Map Point of Interest Query Method, Device, Equipment, Storage Medium and Program Product

Abstract

The present disclosure provides a method, apparatus, electronic device, computer-readable storage medium, and computer program product for querying map points of interest, which relates to the fields of artificial intelligence technologies such as deep learning, natural language processing, knowledge graphs, and intelligent maps. The method includes: receiving an input query request for a point of interest; extracting a target query term included in the query request for the point of interest; using a pre-determined semantic correspondence relationship between the query term and the point of interest to determine a target point of interest corresponding to the target query term; wherein, the query terms and points of interest used to determine the semantic correspondence relationship are extracted from a preset multi-source map knowledge model, and the multi-source map knowledge model records multi-source correspondence relationships between points of interest and multi-source knowledge, and the multi-source knowledge includes: geographical location information, knowledge graph information, and input query terms. Applying this method can improve the accuracy of query results of the point-of-interest query service.

Description

Technical Field

[0001] The present disclosure relates to the technical field of data processing, specifically to artificial intelligence technical fields such as deep learning, natural language processing, knowledge graph, intelligent map, etc., and particularly relates to a method, device, electronic device, computer-readable storage medium and computer program product for querying map points of interest. Background Art

[0002] Navigation electronic map products not only provide travel services for hundreds of millions of users every day, but also serve as the digital foundation of new infrastructure, playing an increasingly important and indispensable role in various industries. For map products, data is the infrastructure for survival, accuracy is the lifeline of the map, and timeliness is the essential ability to depict the real world.

[0003] In addition to a large amount of real data, scientific processing of various types of knowledge involved in the map field can enable products built on map data to better meet the needs of users. Summary of the Invention

[0004] Embodiments of the present disclosure propose a method, device, electronic device, computer-readable storage medium and computer program product for querying map points of interest.

[0005] In a first aspect, embodiments of the present disclosure propose a method for querying map points of interest, including: receiving an input point-of-interest query request; extracting a target query term included in the point-of-interest query request; using a pre-determined semantic correspondence relationship between the query term and the point of interest to determine a target point of interest corresponding to the target query term; wherein, the query term and the point of interest used to determine the semantic correspondence relationship are extracted from a pre-set multi-source map knowledge model, and the multi-source map knowledge model records a multi-source correspondence relationship between the point of interest and multi-source knowledge, and the multi-source knowledge includes: geographical location information, knowledge graph information, and the input query term.

[0006] In a second aspect, embodiments of the present disclosure propose a device for querying map points of interest, including: a point-of-interest query request receiving unit configured to receive an input point-of-interest query request; a target query term extracting unit configured to extract a target query term included in the point-of-interest query request; a target point-of-interest determining unit configured to use a pre-determined semantic correspondence relationship between the query term and the point of interest to determine a target point of interest corresponding to the target query term; wherein, the query term and the point of interest used to determine the semantic correspondence relationship are extracted from a pre-set multi-source map knowledge model, and the multi-source map knowledge model records a multi-source correspondence relationship between the point of interest and multi-source knowledge, and the multi-source knowledge includes: geographical location information, knowledge graph information, and the input query term.

[0007] In a third aspect, embodiments of the present disclosure provide an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to implement the map point of interest query method described in any implementation manner of the first aspect.

[0008] In a fourth aspect, embodiments of the present disclosure provide a non-transitory computer-readable storage medium storing computer instructions, which are used to enable a computer to implement the map point of interest query method described in any implementation manner of the first aspect when executed.

[0009] In a fifth aspect, embodiments of the present disclosure provide a computer program product including a computer program, which is capable of implementing the map point of interest query method described in any implementation manner of the first aspect when executed by a processor.

[0010] In order to provide a better map point of interest query service to users, the present disclosure pre-constructs a multi-source map knowledge model based on points of interest and corresponding multi-source knowledge. Since the multi-source knowledge covers geographical location information, knowledge graph information, and input query terms, the trained multi-source map knowledge model can more comprehensively and accurately determine query terms matching the points of interest. The materials for semantic relevance modeling are exactly the points of interest and matching query terms extracted from the trained multi-source map knowledge model, thereby obtaining a semantic correspondence relationship that can accurately reflect the user's search habits at the semantic level. Finally, with the help of this semantic correspondence relationship, the target point of interest can be accurately determined, improving the accuracy of the query results of the point of interest query service.

[0011] It should be understood that the content described in this part is not intended to identify the key or important features of the embodiments of the present disclosure, nor is it used to limit the scope of the present disclosure. Other features of the present disclosure will become easily understandable through the following description. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] By reading the detailed description of the non-limiting embodiments with reference to the following drawings, other features, objectives, and advantages of the present disclosure will become more apparent:

[0013] Figure 1 is an exemplary system architecture to which the present disclosure can be applied;

[0014] Figure 2 is a flowchart of a map point of interest query method provided by an embodiment of the present disclosure;

[0015] Figure 3 is a flowchart of a method for constructing a multi-source map knowledge model provided by an embodiment of the present disclosure;

[0016] Figure 4 is a schematic diagram of node association constructed based on Figure 3 the provided solution;

[0017] Figure 5 is a flowchart of a method for determining node categories provided by an embodiment of the present disclosure;

[0018] Figure 6 is a flowchart of a method for node annotation provided by an embodiment of the present disclosure;

[0019] Figure 7 is a block diagram of the structure of a map point of interest query device provided by an embodiment of the present disclosure;

[0020] Figure 8 is a schematic diagram of the structure of an electronic device suitable for executing a map point of interest query method provided by an embodiment of the present disclosure. Detailed implementation manners

[0021] The following describes exemplary embodiments of the present disclosure with reference to the accompanying drawings. Various details of the embodiments of the present disclosure are included to assist understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for clarity and conciseness, descriptions of well-known functions and structures are omitted below. It should be noted that, without conflict, the embodiments in the present disclosure and the features in the embodiments can be combined with each other.

