A point of interest retrieval method, apparatus, device and storage medium

Abstract

The application discloses a point-of-interest retrieval method and device, equipment and a storage medium. The application relates to the technical field of maps, and the method comprises the following steps: acquiring user search information comprising search text information and user position information; acquiring at least one piece of point-of-interest feature information corresponding to a plurality of candidate point-of-interest information of the user search information, wherein any piece of point-of-interest feature information is obtained by performing point-of-interest feature identification on point-of-interest identification information, point-of-interest position information and place description information of the corresponding candidate point-of-interest information; inputting the search text information and the user position information into a search feature identification network to perform search feature identification, and obtaining search feature information; and displaying target point-of-interest information in the plurality of candidate point-of-interest information according to feature matching information between the search feature information and the at least one piece of point-of-interest feature information. According to the scheme of the application, the representation accuracy of search features and point-of-interest features can be improved, and therefore the accuracy of point-of-interest retrieval can be improved.

Description

Technical Field

[0001] This application relates to the field of map technology, and in particular to a method, apparatus, device and storage medium for point of interest retrieval. Background Technology

[0002] In electronic maps, important locations such as banks, tourist attractions, companies, hospitals, government agencies, restaurants, and shopping malls are often marked as points of interest (POIs) that users may be interested in. When users search for POIs, they input search text and obtain POI information related to that search text. Traditional POI retrieval methods typically calculate the text similarity between the name identifiers of the search text and candidate POIs during POI matching, and then determine the target POI information corresponding to the search text based on the text similarity.

[0003] However, when searching for complex text information, relying solely on basic search text features and point-of-interest (POI) name identifiers for POI matching fails to capture the deeper semantic relationships between the search text and POI information, resulting in low accuracy of POI retrieval results. Therefore, a more accurate technical solution is needed. Summary of the Invention

[0004] This application provides a method, apparatus, device, and storage medium for point of interest (POI) retrieval. It utilizes search feature information representing search text features and user location features, as well as POI feature information representing POI identifier features, location features, and location description features, to perform feature matching, thereby improving the accuracy of POI retrieval. The technical solution of this application is as follows:

[0005] On the one hand, an interest point retrieval method is provided, the method comprising:

[0006] Obtain user search information, which includes search text information and user location information;

[0007] At least one point of interest feature information corresponding to each of the multiple candidate point of interest information of the user search information is obtained. Any point of interest feature information is obtained after identifying the point of interest feature based on the point of interest identifier information, point of interest location information and location description information of the corresponding candidate point of interest information.

[0008] The search text information and the user location information are input into a search feature recognition network to perform search feature recognition and obtain search feature information.

[0009] Based on the feature matching information between the search feature information and the at least one point of interest feature information, the target point of interest information among the multiple candidate point of interest information is displayed.

[0010] On the other hand, an interest point retrieval device is provided, the method comprising:

[0011] The user search information acquisition module is used to acquire user search information, which includes search text information and user location information.

[0012] The point of interest feature information acquisition module is used to acquire at least one point of interest feature information corresponding to each of the multiple candidate point of interest information in the user search information. Any point of interest feature information is obtained after identifying the point of interest feature based on the point of interest identifier information, point of interest location information and location description information of the corresponding candidate point of interest information.

[0013] The search feature recognition module is used to input the search text information and the user location information into the search feature recognition network to perform search feature recognition and obtain search feature information.

[0014] The target interest point information display module is used to display the target interest point information among the multiple candidate interest point information based on the feature matching information between the search feature information and the at least one interest point feature information.

[0015] On the other hand, an interest point retrieval device is provided, the device including a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the interest point retrieval method as described in the first aspect.

[0016] On the other hand, a computer-readable storage medium is provided, wherein at least one instruction or at least one program is stored therein, the at least one instruction or the at least one program being loaded and executed by a processor to implement the point of interest retrieval method as described in the first aspect.

[0017] On the other hand, a computer program product or computer program is provided, comprising computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the point-of-interest retrieval method as described in the first aspect.

[0018] The point-of-interest (POI) retrieval method, apparatus, device, and storage medium provided in this application have the following technical advantages:

[0019] In the application scenario of point-of-interest (POI) retrieval, this application obtains user search information, including search text information and user location information. Then, it acquires at least one POI feature corresponding to each of the multiple candidate POI information from the user search information. Each POI feature is obtained by identifying the POI identifier, POI location information, and location description information of the corresponding candidate POI information. Utilizing multi-dimensional POI features improves the accuracy of POI feature representation. Next, the search text information and user location information are input into a search feature recognition network for search feature recognition, obtaining search feature information, which further improves the accuracy of search feature representation. Finally, based on the feature matching information between the search feature information and at least one POI feature, the target POI information from the multiple candidate POI information is displayed. Through the deep feature matching relationship between the search text information and the POI information, the accuracy of POI matching is effectively improved. Attached Figure Description

[0020] To more clearly illustrate the technical solutions and advantages 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.

[0021] Figure 1 This is a schematic diagram of an application environment provided in an embodiment of this application;

[0022] Figure 2 This is a flowchart illustrating a point-of-interest (POI) retrieval method provided in an embodiment of this application;

[0023] Figure 3 This is a flowchart illustrating a method for generating at least one initial interest point feature information corresponding to each of multiple initial interest point information provided in an embodiment of this application.

[0024] Figure 4 This application provides a schematic diagram of a process for inputting the target initial identification information, initial location information, initial type information and initial location description information from each initial point of interest information into the point of interest feature recognition network for point of interest feature recognition, and obtaining the target feature information corresponding to the target initial identification information;

[0025] Figure 5 This is a flowchart illustrating a process provided in this application embodiment of inputting a target interest point feature sequence into a second fusion network for fusion processing to obtain target feature information corresponding to the initial target identification information;

[0026] Figure 6This is a schematic diagram of a process provided in this application embodiment to input search text information and user location information into a search feature recognition network for search feature recognition and obtain search feature information;

[0027] Figure 7 This is a schematic diagram of a process provided in this application embodiment to input a target search feature sequence into a first fusion network for fusion processing to obtain search feature information;

[0028] Figure 8 This is a schematic diagram of a process for displaying target interest point information among multiple candidate interest point information based on feature matching information between search feature information and at least one interest point feature information, provided by an embodiment of this application.

[0029] Figure 9 This is a schematic flowchart of a network training method provided in an embodiment of this application;

[0030] Figure 10 This is a schematic diagram of a process for obtaining sample user search information and sample point of interest information corresponding to the sample user search information, provided in an embodiment of this application.

[0031] Figure 11 This is a schematic diagram of an interest point retrieval network provided in an embodiment of this application;

[0032] Figure 12 This is a block diagram of an interest point retrieval device provided in an embodiment of this application;

[0033] Figure 13 This is a schematic diagram of the structure of a point-of-interest retrieval device provided in an embodiment of this application. Detailed Implementation

[0034] 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 of ordinary skill in the art without creative effort are within the scope of protection of this application.

[0035] It should be noted that the terms "comprising" and "having" and any variations thereof in the specification, claims and accompanying drawings of this application are intended to cover non-exclusive inclusion. For example, a process, method, system, product or server that includes a series of steps or units is not necessarily limited to those steps or units that are explicitly listed, but may include other steps or units that are not explicitly listed or that are inherent to such processes, methods, products or devices.

[0036] It is understood that in the specific embodiments of this application, data such as user information are involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of related data must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0037] Please see Figure 1 , Figure 1 This is a schematic diagram of an application environment provided in an embodiment of this application. This application environment may include a client 10 and a server 20, which can be directly or indirectly connected via wired or wireless communication. A user can send a point-of-interest (POI) retrieval request to the server 20 through the client 10. This POI retrieval request carries user search information, which may include search text information and user location information. In response to the POI retrieval request, the server 20 obtains at least one POI feature information corresponding to each of the multiple candidate POI information in the user's search information. Each POI feature information represents the identification feature, type feature, location feature, and location description feature of the corresponding candidate POI information. Then, the server inputs the search text information and user location information into a search feature recognition network for search feature recognition to obtain search feature information. Based on the feature matching information between the search feature information and at least one POI feature information, the server determines the target POI information from the multiple candidate POI information and returns the target POI information to the client 10 so that the client 10 can display the target POI information. It should be noted that... Figure 1 This is just one example.

[0038] The client can be a physical device such as a smartphone, computer (e.g., desktop computer, tablet computer, laptop computer), digital assistant, smart voice interaction device (e.g., smart speaker), smart wearable device, in-vehicle terminal, aircraft, etc., or it can be software running on the physical device, such as a computer program. The corresponding operating system of the client can be Android, iOS (a mobile operating system developed by Apple), Linux (an operating system), Microsoft Windows, etc.

[0039] The server side can be a standalone physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server providing basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, CDN (Content Delivery Network), and big data and artificial intelligence platforms. The server may include network communication units, processors, and memory, etc. The server side can provide backend services to the corresponding clients.

[0040] The aforementioned client 10 and server 20 can be used to build a system for point-of-interest (POI) retrieval, which can be a distributed system. Taking a blockchain system as an example, the distributed system consists of multiple nodes (any form of computing device connected to the network, such as servers or user terminals) and clients. These nodes form a peer-to-peer (P2P) network. The P2P protocol is an application layer protocol running on top of the Transmission Control Protocol (TCP). In a distributed system, any machine, such as a server or terminal, can join and become a node. A node includes a hardware layer, a middleware layer, an operating system layer, and an application layer.

[0041] The functions of each node in the aforementioned blockchain system include:

[0042] 1) Routing: A basic function of nodes used to support communication between nodes.

