An apparatus and method for matching POI entities

Abstract

An apparatus for matching a first point of interest (POI) data set comprising one or more first POI entities in a 2D area with a second POI data set comprising a plurality of second POI entities, each of the first POI entities and each of the second POI entities comprising coordinate data and one or more label tokens indicative of a label of the POI entity. The apparatus comprises processing circuitry configured to: determine, for each of the one or more first POI entities, a subset of the plurality of second POI entities in dependence on the coordinate data of the one or more first POI entities; determine, for each of the one or more first POI entities, a combined matching score with one or more second POI entities of the subset of the plurality of second POI entities, the combined matching score being based on a coordinate matching score and a label matching score; match, for each of the one or more first POI entities, with a second POI entity of the subset of the plurality of second POI entities in dependence on the combined matching score.

Description

Technical Field

[0001] This disclosure relates to information processing technology. More specifically, this disclosure relates to an apparatus and method for matching POI entities, i.e., an apparatus and method for aligning geographic entities from heterogeneous sources. Background Technology

[0002] With the ubiquitous development of cyberspace and internet information gathering technologies, a large number of Points of Interest (POI) entities have been and are currently being aggregated into geospatial databases by sources such as map service providers and social networks. Integrating or fusing this multi-source spatial data is a major challenge facing current web-based information retrieval, spatial analysis, and spatial decision-making applications. This is because geographic information collected from different data sources often contains inconsistencies, redundancies, ambiguities, and conflicts. Furthermore, different platforms typically have different descriptive attributes for the same POI entity. Moreover, the descriptive attributes configured for the same geospatial object on the same platform may differ in terms of temporal precision, location accuracy, and semantic precision. Therefore, improved apparatus and methods are needed for matching or aligning the different attributes of POI entities to obtain more complete information about these objects. Summary of the Invention

[0003] The aim is to provide an improved apparatus and method for matching POI entities.

[0004] According to a first aspect, an apparatus is provided for matching (i.e., aligning) a first POI dataset comprising one or more first points of interest (POIs) entities in a 2D region with a second POI dataset comprising a plurality of second POI entities. Each of the first POI entities includes first coordinate data such as first latitude and longitude data, and one or more first tag tokens indicating a label of the first POI entity such as its name. Similarly, each of the second POI entities includes second coordinate data such as second latitude and longitude data, and one or more second tag tokens indicating a label of the second POI entity such as its name.

[0005] The POI matching device includes a processing circuit configured to: determine (i.e. select) a corresponding subset of candidate POI entities for each of the one or more first POI entities, based on first coordinate data of the one or more first POI entities. When only one first POI entity needs to be matched with multiple second POI entities, the processing circuit selects a subset of the multiple second POI entities as candidate POI entities that may match that one first POI entity. When more than one first POI entity needs to be matched with multiple second POI entities, the processing circuit selects a corresponding number of subsets of the multiple second POI entities as corresponding candidate POI entities that may match the respective first POI entities.

[0006] The processing circuitry of the POI matching device is further configured to: determine a corresponding comprehensive matching score (also known as a similarity score) for each of the one or more first POI entities and one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities. The corresponding comprehensive matching score is based on a coordinate matching score and a tag matching score, wherein the coordinate matching score depends on the geographic distance between the corresponding first POI entity and the corresponding second POI entity as defined by first coordinate data and second coordinate data, and the tag matching score depends on the local contextual relevance of one or more first tag tokens of the corresponding first POI entity. As used herein and described in further detail below, the local contextual relevance of one or more first tag tokens of the corresponding first POI entity is a measure of the relevance of the matching decision within the local context of one or more first tag tokens of the corresponding first POI entity and the multiple tag tokens of the second candidate POI entities of a corresponding subset of the plurality of second POI entities.

[0007] The processing circuitry of the POI matching device is further configured to: match (i.e. align) each of one or more first POI entities with a corresponding subset of second POI entities from a plurality of second POI entities, based on a comprehensive matching score. For example, the processing circuitry of the POI matching device may match a corresponding first POI entity with a second POI entity such that the second POI entity and the corresponding first POI entity together provide the highest comprehensive matching score.

[0008] In another possible implementation of the first aspect, each of the first POI entities further includes one or more first address tokens indicating the address of the first POI entity, such as a street name or city name, and each of the second POI entities further includes one or more second address tokens indicating the address of the second POI entity. In this case, the processing circuitry is configured to: determine a comprehensive matching score based on a coordinate matching score, a label matching score, and an address matching score, wherein the address matching score depends on the local contextual relevance of the one or more first address tokens of the corresponding first POI entity.