[0022] In the technical solution of the present disclosure, the collection, storage, use, processing, transmission, provision, and disclosure of the user's personal information and other processes all comply with the provisions of relevant laws and regulations and do not violate public order and good customs.

[0023] Figure 1 Illustrates an exemplary system architecture 100 of an embodiment to which the map point of interest query method, device, electronic device, and computer-readable storage medium of the present disclosure can be applied.

[0024] As Figure 1 shown, the system architecture 100 may include terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used as a medium to provide a communication link between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or fiber optic cables, etc.

[0025] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc. Various applications for implementing information communication between the two can be installed on terminal devices 101, 102, 103 and server 105, such as map navigation applications, model training applications, instant messaging applications, etc.

[0026] Terminal devices 101, 102, 103 and server 105 can be hardware or software. When terminal devices 101, 102, 103 are hardware, they can be various electronic devices with a display screen, including but not limited to smartphones, tablets, laptop computers, desktop computers, etc.; when terminal devices 101, 102, 103 are software, they can be installed in the above-listed electronic devices, and can be implemented as multiple software or software modules, or can be implemented as a single software or software module, which is not specifically limited here. When server 105 is hardware, it can be implemented as a distributed server cluster composed of multiple servers, or can be implemented as a single server; when the server is software, it can be implemented as multiple software or software modules, or can be implemented as a single software or software module, which is not specifically limited here.

[0027] Server 105 can provide various services through various built-in applications. Taking a map navigation application that can provide map point of interest query services as an example, when server 105 runs this map navigation application, the following effects can be achieved: First, receive the point of interest query requests transmitted by terminal devices 101, 102, 103 via network 104; then, extract the target query term included in the point of interest query request; next, use the pre-determined semantic correspondence relationship between the query term and the point of interest to determine the target point of interest corresponding to the target query term; finally, return the determined target point of interest to terminal devices 101, 102, 103 via network 104.

[0028] Among them, the query terms and points of interest used to determine the semantic correspondence relationship are extracted from a pre-set multi-source map knowledge model, and the multi-source map knowledge model records the multi-source correspondence relationship between the points of interest and multi-source knowledge. The multi-source knowledge includes: geographical location information, knowledge graph information, and the input query terms. Specifically, the multi-source map knowledge model can be pre-trained by server 105 through the built-in model training application.

[0029] It should be noted that, in addition to being obtained from the terminal devices 101, 102, and 103 through the network 104, the point of interest query requests can also be pre-stored locally in the server 105 in various ways. Therefore, when the server 105 detects that these data have been stored locally (such as the pending query tasks left before starting to process), it can choose to directly obtain these data from the local. In this case, the exemplary system architecture 100 may not include the terminal devices 101, 102, 103 and the network 104 either.

[0030] Generally, the map point of interest query method provided in the subsequent embodiments of the present disclosure is executed by the server 105 with relatively strong computing power and more computing resources. Correspondingly, the map point of interest query device is generally also set in the server 105. However, it should also be noted that when the terminal devices 101, 102, and 103 also have the required computing power and computing resources, the terminal devices 101, 102, and 103 can also complete the above operations originally performed by the server 105 through the map navigation applications installed thereon, and then output the same results as the server 105. Especially in the case where there are multiple terminal devices with different computing capabilities at the same time, when the map navigation application determines that the terminal device where it is located has relatively strong computing power and more remaining computing resources, the terminal device can be allowed to execute the above operations, thereby appropriately reducing the computing pressure on the server 105. Correspondingly, the map point of interest query device can also be set in the terminal devices 101, 102, and 103. In this case, the exemplary system architecture 100 may not include the server 105 and the network 104 either.

[0031] It should be understood that Figure 1 the numbers of the terminal devices, the network, and the server in

[0032] are only illustrative. According to the implementation requirements, there can be any number of terminal devices, networks, and servers.

[0032] Please refer to Figure 2 , Figure 2 which is a flowchart of a map point of interest query method provided by an embodiment of the present disclosure. The process 200 includes the following steps:

[0033] Step 201: Receive an input point of interest query request;

[0034] This step aims to receive, by the execution entity of the map point of interest query method (such as Figure 1 the server 105 shown), the point of interest query request transmitted by the user terminal (such as Figure 1 the terminal devices 101, 102, and 103 shown) through the network 104.

[0035] Among them, the point of interest query request is used to represent a user's query demand for a certain point of interest, which at least includes a target query term used as the query basis. The target query term can be directly expressed as plain text information or unencrypted voice signals, or can also be expressed as a ciphertext string or encrypted voice signal to meet security requirements.

[0036] Of course, in addition to including the basic information characterizing the target query term, it can also include the terminal information of the user terminal, the current location when the user initiates the request, and possibly included point of interest query restriction conditions, so as to combine various information that may affect the query result to improve the accuracy of the obtained query result and better meet the user's needs.

[0037] Step 202: Extract the included target query term from the point of interest query request;

[0038] Based on step 201, this step aims to extract the included target query term from the point of interest query request by the above-mentioned execution entity.

[0039] Specifically, considering that the target query term may be expressed in different forms, the correct and matching extraction method also needs to be adopted during extraction. For example, when the target query term is included in the point of interest query request in ciphertext or encrypted form, the correct decryption key or decryption rule also needs to be determined in advance to obtain the correct target query term; when the target query term is expressed as a voice signal, the voice recognition technology also needs to be used to convert it into a text form of the target query term that is convenient for subsequent processing; when the point of interest query request does not directly include the target query term, but includes relevant information that can guide the acquisition of the target query term from other places, the target query term also needs to be correctly obtained from other places according to the relevant information.