[0043] In addition to routing capabilities, nodes can also have the following functions:

[0044] 2) Applications are deployed in the blockchain to implement specific business needs. They record data related to the implementation of functions to form record data, carry digital signatures in the record data to indicate the source of the task data, and send the record data to other nodes in the blockchain system. When other nodes successfully verify the source and integrity of the record data, they add the record data to a temporary block.

[0045] 3) A blockchain consists of a series of blocks that are sequentially generated. Once a new block is added to the blockchain, it will not be removed. The blocks contain the data submitted by the nodes in the blockchain system.

[0046] The following describes a specific embodiment of the point of interest retrieval method provided in this application. Figure 2This is a flowchart illustrating a point-of-interest (POI) retrieval method provided in an embodiment of this application. This application provides the operational steps of the method described in the embodiment or flowchart, but based on conventional or non-inventive methods, more or fewer operational steps may be included. The order of steps listed in the embodiment is merely one possible execution order among many and does not represent the only possible execution order. In actual systems or products, the method can be executed sequentially or in parallel (e.g., in a parallel processor or multi-threaded processing environment) as shown in the embodiment or drawings. Specifically, as... Figure 2 As shown, the method may include:

[0047] S201, Obtain user search information, which includes search text information and user location information.

[0048] In the embodiments of this specification, the aforementioned user search information can be search information corresponding to the user's point-of-interest retrieval request. Specifically, user location information may include: the user's first administrative division information, and the second administrative division information may include, but is not limited to: province identification information, city identification information, district identification information, etc.

[0049] In one specific embodiment, user location information can be determined based on the administrative division search range set by the user or the real-time location information of the user terminal.

[0050] S202, obtain at least one point of interest feature information corresponding to each of the multiple candidate point of interest information of the user's search information, wherein any point of interest feature information is obtained after identifying the point of interest feature based on the point of interest identifier information, point of interest location information and location description information of the corresponding candidate point of interest information.

[0051] In the embodiments of this specification, the candidate point of interest information can be point of interest information obtained by recalling point of interest based on user search information. The point of interest information can be the association information of important points of interest that are preset and marked in the electronic map. In practical applications, locations such as banks, scenic spots, companies, hospitals, government agencies, restaurants, and shopping malls can all be marked as points of interest in the electronic map.

[0052] Specifically, point-of-interest (POI) information may include, but is not limited to: at least one POI identifier, POI location information, POI type information, and location description information.

[0053] In one specific embodiment, at least one point of interest (POI) identification information may include, but is not limited to: the main name of the POI, its former names, and its aliases. Taking the POI "China Media Group (Guanghua Road Office Area)" as an example, the main name of the POI can be China Media Group, and the aliases can include "China Central Television Headquarters Building", "CCTV", "CCTV Building", etc.

[0054] In one specific embodiment, the location information of a point of interest (POI) may include, but is not limited to, the second administrative division information to which the POI belongs, and the spatial location information of the POI. Specifically, the second administrative division information may include, but is not limited to, province identification information, city identification information, and urban area identification information; the spatial location information may include, but is not limited to, the latitude and longitude of the POI, and the regional segmentation identifier of the POI, wherein the regional segmentation identifier of the POI can be obtained by segmenting and converting the latitude and longitude of the POI.

[0055] In practical applications, the latitude and longitude within an administrative division can be pre-divided into regional segments based on administrative divisions, resulting in multiple regional segment identifiers and the latitude and longitude range corresponding to each regional segment identifier. This generates a location mapping relationship between regional segment identifiers and latitude and longitude. Correspondingly, after obtaining the latitude and longitude of a certain point of interest, the latitude and longitude of the point of interest can be segmented and converted based on the location mapping relationship to obtain the regional segment identifier of the point of interest.

[0056] In one specific embodiment, the point of interest type information may include, but is not limited to: schools, government agencies, transportation facilities, daily services, food, shopping, hotels, etc. Specifically, each point of interest type can be further divided into multiple levels based on actual application needs; taking a two-level classification as an example, if the primary type of point of interest is schools, the secondary types may include, but are not limited to: primary school, middle school, university, adult education, etc.

[0057] In one specific embodiment, the location description information may include, but is not limited to: the popularity of point of interest searches, the rating information of point of interest, etc. Specifically, the popularity of point of interest searches can be obtained by statistical analysis based on recent user point of interest search data, and the rating information of point of interest can be set in combination with the type of point of interest. For example, when the type of point of interest is hotel, the rating information of point of interest may include, but is not limited to: star rating, user rating, hotel ranking, etc.

[0058] In one specific embodiment, before obtaining at least one interest point feature information corresponding to each of the multiple candidate interest point information of the user search information, the method may further include:

[0059] 1) Obtain information on multiple initial points of interest;

[0060] 2) Recall points of interest based on user search information, and determine multiple candidate points of interest from multiple initial points of interest information.

[0061] Specifically, multiple initial points of interest (POIs) can be the full set of POIs stored in a pre-configured map platform.

[0062] In an optional embodiment, the above-mentioned point of interest retrieval based on user search information, which determines multiple candidate point of interest information from multiple initial point of interest information, may include: extracting index words from the search text information in the user search information to obtain at least one target index word; performing point of interest indexing on the at least one target index word to obtain initial point of interest information corresponding to each of the at least one target index word; and using the initial point of interest information corresponding to each of the at least one target index word as multiple candidate point of interest information.

[0063] Specifically, the target index terms mentioned above may include prefixes or key words in the search text information.

[0064] In practical applications, multiple index terms can be pre-determined based on the names of points of interest in multiple initial points of interest information. Then, an interest point index table can be constructed using the index terms as keys and the initial points of interest information containing the index terms in the point of interest names as values. Accordingly, the above-mentioned indexing of at least one target index term to obtain the initial points of interest information corresponding to each of the at least one target index term can include: indexing at least one target index term based on the interest point index table to obtain the initial points of interest information corresponding to each of the at least one target index term.

[0065] In another optional embodiment, multiple popular interest points can be pre-determined from multiple initial interest point information based on recent user interest point retrieval data and interest point access data, and a search mapping relationship between the interest point names of the popular interest point information and the search text can be constructed. Correspondingly, the above-mentioned interest point retrieval based on user search information, which determines multiple candidate interest point information from multiple initial interest point information, may include: determining multiple target popular interest point information corresponding to the user search information according to the search mapping relationship, and using the multiple target popular interest point information as multiple candidate interest point information.

[0066] As can be seen from the above embodiments, determining the candidate points of interest information corresponding to the user's search information through index recall or hot word recall can narrow the scope of point of interest retrieval, thereby improving the accuracy and efficiency of point of interest retrieval.

[0067] In the embodiments described in the specification, the point of interest (POI) information may include: POI identification information, POI location information, and location description information. The POI feature information can be obtained by performing POI feature recognition based on the POI identification information, POI location information, and location description information. Generally, the POI feature information can be represented as an POI feature vector.

[0068] In a specific embodiment, the point of interest information may further include: point of interest type information. Correspondingly, the point of interest feature information may be obtained by performing point of interest feature recognition on the point of interest identification information, point of interest location information, point of interest type and location description information. The point of interest feature information may characterize the identification features, type features, location features and location description features of the corresponding point of interest.

[0069] In practical applications, considering the service pressure brought about by generating a large number of interest point feature information online in real time during interest point retrieval, at least one compressed feature information corresponding to each of the multiple initial interest point information can be generated in advance and stored in local preset memory.

[0070] In a specific embodiment, at least one interest point feature information corresponding to each of the multiple candidate interest point information for obtaining user search information may include:

[0071] 1) Obtain at least one compressed feature information corresponding to each of the multiple candidate interest point information from the local preset memory area;

[0072] 2) Decompress at least one compressed feature information to obtain at least one interest point feature information.

[0073] As can be seen from the above embodiments, by pre-storing the compressed feature information of points of interest and then decompressing the compressed feature information in real time during the point of interest retrieval process to obtain the feature information of points of interest, it is possible to reduce the storage pressure of the feature information of points of interest and improve the efficiency of obtaining the feature information of points of interest, thereby improving the efficiency of point of interest retrieval.

[0074] In the embodiments described in this specification, such as Figure 3 As shown, the method for generating at least one initial interest point feature information corresponding to each of the above multiple initial interest point information can include:

[0075] S301, acquire multiple initial point of interest information, each initial point of interest information including: at least one initial identification information, initial location information and initial location description information.

[0076] Specifically, at least one initial identification information can be at least one point of interest identification information for the initial point of interest, the initial location information can be the point of interest location information for the initial point of interest, the initial type information can be the point of interest type information for the initial point of interest, and the initial location description information can be the location description information for the initial point of interest.

[0077] S302, based on the target initial identification information, initial location information and initial location description information in each initial point of interest information, perform point of interest feature recognition to obtain target feature information corresponding to the target initial identification information, where the target initial identification information is any initial identification information in each initial point of interest information.

[0078] S303, take the target feature information corresponding to at least one initial identification information in each initial interest point information as at least one initial interest point feature information corresponding to each initial interest point information.

[0079] Taking the initial point of interest information "China Media Group (Guanghua Road Office Area)" as an example, the initial point of interest information includes four types of initial identifier information. The four types of initial identifier information are: main name: China Media Group, alias 1: CCTV Headquarters Building, alias 2: CCTV, and alias 3: CCTV Building. Specifically, the main name, initial location information, initial type information, and initial location description information are input into the point of interest feature recognition network for point of interest feature recognition to obtain target feature information 1; alias 1, initial location information, initial type information, and initial location description information are input into the point of interest feature recognition network for point of interest feature recognition to obtain target feature information 2; and so on, target feature information 3 and target feature information 4 are obtained based on alias 2 and alias 3 respectively; target feature information 1-4 are used as the four initial point of interest feature information of the initial point of interest information "China Media Group (Guanghua Road Office Area)".