[0009] In another possible implementation of the first aspect, the processing circuitry is configured to: generate a corresponding document for each subset of candidate POI entities of the plurality of second POI entities based on one or more second tag tokens and / or one or more second address tokens of the second POI entities belonging to a corresponding subset of the plurality of second POI entities, and determine the local contextual relevance of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity as the inverse document frequency (IDF) of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity in the corresponding document.

[0010] In yet another possible implementation of the first aspect, the processing circuitry of the POI matching device is configured to: determine the local contextual relevance of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity based on the local IDF-weighted Jaccard similarity (and thus determine the tag matching score and / or address matching score).

[0011] In another possible implementation of the first aspect, the processing circuit of the POI matching device is configured to: determine the comprehensive matching score as a weighted sum of the coordinate matching score, the tag matching score, and / or the address matching score.

[0012] In another possible implementation of the first aspect, the one or more first POI entities include a first POI entity, wherein the processing circuit is configured to: determine (i.e., select) a subset of the plurality of second POI entities, namely the candidate second POI entities, by identifying second POI entities located near the first POI entity, based on first coordinate data of the first POI entity. For example, the processing circuit may be configured to: determine all second POI entities located within a circle centered on the position of the first POI entity. The processing circuit may also be configured to: determine or adjust the radius of the circle based on the type of the first POI entity. For example, for an airport-type first POI entity, the processing circuit may select a larger circle radius compared to a restaurant-type first POI entity.

[0013] In another possible implementation of the first aspect, the processing circuitry is configured to: determine (i.e. select) a corresponding subset of a plurality of second POI entities in the second POI dataset for each of the first POI entities in the first POI dataset by spatially dividing the 2D region into a plurality of partition units based on the first coordinate data of the one or more first POI entities.

[0014] In another possible implementation of the first aspect, for each partition unit, the product of the number of the first POI entity and the number of the second POI entity located in the corresponding partition unit is less than or equal to a predefined threshold N.

[0015] In another possible implementation of the first aspect, the spatial partitioning is a quadtree partitioning of the 2D region, wherein different partitioning units can have different sizes.

[0016] In another possible implementation of the first aspect, each partition unit has a substantially rectangular, particularly substantially square, shape and includes a frame-shaped boundary region surrounding the partition unit, wherein the ratio of the area of ​​the boundary region to the area of ​​the partition unit depends on the local density of the plurality of first POI entities and / or the plurality of second POI entities.

[0017] In yet another possible implementation of the first aspect, the label matching score is further based on a label embedding score between a first label embedding vector and a second label embedding vector, wherein the first label embedding vector is based on one or more first label tokens of the first POI entity, and the second label embedding vector is based on one or more second label tokens of the second POI entity. For example, the label embedding score may be based on a vector similarity measure (e.g., cosine similarity or dot product) between the first and second label embedding vectors.

[0018] In another possible implementation of the first aspect, the processing circuit of the POI matching device is further configured to: reduce the number of second POI entities in each corresponding subset of the candidate POI entities of the plurality of second POI entities by first determining a tag embedding score between each of the one or more first POI entities and one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities, and deleting one or more second candidate POI entities with the lowest tag embedding score from the corresponding subset of the one or more second candidate POI entities.

[0019] In another possible implementation of the first aspect, each of the first POI entities further includes one or more first types, and each of the second POI entities further includes one or more second types, wherein the comprehensive matching score is also based on a type matching score between one or more first types of the first POI entity and one or more second types of the second POI entity.

[0020] In another possible implementation of the first aspect, the type matching score is based on a type embedding score, which indicates the similarity between a first type embedding vector and a second type embedding vector, the first type embedding vector being based on one or more first types of the first POI entity, and the second type embedding vector being based on one or more second types of the second POI entity. In one implementation, the processing circuitry is configured to: determine the type matching score as a type embedding score minus a type downgrade score. The type embedding score itself can be determined based on the dot product between the first and second type embedding vectors.

[0021] In another possible implementation of the first aspect, the processing circuitry is further configured to: reduce the number of second POI entities in each corresponding subset of the candidate POI entities of the plurality of second POI entities by determining a type embedding score between each of the one or more first POI entities and one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities, and deleting one or more second candidate POI entities with the lowest type embedding score from the corresponding subset of the one or more second candidate POI entities.

[0022] In yet another possible implementation of the first aspect, the apparatus is further configured to implement an unsupervised machine learning scheme for determining a first type embedding vector based on one or more first types of the first POI entity and for determining a second type embedding vector based on one or more second types of the second POI entity.

[0023] In yet another possible implementation of the first aspect, the apparatus further includes a memory for storing a second POI dataset comprising the plurality of second POI entities. The memory can operate a database to store the second POI dataset comprising the plurality of second POI entities.

[0024] In yet another possible implementation of the first aspect, the apparatus further includes a communication interface for receiving a first POI dataset including the one or more first POI entities.