[0040] Step 203: Use the pre-determined semantic correspondence relationship between the query term and the point of interest to determine the target point of interest corresponding to the target query term.

[0041] Based on step 202, this step aims to have the above-mentioned execution entity use the determined semantic correspondence relationship to determine the target point of interest that has a correspondence relationship with the target query term and the correspondence relationship is consistent with the semantic correspondence relationship.

[0042] Among them, the query words and points of interest used to determine the semantic correspondence are extracted from a preset multivariate map knowledge model, and the multivariate map knowledge model records the multivariate correspondence between points of interest and multivariate knowledge, and the multivariate knowledge includes: geographic location information, knowledge graph information and input query words. Since the multivariate map knowledge model is centered on points of interest and also integrates knowledge of multiple different dimensions, it can not only better understand the points of interest with the help of these multivariate knowledge, but multivariate knowledge of other dimensions can also help determine the relevance between points of interest and knowledge of a certain dimension. Therefore, the present disclosure chooses to extract points of interest and query words with matching relationships from a model that integrates multivariate knowledge, and based on this, determines their semantic correspondence at the semantic level. While inheriting the benefits brought by multivariate knowledge, it can also simplify the correspondence as much as possible.

[0043] Specifically, a method for predetermining the semantic correspondence between query terms and POIs can be: extracting real POIs and corresponding historical query terms from a multivariate map knowledge model, and then using semantic relevance modeling techniques to determine the semantic correspondence between the real POIs and corresponding historical query terms, thereby obtaining a semantic correspondence. Of course, other techniques that achieve the same or similar effects may also be used to obtain the semantic correspondence.

[0044] It should be understood that the reason for determining the semantic correspondence between matching query terms and POIs is to explain as much as possible the query terms that reflect the user's search habits when trying to search for a POI while integrating multiple knowledge. That is, in actual situations, the query terms are often different from the official text of the POI, but the two have the same semantics. In other words, it is often difficult for users to accurately remember the official name of the POI when searching, but they can often "invent or name" a query term based on the semantics expressed.

[0045] To provide users with a better map point of interest query service, the map point of interest query method provided in the disclosed embodiment pre-constructs a multivariate map knowledge model based on points of interest and corresponding multivariate knowledge. Because multivariate knowledge covers geographic location information, knowledge graph information, and input query terms, the trained multivariate map knowledge model can more comprehensively and accurately determine query terms that match points of interest. The material for semantic relevance modeling is precisely the points of interest and matching query terms extracted from the trained multivariate map knowledge model, which then yields a semantic correspondence that accurately reflects the user's search habits at the semantic level. Ultimately, this semantic correspondence is used to accurately determine the target point of interest, improving the accuracy of the query results of the point of interest query service.

[0046] In order to better understand the multivariate map knowledge model used in process 200, this embodiment uses Figure 3The flowchart shows a method for constructing a multi - source map knowledge model, where process 300 includes the following steps:

[0047] Step 301: Obtain each point of interest within the target map area;

[0048] Step 302: Obtain the geographical location information, knowledge graph information, and the input query term corresponding to the point of interest;

[0049] A specific way to obtain geographical location information, knowledge graph information, and the input query term includes:

[0050] Obtain the geographical location code corresponding to the point of interest; in the preset knowledge graph, obtain the point - of - interest entity information corresponding to the point of interest as the entity to be queried; in the operation log recording user operation information (the reading of this operation log has obtained the corresponding user's authorization in advance), obtain the query term input by the user before the point of interest is selected.

[0051] Among them, the geographical location code can be specifically represented as a string. For example, it can be obtained based on Geohash (a method of address coding that can encode two - dimensional spatial longitude and latitude data into a string) or based on Google - S2; the knowledge graph can be a graph dedicated to recording various knowledge related to the entities of map points of interest. For example, taking a point of interest representing a certain restaurant as an example, this graph can record: guest reviews, ratings for certain dishes, per capita consumption, recommended dishes, floor location, queue waiting time, business hours, etc.; the input query term of the user is used to establish the correspondence between the input query term of the user and the official name of the actually selected point of interest.

[0052] Step 303: Take the point of interest as the main node, and take the geographical location information, knowledge graph information, and the input query term as different slave nodes of the main node respectively, and draw lines between the main node and each slave node;

[0053] Step 304: Determine the association relationship between different points of interest according to the user behavior information, and draw corresponding lines between the main nodes according to the association relationship to obtain a node graph;

[0054] Among them, the line feature of the line is determined based on the confidence level between the two connected nodes. For example, different confidence levels can be represented by different colors of the line, or by the length, thickness, etc. of the line.

[0055] Step 305: Based on the preset training task objective, pre - train the node graph through a graph neural network to obtain a multi - source map knowledge model that meets the requirements of the training task objective.

[0056] That is, the method provided in this embodiment of defining various types of knowledge as nodes and then establishing connections between the nodes can construct a node graph for learning through a graph neural network, such that under the guidance of a preset training task objective, the required multiple correspondence relationships can be learned from the connection relationships recorded in the nodes, and thus a multiple map knowledge model that meets the requirements of the training task objective can be obtained. Figure 4 is a schematic diagram of node associations constructed based on Figure 3 the solution provided.

[0057] Specifically, the training task objective can be expressed as: performing multi-classification on each main node, and the accuracy of the classification results of the multi-classification meets the preset requirements. The classification referred to in the multi-classification can be classification of categories, types, or attributes, and the classification result accuracy is used to represent the accuracy.

