A processing method and device for a POI label

Abstract

The present disclosure provides a processing method and device for POI labels, relates to the field of artificial intelligence, and particularly relates to the field of large language model technology. The specific implementation scheme is as follows: obtaining multiple original labels of a same POI; dividing the multiple original labels into multiple original label sets based on semantic similarity; determining multiple candidate labels for describing a same attribute of the POI or a same service provided by the POI in each original label set by using an aggregated large model, and determining a target label meeting a preset information quality condition from the multiple candidate labels, the target label being used for being displayed in an electronic map to describe the POI or being used for matching POI retrieval information to recall the POI. In this way, the accuracy, representativeness and availability of the target label used for the electronic map display or the POI retrieval are improved.

Description

Technical Field

[0001] This disclosure relates to the field of artificial intelligence, and more particularly to the field of large language model technology. Specifically, this disclosure relates to a method and apparatus for processing POI tags. Background Technology

[0002] Points of Interest (POI) tags are crucial data describing their subcategories, functional attributes, scenario characteristics, service capabilities, and customer / customer reviews. Their quality directly impacts the accuracy of POI retrieval and user experience. Currently collected POI tags frequently exhibit semantic redundancy. Existing methods primarily cluster the collected tags (e.g., K-means) to merge similar tags, but this approach fails to meet the need for precise tag processing. Summary of the Invention

[0003] This disclosure provides a method and apparatus for processing POI tags to achieve accurate processing of the original tags and improve the quality and accuracy of the target tags.

[0004] According to one aspect of this disclosure, a method for processing POI tags is provided, comprising: Retrieve multiple raw labels for the same POI; The multiple original labels are divided into multiple sets of original labels based on semantic similarity; The aggregation model is used to determine multiple candidate tags within each of the original tag sets that describe the same attribute of the POI or the same service provided by the POI. Target tags that meet preset information quality conditions are then determined from the multiple candidate tags. These target tags are used to display the POI on an electronic map to describe the POI, or to match it with POI retrieval information to recall the POI.

[0005] According to another aspect of this disclosure, a processing apparatus for POI tags is provided, comprising: The acquisition unit is configured to acquire multiple raw tags for the same POI; The partitioning unit is configured to divide the plurality of original labels into a plurality of original label sets based on semantic similarity; The target label determination unit is configured to use an aggregated large model to determine multiple candidate labels within each of the original label sets for describing the same attribute of the POI or the same service provided by the POI, and to determine target labels that meet preset information quality conditions from the multiple candidate labels. The target labels are used to be displayed in an electronic map to describe the POI, or to be used to match with POI retrieval information to recall the POI.

[0006] According to another aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the methods described in the embodiments of this disclosure.

[0007] According to another aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the methods described in embodiments of this disclosure.

[0008] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, implements the methods described in the embodiments of this disclosure.

[0009] This disclosure can obtain multiple original tags for the same POI, divide them based on semantic similarity, and then use an aggregation model to filter out target tags that meet the information quality conditions from each set of original tags. This achieves automated aggregation and quality control of massive and messy original tags, solves the problem of tag duplication caused by the lack of deep semantic understanding in traditional methods, and significantly improves the accuracy, representativeness and usability of the target tags finally used for electronic map display or POI retrieval.

[0010] 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

[0011] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure.

[0012] Figure 1 This is the system architecture diagram to which this disclosure applies.

[0013] Figure 2 This is a flowchart of the POI tag processing method provided in this disclosure.

[0014] Figure 3 This is a schematic diagram of the process for determining target labels provided in this publication.

[0015] Figure 4 This is a schematic diagram of the correlation identification process provided in this disclosure.

[0016] Figure 5 This is a schematic diagram of the label classification process provided in this publication.

[0017] Figure 6 This is a schematic diagram of the process for processing POI tags provided in this publication.

[0018] Figure 7 This is a schematic block diagram of a processing apparatus for POI tags provided in this disclosure.

[0019] Figure 8 This is a block diagram of an electronic device used to implement the POI tag processing method of the embodiments of this disclosure. Detailed Implementation

[0020] 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.

[0021] The terminology used in the embodiments of this invention is for the purpose of describing particular embodiments only and is not intended to limit the invention. The singular forms “a,” “the,” and “the” as used in the embodiments of this invention and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise.

[0022] It should be understood that the term "and / or" used in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0023] Depending on the context, the word "if" as used here can be interpreted as "when," "when," "in response to determination," or "in response to detection." Similarly, depending on the context, the phrase "if determination" or "if detection (of the stated condition or event)" can be interpreted as "when determination," "in response to determination," "when detection (of the stated condition or event)," or "in response to detection (of the stated condition or event)."