[0025] According to a second aspect, a method is provided for matching a first POI dataset comprising one or more first points of interest (POI) entities in a 2D region with a second POI dataset comprising a plurality of second POI entities. Each first POI entity includes first coordinate data and one or more first tag tokens indicating a label of the first POI entity. Each second POI entity includes second coordinate data and one or more second tag tokens indicating a label of the second POI entity. The method includes the following steps:

[0026] Based on the first coordinate data of the one or more first POI entities, determine (i.e. select) a corresponding subset of candidate POI entities of the plurality of second POI entities for each of the one or more first POI entities;

[0027] For each of the one or more first POI entities, a corresponding comprehensive matching score is determined for it with one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities, wherein the corresponding comprehensive matching score is based on a coordinate matching score and a tag matching score, wherein the coordinate matching score depends on the geographic distance between the corresponding first POI entity and the corresponding second POI entity, and the tag matching score depends on the local contextual relevance of one or more first tag tokens of the corresponding first POI entity;

[0028] Based on the comprehensive matching score, each of the one or more first POI entities is matched with the second POI entities of the corresponding subset of the plurality of second POI entities.

[0029] In another possible implementation of the second aspect, each of the first POI entities further includes one or more first address tokens indicating the address of the first POI entity, and each of the second POI entities further includes one or more second address tokens indicating the address of the second POI entity, wherein the comprehensive matching score is based on the coordinate matching score, the label matching score and the address matching score, and the address matching score depends on the local contextual relevance of the one or more first address tokens of the corresponding first POI entity.

[0030] The POI matching method according to the second aspect can be executed by the POI matching device according to the first aspect. Therefore, further features of the POI matching method according to the second aspect directly derive from the functionality of the POI matching device described in the first aspect and in its various implementations and embodiments as described above and below.

[0031] According to a third aspect, a computer program product is provided, including a non-transitory computer-readable storage medium for storing program code, which, when executed by a computer or processor, causes the computer or processor to perform the POI matching method according to the second aspect.

[0032] The following figures and description illustrate one or more embodiments in detail. Other features, objects, and advantages are clearly shown in the description, figures, etc. Attached Figure Description

[0033] The embodiments of this disclosure will now be described in detail with reference to the accompanying drawings. In the drawings:

[0034] Figure 1 A schematic diagram of various aspects of the POI matching device according to an embodiment in an online matching scenario is shown;

[0035] Figure 2 A schematic diagram of various aspects of the POI matching device according to an embodiment in an offline matching scenario is shown;

[0036] Figure 3 A schematic diagram showing the positions of a plurality of exemplary first POI entities and second POI entities within a partition unit used by the POI matching device according to an embodiment is shown.

[0037] Figure 4a and Figure 4b Two exemplary distributions of specific types of POI entities in different geographic contexts are shown in an embodiment of a POI matching device for POI matching.

[0038] Figure 5A schematic diagram of an early stage of a scoring process implemented by a POI matching device for reducing the number of POI entity pairs to be matched, according to an embodiment, is shown.

[0039] Figure 6 A flowchart of a POI matching method according to an embodiment is shown.

[0040] In the following text, the same reference numerals refer to the same or at least functionally equivalent features. Detailed Implementation

[0041] In the following description, reference is made to the accompanying drawings, which form part of this disclosure, and which illustrate by way of description specific aspects of embodiments of the disclosure or specific aspects in which embodiments of the disclosure may be used. It should be understood that embodiments of the disclosure can be used in other ways and may include structural or logical variations not depicted in the drawings. Therefore, the following detailed description should not be construed as limiting.

[0042] For example, it should be understood that the disclosure relating to the described method is also applicable to the corresponding device or system for performing the method, and vice versa. For example, if one or more specific method steps are described, the corresponding device may include one or more units, such as functional units, to perform the described one or more method steps (e.g., one unit performs one or more steps, or multiple units perform one or more of multiple steps respectively), even if the one or more units are not explicitly described or shown in the drawings. On the other hand, for example, if a specific apparatus is described based on one or more units, such as functional units, the corresponding method may include a step to perform the function of one or more units (e.g., one step performs the function of one or more units, or multiple steps perform the function of one or more of multiple units respectively), even if the one or more steps are not explicitly described or shown in the drawings. Furthermore, it should be understood that, unless otherwise expressly stated, features of the various exemplary embodiments and / or aspects described herein can be combined with each other.

[0043] The following will describe embodiments of the POI matching device in more detail.