[0058] In this embodiment, a graph neural network is used as the model architecture. Points of interest (usually represented by names in text form) are used as the main nodes, and the corresponding geographical location information, knowledge graph information, and input query words of the points of interest are used as different slave nodes of the main nodes respectively. A node graph is established, which is represented by the connections between the main nodes and the slave nodes and between the main nodes. The node graph is used as a training sample to be trained according to the graph neural network, and thus a multiple map knowledge model that can reflect the multiple knowledge associations of the points of interest as much as possible is obtained.

[0059] It should be understood that in addition to the graph neural network used in this embodiment, other models that can achieve similar effects can also be used as the architecture. This embodiment only uses the graph neural network as a preferred example to describe the whole process. Other application scenarios can select other models according to all possible existing limiting conditions or constraints in the scenario by themselves, and no specific limitation is made here.

[0060] Based on the embodiment shown in process 300, in order to improve the model training effect as much as possible and shorten the model training time-consuming, a secondary pre-training technology can also be adopted. The specific implementation process can be:

[0061] Obtain the trained parameters of the network structure related to the text content from a trained model for natural language processing;

[0062] Use the trained parameters as the initial parameters of the network structure related to the text content in the graph neural network to obtain the graph neural network to be trained;

[0063] Pre-train the constructed node graph through the graph neural network to be trained.

[0064] The reason why the above implementation process is described as secondary pre-training is that the process of training the model (such as models like BERT, ERNIE, GPT-3 commonly used in the field of natural language processing) starting from random parameters is called primary pre-training. Since this embodiment inherits the results of the primary pre-training of the model and does not need to re-start pre-training from random parameters, that is, the training started by the graph neural network after inheriting the trained parameters is called secondary pre-training. Since only the parameters related to the text content are inherited, it does not affect the network structure related to non-text content in the graph neural network, thereby improving the training effect and shortening the training time without bringing negative impacts.

[0065] Considering that various entities in the real world are constantly developing, updating, and the points of interest are constantly increasing, replacing, and changing, based on Figure 3 the multi-source map knowledge model provided by the illustrated embodiment, it can also be used to determine the node category, so as to enrich the node information in the case of helping to determine the node category, thereby further enhancing the relevance between nodes.

[0066] Please refer to Figure 5 , Figure 5 which is a flowchart of a method for determining node categories provided by an embodiment of the present disclosure, where process 500 includes the following steps:

[0067] Step 501: For the first query term already included in the multi-source map knowledge model, obtain the node vector representation of the first query term, determine the probabilities of belonging to different categories according to the node vector representation, and use the category with the highest corresponding probability as the actual category of the first query term;

[0068] Among them, the node vector representation is jointly determined based on the information of other nodes that have established connections or adjacent relationships with itself, so various aspects of information can be combined to correctly represent the node to which it belongs.

[0069] Step 502: For the second query term not included in the multi-source map knowledge model, determine the target first query term similar to the second query term, and use the category of the target first query term with the highest corresponding similarity as the actual category of the second query term.

[0070] Different from the included first query terms, most of the second query terms are newly generated query terms. This embodiment provides an implementation manner of determining the similar target first query term according to similarity and inheriting the category of the target first query term.

[0071] Since the multi-source map knowledge model integrates multi-dimensional knowledge, and the associations between the multi-dimensional knowledge and points of interest can be used to solve other problems. For example, based on the consistency between slave nodes connected to different master nodes, annotation completion or annotation adjustment can be performed according to the consistency results to achieve automatic annotation of unannotated points of interest.

[0072] Please refer to Figure 6 , Figure 6 which is a flowchart of a node annotation method provided by an embodiment of the present disclosure. The process 600 includes the following steps:

[0073] Step 601: Based on the consistency between slave nodes connected to different master nodes, perform an annotation content completion operation on master nodes and / or slave nodes that are unannotated or have a number of annotations less than a preset quantity.

[0074] Step 602: Based on the consistency between slave nodes connected to different master nodes, determine abnormal master nodes and / or abnormal slave nodes with incorrect annotations, and initiate an incorrect annotation inquiry or perform a correction operation on the abnormal master nodes and / or abnormal slave nodes.

[0075] Among them, the annotation content completion operation refers to supplementing the annotation information of nodes with annotations to nodes that are consistent and have no annotation or have a small number of annotations, so as to achieve the completion of annotation information; the incorrect annotation inquiry aims to determine whether there is an error in the consistency judgment by means of inquiry.

[0076] To deepen the understanding, the present disclosure also gives a specific implementation scheme for training a multi-source map knowledge model in combination with the actual situation:

[0077] In order to train a pre-trained model in the map field containing multi-source knowledge on the basis of a pre-trained model in the general field, this embodiment selects to integrate the following key information: 1) basic information of POIs (name, alias, address, category); 2) geographical location information of POIs; 3) knowledge graph information in the POI field; 4) user behavior data related to POIs.

[0078] Among them, the basic information, geographical location information, knowledge graph information, and user behavior information of POIs can be represented in the form of a graph, and are specifically manifested as 4 different types of nodes in the graph, namely POIs, queries (query terms), geographical location information, and labels associated with POIs in the knowledge graph.

