Geographic knowledge graph construction method and device and information generation method and device

Abstract

The invention provides a geographic knowledge graph construction method and device and an information generation method and device, and relates to the technical field of artificial intelligence, in particular to the technical fields of large models, automatic driving, intelligent traffic and the like. According to the specific implementation scheme, entity description information is extracted from target data; determining associated interest points associated with the entity description information; fusing the entity description information corresponding to the same associated interest point to obtain fused description information corresponding to the associated interest point; and constructing a geographic knowledge graph based on the associated interest points and the fusion description information corresponding to the associated interest points. According to the method, the geographic knowledge graph with rich semantics and a stable structure is constructed.

Description

TECHNICAL FIELD

[0001] The present disclosure relates to the technical field of artificial intelligence, in particular to the technical fields of large models, automatic driving, intelligent transportation, and the like, and more particularly to a geographic knowledge graph construction and information generation method and device. BACKGROUND

[0002] Travel guides are the most commonly referenced consumer content form when users plan trips, make instant decisions, and record experiences. With the intuitive presentation effect brought by images and the detailed supplement provided by text, graphic notes have become one of the more attractive forms of many guide information carriers. SUMMARY

[0003] Embodiments of the present disclosure provide a geographic knowledge graph construction and information generation method, device, equipment, and storage medium.

[0004] In a first aspect, embodiments of the present disclosure provide a geographic knowledge graph construction method, which includes: extracting entity description information in target data; determining an associated interest point associated with the entity description information; fusing entity description information corresponding to the same associated interest point to obtain fusion description information corresponding to the associated interest point; and constructing a geographic knowledge graph based on the associated interest point and the fusion description information corresponding to the associated interest point.

[0005] In a second aspect, embodiments of the present disclosure provide an information generation method, which includes: determining image description information and text description information of a target interest point in a geographic knowledge graph according to travel intention information, the geographic knowledge graph being a geographic knowledge graph obtained by the method of any one of the implementations of the first aspect described above; generating a text travel plan based on the travel intention information and the text description information; and generating a graphic text travel plan based on the travel intention information, the text travel plan, and the image description information.

[0006] In a third aspect, embodiments of the present disclosure provide a geographic knowledge graph construction device, which includes an extraction module, an association module, a fusion module, and a construction module. The extraction module is configured to extract entity description information in target data. The association module is configured to determine an associated interest point associated with the entity description information. The fusion module is configured to fuse entity description information corresponding to the same associated interest point to obtain fusion description information corresponding to the associated interest point. The construction module is configured to construct a geographic knowledge graph based on the associated interest point and the fusion description information corresponding to the associated interest point.

[0007] Fourthly, embodiments of this disclosure provide an information generation apparatus, comprising: a determining module, a planning module, and a generating module, wherein the determining module is configured to determine image description information and text description information of a target point of interest in a geographic knowledge graph based on travel intention information, the geographic knowledge graph being a geographic knowledge graph obtained by the method described in any implementation of the first aspect above; the planning module is configured to generate a text travel plan based on the travel intention information and the text description information; and the generating module is configured to generate a graphic-text travel plan based on the travel intention information, the text travel plan, and the image description information.

[0008] Fifthly, embodiments of this disclosure provide an electronic device including one or more processors; and a storage device having one or more programs stored thereon, wherein when the one or more programs are executed by the one or more processors, the one or more processors implement the method described in any implementation of the first or second aspect.

[0009] In a sixth aspect, embodiments of this disclosure provide a computer-readable medium having a computer program stored thereon that, when executed by a processor, implements the method as described in either the first or second aspect.

[0010] In a seventh aspect, embodiments of this disclosure provide a computer program product, including a computer program that, when executed by a processor, implements the method as described in either the first or second aspect.

[0011] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this disclosure, nor is it intended to limit the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0012] Figure 1 This is an exemplary system architecture diagram to which this disclosure can be applied; Figure 2 This is a flowchart of an embodiment of the geographic knowledge graph construction method according to the present disclosure; Figure 3 This is a flowchart of an embodiment of the information generation method according to the present disclosure; Figure 4 This is a schematic diagram of an application scenario of the information generation method according to this disclosure; Figure 5 This is a schematic diagram of one embodiment of a geographic knowledge graph construction apparatus according to the present disclosure; Figure 6 This is a schematic diagram of an embodiment of the information generation apparatus according to the present disclosure; Figure 7This is a schematic diagram of the structure of a computer system suitable for implementing the electronic device of the present disclosure. Detailed Implementation

[0013] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0014] It should be noted that, unless otherwise specified, the embodiments and features described in this disclosure can be combined with each other. This disclosure will now be described in detail with reference to the accompanying drawings and embodiments.

[0015] Figure 1 An exemplary system architecture 100 is shown, in which embodiments of the geographic knowledge graph construction method of this disclosure can be applied.

[0016] like Figure 1 As shown, system architecture 100 may include terminal devices 101, 102, and 103, a network 104, and a server 105. Network 104 serves as the medium for providing communication links between terminal devices 101, 102, and 103 and server 105. Network 104 may include various connection types, such as wired or wireless communication links, or fiber optic cables, etc.

[0017] Users can use terminal devices 101, 102, and 103 to interact with server 105 via network 104 to receive or send messages, etc.

[0018] Terminal devices 101, 102, and 103 can be either hardware or software. When terminal devices 101, 102, and 103 are software, they can be installed in the electronic devices listed above. They can be implemented as multiple software programs or software modules, or as a single software program or software module. No specific limitations are made here.

[0019] Server 105 can be a server that provides various services, such as extracting entity description information from target data; determining associated points of interest (POIs) related to entity description information; fusing entity description information corresponding to the same POI to obtain fused description information corresponding to the POI; and constructing a geographic knowledge graph based on the POIs and the fused description information corresponding to the POIs.

[0020] It should be noted that server 105 can be either hardware or software. When server 105 is hardware, it can be implemented as a distributed server cluster consisting of multiple servers, or as a single server. When server 105 is software, it can be implemented as multiple software programs or software modules (for example, used to provide geographic knowledge graph construction services), or as a single software program or software module. No specific limitations are made here.

[0021] It should be noted that the geographic knowledge graph construction method provided in the embodiments of this disclosure can be executed by server 105, terminal devices 101, 102, and 103, or by server 105 and terminal devices 101, 102, and 103 in cooperation with each other. Accordingly, the various parts (e.g., various units, sub-units, modules, and sub-modules) included in the geographic knowledge graph construction device can all be set in server 105, all of them can be set in terminal devices 101, 102, and 103, or they can be set in server 105 and terminal devices 101, 102, and 103 respectively.

