134] The graph network generation module is used to parse the encyclopedia page of the entity to be processed and construct an edge between the entity to be processed and the corresponding internal link entity to form a graph network.

[0135] The entity sequence extraction module is used to extract the entity sequence corresponding to the entity to be processed by traversing the graph network.

[0136] A network topology embedding generation module is used to determine the network topology embedding representation corresponding to the entity to be processed based on the entity sequence.

[0137] In some example embodiments, based on the above scheme, the link entity determination module 510 is further configured to:

[0138] The query text is vectorized to generate a query text vector corresponding to the query text.

[0139] The query text vector is matched with the preset granularity label embedding representation to obtain the first candidate entity set;

[0140] The query text vector is matched with the entity vectors in the preset entity library to obtain the second candidate entity set;

[0141] The entities in the first candidate entity set and the second candidate entity set are scored and sorted, and the link entities corresponding to the query text are determined based on the sorting results.

[0142] It should be noted that the entity recommendation device provided in the above embodiments is only illustrated by the division of the above functional modules when executing the entity recommendation method. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device can be divided into different functional modules to complete all or part of the functions described above.

[0143] Furthermore, the entity recommendation device and entity recommendation method embodiments provided in the above embodiments belong to the same concept, and their implementation process can be found in the method embodiments, which will not be repeated here.

[0144] This application also provides a computer storage medium that can store multiple instructions. These instructions are adapted to be loaded by a processor and executed as described in the above embodiments. For details of the execution process, please refer to the specific description of the above embodiments, which will not be repeated here.

[0145] This application also provides a computer program product that stores at least one instruction. The at least one instruction is loaded by the processor and executed as the entity recommendation method described in the above embodiments. For the specific execution process, please refer to the detailed description of the above embodiments, which will not be repeated here.

[0146] This application also provides a chip configured to execute the entity recommendation method as described in the above embodiments. For the specific execution process, please refer to the detailed description of the above embodiments, which will not be repeated here.

[0147] In addition, please see Figure 6 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Figure 6 As shown, the electronic device 600 may include: at least one processor 601, at least one communication module 604, input / output interface 603, memory 605, and at least one communication bus 602.

[0148] The communication bus 602 is used to enable communication between these components.

[0149] The input / output interface 603 may include a display screen and a camera. Optional input / output interface 603 may also include a standard wired interface and a wireless interface.

[0150] The communication module 604 may optionally include a standard wired interface or a wireless interface (such as a WIFI interface).

[0151] The processor 601 may include one or more processing cores. The processor 601 connects to various parts within the electronic device 600 using various interfaces and lines, and performs various functions and processes data by running or executing instructions, programs, code sets, or instruction sets stored in the memory 605, and by calling data stored in the memory 605. Optionally, the processor 601 may be implemented using at least one hardware form of Digital Signal Processing (DSP), Field-Programmable Gate Array (FPGA), or Programmable Logic Array (PLA). The processor 601 may integrate one or a combination of several of the following: Central Processing Unit (CPU), Graphics Processing Unit (GPU), and modem. The CPU primarily handles the operating system, user interface, and applications; the GPU is responsible for rendering and drawing the content required for display; and the modem handles wireless communication. It is understood that the modem may also be implemented as a separate chip without being integrated into the processor 601.

[0152] The memory 605 may include random access memory (RAM) or read-only memory. Optionally, the memory 605 may include a non-transitory computer-readable storage medium. The memory 605 can be used to store instructions, programs, code, code sets, or instruction sets. The memory 605 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as touch function, sound playback function, image playback function, etc.), instructions for implementing the above-described method embodiments, etc.; the data storage area may store data involved in the above-described method embodiments, etc. Optionally, the memory 605 may also be at least one storage device located remotely from the aforementioned processor 601. Figure 6 As shown, the memory 605, which serves as a computer storage medium, may include an operating system, a communication module, an input / output interface module, and an entity recommendation application.

[0153] exist Figure 6In the illustrated electronic device 600, the input / output interface 603 is primarily used to provide an interface for user input and to acquire user input data; while the processor 601 can be used to call the entity recommendation program stored in the memory 605, causing the processor 601 to execute steps in the entity recommendation method according to various exemplary embodiments of this disclosure. For example, the processor 601 can execute... Figure 2 The steps shown are as follows: Step S210, in response to the input query text, determine the link entity corresponding to the query text; Step S220, based on the link entity, obtain the corresponding entity set from each of at least two entity embedding representations, wherein the at least two entity embedding representations include at least: a preset granularity label embedding representation and / or a network topology embedding representation; Step S230, determine the recommendation weight corresponding to each entity in the entity set, and make entity recommendations based on the recommendation weights.

[0154] The above is an illustrative embodiment of an electronic device according to this specification. It should be noted that the technical solution of this electronic device and the technical solution of the aforementioned entity recommendation processing method belong to the same concept. Details not described in detail in the technical solution of the electronic device can be found in the description of the technical solution of the aforementioned entity recommendation processing method.

[0155] In the description of this application, it should be understood that the terms "first," "second," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance. In the description of this application, it should be noted that, unless otherwise expressly specified and limited, "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances. Furthermore, in the description of this application, unless otherwise stated, "multiple" means two or more. "And / or" describes the relationship between related objects, indicating that three relationships can exist; for example, A and / or B can represent: A alone, A and B simultaneously, and B alone. The character " / " generally indicates that the preceding and following related objects are in an "or" relationship.

[0156] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The program can be stored in a computer-readable storage medium, and when executed, it can include the processes of the embodiments of the above methods. The storage medium can be a magnetic disk, optical disk, read-only memory, or random access memory, etc.