[0080] S304, compress at least one initial interest point feature information corresponding to each of the multiple initial interest point information to obtain at least one compressed feature information corresponding to each of the multiple initial interest point information.

[0081] In practical applications, considering that directly storing at least one initial interest point feature information corresponding to each of multiple initial interest point information would occupy too much memory space, the above-mentioned at least one initial interest point feature information is compressed to obtain at least one compressed feature information corresponding to each of the multiple initial interest point information.

[0082] In a specific embodiment, at least one compressed feature information corresponding to each of the multiple initial interest point information can be stored as an interest point compression dictionary. Specifically, the interest point compression dictionary stores the encoded identifier of the initial interest point information as the key and the at least one compressed feature information corresponding to the initial interest point information as the value.

[0083] As can be seen from the above embodiments, by performing interest point feature recognition on the identification information, location information, type information and location description information of interest points, interest point feature information can be obtained, which can improve the accuracy and semantic richness of interest point feature representation. Furthermore, when there are multiple identification names for interest points, target feature information corresponding to each identification name can be generated, thereby improving the richness of interest point feature information.

[0084] In a specific embodiment, before compressing at least one initial interest point feature information corresponding to each of the multiple initial interest point information to obtain at least one compressed feature information corresponding to each of the multiple initial interest point information, compression training can be performed on at least one initial interest point feature information corresponding to each of the multiple initial interest point information to determine the target compression parameters, and a preset compression dictionary can be generated based on the target compression parameters.

[0085] Specifically, the target compression parameters may include, but are not limited to: compression frame parameters, compression level, string length of initial interest point feature information, string length of compressed feature information, etc.

[0086] Specifically, the preset compression dictionary can be used as a compression template containing the target compression parameters. During the compression process of other data, the preset compression dictionary is used as a template to compress other data according to the target compression parameters.

[0087] In one specific embodiment, prior to obtaining the user search information, the method further includes:

[0088] Obtain a preset compressed dictionary, which is obtained by compressing and training at least one initial interest point feature information corresponding to each of the multiple initial interest point information.

[0089] Accordingly, the above-mentioned decompression process of at least one compressed feature information to obtain at least one interest point feature information may include:

[0090] Based on a preset compressed dictionary, at least one compressed feature information is decompressed to obtain at least one point of interest feature information.

[0091] As can be seen from the above embodiments, by generating a preset compressed dictionary through compressed training, and then using the preset compressed dictionary to decompress the compressed feature information during the point of interest retrieval process, the decompression efficiency can be effectively improved, thereby improving the efficiency of point of interest retrieval.

[0092] In a specific embodiment, any initial point of interest information further includes: initial type information. The above-mentioned point of interest feature recognition based on the target initial identification information, initial location information, and initial location description information in each initial point of interest information, to obtain target feature information corresponding to the target initial identification information, may include:

[0093] The target initial identification information, initial location information, initial type information and initial location description information in each initial point of interest information are input into the point of interest feature recognition network to perform point of interest feature recognition, and the target feature information corresponding to the target initial identification information is obtained.

[0094] Specifically, the aforementioned interest point feature recognition network can be obtained by training a preset interest point feature recognition network with interest point feature recognition.

[0095] In one specific embodiment, the above-mentioned interest point feature recognition network may include: a second feature extraction network, a second concatenation network, a second semantic recognition network, and a second fusion network, such as... Figure 4 As shown, the target initial identification information, initial location information, initial type information, and initial location description information from each initial point of interest (POI) are input into the POI feature recognition network for POI feature recognition. The target feature information corresponding to the target initial identification information can include:

[0096] S401, the target initial identification information, initial location information, initial type information and initial location description information in each initial point of interest information are respectively input into the second feature extraction network for feature extraction to obtain identification feature information, location feature information, type feature information and location description feature information.

[0097] Specifically, the identification feature information, location feature information, type feature information, and location description feature information can be represented as feature vectors.

[0098] In a specific embodiment, the second feature extraction network may include a word vector model and a one-hot encoding model. Accordingly, the above-mentioned inputting the target initial identification information, initial location information, initial type information, and initial location description information from each initial point of interest information into the second feature extraction network for feature extraction to obtain identification feature information, location feature information, type feature information, and location description feature information may include: inputting the target initial identification information into the word vector model for word feature extraction to obtain identification feature information; and inputting the initial location information, initial type information, and initial location description information into the one-hot encoding model for feature extraction to obtain location feature information, type feature information, and location description feature information.

[0099] S402, input the identification feature information, location feature information, type feature information and location description feature information into the second stitching network for stitching processing to obtain the initial interest point feature sequence.

[0100] S403, input the initial interest point feature sequence into the second semantic recognition network to perform interest point semantic recognition, and obtain the target interest point feature sequence.

[0101] Specifically, the target interest point feature sequence can characterize the contextual semantic features of the initial interest point information. The target interest point feature sequence can include multiple interest point word segmentation feature information. Specifically, the multiple interest point word segmentation feature information can be target identifier feature information, target location feature information, target type feature information, and target location description feature information. Among them, the target identifier feature information can be obtained by semantic recognition of the identifier feature information in combination with the contextual semantics of the initial interest point information; the target location feature information can be obtained by semantic recognition of the location feature information in combination with the contextual semantics of the initial interest point information; the target type feature information can be obtained by semantic recognition of the type feature information in combination with the contextual semantics of the initial interest point information; and the target location description feature information can be obtained by semantic recognition of the location description feature information in combination with the contextual semantics of the initial interest point information.

[0102] In practical applications, the second semantic recognition network may include, but is not limited to, BiGRU (Bidirectional Gated Recurrent Neural Network) and BERT (Transformer-based Bidirectional Encoding Representation Model).

[0103] S404, the target interest point feature sequence is input into the second fusion network for fusion processing to obtain the target feature information corresponding to the initial target identification information.

[0104] Optionally, taking the initial point of interest information i as an example, and the identification feature information ai, the location feature information bi, the type feature information ci, and the location description feature information di as examples, the initial point of interest feature sequence can be Xi = [ai, bi, ci, di], the target point of interest feature sequence can be Yi = [Ai, Bi, Ci, Di], and the target feature information can be Si.

[0105] In a specific embodiment, when the initial location information includes second administrative division information and spatial location information, the aforementioned location feature information may include administrative division feature information and spatial location feature information. Correspondingly, inputting the identifier feature information, location feature information, type feature information, and location description feature information into the second stitching network for stitching processing to obtain the initial point of interest feature sequence may include: inputting the identifier feature information, administrative division feature information, spatial location feature information, type feature information, and location description feature information into the second stitching network for stitching processing to obtain the initial point of interest feature sequence; then inputting the initial point of interest feature sequence into the second semantic recognition network for point of interest semantic recognition to obtain the target point of interest feature sequence; and then inputting the target point of interest feature sequence into the second fusion network for fusion processing to obtain the target feature information corresponding to the target initial identifier information. This target feature information can characterize the identifier features, administrative division features, spatial location features, type features, and location description features of the initial point of interest information.

[0106] As can be seen from the above embodiments, by using the second feature extraction network to extract features from the initial target identification information, initial location information, initial type information, and initial location description information, the following steps are taken: First, the initial point of interest feature sequence obtained by concatenating the identification feature information, location feature information, type feature information, and location description feature information is input into the second semantic recognition network for point of interest semantic recognition to obtain the target point of interest feature sequence. Finally, the target point of interest feature sequence is input into the second fusion network for fusion processing to obtain the target feature information corresponding to the initial target identification information. This can achieve the fusion of identification features, location features, type features, and location description features in the point of interest information, thereby improving the accuracy of the representation of point of interest features.

[0107] In one specific embodiment, the second fusion network described above may include a second feature mapping layer, a second normalization layer, and a second weighting layer, such as... Figure 5 As shown, the target feature information obtained by inputting the target interest point feature sequence into the second fusion network for fusion processing can include:

[0108] S501, input the word segmentation feature information of multiple interest points into the second feature mapping layer for feature mapping processing to obtain multiple interest point mapping feature information.

[0109] Specifically, when the representation of multiple interest point segmentation feature information is multiple interest point segmentation feature vectors, the representation of multiple interest point mapping feature information is multiple interest point mapping feature vectors. The second feature mapping layer can include a mapping matrix, an offset vector, and an activation function. Through the second feature mapping layer, multiple interest point segmentation feature vectors can be linearly transformed to obtain interest point mapping feature vectors of the same dimension.

[0110] S502, input the feature information of multiple interest points into the second normalization layer for normalization processing to obtain the interest point weights of the word segmentation feature information of multiple interest points.

[0111] In one specific embodiment, the second normalization layer may include a normalization function. Multiple interest point mapping feature information is input into the normalization function for normalization calculation to obtain the interest point weights of the multiple interest point word segmentation feature information. Optionally, the normalization function may be a softmax function (normalization exponential function).

[0112] S503, the interest point weights and multiple interest point word segmentation feature information are input into the second weighting layer for weighted processing to obtain the target feature information corresponding to the initial target identifier information.

[0113] In one specific embodiment, the target feature information is obtained by adding the product of each interest point segmentation feature information and the corresponding interest point weight in the multiple interest point segmentation feature information.

[0114] As can be seen from the above embodiments, the fusion network, which includes a feature mapping layer, a normalization layer, and a weighting layer, improves the accuracy of interest point feature representation by capturing the correlation between word segmentation feature information of multiple interest points.

[0115] S203, input the search text information and user location information into the search feature recognition network to perform search feature recognition and obtain search feature information.