[0022] It should be understood that Figure 1 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0023] Figure 2 A flowchart 200 illustrating an embodiment of a geographic knowledge graph construction method is provided. This geographic knowledge graph construction method may specifically include the following steps: Step 201: Extract entity description information from the target data.

[0024] In this embodiment, the execution entity (e.g., Figure 1 The server 105 or terminal devices 101, 102, 103 can extract entity description information from the image data, text data, multimedia data and spatiotemporal data included in the target data.

[0025] Entity description information is a structured and precise collection of information that clearly distinguishes concrete / abstract things (i.e., "entities") in the real world or virtual scenarios. Entity types can include concrete objects (e.g., mobile phones, cars, etc.), people (e.g., Zhang San, Li Si, etc.), organizations / institutions (e.g., enterprises, schools, etc.), geospatial entities (e.g., XX city, XX district, etc.), abstract concepts (e.g., shopping festivals, user orders, etc.), and events (e.g., annual meetings, job fairs, etc.).

[0026] For text data, the executing entity can use a preset text extraction model to extract entity description information and output it in the manner of entity recognition + relation extraction + template calibration.

[0027] The text extraction model is fine-tuned using relevant corpus data, such as notes, comments, and event pages, and has the ability to identify fine-grained entities in short location-related texts.

[0028] Specifically, the entity description information extracted from text data can include location name (e.g., Chengdu XX scenic spot), activity description (e.g., there is a Sichuan opera performance), price information (e.g., 88 yuan per person), waiting / queueing time (e.g., waited for half an hour), user subjective feelings (e.g., the atmosphere is very lively), time clues (8 pm on weekends), etc.

[0029] In addition, it should be noted that for entities that appear infrequently but have high business value (such as specific models, technical terms, and rare brands), the shortcomings of the text extraction model can be supplemented by manually designing regular expressions, thereby improving the recognition rate of these rare categories.

[0030] For image data, a preset image extraction model can be used to extract entity description information from the image data.

[0031] Specifically, the execution entity can adopt the core framework of CLIP (Contrastive Language-Image Pre-training, a pre-training method based on contrastive text-image pairs) + multiple LoRA (Low-Rank Adaptation) + ViT (Vision Transformer) to understand the "visual meaning" of the image (such as "This is a shop sign / West Lake Fish in Vinegar Sauce"); and use OCR (Optical Character Recognition) tools to accurately obtain the "text content" in the image (such as the text on the shop sign, the text of the dish name).

[0032] For example, image data includes store signs, product / dish entities, environmental features, etc. The executing entity can identify store signs, product / dish entities, and environmental features through CLIP + multiple LoRA (Low-Rank Adaptation) + ViT, as well as OCR, and generate entity description information of the image data based on the recognition results.

[0033] For multimedia data, such as video data and audio data, the executing entity can first convert the multimedia data into image data and / or text data, and then extract entity description information from the image data and text data respectively.

[0034] Here, spatiotemporal data, which is a collection of information that simultaneously contains the "time dimension" (when it happened / was recorded) and the "spatial dimension" (where it happened / was recorded), can be generated by semantic understanding of one or more of image data, text data, and multimedia data. Examples include vehicle trajectory data, public transportation operation data, air quality monitoring data, population flow data, and consumption data. Spatiotemporal data can be text data, image data, or multimedia data.

[0035] Step 202: Determine the associated points of interest that are linked to the entity description information.

[0036] In this embodiment, after determining the entity description information, the executing entity can use various methods to determine the associated points of interest associated with the entity description information, that is, to realize the attachment of points of interest. For example, the associated points of interest associated with the entity description information can be determined directly based on the geographical coordinates, name, etc. of the entity description information; firstly, the initial set of points of interest associated with the entity description information can be determined based on the geographical coordinates, name, etc. of the entity description information, and then the associated points of interest associated with the entity description information can be determined based on the association confidence between the entity description information and the corresponding initial points of interest.

[0037] Step 203: Merge the entity description information corresponding to the same associated point of interest to obtain the fused description information corresponding to the associated point of interest.

[0038] In this embodiment, after determining the associated points of interest associated with the entity description information, the executing entity can fuse the entity description information corresponding to the same associated point of interest to obtain the fused description information corresponding to the associated point of interest.

[0039] The specific fusion process may include conflict-free information fusion, i.e. integration integrity, and conflict information resolution, i.e. eliminating contradictions and determining information confidence.

[0040] Here, conflict-free information fusion can include basic information merging and related information supplementation. Basic information merging is used to combine information from different sources into complete information, and related information supplementation is used to complete information missing from a single source.

[0041] For example, the entity description information from source A is: name "coffee", address "No. 88, YY Road"; the entity description information from source B is: name "Coffee (XX store)", address "1st floor of Modern City", and the combined description information is: name "coffee (Coffee XX store)", address "1st floor of Modern City, No. 88, YY Road, N District, M City".

[0042] For example, source A has business hours but no rating, while source B has rating but no business hours. The merged description information includes both business hours and rating.

[0043] Step 204: Construct a geographic knowledge graph based on associated points of interest and the fused description information corresponding to the associated points of interest.

[0044] In this embodiment, the executing entity can determine the associated points of interest as nodes, determine the attributes of the nodes and the relationships between the nodes based on the fusion description information corresponding to the associated points of interest, and construct a geographic knowledge graph based on the nodes, the attributes of the nodes, and the relationships between the nodes.

[0045] The embodiments of this disclosure extract entity description information from target data; determine associated points of interest (POIs) related to the entity description information; fuse entity description information corresponding to the same POI to obtain fused description information corresponding to the POI; and construct a geographic knowledge graph based on the POIs and the fused description information corresponding to the POIs. That is, a geographic knowledge graph is constructed by extracting, associating, and fusing information from multimodal data such as text, images, and videos. This approach considers the association relationships of multimodal data and effectively improves the semantic richness and structural stability of the constructed geographic knowledge graph.

[0046] In some alternative approaches, entity description information is extracted from the target data, including: in response to determining that the multimedia data is video data, initial description information is extracted based on the image frame data and audio data included in the video data; the audio data and image frame data are aligned based on the timestamps included in the video data to obtain aligned data; and the initial description information is verified and supplemented based on the aligned data to generate entity description information.