[0157] The above-disclosed embodiments are merely preferred embodiments of this application and should not be construed as limiting the scope of this application. Therefore, any equivalent variations made in accordance with the claims of this application shall still fall within the scope of this application.

Claims

1. An entity recommendation method, characterized in that, The method includes: In response to the input query text, determine the link entity corresponding to the query text; Based on the link entity, obtain the corresponding entity set from each of the at least two entity embedding representations, wherein the at least two entity embedding representations include at least: a preset granularity label embedding representation and a network topology embedding representation, and the at least two entity embedding representations also include a triplet embedding representation; Determine the recommendation weight for each entity in the entity set, and make entity recommendations based on the recommendation weight; The step of obtaining the corresponding entity set from each of at least two entity embedding representations based on the linked entity includes: Based on the link entity, obtain the corresponding first entity set from the preset granularity tag embedding representation; Based on the link entity, obtain the corresponding second entity set from the network topology embedding representation; Based on the linked entity, the corresponding set of third entities is obtained from the triple embedding representation; Determining the recommendation weight corresponding to each entity in the entity set includes: The first entity set, the second entity set, and the third entity set are merged and deduplicated to generate a merged entity set. Determine the frequency of occurrence of each entity in the merged entity set within the first entity set, the second entity set, and the third entity set. Based on the frequency of occurrence of each entity in the merged entity set, the entity weight corresponding to each entity in the merged entity set is determined; The step of determining the link entity corresponding to the query text includes: The query text is vectorized to generate a query text vector corresponding to the query text. The query text vector is matched with the preset granularity label embedding representation to obtain the first candidate entity set; The query text vector is matched with the entity vectors in the preset entity library to obtain the second candidate entity set; The entities in the first candidate entity set and the second candidate entity set are scored and sorted, and the link entities corresponding to the query text are determined based on the sorting results.

2. The method according to claim 1, characterized in that, The method further includes: The preset granularity tags of each entity are extracted from the unstructured text using a predetermined entity extraction method, and a tuple of <entity, preset granularity tag> is generated. Clustering operations are performed on the preset granularity labels of the binary tuples of each entity to form a set of preset granularity labels corresponding to different meanings under the same entity. The preset granularity tag set corresponding to different meanings is used as text, and the preset granularity tag embedding representation is obtained by using a preset vector representation method.

3. The method according to claim 2, characterized in that, The method further includes: Determine the tag frequency of the preset granularity tags corresponding to each definition of the same entity; The frequency of the tags is normalized, and the tag weights of each sense of the same entity are determined based on the result of the normalization process. Based on the <entity, preset granularity label> tuple and the label weight, generate the <entity, preset granularity label, label weight> triple corresponding to each sense of the same entity.

4. The method according to claim 1, characterized in that, The recommendation weights include entity weights and tag weights. Determining the recommendation weights for each entity in the entity set includes: Determine the preset granularity label corresponding to each entity in the merged entity set in the <entity, preset granularity label, label weight> triple, and obtain the preset granularity label set corresponding to each entity in the merged entity set; Based on each preset granularity label of the preset granularity label set, the entities in the merged entity set are classified to obtain the entity set corresponding to each preset granularity label; Based on the entity weights of each entity in the entity set corresponding to each preset granularity label, the total entity weight of each preset granularity label is determined. Based on the total entity weight and the tag weight corresponding to each preset granularity tag, the recommendation weight corresponding to each preset granularity tag is determined.

5. The method according to claim 1, characterized in that, The method further includes: By parsing the encyclopedia page of the entity to be processed, an edge is constructed between the entity to be processed and the corresponding internal link entity, forming a graph network; By traversing the graph network, the entity sequence corresponding to the entity to be processed is extracted; The network topology embedding representation corresponding to the entity to be processed is determined based on the entity sequence.

6. A physical recommendation device, characterized in that, The device includes: The link entity determination module is used to determine the link entity corresponding to the input query text in response to the input query text; The entity set determination module is used to obtain the corresponding entity set based on the linked entity from each of at least two entity embedding representations, wherein the at least two entity embedding representations include at least: a preset granularity label embedding representation and a network topology embedding representation, and the at least two entity embedding representations also include a triplet embedding representation; The entity recommendation module is used to determine the recommendation weight of each entity in the entity set and to make entity recommendations based on the recommendation weight. The entity set determination module is also used for: Based on the link entity, obtain the corresponding first entity set from the preset granularity tag embedding representation; Based on the link entity, obtain the corresponding second entity set from the network topology embedding representation; Based on the linked entity, the corresponding set of third entities is obtained from the triple embedding representation; The entity recommendation module is also used for: The first entity set, the second entity set, and the third entity set are merged and deduplicated to generate a merged entity set. Determine the frequency of occurrence of each entity in the merged entity set within the first entity set, the second entity set, and the third entity set. Based on the frequency of occurrence of each entity in the merged entity set, the entity weight corresponding to each entity in the merged entity set is determined; The link entity determination module is also used for: The query text is vectorized to generate a query text vector corresponding to the query text. The query text vector is matched with the preset granularity label embedding representation to obtain the first candidate entity set; The query text vector is matched with the entity vectors in the preset entity library to obtain the second candidate entity set; The entities in the first candidate entity set and the second candidate entity set are scored and sorted, and the link entities corresponding to the query text are determined based on the sorting results.

7. A computer storage medium storing a plurality of instructions adapted for loading by a processor and performing the steps of the method as claimed in any one of claims 1 to 5.

8. An electronic device, characterized in that, include: A processor and a memory, the memory storing a computer program adapted to be loaded by the processor and to execute the steps of the method as claimed in any one of claims 1 to 5.

Images