[0044] Figure 1A schematic diagram of a POI matching device according to an embodiment is shown in an online matching scenario, where there is an existing POI dataset (hereinafter referred to as the dataset of the second POI entity 103), and a new POI entity (hereinafter referred to as the first POI entity 101) needs to be matched with the existing POI entities. A common application of this use case is adding user-generated POI entities to an existing POI dataset. Before actually adding a new POI entity, it is necessary to check whether the new POI entity already exists to avoid duplication. In the embodiment, the POI matching device can be implemented as a cloud server, which may further include a communication interface for receiving, for example, a first POI dataset including one or more first POI entities 101, via the Internet.

[0045] Figure 2 A schematic diagram of a POI matching apparatus according to yet another embodiment in an offline matching scenario is shown. In this offline scenario, the POI matching apparatus is used to match two large POI datasets with each other at a global scale. A good example of this is when a map provider wants to expand its dataset of POI entities by merging datasets obtained from external providers. In this setting, there may be some overlap between the two datasets, and not all POI entities in the two datasets can be matched. Typically, the aligned POI datasets are heterogeneous and have varying degrees of incompleteness. Some major differences between the datasets may lie in their type hierarchy, label and address formats, and coordinate availability. Seeking matches at a global scale can be challenging because the datasets may each have tens of millions or hundreds of millions of entities, resulting in trillions of possible matching pairs. Therefore, in an embodiment, the POI matching apparatus may include a memory for storing a first POI dataset and a second POI dataset comprising a plurality of first POI entities 101 and a second POI entity 103. The memory may operate on a database for storing the first POI dataset and the second POI dataset.

[0046] Therefore, the POI matching device is generally used to match (i.e., align) a first POI dataset comprising one or more first Points of Interest (POI) entities 101 in a 2D spatial region such as a city with a second POI dataset comprising a plurality of second POI entities 103. Each of the one or more first POI entities 101 includes first coordinate data such as first latitude and longitude data, and one or more first tag tokens indicating a label of the first POI entity such as its name. Similarly, each of the second POI entities 103 includes second coordinate data such as second latitude and longitude data, and one or more second tag tokens indicating a label of the second POI entity such as its name.

[0047] The POI matching device includes a processing circuit configured to: determine (i.e. select) a corresponding subset of candidate POI entities of the plurality of second POI entities 103 for each of the one or more first POI entities 101, based on first coordinate data of the one or more first POI entities 101. Figure 1 In the online matching scenario shown, when only one (or several) first POI entities 101 need to be matched with multiple second POI entities 103, the processing circuit will select a subset of the multiple second POI entities 103 as candidate POI entities that may match a first POI entity. Figure 2 In the offline matching scenario shown, when there is more than one first POI entity 101 to be matched with multiple second POI entities 103, the processing circuit will select a subset of the corresponding number of multiple second POI entities 103 as the corresponding candidate POI entities that may be matched with the corresponding first POI entity 101.

[0048] The processing circuitry of the POI matching device is further configured to: determine a corresponding comprehensive matching score (also referred to herein as a similarity score) for each of the one or more first POI entities 101 with one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities 103. The corresponding comprehensive matching score is based on a coordinate matching score and a tag matching score, wherein the coordinate matching score depends on the geographic distance between the corresponding first POI entity and the corresponding second POI entity (defined by first coordinate data and second coordinate data), and the tag matching score depends on the local contextual relevance of one or more first tag tokens of the corresponding first POI entity. As used herein and further described in detail below, the local contextual relevance of one or more first tag tokens of the corresponding first POI entity is a measure of the relevance of the matching score / decision within the local context of the one or more first tag tokens of the corresponding first POI entity to the multiple tag tokens of the second candidate POI entities of a corresponding subset of the plurality of second POI entities 103.

[0049] The processing circuit of the POI matching device is further configured to: match (i.e. align) each of one or more first POI entities 101 with a corresponding subset of second POI entities of a plurality of second POI entities 103, based on a comprehensive matching score. For example, the processing circuit of the POI matching device may match a corresponding first POI entity with a second POI entity that together provides the highest comprehensive matching score.

[0050] As mentioned above, Figure 1 The diagram illustrates various aspects of the POI matching apparatus according to an embodiment in an online matching scenario, where only a single first POI entity 101 may need to be matched with an existing dataset of second POI entities 103 (in... Figure 1 Matching is performed on the "geographical items" (referred to as "geographical items" in the text). In this case, the processing circuit can be used to: determine (i.e., retrieve) a subset of the multiple second POI entities 103, namely candidate second POI entities 107, in the "retrieval candidates" level 105 by identifying second POS entities of multiple second POI entities 103 located near the first POI entity 101, based on the first coordinate data of the first POI entity 101. For example, as Figure 1 As shown, the processing circuit can be used to: determine the location of the first POI entity 101 ( Figure 1In the example, all second POI entities 103 are located within a circle 102 centered on the first POI entity 101 (e.g., coordinates 56.054, 3.345). In this embodiment, the processing circuitry can also be used to determine or adjust the radius of the circle 102 based on the type of the first POI entity 101. For example, for an airport-type first POI entity 101, the processing circuitry may select a larger radius for the circle 102 compared to a restaurant-type first POI entity 101. As described above, a subset or list of the second candidate POI entities 107 is used to calculate a match or similarity score and to sort the results of the matching (see [link to documentation]). Figure 1 Processing level 109 and processing level 111).