[0047] In this implementation, the executing entity can determine the type of multimedia data. If the type of multimedia data is video data, the initial description information can be extracted from the image frame data and audio data included in the video data.

[0048] Since the entity description information extracted based on single-modal data, such as image frame data or audio data, may be missing, the executing entity can align the audio data and image frame data according to the timestamps included in the video data to obtain aligned data; and then verify and supplement the initial description information based on the aligned data.

[0049] For example, if the initial description information of an image frame data is that the user points to a scenic spot, and if the information determined based on the audio data in the aligned data corresponding to the image frame data includes the name of the scenic spot, then the name of the scenic spot can be added to the initial description information of the image frame data to obtain the entity description information.

[0050] For example, if the initial description information of the audio data is "A store", and the information determined based on the image frame data in the aligned data corresponding to the audio data includes the environmental features of the store, then the environmental features of the store can be added to the initial description information of the audio data to obtain the entity description information.

[0051] For example, if the initial description information of the audio data is "AA store", and the information determined from the image frame data in the aligned data corresponding to the audio data includes the store name, such as "BB store", then the initial description information of the audio data can be corrected based on the determined store name, that is, the store name can be corrected from "AA store" to "BB store".

[0052] This implementation method, in response to determining that the multimedia data is video data, extracts initial description information based on the image frame data and audio data included in the video data; aligns the audio data and image frame data based on the timestamps included in the video data to obtain aligned data; and verifies and supplements the initial description information based on the aligned data to generate entity description information. This achieves the supplementation and correction of entity description information generated from multimedia data, improving the accuracy of the generated entity description information.

[0053] Among some alternative methods, points of interest associated with entity description information are identified, including: Based on at least one of the geographical coordinates, name, and administrative region code of the entity description information, determine the initial set of points of interest associated with the entity description information; based on the feature information of the entity description information and the initial points of interest, determine the association confidence between the entity description information and the initial points of interest. Based on the association confidence score, associated interest points that are related to the entity description information are identified in the initial set of interest points.

[0054] In this implementation, if the entity description information includes geographic coordinates, such as location information or trajectory information, then the interest points whose geographic coordinates are less than or equal to the geographic coordinates of the entity description information can be determined as initial interest points. That is, interest points located within the area centered on the geographic coordinates of the entity description information and with a radius equal to the preset distance value are determined as initial interest points.

[0055] If the entity description information includes a name, then interest points whose name has a similarity to the name included in the entity description information that is greater than or equal to a preset similarity threshold can be determined as initial interest points.

[0056] There are various ways to determine similarity, such as calculating edit distance, Jaccard similarity, cosine similarity, etc.

[0057] If the entity description information includes administrative region codes, then interest points whose administrative region codes completely match the preset positions (e.g., the first 4 digits, the first 3 digits, etc.) of the administrative region codes included in the entity description information can be determined as initial interest points.

[0058] Specifically, if the entity description information includes geographic coordinates, name, and administrative region code, then the interest points that simultaneously satisfy the following conditions can be determined as initial interest points, thus obtaining the initial interest point set.

[0059] After determining the initial set of points of interest, the executing entity can input entity description information and feature information of the initial points of interest, such as address, tags, and historical comments, into a preset association model to determine the association confidence between the entity description information and the initial points of interest.

[0060] Here, the association confidence can be represented by the association confidence value. The higher the association confidence value, the higher the association confidence between the initial point of interest and the entity description information.

[0061] Furthermore, the executing entity can sort the initial points of interest in descending order of their association confidence values, and determine one or more of the top-ranked initial points of interest as associated points of interest for the entity description information.

[0062] It should be noted that for initial points of interest with a correlation confidence value lower than the specified threshold or similar correlation confidence values ​​(such as a difference less than the preset value), manual confirmation can be carried out by content operation personnel.

[0063] Furthermore, after determining the entity description information and the associated points of interest, the executing entity can identify the entity description information and the associated entity description information as entity-point of interest pairs, and save them to the cache pool after summarizing and indexing. When it is necessary to determine the associated points of interest associated with the entity description information again, the matching can be directly searched in the cache pool, which can greatly reduce repeated calculations and improve response speed.

[0064] This implementation method determines an initial set of points of interest associated with entity description information by using at least one of the following: geographic coordinates, name, and administrative region code. Based on the feature information of the entity description information and the initial points of interest, it determines the association confidence between the entity description information and the initial points of interest. Based on the association confidence, it identifies associated points of interest from the initial set of points of interest that are related to the entity description information. In other words, it performs a rough screening using geographic coordinates, name, etc., to determine the initial points of interest, and then a fine screening based on the association confidence to determine the associated points of interest, thus improving the comprehensiveness and accuracy of the identified associated points of interest.

[0065] In some optional approaches, associated interest points that are associated with entity description information are identified in the initial set of interest points based on association confidence. This includes: identifying associated interest points that are associated with entity description information in the initial set of interest points based on association confidence, the current popularity of the initial interest points, and the historical association frequency between the initial interest points and entity description information.

[0066] In this implementation, after determining the association confidence of the initial point of interest, the executing entity can further consider the current popularity of the initial point of interest and the historical association frequency between the initial point of interest and entity description information.

[0067] The current popularity of the initial point of interest is used to indicate the popularity or attention of the initial point of interest in the current time period. It can be determined by the weighted value of data such as the number of views, interactions, and check-ins of the initial point of interest on social platforms.

[0068] The historical association frequency between the initial point of interest and entity description information is used to indicate how often the initial point of interest has been historically associated with entity description information.

[0069] The executing entity can perform a weighted summation of the association confidence, the current popularity of the initial point of interest, and the historical association frequency between the initial point of interest and the entity description information to determine the ranking value of the initial point of interest; and sort the initial points of interest in descending order of the ranking value, and determine one or more initial points of interest with the highest ranking as the associated points of interest of the entity description information.

[0070] This implementation method determines the associated interest points that are associated with the entity description information from the initial interest point set by considering the association confidence, the current popularity of the initial interest points, and the historical association frequency between the initial interest points and the entity description information. In other words, the influence of the popularity and historical association frequency of the historical interest points is further considered in the process of determining the associated interest points, which further improves the accuracy of the determined associated interest points.