[0116] In the embodiments of this specification, search feature information can characterize search text features and user location features. Specifically, the search feature information can be represented as a feature vector.

[0117] Specifically, the aforementioned search feature recognition network can be obtained by training a preset search feature recognition network with search feature recognition.

[0118] In one specific embodiment, the search feature recognition network described above may include: a first feature extraction network, a first concatenation network, a first semantic recognition network, and a first fusion network, such as... Figure 6 As shown, the search text information and user location information are input into the search feature recognition network to perform search feature recognition, and the resulting search feature information may include:

[0119] S601, input the search text information into the first feature extraction network to extract word segmentation features and obtain the text segmentation feature sequence.

[0120] Specifically, the text segmentation feature sequence can represent the word features of multiple text segments of the search text information. The text segmentation feature sequence can be represented as a feature matrix, which can include the feature vectors of multiple text segments.

[0121] In one specific embodiment, the first feature extraction network may include a word vector model, wherein multiple word segments of the search text information are respectively input into the word vector model for word segmentation feature extraction to obtain a text word segmentation feature sequence.

[0122] S602, the user location information is input into the first feature extraction network to extract location features and obtain user location feature information.

[0123] Specifically, user location feature information can be represented as a feature vector.

[0124] In one specific embodiment, the first feature extraction network may include a one-hot coding model, wherein the user location information is input into the one-hot coding model for location feature extraction to obtain the user location feature information.

[0125] S603, input the text segmentation feature sequence and user location feature information into the first concatenation network for concatenation processing to obtain the initial search feature sequence.

[0126] S604, input the initial search feature sequence into the first semantic recognition network for search semantic recognition to obtain the target search feature sequence.

[0127] Specifically, the target search feature sequence can characterize the contextual semantic features of the user's search information. The target search feature sequence can include multiple search word segmentation feature information, specifically, the multiple search word segmentation feature information can include: target text word segmentation feature sequence and target user location feature information. The target text word segmentation feature sequence can be obtained by semantic recognition of the text word segmentation feature sequence in combination with the contextual semantics of the user's search information, and the target user location feature information can be obtained by semantic recognition of the user location feature information in combination with the contextual semantics of the user's search information.

[0128] In practical applications, the first semantic recognition network may include, but is not limited to, BiGRU (Bidirectional Gated Recurrent Neural Network) and BERT (Transformer-based Bidirectional Encoding Representation Model).

[0129] S605, the target search feature sequence is input into the first fusion network for fusion processing to obtain search feature information.

[0130] As can be seen from the above embodiments, by inputting the search text information and user location information into the first feature extraction network for feature extraction, the text segmentation feature sequence and user location feature information are obtained. Then, the initial search feature sequence obtained by concatenating the text segmentation feature sequence and user location feature information is input into the first semantic recognition network for search semantic recognition to obtain the target search feature sequence. Finally, the target search feature sequence is input into the first fusion network for fusion processing to obtain search feature information. This can realize the fusion of search text features and user location features, thereby improving the accuracy of the representation of search features.

[0131] In one specific embodiment, the target search feature sequence may include multiple search word segmentation feature information, and the aforementioned first fusion network may include: a first feature mapping layer, a first normalization layer, and a first weighting layer, such as... Figure 7 As shown, the above-mentioned input of the target search feature sequence into the first fusion network for fusion processing, to obtain search feature information, may include:

[0132] S701, input multiple search word segmentation feature information into the first feature mapping layer for feature mapping processing to obtain multiple search mapping feature information.

[0133] S702, input multiple search mapping feature information into the first normalization layer for normalization processing to obtain the search weights of multiple search word segmentation feature information.

[0134] S703 inputs the search weights and multiple search word segmentation feature information into the first weighting layer for weighted processing to obtain search feature information.

[0135] Specifically, the process of inputting the target search feature sequence into the first fusion network for fusion processing to obtain search feature information is similar to the process of "inputting the target interest point feature sequence into the second fusion network for fusion processing to obtain target feature information corresponding to the initial target identification information" in step S404 above. For detailed information, please refer to the detailed information of step S404 above, which will not be repeated here.

[0136] As can be seen from the above embodiments, the fusion network, which includes a feature mapping layer, a normalization layer, and a weighting layer, improves the accuracy of representing user search features by capturing the correlation between multiple search word segmentation feature information.

[0137] S204, based on the feature matching information between the search feature information and at least one point of interest feature information, display the target point of interest information among multiple candidate point of interest information.

[0138] In a specific embodiment, such as Figure 8As shown, displaying the target interest point information from multiple candidate interest point information based on feature matching information between search feature information and at least one interest point feature information can include:

[0139] S801, perform feature matching analysis on the search feature information and at least one point of interest feature information to obtain feature matching information.

[0140] In the embodiments of this specification, feature matching information may include numerical values ​​that quantify the similarity between search feature information and point of interest feature information according to certain rules.

[0141] In a specific embodiment, when the search feature information is a search feature vector and the interest point feature information is an interest point feature vector, the feature matching information may include: vector similarity, which is a quantified value of the similarity between the search feature vector and the interest point feature vector. Specifically, the matching degree between feature vectors is determined by calculating the distance between them. Specifically, the distance here may include, but is not limited to, Euclidean distance, cosine distance, Manhattan distance, etc. Accordingly, the feature matching information obtained by performing feature matching analysis on the search feature information and at least one interest point feature may include:

[0142] The similarity between the search feature vector and at least one interest point feature vector is calculated to obtain the vector similarity between the search feature vector and each interest point feature vector.

[0143] S802, based on the feature matching information corresponding to at least one feature information of multiple candidate interest points, determine the target matching information of multiple candidate interest points.

[0144] Specifically, target matching information can characterize the degree of matching between any candidate point of interest information and the user's search information.

[0145] In an optional embodiment, if any candidate point of interest information has only one point of interest feature information, the feature matching information corresponding to that one point of interest feature information is used as the target matching information of the candidate point of interest information.

[0146] In another optional embodiment, when any candidate point of interest information has multiple points of interest feature information, the feature matching information with the highest matching degree among the feature matching information corresponding to the multiple points of interest feature information is taken as the target matching information of the candidate point of interest information. Specifically, when the point of interest feature information is a point of interest feature vector and the feature matching information is vector similarity, the maximum value among the vector similarities corresponding to the multiple point of interest feature vectors is taken as the target matching information of the candidate point of interest information.

[0147] S803: Based on the target matching information, sort the information of multiple candidate points of interest to obtain the matching order information.

[0148] In an optional embodiment, the above-mentioned multiple candidate interest point information are sorted according to the target matching information in descending order to obtain matching order information.

[0149] S804 determines the target interest point information from multiple candidate interest point information based on the matching order information.

[0150] Specifically, the target interest point information can be a target number of optimized interest point information.

[0151] In practical applications, the number of targets can be preset based on the actual needs of point of interest retrieval.

[0152] In an optional embodiment, the target number of candidate interest points that appear first in the multiple candidate interest point information can be selected as the target interest point information based on the matching order information.

[0153] S805 displays target point of interest information based on matching order information.

[0154] As can be seen from the above embodiments, determining the matching order information through feature matching information and then determining the target point of interest information based on the matching order information as the point of interest retrieval result for display can improve the accuracy of the point of interest retrieval result.

[0155] In the embodiments of this specification, the preset interest point feature recognition network and the preset search feature recognition network can be trained using sample user search information and sample interest point information corresponding to the sample user search information to obtain the above-mentioned interest point feature recognition network and the above-mentioned search feature recognition network.

[0156] In a specific embodiment, such as Figure 9 As shown, Figure 9 This is a flowchart illustrating a network training method provided in an embodiment of this application. Specifically, it may include:

[0157] S901, obtain sample user search information and sample point of interest information corresponding to the sample user search information. The sample user search information includes: sample search text information and sample user location information. The sample point of interest information includes: at least one sample identification information, sample location information, sample type information and sample location description information.

[0158] In a specific embodiment, the above-mentioned sample interest point information may include: positive sample interest point information, correspondingly, such as Figure 10 As shown, the above-mentioned acquisition of sample user search information and corresponding sample user search information includes:

[0159] S1001, Obtain the initial sample search information and the initial sample interest point information corresponding to the initial sample search information.

[0160] Specifically, historical user search information and frequently selected points of interest information corresponding to historical user search information can be extracted from a large amount of historical user search data. Historical user search information is used as initial sample search information, and frequently selected points of interest information is used as initial sample points of interest information.

[0161] S1002, perform text matching processing on the sample search text information in the initial sample search information and the sample identification information in the initial sample interest point information to obtain the text matching result.

[0162] S1003, the initial sample interest point information corresponding to the sample identifier information that meets the preset matching conditions in the text matching result is taken as the positive sample interest point information.

[0163] Specifically, text matching results can characterize the degree of text matching between sample search text information and sample identifier information. In practical applications, preset matching conditions can be pre-set according to actual text matching needs.

[0164] In an optional embodiment, the text matching result may include text repetition. Specifically, text repetition can be the ratio of the number of repeated characters between the sample search text information and the sample identifier information to the total number of characters in the sample search text information. The preset matching condition may include: the text repetition is greater than a preset repetition threshold, for example, the preset repetition threshold may be 25%. Correspondingly, the above-mentioned use of the initial sample interest point information corresponding to the sample identifier information whose text matching result meets the preset matching condition as positive sample interest point information may include: using the initial sample interest point information corresponding to the sample identifier information whose text repetition is greater than the preset repetition threshold as positive sample interest point information.

[0165] S1004, extract core words from the sample search text information in the initial sample search information to obtain the sample user search information.