[0079] Each node has its own features, mainly text features. Nodes are connected by edges. The graph mainly contains the following types of edges: 1) The path relationship between POIs, indicating that a user has traveled from one POI to another; 2) The click relationship between a query and a POI; 3) The relationship between a POI and a graph spectrum label; 4) The relationship between a POI and a geographical location. The constructed node graph can be seen in Figure 4 。

[0080] To better model the relationships between nodes in the graph, this embodiment proposes a pre-training model based on a graph neural network. When the pre-training model is applied to downstream tasks, it can either provide a pre-training model similar to the general domain for downstream tasks to model the text, or learn a vector representation containing rich information for each type of node. For example, the POI vector representation can fully integrate the basic attribute information of the POI, geographical location, the association between POIs, and the association between POIs and queries. These vectors can be used for tasks such as semantic recall in map retrieval, query analysis, POI retrieval, and personalized recommendation of POIs. The specific technical solution is as follows:

[0081] In a graph neural network, the vector representation of each node is jointly determined by its own feature information and its neighbor nodes in the graph. The features of the node itself are mainly text information. For example, the features of a POI node mainly include name, alias, address, and category. The features of a query node are mainly the query terms. The features of a graph spectrum label node include label type, content, and related description text. Geographical location information can be represented as a string, such as obtained based on Geohash or Google-S2.

[0082] Since the main feature of the node is text, and the pre-training models in the general domain have very good effects on text modeling, existing pre-training models such as BERT, ERNIE, GPT-3 (all are deep learning models commonly used in natural language processing) can be used to model the text features of the nodes. The simplest way is to take the vector corresponding to the output layer CLS (the full English name is Classification, which can be understood as being used for downstream classification tasks) as the initial vector representation of the node.

[0083] The update of the node vector is based on the vector representation of the node itself and the vector representation of its neighbor nodes in the previous iteration, which can be expressed by the following formula:

[0084] where N(u) represents the neighbor nodes of u in the graph, h u represents the vector representation of node u, represents the initial vector representation of node u, Denotes the vector representation of node u after k iterations.

[0085] The aggregate function in the formula is responsible for aggregating the representations of the neighbor nodes of node u in the graph. When designing the aggregate, the specific characteristics of the heterogeneous graph in the map domain need to be considered. Since the graph contains different types of nodes and relationships, when aggregating the representations of neighbor nodes, in order to better retain the information of different types of nodes, the information of each type of node can be aggregated separately, and then the information of different types of nodes can be summarized. When aggregating the information of each type of neighbor node, the representations of the neighbor nodes can be added, averaged, or an attention mechanism can be introduced to weight according to the importance of different nodes to the current node. Since the edges in the graph naturally have confidence, the vector representations of neighbor nodes can also be weighted and summed according to the weights of the edges. The update function updates the vector representation of each node based on the vector representation of the node itself and the aggregated vector representation of the neighbor nodes in the previous iteration. There can be various design schemes for this function. For example, the node representation and the representation of the neighbor nodes are respectively passed through a linear layer and then added together, and then passed through a non-linear transformation to obtain the updated vector representation.

[0086] The number of iterations k indicates that at most the k-hop neighbor information of one node is used. The final representation of each node in the graph can be individually taken as the node representation after the k-th iteration. It can also be combined with the initial (such as concatenation), or combine the representation from to as the final node representation. The advantage of this method is that it can retain more important feature information of the node itself, such as text description information.

[0087] The pre-training task designed in this embodiment is node prediction, and the goal is to predict which category a POI node belongs to (the map domain classifies POIs, and each POI is labeled with a category). For node prediction, it can be regarded as a classification task. The representation of the node to be predicted is passed through a simple network structure, such as a linear layer, and then through a softmax (normalization) layer to predict the category it belongs to. The loss function for model training can use the cross-entropy loss function.

[0088] For further reference Figure 7 , as an implementation of the methods shown in the above figures, the present disclosure provides an embodiment of a map point of interest query device. This device embodiment corresponds to the method embodiment shown in Figure 2 and this device can be specifically applied to various electronic devices.

[0089] Such asFigure 7 As shown in Figure 7 , the map point of interest query device 700 in this embodiment may include: a point of interest query request receiving unit 701, a target query term extraction unit 702, and a target point of interest determination unit 703. Among them, the point of interest query request receiving unit 701 is configured to receive an input point of interest query request; the target query term extraction unit 702 is configured to extract the included target query term from the point of interest query request; the target point of interest determination unit 703 is configured to use a pre-determined semantic correspondence relationship between the query term and the point of interest to determine the target point of interest corresponding to the target query term; among them, the query term and the point of interest used to determine the semantic correspondence relationship are extracted from a preset multi-source map knowledge model, and the multi-source map knowledge model records the multi-source correspondence relationship between the point of interest and multi-source knowledge, and the multi-source knowledge includes: geographical location information, knowledge graph information, and the input query term.

[0090] In this embodiment, in the map point of interest query device 700: the specific processing of the point of interest query request receiving unit 701, the target query term extraction unit 702, and the target point of interest determination unit 703 and the technical effects brought by them can be respectively referred to Figure 2 the relevant descriptions of steps 201-203 in the corresponding embodiment, which will not be elaborated here.

[0091] In some optional implementation manners of this embodiment, the map point of interest query device 700 may further include:

[0092] An information extraction unit, configured to extract the real point of interest and the corresponding historical query term from the multi-source map knowledge model;

[0093] A semantic relevance modeling unit, configured to use semantic relevance modeling technology to determine the correspondence relationship at the semantic level between the real point of interest and the corresponding historical query term, and obtain the semantic correspondence relationship.

[0094] In some optional implementation manners of this embodiment, the map point of interest query device 700 may further include:

[0095] A point of interest text information acquisition unit, configured to acquire each point of interest within the target map area;

[0096] A multi-source knowledge acquisition unit, configured to acquire the geographical location information, knowledge graph information, and the input query term corresponding to the point of interest;

[0097] A master-slave node determination and connection unit, configured to use the point of interest as the master node, and use the geographical location information, knowledge graph information, and the input query term as different slave nodes of the master node respectively, and make connections between the master node and each slave node;

[0098] The main node connection unit is configured to determine the association relationship between different points of interest according to user behavior information, and make corresponding connections between the main nodes according to the association relationship to obtain a node graph; wherein, the line features of the connections are determined based on the confidence between the two connected nodes.