[0071] In some optional methods, entity description information corresponding to the same associated point of interest is fused to obtain fused description information corresponding to the associated point of interest, including: determining the information confidence of entity description information corresponding to the same associated point of interest based on at least one of source credibility, information timeliness, and consistency of multi-source feedback; and fusing entity description information whose information confidence meets preset conditions to obtain fused description information corresponding to the associated point of interest.

[0072] In this implementation, since the sources of entity description information are wide-ranging, including data uploaded by industry users as well as notes, comments, and short video content from general users, the executing entity can determine the information confidence level of entity description information based on at least one of the following: source credibility, information timeliness, and consistency of multi-source feedback, for entity description information corresponding to the same related point of interest.

[0073] Information confidence is positively correlated with source credibility, information timeliness, and consistency of multi-source feedback; that is, the higher the source credibility, the higher the information timeliness, and the higher the consistency of multi-source feedback, the higher the information confidence. Source credibility, information timeliness, and consistency of multi-source feedback can be represented by weight values.

[0074] Here, the executing entity can determine the source confidence level based on the source type of the entity description information and a preset mapping table between source types and weight values. Different source types typically correspond to different weight values.

[0075] Specifically, the source types can include data uploaded by industry users, user-generated content screened by platform operations, and content automatically crawled by platforms such as social media, with the weight values ​​of the three decreasing in that order.

[0076] Here, the executing entity can determine the timeliness of information based on the update time of the entity description information and a preset mapping table between update times and weight values. Different update times typically correspond to different weight values. Generally, the closer the update time is to the current time, the higher the weight value corresponding to that update time.

[0077] Here, the executing entity can determine multi-source feedback consistency based on the frequency of entity description information in different data sources and a preset mapping table between frequency and weight values. Different frequencies typically correspond to different weight values. Generally, the higher the frequency, the higher the corresponding weight value.

[0078] Furthermore, the implementing entity can determine the information confidence level of entity description information corresponding to the same associated point of interest by summing the weights corresponding to the source confidence level, the information timeliness, and the consistency of multi-source feedback.

[0079] Entity description information whose confidence level meets preset conditions is fused to obtain fused description information corresponding to associated points of interest.

[0080] The preset conditions can be set according to actual needs, such as information confidence level being greater than or equal to a preset value, or information confidence level being the highest.

[0081] Specifically, the executing entity can sort the entity description information in descending order of information confidence, and then merge the top three entity description information to obtain the fused description information corresponding to the associated points of interest.

[0082] It should be noted that for multiple entity descriptions with similar confidence levels (e.g., the difference in confidence levels is less than a preset value) or with conflicting data logic (e.g., "open for business" and "closed for business" appearing within the same time period), a small-scale "crowdsourced verification task" can be initiated for some users. For example, a voting request can be sent to nearby active users: "Please confirm whether location XX is currently open." The confidence level can be adjusted based on the user voting results. After adjustment, the decision path and original data can be recorded for subsequent learning and rule optimization.

[0083] This implementation determines the information confidence level of entity description information corresponding to the same related interest point based on at least one of the following: source credibility, information timeliness, and consistency of multi-source feedback. Entity description information whose information confidence level meets the preset conditions is fused to obtain fused description information corresponding to the related interest point. This further filters the entity description information corresponding to the same related interest point and improves the accuracy of the determined fused description information.

[0084] In some alternative approaches, the method further includes: directly storing the structured data in the associated points of interest and the fused description information corresponding to the associated points of interest into a graph database; encoding the unstructured data in the associated points of interest and the fused description information corresponding to the associated points of interest into vectors, and storing the vectors into a vector database.

[0085] In this implementation, the executing entity can store the structured data (such as business hours, price, average consumption per person, address, etc.) in the associated points of interest and the fusion description information corresponding to the associated points of interest into a graph database, and use a graph database model (e.g., attribute graph model, resource description framework triple model, etc.) to organize nodes and relationships.

[0086] Specifically, the executing entity can adopt a property graph model to organize nodes and relationships. Each associated point of interest corresponds to an independent node, which forms a connection relationship with location type, business district, activity information, etc., which facilitates graph structure traversal and multi-hop queries, and supports linked queries with logical paths such as "find attractions from the region and then find related activities".

[0087] The executing entity can encode unstructured data (such as user comments, image clips, short video scene descriptions, etc.) in the associated points of interest and the fused description information corresponding to the associated points of interest into vectors and store them in a vector database.

[0088] The vector database may include systems such as Milvus and Faiss (Facebook AI Similarity Search).

[0089] Milvus is an open-source vector database primarily used for storing, indexing, and managing large-scale vector data, commonly found in artificial intelligence fields (such as large model retrieval enhancement, image / speech recognition, etc.). Faiss, originally developed by Facebook, is an open-source vector similarity search and clustering library used for efficiently calculating the similarity between vectors.

[0090] Specifically, the executing entity can encode the unstructured data in the associated points of interest and the fused description information corresponding to the associated points of interest into vectors and store them in the Milvus system. The Milvus system has a high-throughput near nearest neighbor search capability, which supports the rapid location of related content in a large-scale semantic vector space. For example, when a user searches for "outdoor places suitable for children", the system can find associated points of interest with children's entertainment elements in the description through vector similarity matching.

[0091] Furthermore, to enhance the geographic awareness capabilities of the Milvus system, a spatial indexing mechanism can be introduced. By building a custom R-Tree (R-tree structure), the latitude and longitude information of all associated points of interest is regionally divided and hierarchically organized, establishing a spatial index structure covering the entire country. This mechanism supports "range queries" (such as finding cafes within 500 meters of a location) and "nearby searches" (such as returning the nearest public restroom to the current location), greatly improving the response speed and accuracy of spatial search requests.

[0092] This implementation achieves a dual storage strategy for geographic knowledge graphs by directly storing the structured data in the associated points of interest and the fused description information corresponding to the associated points of interest into a graph database; and encoding the unstructured data in the associated points of interest and the fused description information corresponding to the associated points of interest into vectors and storing the vectors into a vector database. This strategy ensures efficient processing of clearly structured information while supporting high-dimensional semantic retrieval of complex and difficult-to-structure content.

[0093] In some alternative approaches, entity description information is extracted from the target data: the target data is filtered based on at least one of the following: consistency of target data associated with the same point of interest, consistency between the target data and the associated point of interest, and timestamp deviation of the target data; entity description information is then extracted from the filtered target data.