[0166] Specifically, administrative divisions and prefixes / suffixes can be removed from the sample search text information to extract core words. For example, filtering out pure administrative place names, chain store words, type words, etc. Taking the sample search text information "Beijing Haidian District Experimental Primary School" as an example, the core word is "primary school".

[0167] As can be seen from the above examples, by performing weakly related text filtering and core word extraction on the samples, the sample data can be enhanced, thereby improving the effect of network training.

[0168] S902, input the sample search text information and sample user location information into the preset search feature recognition network to perform search feature recognition and obtain sample search feature information.

[0169] Specifically, the process of inputting the sample search text information and sample user location information into a preset search feature recognition network to obtain sample search feature information is similar to the process of "inputting the search text information and user location information into the search feature recognition network to obtain search feature information" in step S203 above. For detailed information, please refer to the detailed information in step S203 above, which will not be repeated here.

[0170] S903, at least one sample identification information, sample location information, sample type information and sample location description information are input into a preset interest point feature recognition network to perform interest point feature recognition, and at least one sample interest point feature information corresponding to the sample interest point information is obtained.

[0171] In a specific embodiment, the above-mentioned inputting at least one sample identification information, sample location information, sample type information, and sample location description information into a preset interest point feature recognition network for interest point feature recognition, to obtain at least one sample interest point feature information corresponding to the sample interest point information, may include:

[0172] 1) Input the target sample identification information, sample location information, sample type information and sample location description information into the preset interest point feature recognition network to perform interest point feature recognition, and obtain the target sample feature information corresponding to the target sample identification information. The target sample identification information is any sample identification information among at least one sample identification information.

[0173] Specifically, the process of inputting the target sample identification information, sample location information, sample type information, and sample location description information into the preset interest point feature recognition network to perform interest point feature recognition and obtain the target sample feature information corresponding to the target sample identification information is similar to the process of step S302 above, which involves inputting the target initial identification information, initial location information, initial type information, and initial location description information from each initial interest point information into the interest point feature recognition network to perform interest point feature recognition and obtain the target feature information corresponding to the target initial identification information. For detailed information, please refer to the detailed information of step S302 above, which will not be repeated here.

[0174] 2) Use the target sample feature information corresponding to at least one sample identification information as at least one sample interest point feature information.

[0175] S904, Perform feature matching analysis on the sample search feature information and at least one sample interest point feature information to obtain sample feature matching information between the sample search feature information and at least one sample interest point feature information.

[0176] In a specific embodiment, the above-mentioned feature matching analysis of sample search feature information and at least one sample interest point feature information to obtain sample feature matching information between sample search feature information and at least one sample interest point feature information may include:

[0177] Feature matching analysis is performed on the sample search feature information and the interest point feature information of each sample to obtain the feature matching information between the sample search feature information and the interest point feature information of each sample.

[0178] Based on the feature matching information corresponding to at least one sample interest point feature information, the sample feature matching information is determined.

[0179] In an optional embodiment, if there is only one sample interest point feature information, the feature matching information corresponding to that one sample interest point feature information is used as the sample feature matching information.

[0180] In another optional embodiment, when there are multiple sample interest point feature information, the feature matching information with the highest matching degree among the feature matching information corresponding to the multiple sample interest point feature information is taken as the sample feature matching information.

[0181] S905 determines the target loss information based on sample feature matching information.

[0182] S906, based on target loss information, train a preset search feature recognition network and a preset interest point feature recognition network to obtain the search feature recognition network and the interest point feature recognition network.

[0183] In an optional embodiment, the target loss information may include feature matching loss; correspondingly, determining the target loss information based on sample feature matching information may include: determining feature matching loss based on sample feature matching information.

[0184] In a specific embodiment, when the sample interest point information includes positive sample interest point information and a preset number of negative sample interest point information, the sample feature matching information may include positive sample feature matching information corresponding to the positive sample interest point information and negative sample feature matching information corresponding to each of the preset number of negative sample interest point information. Accordingly, the above-mentioned determination of feature matching loss based on sample feature matching information includes:

[0185] 1) Based on the feature matching information of positive samples and the feature matching information of negative samples, perform probability analysis to obtain the posterior probability information of the interest point information of positive samples.

[0186] Specifically, the posterior probability information can characterize the probability that a sample user selects the positive sample interest point information when there is positive sample interest point information and a preset number of negative sample interest point information.

[0187] 2) Determine the feature matching loss based on posterior probability information and a preset loss function.

[0188] In a specific embodiment, the preset loss function may include, but is not limited to, the maximum likelihood loss function.

[0189] In an optional embodiment, based on target loss information, training a preset search feature recognition network and a preset interest point feature recognition network to obtain the search feature recognition network and interest point feature recognition network may include:

[0190] S1101, based on the target loss information, update the network parameters of the preset search feature recognition network and the network parameters of the preset interest point feature recognition network;

[0191] S1102, based on the updated preset search feature recognition network and the updated preset interest point feature recognition network, repeat the feature matching training iteration operation from S902 to S905 and S1101 until the feature matching convergence condition is met; use the preset search feature recognition network and the preset interest point feature recognition network obtained when the feature matching convergence condition is met as the search feature recognition network and the interest point feature recognition network.

[0192] In an optional embodiment, the aforementioned feature matching convergence condition can be that the number of training iterations reaches a preset number of training iterations. Optionally, the feature matching convergence condition can also be that the target loss information is less than a specified threshold. In the embodiments of this specification, the preset number of training iterations and the specified threshold can be preset in conjunction with the training speed and accuracy of the network in practical applications.

[0193] In a specific embodiment, such as Figure 11 As shown, an interest point matching network is established, which includes the above-mentioned search feature recognition network, the above-mentioned interest point feature recognition network, and the feature matching network. The user search information and interest point information are input into the interest point matching network to obtain the feature matching information between the user search information and the interest point information.

[0194] As can be seen from the above embodiments, joint training of the network based on sample user search information and sample interest point information yields an interest point feature recognition network and a search feature recognition network with high generalization ability; in addition, joint training of the above-mentioned preset networks improves training efficiency.

[0195] As can be seen from the technical solutions provided in the embodiments of this application above, in the application scenario of point of interest retrieval, this application determines the candidate point of interest information corresponding to the user's search information through index recall or hot word recall, which can narrow the scope of point of interest retrieval. Then, by pre-storing compressed feature information of point of interest, the compressed feature information is decompressed in real time during the point of interest retrieval process to obtain the feature information of point of interest. This can reduce the storage pressure of the feature information of point of interest while improving the efficiency of obtaining the feature information of point of interest, thereby improving the efficiency of point of interest retrieval. In addition, by inputting the identification information, location information, type information and location description information of the point of interest into the point of interest feature recognition network for point of interest feature recognition, the feature information of point of interest can be obtained, which can improve the accuracy of the representation of the feature of point of interest and the semantic richness. Furthermore, when there are multiple identification names for the point of interest, a target corresponding to each identification name is generated. Feature information is enriched by inputting search text information and user location information into the search feature recognition network for search feature recognition, thereby achieving the fusion of search text features and user location features and improving the accuracy of search feature representation. Furthermore, the matching order of candidate interest points is determined by feature matching information between search feature information, and the target interest point information is determined as the interest point retrieval result based on the matching order information, thus improving the accuracy of interest point retrieval results. Additionally, joint training of the network based on sample user search information and sample interest point information yields an interest point feature recognition network and a search feature recognition network with high generalization ability. Weakly related text filtering and core word extraction of samples further enhance the sample data and improve the network training effect.

[0196] This application also provides a point-of-interest (POI) retrieval device, such as... Figure 12 As shown, the point-of-interest retrieval device may include:

[0197] User search information acquisition module 1210 is used to acquire user search information, which includes search text information and user location information.

[0198] The point of interest feature information acquisition module 1220 is used to acquire at least one point of interest feature information corresponding to each of the multiple candidate point of interest information in the user's search information. Any point of interest feature information is obtained after identifying the point of interest feature based on the point of interest identifier information, point of interest location information and location description information of the corresponding candidate point of interest information.

[0199] The search feature recognition module 1230 is used to input search text information and user location information into the search feature recognition network to perform search feature recognition and obtain search feature information.

[0200] The target interest point information display module 1240 is used to display the target interest point information from multiple candidate interest point information based on the feature matching information between the search feature information and at least one interest point feature information.

[0201] In one specific embodiment, the above-described apparatus may further include:

[0202] The initial point of interest information acquisition module is used to acquire information on multiple initial points of interest.

[0203] The Point of Interest (POI) recall module is used to recall POIs based on user search information, and to determine multiple candidate POIs from multiple initial POI information.

[0204] In a specific embodiment, the above-mentioned interest point feature information acquisition module may include:

[0205] The compressed feature information acquisition unit is used to acquire at least one compressed feature information corresponding to each of the multiple candidate interest point information from a local preset memory area.

[0206] The decompression processing unit is used to decompress at least one compressed feature information to obtain at least one interest point feature information.

[0207] In the embodiments described in this specification, the above-mentioned apparatus may further include:

[0208] The initial point of interest information acquisition module is used to acquire multiple initial point of interest information, each of which includes: at least one initial identification information, initial location information, and initial location description information;

[0209] The point of interest feature recognition module is used to perform point of interest feature recognition based on the target initial identification information, initial location information and initial location description information in each initial point of interest information, and to obtain the target feature information corresponding to the target initial identification information. The target initial identification information is any initial identification information in each initial point of interest information.

[0210] The initial interest point feature information module is used to take the target feature information corresponding to at least one initial identification information in each initial interest point information as at least one initial interest point feature information corresponding to each initial interest point information.

[0211] The compression processing module is used to compress at least one initial interest point feature information corresponding to each of the multiple initial interest point information to obtain at least one compressed feature information corresponding to each of the multiple initial interest point information.