[0051] As mentioned above, Figure 2 The POI matching apparatus according to the embodiment is shown in an offline matching scenario, i.e., two large POI datasets (in Figure 2 The diagram illustrates various aspects of a scenario where sources 1 (101) and 2 (103) are mutually matched. In an embodiment, the processing circuitry of the POI matching device can be used to: implement a preprocessing and loading stage 201 for multiple first POI entities 101 and multiple second POI entities 103.

[0052] like Figure 2 As shown by reference numeral 203 in the accompanying drawings, the processing circuit can be used to: determine (i.e. select) a corresponding subset of a plurality of second POI entities 103 in the second POI dataset for each of the plurality of first POI entities 101 in the first POI dataset by dividing a 2D spatial region into a plurality of partition units in space to provide partitioned second POI entities 103 in 207 according to the preprocessed second POI entities from block 205, based on the first coordinate data of the corresponding first POI entity 101. Figure 3 An example of such a partitioning unit 301 is shown, comprising a pair of first POI entities 101 and a pair of second POI entities 103 (which may further define bounding boxes). In an embodiment, spatial partitioning may be a quadtree partitioning of a 2D region, wherein different partitioning units 301 may have different sizes. Figure 2 and Figure 3As shown, each partition unit 301 may have a rectangular, particularly square, shape and may include a frame-shaped boundary region 303 surrounding the partition unit 301. In embodiments, POI entities located within the boundary region 303 of the partition unit 301 may also be considered to belong to the partition unit 301. In embodiments, the ratio of the area of ​​the boundary region 303 to the area of ​​the partition unit 301 may depend on the local density of the plurality of first POI entities and / or the plurality of second POI entities. For example, the processing circuit may select a smaller boundary region 303 in POI-dense regions and a larger boundary region 303 in regions with fewer POI entities. In yet another embodiment, the processing circuit of the POI matching device may select the size of each partition unit 301 such that the product of the number of first POI entities 101 and second POI entities 103 located in the corresponding partition unit 301 is less than or equal to a predefined threshold N. This limits the computational complexity of the matching process.

[0053] exist Figure 2 In the illustrated embodiment, the POI matching device implements the following additional processing levels (some of which have been described above): a back-partition-based matching level 208 for processing POI entities 206 without coordinate data, and a partition-based matching level 209 described above for implementing the matching process based on a comprehensive matching score. The output 210 of the back-partition-based matching level 208 and the output 211 of the partition-based matching level 209 can be post-processed 213 and output as the final alignment, i.e., the match 215.

[0054] In an embodiment, each of the one or more first POI entities 101 may include, in addition to one or more first tag tokens, one or more first address tokens indicating the address of the corresponding first POI entity 101 (e.g., the street name or city name where the first POI entity 101 is located). Similarly, each of the plurality of second POI entities 103 may also include one or more second address tokens indicating the address of the second POI entity. In this case, the processing circuitry of the POI matching device is configured to: determine a comprehensive matching score based on a coordinate matching score, a tag matching score, and an address matching score, wherein the address matching score depends on the local context relevance of one or more first address tokens of the corresponding first POI entity within a local context defined by the multiple address tokens of the plurality of second POI entities 103.

[0055] Local contextual relevance is important when calculating tag and / or address similarity because the same token may have different relevance levels in different geographic contexts. Figure 4a and Figure 4b The example shown illustrates this point. Figure 4a and Figure 4b It shows West Jerusalem ( Figure 4a ) and Edinburgh ( Figure 4b The frequency of POI entities tagged "Sycamore". In an embodiment, the processing circuit of the POI matching device is configured to: generate a corresponding document for each subset of candidate POI entities of the plurality of second POI entities based on one or more second tag tokens and / or one or more second address tokens belonging to a corresponding subset of the plurality of second POI entities, and determine the local context relevance of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity 101 as the inverse document frequency (IDF) of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity in the corresponding document. In other words, in an embodiment, the concatenation of multiple tags and / or address tokens of candidate second POI entities is considered as a document. ,in This represents the set of documents contained within the geographic context. In an embodiment, the processing circuitry can be used to determine the IDF based on the following equation:

[0056] ,

[0057] In the formula, The corresponding token represents the first POI entity 101. In yet another embodiment, the processing circuitry of the POI matching device can be used to: determine the local context relevance of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity 101 based on local IDF-weighted Jaccard similarity (and thus determine the tag matching score and / or address matching score).