[0094] In this implementation, for target data from industry users and general users, namely text data, image data, multimedia data and spatiotemporal data, the executing entity can first perform preliminary screening of the target data according to preset rules to obtain the target data after initial screening, that is, the target data that conforms to the preset rules.

[0095] Industry users can include scenic area operators, merchants, OTA (Online Travel Agency) platforms, and lifestyle service partners. Target data from industry users is characterized by high structure, fixed update cycles, and high credibility, but it lacks fine-grained experience descriptions. Public users can include map-based UGC (User Generated Content), public posts on social media, short video comments, and image reviews. Target data from public users has rich on-site experience perception signals (queue length, atmosphere, day's activities, etc.), but it is noisy and highly homogeneous.

[0096] Here, prediction rules can include a variety of rules, such as: field integrity verification: the missing rate of required fields (name, coordinates, timestamp) is less than or equal to the preset missing rate threshold; coordinate validity: latitude and longitude range and reverse geocoding conform to preset standards and specifications; business status consistency: consistent with the merchant's official timetable and historical trajectory; multi-field joint verification: consistency of multiple attribute fields, that is, the data of the field conforms to the preset rules and constraints at any point in time.

[0097] For the target data after initial screening, the executing entity can further use anomaly detection algorithms to identify abnormal data (such as abnormal data in value range, abnormal data in distribution, etc.) in the target data after initial screening, and remove the abnormal data to obtain the target data after secondary screening.

[0098] Anomaly detection algorithms can include Isolation Forest and AutoEncoder algorithms.

[0099] Furthermore, the executing entity can filter the target data after secondary screening based on at least one of the following: consistency of target data associated with the same point of interest, consistency between target data and associated point of interest, and timestamp deviation of target data, to obtain the filtered target data, which is the final filtered target data.

[0100] Among them, the consistency of target data associated with the same point of interest can be characterized by the similarity of target data associated with the same point of interest, the consistency between target data and associated point of interest can be characterized by the similarity between target data and associated point of interest, and the timestamp deviation of target data can be characterized by the difference between the timestamp carried by the target data and the actual time of occurrence / generation / processing of the data (or the preset standard time).

[0101] The executing entity can directly filter out target data with similarity less than the first preset value or difference less than the second preset value from the target data after secondary filtering, and obtain the filtered target data.

[0102] If the objects for similarity calculation are all text data, algorithms such as Jaccard similarity and cosine similarity can be used. If the objects for similarity calculation are both text data and image data, multimodal models such as CLIP model and DINOv2 (Data-efficient Image Transformer v2, two-stage implicit object-oriented network) model can be used.

[0103] Furthermore, for conflicting or complex similarity or difference calculation results, crowdsourced voting can be triggered, and the final similarity or difference will be determined by combining the "weighted voting" results.

[0104] Furthermore, the implementing entity can construct a data access system to perform data access, data cleaning, and spatiotemporal field correction on the filtered target data.

[0105] Data access methods may include API (Application Programming Interface) integration, incremental subscription, map-based data tracking, asynchronous data retrieval, and UGC data collection.

[0106] Data cleaning methods can include text cleaning and image cleaning. Text cleaning can include deduplication (e.g., using MinHash+Locality Sensitive Hashing, i.e., minimum hash + locality sensitive hash to quickly remove duplicate records), error correction (e.g., spell correction and rectification based on language models), and sensitive information desensitization (e.g., using custom dictionaries and regular expressions to clean personal privacy information, such as mobile phone numbers and ID cards, and combining text classifiers to detect potentially sensitive text).

[0107] Spatiotemporal field correction methods can include various approaches, such as trajectory fitting and business hours correction. Trajectory fitting methods can include various approaches, such as using Kalman filtering or SMA (Simple Moving Average) curves to smooth and fill in missing coordinates on GPS (Global Positioning System) trajectories uploaded by users. Business hours correction methods can include various approaches, such as combining historical store opening and closing time series and surrounding business district behavior patterns with LSTM (Long Short-Term Memory) network predictions to correct abnormal time periods.

[0108] Furthermore, the executing entity can format the target data after data access, data cleaning, and spatiotemporal field correction, that is, achieve structural uniformity through relevant field mapping and standardization, which may include address hierarchical and category uniformity.

[0109] Specifically, address hierarchy can be achieved by using an address resolution library to break down the levels into provinces, cities, towns, and streets; category unification can be achieved by adopting an industry classification system and building a second- and third-level category mapping rule library.

[0110] Furthermore, entity description information is extracted from the formatted target data.

[0111] This implementation method filters target data based on at least one of the following: consistency of target data associated with the same point of interest, consistency between target data and associated points of interest, and timestamp deviation of target data. Entity description information is extracted from the filtered target data, so that entity description information can be generated based on target data with broad coverage, low noise, and uniform structure, thereby improving the accuracy and reliability of the extracted entity description information.

[0112] In some alternative approaches, the method also includes updating the geographic knowledge graph based on a streaming pipeline.

[0113] In this implementation, after generating the geographic knowledge graph, the executing entity can update the geographic knowledge graph using a streaming pipeline-based processing method.

[0114] Here, the core of updating the geographic knowledge graph using a streaming pipeline-based processing approach is to build a real-time / near real-time, continuously automated processing workflow that transforms dynamically generated streaming data (such as logs, sensor data, real-time API feedback, database changes, etc.) into data acceptable to the knowledge graph and completes incremental updates (additions, modifications, and deletions) to the existing graph.

[0115] Specifically, when new data (such as new comments uploaded by users, updated business hours of merchants, new activity information captured from social platforms, etc.), i.e. streaming data, is generated, the data processing flow is triggered. Processing steps such as data cleaning, content extraction, interest point attachment, association confidence, and information confidence update are organized in sequence. That is, the above-mentioned geographic knowledge graph construction steps are performed on the new data, and an appropriate update strategy is automatically selected according to the data type.

[0116] To ensure the concurrency and reliability of data processing, a multi-message queue collaboration mechanism can be introduced for data processing in high-concurrency scenarios. For example, when a landmark receives a large number of comments from users simultaneously saying "Closed today," the surge in popularity can be detected through message backlog, and the attributes of relevant geographic knowledge graph nodes can be updated immediately, while generating the latest text and image notes.

[0117] In addition, the update process can record processing status and context information, and replay data after an anomaly occurs to ensure the final consistency and integrity of the geographic knowledge graph.

[0118] This approach achieves timely and effective updates to the geographic knowledge graph by updating it using a streaming pipeline.