[0212] In one specific embodiment, the above-described apparatus may further include:

[0213] The preset compressed dictionary acquisition module is used to acquire a preset compressed dictionary, which is obtained by compressing and training at least one initial interest point feature information corresponding to each of multiple initial interest point information.

[0214] Accordingly, the decompression processing unit mentioned above may include:

[0215] The dictionary decompression unit is used to decompress at least one compressed feature information according to a preset compressed dictionary to obtain at least one interest point feature information.

[0216] In one specific embodiment, any initial point of interest information further includes: initial type information, and the aforementioned point of interest feature recognition module may include:

[0217] The point of interest feature recognition network unit is used to input the target initial identification information, initial location information, initial type information and initial location description information from each initial point of interest information into the point of interest feature recognition network to perform point of interest feature recognition and obtain the target feature information corresponding to the target initial identification information.

[0218] In one specific embodiment, the above-mentioned interest point feature recognition network may include: a second feature extraction network, a second concatenation network, a second semantic recognition network, and a second fusion network, wherein the above-mentioned interest point feature recognition network unit may include:

[0219] The feature extraction unit is used to input the target initial identification information, initial location information, initial type information and initial location description information from each initial point of interest information into the second feature extraction network for feature extraction, so as to obtain identification feature information, location feature information, type feature information and location description feature information;

[0220] The point of interest feature stitching unit is used to input the identification feature information, location feature information, type feature information and location description feature information into the second stitching network for stitching processing to obtain the initial point of interest feature sequence;

[0221] The interest point semantic recognition unit is used to input the initial interest point feature sequence into the second semantic recognition network to perform interest point semantic recognition and obtain the target interest point feature sequence.

[0222] The interest point feature fusion unit is used to input the target interest point feature sequence into the second fusion network for fusion processing to obtain the target feature information corresponding to the target initial identification information.

[0223] In one specific embodiment, the second fusion network may include a second feature mapping layer, a second normalization layer, and a second weighting layer, and the interest point feature fusion unit may include:

[0224] The second feature mapping unit is used to input the word segmentation feature information of multiple interest points into the second feature mapping layer for feature mapping processing to obtain multiple interest point mapping feature information.

[0225] The second normalization unit is used to input the mapping feature information of multiple interest points into the second normalization layer for normalization processing, and obtain the interest point weights of the word segmentation feature information of multiple interest points.

[0226] The second weighting unit is used to input the interest point weights and multiple interest point word segmentation feature information into the second weighting layer for weighting processing, so as to obtain the target feature information corresponding to the initial target identification information.

[0227] In one specific embodiment, the search feature recognition network may include: a first feature extraction network, a first concatenation network, a first semantic recognition network, and a first fusion network; the search feature recognition module may include:

[0228] The word segmentation feature extraction unit is used to input the search text information into the first feature extraction network to extract word segmentation features and obtain the text word segmentation feature sequence.

[0229] The location feature extraction unit is used to input the user's location information into the first feature extraction network to extract location features and obtain the user's location feature information.

[0230] The search feature concatenation unit is used to input the text segmentation feature sequence and user location feature information into the first concatenation network for concatenation processing to obtain the initial search feature sequence.

[0231] The search semantic recognition unit is used to input the initial search feature sequence into the first semantic recognition network for search semantic recognition to obtain the target search feature sequence.

[0232] The search feature fusion unit is used to input the target search feature sequence into the first fusion network for fusion processing to obtain search feature information.

[0233] In one specific embodiment, the first fusion network may include: a first feature mapping layer, a first normalization layer, and a first weighting layer; the search feature fusion unit may include:

[0234] The first feature mapping unit is used to input multiple search word segmentation feature information into the first feature mapping layer for feature mapping processing to obtain multiple search mapping feature information.

[0235] The first normalization unit is used to input multiple search mapping feature information into the first normalization layer for normalization processing to obtain the search weights of multiple search word segmentation feature information.

[0236] The first weighting unit is used to input the search weights and multiple search word segmentation feature information into the first weighting layer for weighting processing to obtain search feature information.

[0237] In one specific embodiment, the above-mentioned target point of interest information display module may include:

[0238] The feature matching analysis unit is used to perform feature matching analysis on the search feature information and at least one point of interest feature information to obtain feature matching information;

[0239] The target matching information unit is used to determine the target matching information of multiple candidate interest points based on the feature matching information corresponding to at least one interest point feature information of multiple candidate interest point information.

[0240] The sorting unit is used to sort multiple candidate interest point information based on target matching information to obtain matching order information;

[0241] The target interest point information determination unit is used to determine the target interest point information from multiple candidate interest point information based on the matching order information;

[0242] The target interest point information display unit is used to display target interest point information based on matching order information.

[0243] In one specific embodiment, the above-described apparatus may further include:

[0244] The sample acquisition module is used to acquire sample user search information and sample point of interest information corresponding to the sample user search information. The sample user search information includes: sample search text information and sample user location information. The sample point of interest information includes: at least one sample identification information, sample location information, sample type information and sample location description information.

[0245] The sample search feature recognition module is used to input sample search text information and sample user location information into a preset search feature recognition network to perform search feature recognition and obtain sample search feature information.

[0246] The sample interest point feature recognition module is used to input at least one sample identification information, sample location information, sample type information and sample location description information into a preset interest point feature recognition network to perform interest point feature recognition and obtain at least one sample interest point feature information corresponding to the sample interest point information.

[0247] The sample feature matching analysis module is used to perform feature matching analysis on sample search feature information and at least one sample interest point feature information to obtain sample feature matching information between sample search feature information and at least one sample interest point feature information.

[0248] The target loss information determination module is used to determine the target loss information based on sample feature matching information;

[0249] The network training module is used to train a preset search feature recognition network and a preset interest point feature recognition network based on the target loss information, thereby obtaining the search feature recognition network and the interest point feature recognition network.

[0250] In one specific embodiment, the above-mentioned sample interest point information may include: positive sample interest point information, and the above-mentioned sample acquisition module may include:

[0251] The initial sample acquisition unit is used to acquire initial sample search information and the initial sample interest point information corresponding to the initial sample search information.

[0252] The text matching processing unit is used to perform text matching processing on the sample search text information in the initial sample search information and the sample identification information in the initial sample interest point information to obtain the text matching result.

[0253] The positive sample interest point information unit is used to take the initial sample interest point information corresponding to the sample identification information that meets the preset matching conditions in the text matching result as the positive sample interest point information.

[0254] The core word extraction unit is used to extract core words from the sample search text information in the initial sample search information to obtain the sample user search information.

[0255] In one specific embodiment, the above-mentioned sample interest point feature recognition module may include:

[0256] The target sample feature information unit is used to input the target sample identification information, sample location information, sample type information and sample location description information into a preset interest point feature recognition network to perform interest point feature recognition, and obtain the target sample feature information corresponding to the target sample identification information. The target sample identification information is any sample identification information among at least one sample identification information.

[0257] The sample interest point feature information unit is used to take the target sample feature information corresponding to at least one sample identification information as at least one sample interest point feature information.

[0258] It should be noted that the apparatus in the device embodiment and the method embodiment are based on the same inventive concept.

[0259] This application provides a point of interest (POI) retrieval device, which includes a processor and a memory. The memory stores at least one instruction or at least one program segment, which is loaded and executed by the processor to implement the POI retrieval method provided in the above method embodiments.

[0260] Furthermore, Figure 13 This diagram illustrates a hardware structure of a point-of-interest (POI) retrieval device for implementing the POI retrieval method provided in the embodiments of this application. The POI retrieval device can participate in or include the POI retrieval apparatus provided in the embodiments of this application. Figure 13 As shown, the point-of-interest (POI) retrieval device 130 may include one or more processors 1302 (shown as 1302a, 1302b, ..., 1302n in the figure) (processor 1302 may include, but is not limited to, a microprocessor MCU or a programmable logic device FPGA, etc.), a memory 1304 for storing data, and a transmission device 1306 for communication functions. In addition, it may also include: a display, an input / output interface (I / O interface), a universal serial bus (USB) port (which may be included as one of the ports of the I / O interface), a network interface, a power supply, and / or a camera. Those skilled in the art will understand that... Figure 13 The structure shown is for illustrative purposes only and does not limit the structure of the aforementioned electronic device. For example, the point-of-interest retrieval device 130 may also include a... Figure 13 The more or fewer components shown, or having the same Figure 13 The different configurations shown.

[0261] It should be noted that the aforementioned one or more processors 1302 and / or other data processing circuitry are generally referred to herein as "data processing circuitry". This data processing circuitry may be embodied, in whole or in part, in software, hardware, firmware, or any other combination thereof. Furthermore, the data processing circuitry may be a single, independent processing module, or may be wholly or partially integrated into any other element within the point-of-interest retrieval device 130 (or mobile device). As involved in the embodiments of this application, this data processing circuitry serves as a processor control mechanism (e.g., selection of a variable resistor termination path connected to an interface).

[0262] The memory 1304 can be used to store software programs and modules of application software, such as the program instructions / data storage device corresponding to the point of interest retrieval method in this embodiment. The processor 1302 executes various functional applications and data processing by running the software programs and modules stored in the memory 1304, thereby implementing the aforementioned point of interest retrieval method. The memory 1304 may include high-speed random access memory, and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 1304 may further include memory remotely located relative to the processor 1302, and these remote memories can be connected to the point of interest retrieval device 130 via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0263] The transmission device 1306 is used to receive or send data via a network. Specific examples of the network described above may include a wireless network provided by the communication provider of the point-of-interest retrieval device 130. In one example, the transmission device 1306 includes a network interface controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In one embodiment, the transmission device 1306 may be a radio frequency (RF) module for wireless communication with the Internet.