[0099] The pre-training unit is configured to pre-train the node graph through a graph neural network based on a preset training task objective to obtain a multi-source map knowledge model that meets the requirements of the training task objective.

[0100] In some optional implementation manners of this embodiment, the multi-source knowledge acquisition unit may be further configured to:

[0101] Obtain the geographical location encoding corresponding to the point of interest.

[0102] In a preset knowledge graph, obtain the point of interest entity information corresponding to the point of interest as the entity to be queried.

[0103] In the operation log, obtain the query terms input by the user before the point of interest is selected.

[0104] In some optional implementation manners of this embodiment, the training task objective includes: performing multi-classification on each main node, and the accuracy of the classification results of the multi-classification meets the preset requirements.

[0105] In some optional implementation manners of this embodiment, the map point of interest query device 700 may further include:

[0106] The included query term category determination unit is configured to, for the first query term included in the multi-source map knowledge model, obtain the node vector representation of the first query term, determine the probabilities of belonging to different categories according to the node vector representation, and use the category with the highest corresponding probability as the actual category of the first query term; wherein, the node vector representation is jointly determined based on the information of other nodes that have connections or adjacent relationships with itself.

[0107] The unincluded query term category determination unit is configured to, for the second query term not included in the multi-source map knowledge model, determine the target first query term similar to the second query term, and determine the category of the target first query term with the highest corresponding similarity as the actual category of the second query term.

[0108] In some optional implementation manners of this embodiment, the map point of interest query device 700 may further include:

[0109] The node annotation content completion unit is configured to perform an annotation content completion operation on the main node and / or the slave node that is not annotated or has less than the preset number of annotations based on the consistency between the slave nodes connected to different main nodes.

[0110] An abnormal annotation content processing unit is configured to determine an abnormal master node and / or an abnormal slave node with annotation errors based on the consistency among slave nodes connected to different master nodes, and initiate an error annotation inquiry or perform a correction operation on the abnormal master node and / or the abnormal slave node.

[0111] In some alternative implementation manners of this embodiment, the pre-training unit may be further configured to:

[0112] Obtain the trained parameters of the network structure related to the text content from a trained model for natural language processing;

[0113] Use the trained parameters as the initial parameters of the network structure related to the text content in the graph neural network to obtain the graph neural network to be trained;

[0114] Pre-train the constructed node graph through the graph neural network to be trained.

[0115] This embodiment exists as a device embodiment corresponding to the above method embodiment.

[0116] In order to provide users with better map point of interest query services, the map point of interest query device provided in the embodiments of the present disclosure pre-constructs a multi-source map knowledge model based on points of interest and corresponding multi-source knowledge. Since the multi-source knowledge covers geographical location information, knowledge graph information, and input query terms, the trained multi-source map knowledge model can more comprehensively and accurately determine the query terms matching the points of interest. The materials for semantic relevance modeling are exactly the points of interest and the matching query terms extracted from the trained multi-source map knowledge model, so as to obtain a semantic correspondence relationship that can accurately reflect the user's search habits at the semantic level. Finally, the target point of interest is accurately determined with the help of this semantic correspondence relationship, improving the accuracy of the query results of the point of interest query service.

[0117] According to an embodiment of the present disclosure, the present disclosure also provides an electronic device, which includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein, the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor can implement the map point of interest query method described in any of the above embodiments.

[0118] According to an embodiment of the present disclosure, the present disclosure also provides a readable storage medium, which stores computer instructions for enabling a computer to implement the map point of interest query method described in any of the above embodiments when executed.

[0119] According to an embodiment of the present disclosure, the present disclosure also provides a computer program product, which, when executed by a processor, can implement the map point of interest query method described in any of the above embodiments.

[0120] Figure 8 FIG. shows a schematic block diagram of an exemplary electronic device 800 that can be used to implement embodiments of the present disclosure. The electronic device is intended to represent various forms of digital computers, such as, for example, a laptop computer, a desktop computer, a workbench, a personal digital assistant, a server, a blade server, a mainframe computer, and other suitable computers. The electronic device can also represent various forms of mobile devices, such as, for example, a personal digital processor, a cellular phone, a smart phone, a wearable device, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0121] As Figure 8 shown, the device 800 includes a computing unit 801, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 802 or a computer program loaded from a storage unit 808 into a random access memory (RAM) 803. In the RAM 803, various programs and data required for the operation of the device 800 can also be stored. The computing unit 801, the ROM 802, and the RAM 803 are connected to each other via a bus 804. An input / output (I / O) interface 805 is also connected to the bus 804.

[0122] A plurality of components in the device 800 are connected to the I / O interface 805, including: an input unit 806, such as a keyboard, a mouse, etc.; an output unit 807, such as various types of displays, speakers, etc.; a storage unit 808, such as a magnetic disk, an optical disk, etc.; and a communication unit 809, such as a network card, a modem, a wireless communication transceiver, etc. The communication unit 809 allows the device 800 to exchange information / data with other devices via a computer network such as the Internet and / or various telecommunication networks.

[0123] The computing unit 801 can be various general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various dedicated artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 801 executes the various methods and processes described above, such as the map point of interest query method. For example, in some embodiments, the map point of interest query method can be implemented as a computer software program that is tangibly contained in a machine-readable medium, such as the storage unit 808. In some embodiments, part or all of the computer program can be loaded and / or installed onto the device 800 via the ROM 802 and / or the communication unit 809. When the computer program is loaded into the RAM 803 and executed by the computing unit 801, one or more steps of the map point of interest query method described above can be executed. Alternatively, in other embodiments, the computing unit 801 can be configured to execute the map point of interest query method in any other suitable way (e.g., by means of firmware).