[0058] In an embodiment, the processing circuit of the POI matching device can be used to: determine the comprehensive matching score as a weighted sum of the coordinate matching score, the label matching score, and / or the address matching score.

[0059] In an embodiment, each of the first POI entities 101 further includes one or more first types, and each of the second POI entities 103 further includes one or more second types, wherein the comprehensive matching score is also based on the type matching score between one or more first types of the first POI entity and one or more second types of the second POI entity.

[0060] In an embodiment, the label matching score is further based on a label embedding score between a first label embedding vector and a second label embedding vector, wherein the first label embedding vector is based on one or more first label tokens of the first POI entity 101, and the second label embedding vector is based on one or more second label tokens of the second POI entity 103. For example, the label embedding score may be based on a vector similarity measure (e.g., cosine similarity or dot product) between the first label embedding vector and the second label embedding vector.

[0061] In an embodiment, the type matching score is based on a type embedding score and a type downgrading score. The type embedding score indicates the similarity between a first type embedding vector based on one or more first types of the first POI entity 101 and a second type embedding vector based on one or more second types of the second POI entity 103. In an embodiment, the processing circuitry of the POI matching device is configured to determine the type matching score as a type embedding score minus the type downgrading score. The type embedding score itself can be determined based on the dot product between the first and second type embedding vectors.

[0062] In an embodiment, the POI matching device is further configured to implement an unsupervised machine learning scheme, which is configured to determine a first type embedding vector based on one or more first types of the first POI entity 101 and to determine a second type embedding vector based on one or more second types of the second POI entity 103.

[0063] In an embodiment, the processing circuit of the POI matching device can also be used to: reduce the number of second POI entities 103 in each corresponding subset of the candidate POI entities of the plurality of second POI entities 103 by first determining the tag embedding score between each of the one or more first POI entities 101 and one or more second candidate POI entities in a corresponding subset of the plurality of second POI entities 103, and deleting one or more second candidate POI entities with the lowest tag embedding score from the corresponding subset of the one or more second candidate POI entities.

[0064] Alternatively, the processing circuitry of the POI matching device may also be used to: reduce the number of second POI entities in each corresponding subset of the candidate POI entities of the plurality of second POI entities 103 by determining a type embedding score between each of the one or more first POI entities and one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities 103, and deleting one or more second candidate POI entities with the lowest type embedding score from the corresponding subset of the one or more second candidate POI entities.

[0065] The reduced number of second candidate POIs (103) lowers the computational complexity of the matching process, and through... Figure 5 The example shown provides further illustration. At the first level, for example, located in a partition unit... Figure 3 The first POI entity 101 and the second POI entity 103 within the exemplary partition unit 301 shown calculate tag embedding vectors and address embedding vectors. These vectors are combined in matrix calculation and preliminary matching, i.e., a similarity score is determined based on them. The three first POI entities 101 and the three second POI entities 103 with the highest preliminary matching scores are selected, and the total comprehensive matching score is determined only for these entities in the manner described above.

[0066] Figure 6 A flowchart of a POI matching method 600 according to an embodiment is shown. Method 600 includes the following steps:

[0067] Based on the first coordinate data of the one or more first POI entities 101, a corresponding subset of candidate POI entities of the plurality of second POI entities 103 is determined (i.e. selected) for each of the one or more first POI entities 101.

[0068] For each of the one or more first POI entities 101, a corresponding comprehensive matching score is determined for it with one or more second candidate POI entities of a corresponding subset of the plurality of second POI entities 103, wherein the corresponding comprehensive matching score is based on a coordinate matching score and a tag matching score, wherein the coordinate matching score depends on the geographic distance between the corresponding first POI entity 101 and the corresponding second POI entity 103, and the tag matching score depends on the local contextual relevance of one or more tag tokens of the corresponding first POI entity 101;

[0069] Based on the comprehensive matching score, each of the one or more first POI entities 101 is matched with the second POI entities of the corresponding subset of the plurality of second POI entities 103 (605).

[0070] In another embodiment of method 600, each of the first POI entities further includes one or more first address tokens indicating the address of the first POI entity, and each of the second POI entities further includes one or more second address tokens indicating the address of the second POI entity, wherein the comprehensive matching score is based on the coordinate matching score, the label matching score, and the address matching score, and the address matching score depends on the local contextual relevance of the one or more first address tokens of the corresponding first POI entity.

[0071] The POI matching method 600 can be executed by a POI matching device. Therefore, further features of the POI matching method 600 come directly from the functionality of the POI matching device and its various embodiments described above and below.