[0264] The display may be, for example, a touchscreen liquid crystal display (LCD) that allows the user to interact with the user interface of the point of interest retrieval device 130 (or mobile device).

[0265] The embodiments of this application also provide a computer-readable storage medium, which can be disposed in a point-of-interest (POI) retrieval device to store at least one instruction or at least one program related to implementing the POI retrieval method in the method embodiments. The at least one instruction or the at least one program is loaded and executed by the processor to implement the POI retrieval method provided in the above method embodiments.

[0266] Optionally, in this embodiment, the storage medium may be located in at least one of the multiple network servers in a computer network. Optionally, in this embodiment, the storage medium may include, but is not limited to, various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0267] Embodiments of this application also provide a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the computer device to perform the point-of-interest retrieval method provided in the method embodiments. Optionally, the computer program product may include, but is not limited to, a map retrieval system and a map navigation system.

[0268] It should be noted that the order of the embodiments described above is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. Furthermore, the above description focuses on specific embodiments of this application. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims can be performed in a different order than that shown in the embodiments and still achieve the desired results. Additionally, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some implementations, multitasking and parallel processing are also possible or may be advantageous.

[0269] The various embodiments in this application are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the device and apparatus embodiments are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0270] Those skilled in the art will understand that all or part of the steps of the above embodiments can be implemented by hardware, or by a program instructing related hardware. The program can be stored in a computer-readable storage medium, such as a read-only memory, a disk, or an optical disk.

[0271] The above are merely preferred embodiments of this application and are not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for retrieving points of interest, characterized in that, The method includes: Obtain user search information, which includes search text information and user location information; At least one point of interest feature information corresponding to each of the multiple candidate point of interest information of the user search information is obtained. Any point of interest feature information is obtained after identifying the point of interest feature based on the point of interest identifier information, point of interest location information and location description information of the corresponding candidate point of interest information. The search text information is input into the first feature extraction network in the search feature recognition network for word segmentation feature extraction to obtain a text word segmentation feature sequence. The user location information is input into the first feature extraction network to extract location features, thereby obtaining user location feature information; The text segmentation feature sequence and the user location feature information are input into the first concatenation network in the search feature recognition network for concatenation processing to obtain the initial search feature sequence. The initial search feature sequence is input into the first semantic recognition network in the search feature recognition network for search semantic recognition to obtain the target search feature sequence; The target search feature sequence is input into the first fusion network in the search feature recognition network for fusion processing to obtain search feature information; Based on the feature matching information between the search feature information and the at least one point of interest feature information, the target point of interest information among the multiple candidate point of interest information is displayed.

2. The method according to claim 1, characterized in that, The target search feature sequence includes multiple search word segmentation feature information. The first fusion network includes: a first feature mapping layer, a first normalization layer, and a first weighting layer. The step of inputting the target search feature sequence into the first fusion network of the search feature recognition network for fusion processing yields search feature information including: The multiple search word segmentation feature information is respectively input into the first feature mapping layer for feature mapping processing to obtain multiple search mapping feature information; The multiple search mapping feature information is input into the first normalization layer for normalization processing to obtain the search weights of the multiple search word segmentation feature information; The search weights and the multiple search word segmentation feature information are input into the first weighting layer for weighting processing to obtain the search feature information.

3. The method according to claim 1, characterized in that, The at least one interest point feature information corresponding to each of the multiple candidate interest point information of the user search information includes: Obtain at least one compressed feature information corresponding to each of the multiple candidate interest point information from a local preset memory area; The at least one compressed feature information is decompressed to obtain the at least one point of interest feature information.

4. The method according to claim 3, characterized in that, Prior to obtaining user search information, the method further includes: A preset compressed dictionary is obtained, which is obtained by compressing and training at least one initial interest point feature information corresponding to each of the multiple initial interest point information. The step of decompressing the at least one compressed feature information to obtain the at least one interest point feature information includes: According to the preset compression dictionary, the at least one compressed feature information is decompressed to obtain the at least one point of interest feature information.

5. The method according to claim 1, characterized in that, Before obtaining at least one point of interest feature information corresponding to each of the multiple candidate point of interest information of the user search information, the method further includes: Acquire information on multiple initial points of interest; Based on the user search information, interest point recall is performed, and multiple candidate interest point information is determined from the multiple initial interest point information.

6. The method according to claim 1, characterized in that, The method further includes: Acquire multiple initial points of interest (POIs), where each POI includes at least one initial identifier, initial location information, and initial location description information. Based on the target initial identifier information, initial location information and initial location description information in each initial point of interest information, point of interest feature recognition is performed to obtain target feature information corresponding to the target initial identifier information, wherein the target initial identifier information is any initial identifier information in each initial point of interest information. Each initial point of interest information shall be used as at least one initial point of interest feature information corresponding to at least one initial point of interest feature information. At least one initial interest point feature information corresponding to each of the plurality of initial interest point information is compressed to obtain at least one compressed feature information corresponding to each of the plurality of initial interest point information.

7. The method according to claim 6, characterized in that, The initial point of interest information further includes: initial type information. The step of performing point of interest feature recognition based on the target initial identifier information, initial location information, and initial location description information in each initial point of interest information to obtain the target feature information corresponding to the target initial identifier information includes: The target initial identification information, initial location information, initial type information and initial location description information in each initial point of interest information are input into the point of interest feature recognition network for point of interest feature recognition to obtain the target feature information corresponding to the target initial identification information.

8. The method according to claim 7, characterized in that, The interest point feature recognition network includes: a second feature extraction network, a second concatenation network, a second semantic recognition network, and a second fusion network. The initial target identifier information, initial location information, initial type information, and initial location description information from each initial interest point information are input into the interest point feature recognition network for interest point feature recognition. The target feature information corresponding to the initial target identifier information includes: The target initial identification information, initial location information, initial type information and initial location description information in each initial point of interest information are respectively input into the second feature extraction network for feature extraction to obtain identification feature information, location feature information, type feature information and location description feature information; The identification feature information, the location feature information, the type feature information, and the location description feature information are input into the second stitching network for stitching processing to obtain an initial interest point feature sequence; The initial interest point feature sequence is input into the second semantic recognition network for interest point semantic recognition to obtain the target interest point feature sequence. The target interest point feature sequence is input into the second fusion network for fusion processing to obtain the target feature information corresponding to the target initial identification information.

9. The method according to claim 8, characterized in that, The target interest point feature sequence includes multiple interest point word segmentation feature information. The second fusion network includes a second feature mapping layer, a second normalization layer, and a second weighting layer. The step of inputting the target interest point feature sequence into the second fusion network for fusion processing to obtain the target feature information corresponding to the target initial identification information includes: The word segmentation feature information of the multiple interest points is respectively input into the second feature mapping layer for feature mapping processing to obtain multiple interest point mapping feature information; The multiple interest point mapping feature information is input into the second normalization layer for normalization processing to obtain the interest point weights of the multiple interest point word segmentation feature information; The interest point weights and the word segmentation feature information of the multiple interest points are input into the second weighting layer for weighting processing to obtain the target feature information corresponding to the initial target identification information.

10. The method according to claim 1, characterized in that, The step of displaying the target interest point information from the plurality of candidate interest point information based on the feature matching information between the search feature information and the at least one interest point feature information includes: The search feature information is matched with the at least one point of interest feature information to obtain the feature matching information. Based on the feature matching information corresponding to at least one interest point feature information of the plurality of candidate interest point information, the target matching information of the plurality of candidate interest point information is determined. Based on the target matching information, the multiple candidate interest point information are sorted to obtain matching order information; Based on the matching order information, the target interest point information is determined from the plurality of candidate interest point information; Based on the matching order information, the target point of interest information is displayed.

11. The method according to any one of claims 1 to 10, characterized in that, The search feature recognition network and interest point feature recognition network are trained in the following manner: Obtain sample user search information and sample point of interest information corresponding to the sample user search information. The sample user search information includes: sample search text information and sample user location information. The sample point of interest information includes: at least one sample identification information, sample location information, sample type information and sample location description information. The sample search text information and the sample user location information are input into a preset search feature recognition network to perform search feature recognition and obtain sample search feature information. The at least one sample identification information, the sample location information, the sample type information, and the sample location description information are input into a preset interest point feature recognition network to perform interest point feature recognition, thereby obtaining at least one sample interest point feature information corresponding to the sample interest point information; Perform feature matching analysis on the sample search feature information and the at least one sample interest point feature information to obtain sample feature matching information between the sample search feature information and the at least one sample interest point feature information; Based on the sample feature matching information, the target loss information is determined; Based on the target loss information, the preset search feature recognition network and the preset interest point feature recognition network are trained to obtain the search feature recognition network and the interest point feature recognition network.

12. The method according to claim 11, characterized in that, The sample interest point information includes positive sample interest point information, and the acquisition of sample user search information and the sample interest point information corresponding to the sample user search information includes: Obtain initial sample search information and the initial sample interest point information corresponding to the initial sample search information; The sample search text information in the initial sample search information and the sample identification information in the initial sample interest point information are subjected to text matching processing to obtain the text matching result; The initial sample interest point information corresponding to the sample identifier information that satisfies the preset matching conditions in the text matching results is taken as the positive sample interest point information. Core words are extracted from the initial sample search information to obtain the sample user search information.