[0024] To sift through massive amounts of raw tags to extract accurate and standardized target tags, related technologies typically employ clustering schemes. Specifically, this scheme encodes the raw tags into static text vectors and uses traditional clustering algorithms (such as K-means) to group the text vectors based on the distance between them, thus merging semantically repetitive tags. Its design aims to leverage the automated processing capabilities of clustering algorithms to achieve basic tag cleaning and deduplication with relatively low computational cost.

[0025] However, this approach falls short when applied to addressing the quality issues of tags with diverse expressions, complex contexts, and implicit deep semantic relationships. Specifically, when distinguishing between semantically similar but differently expressed tags such as "children's playground" and "family playground," geometric distance-based clustering methods struggle to make accurate judgments. Therefore, traditional methods rely on shallow semantic representations and lack a deep understanding of the ambiguity and contextual dependence of natural language, failing to meet the demands for high-quality and accurate processing of original tags.

[0026] In view of this, this disclosure provides a new approach. To facilitate understanding of this disclosure, the system architecture on which this disclosure is based will first be described. Figure 1 Exemplary system architectures that can be applied to embodiments of this disclosure are shown, such as Figure 1 As shown, the system architecture may include: a client and a server.

[0027] The server side and the client side are the two main components of an application service. The server side uses a server as its primary hardware infrastructure and may include one or more software service modules. The server side and the client side form a collaborative front-end and back-end.

[0028] The client can be set on the terminal device. In this embodiment of the disclosure, the client can be a local application, a mini-program, or a web application running through a browser on the terminal device.

[0029] Terminal devices can include, but are not limited to, smart mobile terminals, wearable devices, PCs (Personal Computers), and smart home devices. Smart mobile devices can include devices such as mobile phones, tablets, laptops, PDAs (Personal Digital Assistants), and connected car terminals. Wearable devices can include devices such as smartwatches, smart glasses, smart bracelets, VR (Virtual Reality) devices, AR (Augmented Reality) devices, and mixed reality devices (devices that support both virtual and augmented reality). Smart home devices can include devices such as smart TVs and smart refrigerators with displays.

[0030] A server can be a single server, a server cluster consisting of multiple servers, or a cloud server. A cloud server, also known as a cloud computing server or cloud host, is a hosting product in the cloud computing service system, designed to address the shortcomings of traditional physical hosts and Virtual Private Servers (VPS) services, such as high management difficulty and weak service scalability.

[0031] It should be understood that Figure 1 The number of client and server components shown is merely illustrative. Depending on implementation needs, there can be any number of client and server components.

[0032] As one implementation, the server can obtain multiple original tags for the same POI, divide these tags into multiple sets based on semantic similarity, and then use an aggregation model to determine multiple candidate tags within each set that describe the same attribute of the POI or the same service provided by the POI. Finally, it determines the target tag that meets preset information quality conditions from among these candidate tags. The client responds to the user triggering the POI details display function on the electronic map, obtains the target tag of the POI from the server, and displays it.

[0033] Figure 2 This is a flowchart illustrating a method for processing POI tags provided in an embodiment of this disclosure. This method for processing POI tags can be implemented by... Figure 1 The server-side execution in the system shown. For example... Figure 2 As shown, the method may include the following steps: Step 201: Obtain multiple original labels for the same POI.

[0034] Step 202: Divide the multiple original tags into multiple original tag sets based on semantic similarity.

[0035] Step 203: Use the aggregated large model to determine multiple candidate tags in each original tag set that describe the same attribute of POI or the same service provided by POI, and determine the target tag that meets the preset information quality conditions from the multiple candidate tags. The target tag is used to display in the electronic map to describe POI, or to match with POI retrieval information to recall POI.

[0036] As can be seen from the above process, this disclosure can obtain multiple original tags of the same POI, divide them based on semantic similarity, and then use an aggregation model to filter out target tags that meet the information quality conditions from each set of original tags. This achieves automated aggregation and quality control of massive and messy original tags, solves the problem of tag duplication caused by the lack of deep semantic understanding in traditional methods, and significantly improves the accuracy, representativeness and usability of the target tags finally used for electronic map display or POI retrieval.

[0037] The following describes in detail each step of the above process and the effects that can be further produced, with reference to the embodiments.

[0038] First, the above step 201, namely "obtaining multiple original tags of the same POI", will be described in detail with reference to the embodiments.

[0039] In the embodiments of this disclosure, a POI (Point of Interest) refers to a geographic entity with specific geographic coordinates and attribute information. It is widely used in geographic information systems, navigation, LBS (Location-Based Services), and other fields. A POI can be a restaurant, gas station, tourist attraction, hospital, school, etc.

[0040] A POI's tag can refer to any textual information used to describe or characterize a POI, aiming to enrich the POI's attribute description, service capabilities, or user perception. POI tags can originate from extracting keywords from user reviews, filtering key attributes from the POI's basic information, or manual annotation of the POI by business personnel.