[0119] Further reference Figure 3 This illustrates a flow 300 of an embodiment of the information generation method. In this embodiment, the flow 300 of the information generation method may include the following steps: Step 301: Based on the travel intention information, determine the image description information and text description information of the target point of interest in the geographic knowledge graph.

[0120] In this embodiment, the executing entity can directly obtain the travel intention information input by the user, or it can perform semantic parsing based on the travel theme input by the user, such as "family day trip", "hiking and photography", "late-night food route", etc., to determine the travel intention information. This application does not limit this.

[0121] Here, the implementing entity can determine travel intent information solely based on the semantic parsing results of the travel theme, or it can determine travel intent information jointly based on the semantic parsing results of the travel theme, the user's interests and preferences, the user's consumption habits, and historical online behavior (such as browsing history, collection history, and shopping history).

[0122] The travel intention information may include points of interest, time of day, budget, and transportation preferences.

[0123] Furthermore, the implementing entity can retrieve geographic knowledge graphs based on travel intent information to determine the fused description information of target points of interest. The fused description information can include image description information and text description information of the target points of interest.

[0124] Here, a geographical knowledge graph is like this: Figure 2 The geographic knowledge graph obtained by the method described in the corresponding implementation example will not be repeated here.

[0125] Step 302: Generate a text-based travel plan based on travel intention information and text description information.

[0126] In this embodiment, the executing entity can directly constrain and guide the text description information based on the travel intention information to generate a text travel plan; alternatively, the travel intention information and the text description information can be input into a preset first language model to generate a text travel plan, and this application does not limit this.

[0127] Here, in the process of constraining and guiding the text description information based on travel intention information, the implementing entity can further introduce preset content templates (such as single point of interest focus, multi-point of interest route connection, real-time event tracking and broadcasting, etc.) to guide the generation of travel planning in the constrained text.

[0128] The content templates can be used to constrain the chapter order of travel plans in texts, and to emphasize key points.

[0129] Step 303: Generate a text-based travel plan based on travel intention information, text travel planning, and image description information.

[0130] In this embodiment, the executing entity can directly constrain and guide the image description information based on the travel intention information and text travel plan to generate a text-image travel plan; alternatively, the travel intention information, text travel plan, and image description information can be input into a preset second language model to generate a text-image travel plan. This application does not limit this approach.

[0131] Here, in the process of constraining and guiding text-based travel planning and image description information based on travel intention information, the implementing entity can further introduce preset image and text templates and copyright requirements to guide and constrain the generation of image and text travel plans.

[0132] Among them, the graphic templates can be used to constrain the graphic structure, image location, and number of images in graphic travel planning; the copyright requirements are used for copyright protection and can be achieved by embedding specific identification information in the image, such as automatically overlaying "small text copyright watermark + generation mark" in the lower left or lower right corner of the image.

[0133] The above embodiments of this disclosure determine the image description information and text description information of the target point of interest in the geographic knowledge graph based on the travel intention information; generate a text travel plan based on the travel intention information and text description information; and generate a graphic travel plan based on the travel intention information, text travel plan, and image description information. That is, the text travel plan is generated first, and then the image description information is constrained and guided by the full text of the text travel plan to generate a graphic travel plan. This takes into account the overall impact of the text travel plan on image location, number of images, etc., and improves the accuracy of the generated graphic travel plan.

[0134] In some optional approaches, text-based travel planning is generated based on travel intent information and text description information, including: inputting travel intent information and text description information into a first language model to generate text-based travel planning; and image-based travel planning is generated based on travel intent information, text-based travel planning, and image description information, including: inputting travel intent information, text-based travel planning, and image description information into a second language model to generate image-based travel planning.

[0135] In this implementation, the executing entity can input travel intent information, text description information, and preset content templates into the first language model to generate a text travel plan; and input travel intent information, text travel plan, image description information, preset image and text templates, and preset copyright requirements into the second language model to generate an image and text travel plan.

[0136] Among them, the first and second largest language models are both large language models.

[0137] Large language models typically have a large number of parameters and learn the structure and semantics of language by training on large amounts of text data. A large language model can include an input layer, an embedding layer, an encoder / decoder, a feedforward neural network, and an output layer.

[0138] The input layer decomposes the input information into tokens and maps them to token identifiers, which are then input into the embedding layer. The embedding layer transforms the token identifiers into low-dimensional dense vectors and adds positional encoding to obtain processed data, which is then input into the encoder / decoder. The encoder / decoder uses a multi-layered stacked attention mechanism and feedforward neural network to deeply mine the contextual semantic relationships of the text, and inputs the mined data into the feedforward neural network. The feedforward neural network performs nonlinear transformations on the mined data and inputs the nonlinearly transformed data into the output layer. The output layer uses a linear layer and a softmax function to transform the nonlinearly transformed data into a token probability distribution, and uses the token with the highest probability as the generated result.

[0139] In addition, large language models can also leverage RAG (Retrieval-Augmented Generation) to enhance their generation capabilities. For the first large language model, RAG can be used to enhance the generated text to ensure key facts, such as business hours, tickets, transportation methods, and the latest events, with zero illusions. For the second large language model, RAG can be used to complete the generation of images from specific perspectives or scenes that are missing in the image description information.

[0140] Furthermore, for the second largest language model, existing map open platforms can be used to draw route / heatmap views and render them as images.

[0141] This approach involves inputting travel intent information and textual description information into a first large language model to generate a text-based travel plan, and then inputting the travel intent information, textual travel plan, and image description information into a second large language model to generate a graphic-text travel plan. This fully leverages the large model's superior understanding and processing capabilities for complex language structures, thereby improving the accuracy of the generated graphic-text travel plan.

[0142] In some alternative approaches, the method further includes: generating quality assessment information for the map-based travel plan based on user feedback; and publishing the map-based travel plan to a map application in response to determining that the quality assessment information meets specified conditions.

[0143] In this implementation, after generating a text and image travel plan, the executing entity can present the text and image travel plan and receive feedback from users regarding the presented text and image travel plan.

[0144] Feedback information can include online user feedback and offline human evaluation. Online user feedback can include dwell time, like rate, and error correction rate; offline human evaluation can include the content's accuracy in matching facts.

[0145] Furthermore, the implementing entity can determine whether the quality assessment information meets the specified conditions. If it does, the travel plan can be published to the map application's recommendation, guide, attachment, and other channels.