13. The method according to claim 11, characterized in that, The step of inputting the at least one sample identifier information, the sample location information, the sample type information, and the sample location description information into a preset interest point feature recognition network for interest point feature recognition, and obtaining at least one sample interest point feature information corresponding to the sample interest point information, includes: The target sample identification information, the sample location information, the sample type information, and the sample location description information are input into the preset interest point feature recognition network to perform interest point feature recognition, thereby obtaining the target sample feature information corresponding to the target sample identification information, wherein the target sample identification information is any one of the at least one sample identification information; The target sample feature information corresponding to each of the at least one sample identification information is used as the at least one sample interest point feature information.

14. A point-of-interest (POI) retrieval device, characterized in that, The device includes: The user search information acquisition module is used to acquire user search information, which includes search text information and user location information. The point of interest feature information acquisition module is used to acquire at least one point of interest feature information corresponding to each of the multiple candidate point of interest information in the user search information. Any point of interest feature information is obtained after identifying the point of interest feature based on the point of interest identifier information, point of interest location information and location description information of the corresponding candidate point of interest information. The word segmentation feature extraction unit is used to input the search text information into the first feature extraction network in the search feature recognition network to extract word segmentation features and obtain a text word segmentation feature sequence. The location feature extraction unit is used to input the user location information into the first feature extraction network to extract location features and obtain user location feature information. The search feature splicing unit is used to input the text segmentation feature sequence and the user location feature information into the first splicing network in the search feature recognition network for splicing processing to obtain the initial search feature sequence; The search semantic recognition unit is used to input the initial search feature sequence into the first semantic recognition network in the search feature recognition network to perform search semantic recognition and obtain the target search feature sequence. The search feature fusion unit is used to input the target search feature sequence into the first fusion network in the search feature recognition network for fusion processing to obtain search feature information; The target interest point information display module is used to display the target interest point information among the multiple candidate interest point information based on the feature matching information between the search feature information and the at least one interest point feature information.

15. The apparatus according to claim 14, characterized in that, The target search feature sequence includes multiple search word segmentation feature information. The first fusion network includes: a first feature mapping layer, a first normalization layer, and a first weighting layer. The search feature fusion unit includes: The first feature mapping unit is used to input the multiple search word segmentation feature information into the first feature mapping layer for feature mapping processing to obtain multiple search mapping feature information. The first normalization unit is used to input the multiple search mapping feature information into the first normalization layer for normalization processing to obtain the search weights of the multiple search word segmentation feature information. The first weighting unit is used to input the search weight and the multiple search word segmentation feature information into the first weighting layer for weighting processing to obtain the search feature information.

16. The apparatus according to claim 14, characterized in that, The interest point feature information acquisition module includes: The compressed feature information acquisition unit is used to acquire at least one compressed feature information corresponding to each of the plurality of candidate interest point information from a local preset memory area; The decompression processing unit is used to decompress the at least one compressed feature information to obtain the at least one point of interest feature information.

17. The apparatus according to claim 16, characterized in that, The device further includes: The preset compressed dictionary acquisition module is used to acquire a preset compressed dictionary, which is obtained by compressing and training at least one initial interest point feature information corresponding to each of multiple initial interest point information. The decompression processing unit includes: The dictionary decompression unit is used to decompress the at least one compressed feature information according to the preset compressed dictionary to obtain the at least one point of interest feature information.

18. The apparatus according to claim 14, characterized in that, The device further includes: The first initial point of interest information acquisition module is used to acquire information on multiple initial points of interest. The point of interest recall module is used to recall points of interest based on the user search information and determine the multiple candidate points of interest information from the multiple initial points of interest information.

19. The apparatus according to claim 14, characterized in that, The device further includes: The second initial point of interest information acquisition module is used to acquire multiple initial point of interest information, wherein any initial point of interest information includes: at least one initial identification information, initial location information and initial location description information; The point of interest feature recognition module is used to perform point of interest feature recognition based on the target initial identification information, initial location information and initial location description information in each initial point of interest information, and to obtain the target feature information corresponding to the target initial identification information, wherein the target initial identification information is any initial identification information in each initial point of interest information; The initial interest point feature information module is used to take the target feature information corresponding to each of the at least one initial identification information in each initial interest point information as at least one initial interest point feature information corresponding to each initial interest point information. The compression processing module is used to compress at least one initial interest point feature information corresponding to each of the plurality of initial interest point information to obtain at least one compressed feature information corresponding to each of the plurality of initial interest point information.

20. The apparatus according to claim 19, characterized in that, The initial point of interest information further includes: initial type information, and the point of interest feature recognition module includes: The point of interest feature recognition network unit is used to input the target initial identification information, initial location information, initial type information and initial location description information in each initial point of interest information into the point of interest feature recognition network to perform point of interest feature recognition, and obtain the target feature information corresponding to the target initial identification information.

21. The apparatus according to claim 20, characterized in that, The interest point feature recognition network includes: a second feature extraction network, a second concatenation network, a second semantic recognition network, and a second fusion network. The interest point feature recognition network unit includes: The feature extraction unit is used to input the target initial identification information, initial location information, initial type information and initial location description information from each initial point of interest information into the second feature extraction network for feature extraction, so as to obtain identification feature information, location feature information, type feature information and location description feature information; The point of interest feature stitching unit is used to input the identification feature information, the location feature information, the type feature information and the location description feature information into the second stitching network for stitching processing to obtain an initial point of interest feature sequence; An interest point semantic recognition unit is used to input the initial interest point feature sequence into the second semantic recognition network to perform interest point semantic recognition and obtain the target interest point feature sequence. The interest point feature fusion unit is used to input the target interest point feature sequence into the second fusion network for fusion processing to obtain the target feature information corresponding to the target initial identification information.

22. The apparatus according to claim 21, characterized in that, The target interest point feature sequence includes multiple interest point word segmentation feature information. The second fusion network includes a second feature mapping layer, a second normalization layer, and a second weighting layer. The interest point feature fusion unit includes: The second feature mapping unit is used to input the word segmentation feature information of the multiple interest points into the second feature mapping layer for feature mapping processing to obtain multiple interest point mapping feature information. The second normalization unit is used to input the multiple interest point mapping feature information into the second normalization layer for normalization processing to obtain the interest point weights of the multiple interest point word segmentation feature information. The second weighting unit is used to input the interest point weights and the word segmentation feature information of the multiple interest points into the second weighting layer for weighting processing to obtain the target feature information corresponding to the initial target identification information.

23. The apparatus according to claim 14, characterized in that, The target point of interest information display module includes: A feature matching analysis unit is used to perform feature matching analysis on the search feature information and the at least one point of interest feature information to obtain the feature matching information; The target matching information unit is used to determine the target matching information of the multiple candidate interest points based on the feature matching information corresponding to at least one interest point feature information of the multiple candidate interest point information. The sorting unit is used to sort the multiple candidate interest point information based on the target matching information to obtain matching order information; The target interest point information determination unit is used to determine the target interest point information from the plurality of candidate interest point information based on the matching order information; The target interest point information display unit is used to display the target interest point information based on the matching order information.

24. The apparatus according to any one of claims 14 to 23, characterized in that, The search feature recognition network and the interest point feature recognition network are trained using the following device: The sample acquisition module is used to acquire sample user search information and sample point of interest information corresponding to the sample user search information. The sample user search information includes: sample search text information and sample user location information. The sample point of interest information includes: at least one sample identification information, sample location information, sample type information and sample location description information. The sample search feature fusion module is used to input the sample search text information and the sample user location information into a preset search feature recognition network to perform search feature recognition and obtain sample search feature information. The sample interest point feature recognition module is used to input the at least one sample identification information, the sample location information, the sample type information and the sample location description information into a preset interest point feature recognition network to perform interest point feature recognition, and obtain at least one sample interest point feature information corresponding to the sample interest point information; The sample feature matching analysis module is used to perform feature matching analysis on the sample search feature information and the at least one sample interest point feature information to obtain sample feature matching information between the sample search feature information and the at least one sample interest point feature information; The target loss information determination module is used to determine the target loss information based on the sample feature matching information; The network training module is used to train the preset search feature recognition network and the preset interest point feature recognition network based on the target loss information, so as to obtain the search feature recognition network and the interest point feature recognition network.

25. The apparatus according to claim 24, characterized in that, The sample interest point information includes positive sample interest point information, and the sample acquisition module includes: An initial sample acquisition unit is used to acquire initial sample search information and initial sample interest point information corresponding to the initial sample search information; The text matching processing unit is used to perform text matching processing on the sample search text information in the initial sample search information and the sample identification information in the initial sample interest point information to obtain the text matching result. The positive sample interest point information unit is used to take the initial sample interest point information corresponding to the sample identifier information that satisfies the preset matching conditions in the text matching result as the positive sample interest point information. The core word extraction unit is used to extract core words from the initial sample search information to obtain the sample user search information.

26. The apparatus according to claim 24, characterized in that, The sample interest point feature recognition module includes: The target sample feature information unit is used to input the target sample identification information, the sample location information, the sample type information and the sample location description information into the preset interest point feature recognition network to perform interest point feature recognition, and obtain the target sample feature information corresponding to the target sample identification information, wherein the target sample identification information is any sample identification information among the at least one sample identification information; The sample interest point feature information unit is used to take the target sample feature information corresponding to each of the at least one sample identification information as the at least one sample interest point feature information.

27. A point-of-interest (POI) retrieval device, characterized in that, The device includes a processor and a memory, the memory storing at least one instruction or at least one program, the at least one instruction or the at least one program being loaded and executed by the processor to implement the point of interest retrieval method as described in any one of claims 1 to 13.

28. A computer-readable storage medium, characterized in that, The storage medium stores at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the point of interest retrieval method as described in any one of claims 1 to 13.

29. A computer program product, characterized in that, The computer program product includes at least one instruction or at least one program segment, which is loaded and executed by a processor to implement the point of interest retrieval method as described in any one of claims 1 to 13.

Images