[0072] Those skilled in the art will understand that “blocks” (“units”) in the various drawings (methods and apparatuses) represent or describe the functionality of embodiments of this disclosure (and are not necessarily independent “units” in hardware or software), thereby equally describing the functionality or features (unit = step) of apparatus embodiments and method embodiments.

[0073] In the several embodiments provided in this application, it should be understood that the disclosed systems, apparatuses, and methods can be implemented in other ways. For example, the described embodiments of the apparatus are merely exemplary. For example, the unit division is only a logical functional division, and other division methods may be used in actual implementation. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not performed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be implemented through some interfaces. Direct coupling or communication connection between devices or units can be implemented electronically, mechanically, or in other ways.

[0074] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units, and may be located in one location or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of the embodiment according to actual needs.

[0075] In addition, the functional units in the embodiments of the present invention can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

Claims

1. An apparatus for matching a first POI dataset comprising one or more first POI entities (101) in a 2D region with a second POI dataset comprising a plurality of second POI entities (103), characterized in that, Each of the first POI entities (101) includes first coordinate data and one or more first tag tokens indicating a label of the first POI entity (101), and each of the second POI entities (103) includes second coordinate data and one or more second tag tokens indicating a label of the second POI entity (103), wherein the device includes processing circuitry for: Based on the first coordinate data of the one or more first POI entities (101), a subset of the plurality of second POI entities (103) is determined for each of the one or more first POI entities (101); For each of the one or more first POI entities (101), a comprehensive matching score is determined for it with one or more second candidate POI entities (107) of a subset of the plurality of second POI entities (103), wherein the comprehensive matching score is based on a coordinate matching score and a tag matching score, wherein the coordinate matching score depends on the geographic distance between the corresponding first POI entity (101) and the corresponding second POI entity (103), and the tag matching score depends on the local context relevance of one or more first tag tokens of the corresponding first POI entity (101), wherein the local context relevance of one or more first tag tokens of the corresponding first POI entity (101) is a measure of the relevance of the matching score or decision of the tag tokens defined by the tag tokens of the corresponding subset of the plurality of second candidate POI entities (107) of the plurality of second POI entities (103); Based on the comprehensive matching score, each of the one or more first POI entities (101) is matched with the second candidate POI entities (107) of a subset of the plurality of second POI entities (103).

2. The apparatus according to claim 1, characterized in that, Each of the one or more first POI entities (101) further includes one or more first address tokens, and each of the plurality of second POI entities (103) further includes one or more second address tokens, wherein the comprehensive matching score is based on the coordinate matching score, the label matching score and the address matching score, and the address matching score depends on the local context relevance of the one or more first address tokens of the corresponding first POI entity (101), and the local context relevance of the one or more first address tokens of the corresponding first POI entity (101) is the local context relevance of the one or more first address tokens of the corresponding first POI entity (101) within the local context defined by the multiple address tokens of the plurality of second POI entities (103).

3. The apparatus according to claim 2, characterized in that, The processing circuit is configured to: generate a document for each subset of the plurality of second POI entities (103) based on one or more second tag tokens and / or one or more second address tokens of a second candidate POI entity (107) belonging to a subset of the plurality of second POI entities (103), and determine the local context relevance of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity (101) as the inverse document frequency (IDF) of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity (101) in the corresponding document.

4. The apparatus according to claim 3, characterized in that, The processing circuitry is used to: determine the local context relevance of one or more first tag tokens and / or one or more first address tokens of the corresponding first POI entity (101) based on the local IDF-weighted Jaccard similarity.

5. The apparatus according to claim 2 or 3, characterized in that, The processing circuit is used to: determine the comprehensive matching score as a weighted sum of the coordinate matching score, the label matching score, and / or the address matching score.

6. The apparatus according to any one of claims 1 to 3, characterized in that, The one or more first POI entities (101) include a first POI entity (101), and the processing circuit is configured to: determine a subset of the plurality of second POI entities (103) by determining second POI entities (103) of the plurality of second POI entities (103) located near the first POI entity (101) based on the first coordinate data of the first POI entity (101).

7. The apparatus according to any one of claims 1 to 3, characterized in that, The processing circuit is used to: determine a subset of multiple second POI entities (103) of the second POI dataset for each of the first POI entities (101) in the first POI dataset by dividing the 2D region into multiple partition units (301) in space, based on the first coordinate data of the one or more first POI entities (101) of the first POI dataset.

8. The apparatus according to claim 7, characterized in that, For each partition unit (301), the product of the number of the first POI entity (101) and the second POI entity (103) located in the partition unit (301) is less than or equal to a predefined threshold.

9. The apparatus according to claim 7, characterized in that, The spatial partitioning is a quadtree partitioning of the 2D region.