[0146] The specified conditions can be set according to actual needs, such as matching ratio greater than preset value, dwell time greater than preset time, like rate greater than preset like rate threshold, error correction rate less than preset error correction rate threshold, etc.

[0147] This implementation generates quality assessment information for the map and text travel plan based on user feedback. In response to determining that the quality assessment information meets specified conditions, the map and text travel plan is published to the map application. That is, the map and text travel plan is only published when it meets the specified conditions, thus improving the accuracy of the published map and text travel plan.

[0148] See also Figure 4 , Figure 4 This is a schematic diagram of an application scenario of the information generation method according to this embodiment.

[0149] During the data mining and collection phase, for the target data 403 from industry users 401 and general users 402, namely text data, image data, multimedia data and spatiotemporal data, the executing entity can filter the target data 404 to obtain the filtered target data.

[0150] In the knowledge graph construction phase, 405 entity description information is extracted from the filtered target data. Further, associated points of interest (POIs) are identified based on the entity description information, i.e., POI linking is performed 406, and entity description information corresponding to the same POI is fused 407 to obtain fused description information corresponding to the POI. A geographic knowledge graph is then constructed based on the POIs and their corresponding fused description information 408. Here, POIs are nodes in the geographic knowledge graph, and the fused description information corresponding to the POIs is used to determine the relationships between nodes. The fused description information can include both textual and image description information.

[0151] The text description information can include various types, such as name, coordinates, status, ticket, recommended dishes, reservation status, travel reminders, travel suggestions, precautions, reviews, rankings, routes, etc.; the image description information can include various types, such as images of attractions, images of dishes, etc.

[0152] During the image and text travel planning generation stage, after obtaining the user's travel theme 409, the executing entity can perform semantic analysis based on the user's input travel theme, such as "family day trip", "hiking and photography", "late-night food route", etc., to determine the travel intention information, and then search the geographic knowledge graph 410 based on the travel intention information to determine the image description information and text description information of the target point of interest.

[0153] Furthermore, based on travel intention information, text description information, image description information, image and text templates, and a preset large language model 411, an image and text travel plan 412 is generated.

[0154] Further reference Figure 5 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of a geographic knowledge graph construction device, which is similar to... Figure 2 The method embodiments shown correspond to those described.

[0155] like Figure 5 As shown, the geographic knowledge graph construction device 500 in this embodiment includes: an extraction module 501, an association module 502, a fusion module 503, and a construction module 504.

[0156] The extraction module 501 can be configured to extract entity description information from the target data.

[0157] The association module 502 can be configured to determine the associated points of interest associated with the entity description information.

[0158] The fusion module 503 can be configured to fuse entity description information corresponding to the same associated point of interest to obtain fused description information corresponding to the associated point of interest.

[0159] Module 504 can be configured to construct a geographic knowledge graph based on associated points of interest and the fused description information corresponding to the associated points of interest.

[0160] In some optional embodiments of this example, the extraction module further includes a data extraction unit, a data alignment unit, and a data correction unit. The data extraction unit is configured to extract initial description information based on the image frame data and audio data included in the video data in response to determining that the multimedia data is video data. The data alignment unit is configured to align the audio data and image frame data based on the timestamps included in the video data to obtain aligned data. The data correction unit is configured to correct and supplement the initial description information based on the aligned data to generate entity description information.

[0161] In some optional embodiments, the association module further includes a first determining unit, a second determining unit, and a data association unit. The first determining unit is configured to determine an initial set of points of interest associated with the entity description information based on at least one of the geographic coordinates, name, and administrative region code of the entity description information. The second determining unit is configured to determine the association confidence between the entity description information and the initial points of interest based on the feature information of the entity description information and the initial points of interest. The data association unit is configured to determine the associated points of interest associated with the entity description information from the initial set of points of interest based on the association confidence.

[0162] In some optional embodiments of this example, the data association unit is further configured to determine the associated interest points that are associated with the entity description information from the set of initial interest points based on the association confidence, the current popularity of the initial interest points, and the historical association frequency between the initial interest points and the entity description information.

[0163] In some optional embodiments of this example, the fusion module further includes a data determination unit and a data fusion unit. The data determination unit is configured to determine the information confidence level of entity description information corresponding to the same associated point of interest based on at least one of source credibility, information timeliness, and consistency of multi-source feedback. The data fusion unit is configured to fuse entity description information whose information confidence level meets preset conditions to obtain fused description information corresponding to the associated point of interest.

[0164] In some optional embodiments of this example, the extraction module includes a data filtering unit and a target extraction unit. The data filtering unit can be configured to filter the target data based on at least one of the following: consistency of target data associated with the same point of interest, consistency between the target data and the associated point of interest, and timestamp deviation of the target data. The target extraction unit can be configured to extract entity description information from the filtered target data.

[0165] In some optional embodiments of this invention, the device further includes a first storage module and a second storage module. The first storage module is configured to directly store the structured data in the associated points of interest and the fused description information corresponding to the associated points of interest into a graph database. The second storage module is configured to encode the unstructured data in the associated points of interest and the fused description information corresponding to the associated points of interest into vectors and store the vectors into a vector database.

[0166] In some alternative embodiments of this embodiment, the apparatus further includes an update module configured to update the geographic knowledge graph based on a streaming pipeline.

[0167] Further reference Figure 6 As an implementation of the methods shown in the above figures, this disclosure provides an embodiment of an information generation apparatus, which is similar to...Figure 4 The method embodiments shown correspond to those described.

[0168] like Figure 6 As shown, the information generation device 600 of this embodiment includes: a determination module 601, a planning module 602, and a generation module 603.

[0169] The acquisition module 601 can be configured to determine the image and text description information of the target point of interest in a geographic knowledge graph based on travel intention information. The geographic knowledge graph is as follows: Figure 2 The geographic knowledge graph obtained by the method described in the corresponding implementation example will not be repeated here.

[0170] The planning module 602 can be configured to generate a text travel plan based on travel intention information and the text description information.

[0171] The generation module 603 can be configured to generate a graphic travel plan based on travel intention information, text travel planning, and image description information.