[0124] Various embodiments of the systems and techniques described above in this document can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-chip systems (SOCs), complex programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments can include: being implemented in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which can be a dedicated or general-purpose programmable processor, receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting the data and instructions to the storage system, the at least one input device, and the at least one output device.

[0125] The program code for implementing the methods of the present disclosure can be written in any combination of one or more programming languages. These program codes can be provided to a processor or controller of a general-purpose computer, a special-purpose computer, or other programmable data processing device, such that when the program codes are executed by the processor or controller, the functions / operations specified in the flowcharts and / or block diagrams are implemented. The program codes can be executed entirely on the machine, partially on the machine, as an independent software package partially on the machine and partially on a remote machine, or entirely on a remote machine or server.

[0126] In the context of this disclosure, a machine-readable medium can be a tangible medium that can contain or store a program for use by or in connection with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of a machine-readable storage medium would include an electrical connection based on one or more wires, a portable computer diskette, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0127] In order to provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor) for displaying information to the user; and a keyboard and a pointing device (e.g., a mouse or a trackball) by which the user can provide input to the computer. Other kinds of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including acoustic input, voice input, or tactile input).

[0128] The systems and techniques described herein can be implemented in a computing system that includes backend components (e.g., as a data server), or a computing system that includes middleware components (e.g., an application server), or a computing system that includes frontend components (e.g., a user computer having a graphical user interface or a web browser through which the user can interact with an implementation of the systems and techniques described herein), or a computing system that includes any combination of such backend components, middleware components, or frontend components. The components of the system can be interconnected by any form or medium of digital data communication (e.g., a communication network). Examples of communication networks include: a local area network (LAN), a wide area network (WAN), and the Internet.

[0129] A computer system may include a client and a server. The client and the server are generally far from each other and usually interact via a communication network. The relationship between the client and the server is created by computer programs running on respective computers and having a client-server relationship with each other. The server may be a cloud server, also known as a cloud computing server or a cloud host, which is a host product in the cloud computing service system to address the defects of difficult management and weak business scalability existing in traditional physical hosts and virtual private server (VPS) services.

[0130] To provide users with a better map point of interest (POI) query service, embodiments of the present disclosure pre-construct a multi-source map knowledge model based on POIs and corresponding multi-source knowledge. Since the multi-source knowledge covers geographical location information, knowledge graph information, and the input query terms, the trained multi-source map knowledge model can more comprehensively and accurately determine the query terms that match the POIs. The materials for semantic relevance modeling are exactly the POIs and the matching query terms extracted from the trained multi-source map knowledge model, thereby obtaining a semantic correspondence relationship that can accurately reflect the user's search habits at the semantic level. Finally, with the help of this semantic correspondence relationship, the target POI can be accurately determined, improving the accuracy of the query results of the POI query service.

[0131] It should be understood that various forms of the processes shown above can be used, with steps reordered, added, or deleted. For example, the steps described in the present disclosure can be executed in parallel, sequentially, or in a different order, as long as the desired results of the technical solutions disclosed in the present disclosure can be achieved, and no limitations are imposed herein.

[0132] The above specific embodiments do not constitute a limitation on the protection scope of the present disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present disclosure shall be included within the protection scope of the present disclosure.

Claims

1. A method for querying map points of interest, comprising: Receiving an input query request for points of interest; Extracting a target query term included in the query request for points of interest; Using a pre-determined semantic correspondence relationship between query terms and points of interest to determine a target point of interest corresponding to the target query term; wherein, the query terms and points of interest used to determine the semantic correspondence relationship are extracted from a preset multi-source map knowledge model, and the multi-source map knowledge model records a multi-source correspondence relationship between points of interest and multi-source knowledge, and the multi-source knowledge includes: geographical location information, knowledge graph information, and input query terms. The geographical location information, knowledge graph information, and input query terms are represented as nodes in a graph, and the vector representation of the nodes is jointly determined by their own feature information and their neighbor nodes in the graph. The knowledge graph information uses a knowledge graph that records various knowledge related to the entities of map points of interest.

2. The method according to claim 1, further comprising: Extracting real points of interest and corresponding historical query terms from the multi-source map knowledge model; Using semantic correlation modeling technology to determine the correspondence relationship at the semantic level between the real points of interest and the corresponding historical query terms, and obtaining the semantic correspondence relationship.

3. The method according to claim 1, further comprising: Obtaining each point of interest within a target map area; Obtaining the geographical location information, knowledge graph information, and input query terms corresponding to the point of interest; Taking the point of interest as the main node, and taking the geographical location information, the knowledge graph information, and the input query term as different slave nodes of the main node respectively, and making connections between the main node and each slave node; Determining the association relationship between different points of interest according to user behavior information, and making corresponding connections between the main nodes according to the association relationship to obtain a node graph; wherein, the line feature of the connection is determined based on the confidence between the two connected nodes; Based on a preset training task objective, pre-training the node graph through a graph neural network to obtain a multi-source map knowledge model that meets the requirements of the training task objective.

4. The method according to claim 3, wherein The obtaining the geographical location information, knowledge graph information, and input query terms corresponding to the point of interest includes: Obtaining a geographical location encoding corresponding to the point of interest; In a preset knowledge graph, obtaining interest point entity information corresponding to the interest point as the entity to be queried; In the operation log, obtaining the query term input by the user before the moment when the interest point is selected.

5. The method according to claim 3, wherein The training task objective includes: performing multi-classification on each main node, and the accuracy of the classification results of the multi-classification meets the preset requirements.