10. The apparatus according to claim 9, characterized in that, Each partition unit (301) has a rectangular shape and includes a frame-shaped boundary region (303) surrounding the partition unit (301), wherein the ratio of the area of ​​the boundary region (303) to the area of ​​the partition unit (301) depends on the local density of the plurality of first POI entities (101) and / or the plurality of second POI entities (103).

11. The apparatus according to any one of claims 1 to 3, characterized in that, The tag matching score is also based on the tag embedding score between the first tag embedding vector and the second tag embedding vector, wherein the first tag embedding vector is based on one or more first tag tokens of the first POI entity (101) and the second tag embedding vector is based on one or more second tag tokens of the second POI entity (103).

12. The apparatus according to claim 11, characterized in that, The processing circuit is further configured to: reduce the number of second candidate POI entities (107) in each subset of the plurality of second POI entities (103) by determining a tag embedding score between each of the one or more first POI entities (101) and one or more second candidate POI entities (107) in a subset of the plurality of second POI entities (103), and deleting at least one second candidate POI entity (107) with the lowest tag embedding score from the subset of the one or more second candidate POI entities (107).

13. The apparatus according to any one of claims 1 to 3, characterized in that, Each of the first POI entities (101) further includes one or more first types, and each of the second POI entities (103) further includes one or more second types, wherein the comprehensive matching score is also based on the type matching score between one or more first types of the first POI entity (101) and one or more second types of the second POI entity (103).

14. The apparatus according to claim 13, characterized in that, The type matching score is based on a type embedding score, which indicates the similarity between a first type embedding vector and a second type embedding vector, wherein the first type embedding vector is based on one or more first types of the first POI entity (101), and the second type embedding vector is based on one or more second types of the second POI entity (103).

15. The apparatus according to claim 14, characterized in that, The processing circuit is further configured to: reduce the number of second candidate POI entities (107) in each subset of the plurality of second POI entities (103) by determining a type embedding score between each of the one or more first POI entities (101) and one or more second candidate POI entities (107) in a subset of the plurality of second POI entities (103), and deleting at least one second candidate POI entity (107) with the lowest type embedding score from the subset of the one or more second candidate POI entities (107).

16. The apparatus according to claim 14 or 15, characterized in that, The apparatus is also used to implement an unsupervised machine learning scheme for determining a first type embedding vector based on one or more first types of the first POI entity (101) and determining a second type embedding vector based on one or more second types of the second POI entity (103).

17. The apparatus according to any one of claims 1 to 3, characterized in that, The device further includes a memory for storing a second POI dataset including the plurality of second POI entities (103).

18. The apparatus according to any one of claims 1 to 3, characterized in that, The device further includes a communication interface for receiving a first POI dataset including the one or more first POI entities (101).

19. A method for matching a first POI dataset comprising one or more first POI entities (101) in a 2D region with a second POI dataset comprising a plurality of second POI entities (103), characterized in that, Each of the first POI entities (101) includes first coordinate data and one or more first tag tokens indicating a label of the first POI entity (101), and each of the second POI entities (103) includes second coordinate data and one or more second tag tokens indicating a label of the second POI entity (103), wherein the method includes: Based on the first coordinate data of the one or more first POI entities (101), a subset of the plurality of second POI entities (103) is determined for each of the one or more first POI entities (101); For each of the one or more first POI entities (101), a comprehensive matching score is determined for it with one or more second candidate POI entities (107) of a subset of the plurality of second POI entities (103), wherein the comprehensive matching score is based on a coordinate matching score and a tag matching score, wherein the coordinate matching score depends on the geographic distance between the corresponding first POI entity (101) and the corresponding second POI entity (103), and the tag matching score depends on the local context relevance of one or more first tag tokens of the corresponding first POI entity (101), wherein the local context relevance of one or more first tag tokens of the corresponding first POI entity (101) is a measure of the relevance of the matching score or decision of the tag tokens defined by the tag tokens of the corresponding subset of the plurality of second candidate POI entities (107) of the plurality of second POI entities (103); Based on the comprehensive matching score, each of the one or more first POI entities (101) is matched with the second candidate POI entities (107) of a subset of the plurality of second POI entities (103).

20. The method according to claim 19, characterized in that, Each of the one or more first POI entities (101) further includes one or more first address tokens, and each of the plurality of second POI entities (103) further includes one or more second address tokens, wherein the comprehensive matching score is based on the coordinate matching score, the label matching score and the address matching score, and the address matching score depends on the local context relevance of the one or more first address tokens of the corresponding first POI entity (101), and the local context relevance of the one or more first address tokens of the corresponding first POI entity (101) is the local context relevance of the one or more first address tokens of the corresponding first POI entity (101) within the local context defined by the multiple address tokens of the plurality of second POI entities (103).

Images