[0172] The acquisition, storage, and application of user personal information involved in the technical solution disclosed herein comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0173] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0174] Figure 7 A schematic block diagram of an example electronic device 700 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0175] like Figure 7As shown, device 700 includes a computing unit 701, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 702 or a computer program loaded from storage unit 708 into random access memory (RAM) 703. RAM 703 may also store various programs and data required for the operation of device 700. The computing unit 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0176] Multiple components in device 700 are connected to I / O interface 705, including: input unit 706, such as keyboard, mouse, etc.; output unit 707, such as various types of monitors, speakers, etc.; storage unit 708, such as disk, optical disk, etc.; and communication unit 709, such as network card, modem, wireless transceiver, etc. Communication unit 709 allows device 700 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0177] The computing unit 701 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 701 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 701 performs the various methods and processes described above, such as methods for constructing geographic knowledge graphs. For example, in some embodiments, the methods for constructing geographic knowledge graphs may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 708. In some embodiments, part or all of the computer program may be loaded and / or installed on device 700 via ROM 702 and / or communication unit 709. When the computer program is loaded into RAM 703 and executed by the computing unit 701, one or more steps of the methods for constructing geographic knowledge graphs described above may be performed. Alternatively, in other embodiments, computing unit 701 may be configured to perform methods for constructing geographic knowledge graphs by any other suitable means (e.g., by means of firmware).

[0178] Various embodiments of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), systems-on-a-chip (SoCs), payload-programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0179] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0180] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0181] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0182] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., a communication network). Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0183] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is established by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, also known as cloud computing servers or cloud hosts, which are hosting products within the cloud computing service ecosystem to address the management difficulties and weak business scalability inherent in traditional physical hosts and Virtual Private Servers (VPS) services. Servers can also be categorized as distributed system servers or servers incorporating blockchain technology.

[0184] According to the technical solution of the embodiments of this disclosure, a semantically rich and structurally stable geographic knowledge graph is constructed.

[0185] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution provided in this disclosure can be achieved, and this is not limited herein.

[0186] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for constructing a geographic knowledge graph, comprising: extracting entity description information from target data, the target data comprising image data, text data, multimedia data, and spatiotemporal data generated by semantic understanding of at least one of the image data, the text data, and the multimedia data; determining an associated interest point associated with the entity description information; fusing entity description information corresponding to the same associated interest point to obtain fused description information corresponding to the associated interest point; constructing a geographic knowledge graph based on the associated interest point and the fused description information corresponding to the associated interest point. 2.The method of claim 1, wherein the determining an associated interest point associated with the entity description information comprises: determining an initial interest point set associated with the entity description information based on at least one of geographical coordinates, a name, and an administrative code of the entity description information; determining an association confidence of the entity description information and the initial interest point based on the entity description information and feature information of the initial interest point; determining the associated interest point associated with the entity description information from the initial interest point set based on the association confidence.

3. The method of claim 2, wherein, The determining the associated interest point associated with the entity description information from the initial interest point set based on the association confidence comprises: determining the associated interest point associated with the entity description information from the initial interest point set based on the association confidence, a current popularity of the initial interest point, and a historical association frequency of the initial interest point and the entity description information.

4. The method of claim 1, wherein, The extracting entity description information from target data comprises: in response to determining that the multimedia data is video data, extracting initial description information based on image frame data and audio data included in the video data; aligning the audio data and the image frame data based on a timestamp included in the video data to obtain aligned data; correcting and supplementing the initial description information based on the aligned data to generate entity description information.

5. The method of claim 1, wherein, The fusing entity description information corresponding to the same associated interest point to obtain fused description information corresponding to the associated interest point comprises: determining an information confidence of the entity description information corresponding to the same associated interest point based on at least one of source credibility, information timeliness, and multi-source feedback consistency; fusing the entity description information whose information confidence meets a preset condition to obtain the fused description information corresponding to the associated interest point.

6. The method of claim 1, wherein, The extracting entity description information from target data comprises: filtering target data based on at least one of target data consistency associated with the same interest point, consistency between the target data and the associated interest point, and timestamp deviation of the target data; extracting entity description information from the filtered target data. 7.The method of claim 1, further comprising: storing structured data in the associated interest point and the fused description information corresponding to the associated interest point directly into a graph database. Encode the associated interest point and the unstructured data in the fusion description information corresponding to the associated interest point into a vector, and store the vector into a vector database.

8. The method of any one of claims 1-7, further comprising: updating the geographic knowledge graph based on a streaming pipeline.

9. An information generation method, comprising: determining image description information and text description information of a target interest point in a geographic knowledge graph according to travel intention information, the geographic knowledge graph being obtained by the method of any one of claims 1-8; generating a text travel plan based on the travel intention information and the text description information; generating a graphic-text travel plan based on the travel intention information, the text travel plan, and the image description information.

10. The method of claim 9, wherein, The generating a text travel plan based on the travel intention information and the text description information comprises: inputting the travel intention information and the text description information into a first large language model to generate a text travel plan; and The generating a graphic-text travel plan based on the travel intention information, the text travel plan, and the image description information comprises: inputting the travel intention information, the text travel plan, and the image description information into a second large language model to generate a graphic-text travel plan.

11. The method of any one of claims 9-10, further comprising: generating quality evaluation information of the graphic-text travel plan based on feedback information of the graphic-text travel plan from a user; in response to determining that the quality evaluation information meets a specified condition, publishing the graphic-text travel plan to a map application.

12. A geographic knowledge graph construction apparatus, comprising: an extraction module configured to extract entity description information in target data, the target data comprising image data, text data, multimedia data, and spatiotemporal data generated by semantic understanding of at least one of the image data, the text data, and the multimedia data; an association module configured to determine an associated interest point associated with the entity description information; a fusion module configured to fuse entity description information corresponding to the same associated interest point to obtain fusion description information corresponding to the associated interest point; a construction module configured to construct a geographic knowledge graph based on the associated interest point and the fusion description information corresponding to the associated interest point.

13. An information generation apparatus, comprising: a determination module configured to determine image description information and text description information of a target interest point in a geographic knowledge graph according to travel intention information, the geographic knowledge graph being obtained by the method of any one of claims 1-8; a plan module configured to generate a text travel plan based on the travel intention information and the text description information; a generation module configured to generate a graphic-text travel plan based on the travel intention information, the text travel plan, and the image description information.

14. An electronic device, comprising: comprising: at least one processor; and a memory communicatively connected to the at least one processor; wherein, The memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-11.

15. A non-transitory computer-readable storage medium having stored thereon computer instructions, wherein, The computer instructions are for causing the computer to perform the method of any one of claims 1-11.

16. A computer program product comprising a computer program which, when executed by a processor, implements the method of any one of claims 1-11.

Images