6. The method according to claim 3, further comprising: For a first query term already included in the multi-source map knowledge model, obtaining the node vector representation of the first query term, determining the probabilities of belonging to different categories according to the node vector representation, and taking the category with the largest corresponding probability as the actual category of the first query term; wherein, the node vector representation is jointly determined based on the information of other nodes that have established connections or have adjacent relationships with itself. For a second query term not included in the multi-source map knowledge model, determine a target first query term similar to the second query term, and determine the category of the target first query term with the highest corresponding similarity as the actual category of the second query term.

7. The method according to claim 3, further comprising: Based on the consistency among slave nodes connected to different master nodes, perform an operation of supplementing annotation content on master nodes and / or slave nodes with no annotation or less than a preset number of annotations. Based on the consistency among slave nodes connected to different master nodes, determine abnormal master nodes and / or abnormal slave nodes with annotation errors, and initiate an inquiry about incorrect annotations or perform a correction operation on the abnormal master nodes and / or the abnormal slave nodes.

8. The method according to any one of claims 3-7, wherein, The pre-training of the constructed node graph through a graph neural network includes: Obtain the trained parameters of the network structure related to the text content from a trained model for natural language processing. Use the trained parameters as the initial parameters of the network structure related to the text content in the graph neural network to obtain a graph neural network to be trained. Pre-train the constructed node graph through the graph neural network to be trained.

9. A map point of interest query device, comprising: A point of interest query request receiving unit configured to receive an input point of interest query request. A target query term extraction unit configured to extract a target query term included in the point of interest query request. A target point of interest determination unit configured to use a pre-determined semantic correspondence relationship between a query term and a point of interest to determine a target point of interest corresponding to the target query term; wherein, the query terms and points of interest used to determine the semantic correspondence relationship are extracted from a preset multi-source map knowledge model, and the multi-source map knowledge model records a multi-source correspondence relationship between points of interest and multi-source knowledge, and the multi-source knowledge includes: geographical location information, knowledge graph information, and input query terms. The geographical location information, knowledge graph information, and input query terms are represented as nodes in a graph, and the vector representation of the nodes is determined by their own feature information and their neighbor nodes in the graph. The knowledge graph information uses a knowledge graph recording various knowledge related to entities of map points of interest.

10. The device according to claim 9, further comprising: An information extraction unit configured to extract real points of interest and corresponding historical query terms from the multi-source map knowledge model. A semantic relevance modeling unit configured to use semantic relevance modeling technology to determine the correspondence relationship at the semantic level between real points of interest and corresponding historical query terms to obtain the semantic correspondence relationship.

11. The device according to claim 9, further comprising: A point of interest text information acquisition unit configured to acquire each point of interest within a target map area. A multi-source knowledge acquisition unit configured to acquire geographical location information, knowledge graph information, and input query terms corresponding to the point of interest. The master-slave node determination and connection unit is configured to use the point of interest as the master node, and use the geographical location information, the knowledge graph information, and the input query term as different slave nodes of the master node respectively, and make connections between the master node and each of the slave nodes; The master node interconnection unit is configured to determine the association relationship between different points of interest according to the user behavior information, and make corresponding connections between the master nodes according to the association relationship to obtain a node graph; wherein, the line feature of the connection is determined based on the confidence between the two connected nodes; The pre-training unit is configured to pre-train the node graph through a graph neural network based on a preset training task objective to obtain a multi-source map knowledge model that meets the requirements of the training task objective.

12. The apparatus according to claim 11, wherein, The multi-source knowledge acquisition unit is further configured to: Obtain the geographical location encoding corresponding to the point of interest; In a preset knowledge graph, obtain the point of interest entity information corresponding to the point of interest as the entity to be queried; In the operation log, obtain the query term input by the user before the moment when the point of interest is selected.

13. The apparatus according to claim 11, wherein The training task objective includes: performing multi-classification on each of the master nodes, and the accuracy of the classification results of the multi-classification meets the preset requirements.

14. The device according to claim 11, further comprising: The included query term category determination unit is configured to, for the first query term included in the multi-source map knowledge model, obtain the node vector representation of the first query term, determine the probabilities of belonging to different categories according to the node vector representation, and use the category with the highest corresponding probability as the actual category of the first query term; wherein, the node vector representation is jointly determined based on the information of other nodes that have established connections or adjacent relationships with itself; The unincluded query term category determination unit is configured to, for the second query term not included in the multi-source map knowledge model, determine the target first query term similar to the second query term, and determine the category of the target first query term with the highest corresponding similarity as the actual category of the second query term.

15. The device according to claim 11, further comprising: The node annotation content completion unit is configured to perform an annotation content completion operation on the master node and / or the slave node that is not annotated or has less than the preset number of annotations based on the consistency between the slave nodes connected to different master nodes; The abnormal annotation content processing unit is configured to determine the abnormal master node and / or the abnormal slave node with incorrect annotations based on the consistency between the slave nodes connected to different master nodes, and initiate an incorrect annotation inquiry or perform a correction operation on the abnormal master node and / or the abnormal slave node.

16. The device according to any one of claims 11-15, wherein, The pre-training unit is further configured to: Obtain the trained parameters of the network structure related to the text content from a trained model for natural language processing; Use the trained parameters as the initial parameters of the network structure related to the text content in the graph neural network to obtain a graph neural network to be trained; Pre-train the constructed node graph through the graph neural network to be trained.

17. An electronic device, comprising: At least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and when the instructions are executed by the at least one processor, the at least one processor is enabled to execute the map point of interest query method according to any one of claims 1-8.

18. A non-transitory computer-readable storage medium storing computer instructions for causing a computer to execute the map point of interest query method according to any one of claims 1-8.

19. A computer program product comprising a computer program which, when executed by a processor, implements the steps of the map point of interest query method according to any one of claims 1-8.

Images