[0041] Specifically, a POI's tag can be a word (such as "Wi-Fi"), a phrase (such as "good for taking photos"), or a short sentence (such as "the signature dish is Peking duck"). Its content can involve the POI's subcategories, functional attributes, scene characteristics, service capabilities, and customer reviews. For example, for a restaurant POI, its tags can include "Sichuan cuisine", "reservations available", "private rooms available", "good service", "150 yuan per person", etc.

[0042] As an example, user reviews of the same POI can be collected first, and then keywords describing the POI can be extracted based on the review information. These keywords can then be used as the original tags for the POI.

[0043] In some embodiments, after obtaining the original tags for the same POI, they can be preprocessed. Preprocessing may include data cleaning and / or standardization. Data cleaning may include removing at least one of invalid characters (such as anomalous symbols), garbled text, or stop words, while standardization may include case conversion (such as converting all English letters to lowercase) and / or traditional / simplified Chinese conversion (such as converting traditional Chinese tags to simplified Chinese). These preprocessing procedures help reduce noise and provide cleaner, more accurate input for subsequent semantic analysis.

[0044] The following describes step 202, namely "dividing multiple original tags into multiple sets of original tags based on semantic similarity", in detail with reference to the embodiments.

[0045] In this embodiment of the disclosure, multiple original tags can be divided into multiple sets of original tags based on semantic similarity. For example, the semantic similarity between two original tags can be determined by whether there are repeated words or the number of repeated words, or it can be determined based on the distance between two original tags. Then, based on the pre-set semantic similarity, multiple original tags are divided into multiple sets of original tags. The purpose is to initially group semantically similar or related original tags, creating conditions for subsequent fine aggregation.

[0046] Specifically, multiple original labels can be semantically encoded to obtain multiple semantic vectors corresponding to the original labels, and the semantic similarity between the multiple semantic vectors can be determined. Then, based on the semantic similarity, the multiple original labels can be divided into multiple sets of original labels.

[0047] As one feasible implementation, multiple original labels can be input into the embedding model to obtain multiple semantic vectors corresponding to the multiple original labels. The semantic similarity between semantic vectors is determined by calculating the cosine similarity between the semantic vectors, and an original label set is formed based on a pre-set first similarity threshold.

[0048] As another feasible implementation, multiple original labels can be input into the embedding model to obtain multiple semantic vectors corresponding to the multiple original labels. Then, clustering algorithms (such as K-means, hierarchical clustering, DBSCAN, etc.) are used to cluster the multiple semantic vectors, and each cluster constitutes a set of original labels.

[0049] By semantically encoding the original labels and dividing them based on the semantic similarity between semantic vectors, the abstract semantic similarity is transformed into a quantifiable calculation, making the division process more objective, accurate and efficient. This provides high-quality input for the subsequent fine processing of large aggregate models and improves the automation and processing efficiency of the entire process.

[0050] The following describes in detail step 203, namely, "using the aggregated large model to determine multiple candidate tags in each original tag set that describe the same attribute of the POI or the same service provided by the POI, and determining the target tag that meets the preset information quality conditions from the multiple candidate tags", with reference to the embodiments.

[0051] In the embodiments disclosed herein, such as Figure 3 As shown, in step 202, multiple original tags are divided into multiple original tag sets based on semantic similarity. Then, the aggregation model is used to determine multiple candidate tags in each original tag set that describe the same attribute of the POI or the same service provided by the POI. And from the multiple candidate tags, the target tag that meets the preset information quality conditions is determined.

[0052] In simple terms, the aggregation model can understand the true semantics of the original tags within a set of original tags, and then determine whether multiple original tags within a set of original tags truly point to the same characteristic of the POI (such as "has private rooms" and "set up private rooms" both describe the existence of private room facilities), and select one or more optimal expressions as output from the original tags that point to the same characteristic.

[0053] It should be noted that information quality conditions can be used to limit the standardization of target labels, that is, the target labels are labels with consistent expression and clear logic. They can also be used to limit the representativeness of target labels, that is, the target labels can reflect the overall meaning of multiple candidate labels.

[0054] As one implementation approach, the aggregated large model can refer to a base large language model (such as Qwen2.5-7B). It can first construct prompt words, for example, the prompt words could be: "Complete one of the following tasks: determine whether multiple original tags in an original tag set describe the same attribute of a POI or the same service provided by the POI; if so, output one or more of the most representative, most standardized, most commonly used, or most concise original tags; if not, determine multiple candidate tags within the original tag set used to describe the same attribute of a POI or the same service provided by the POI, and select one or more of the most representative, most standardized, most commonly used, or most concise candidate tags from among the multiple candidate tags, or output a unified tag that has been merged or restated based on multiple candidate tags as the target tag." Then, the original tags included in an original tag set are input into the aggregated large model, so that the aggregated large model outputs the target tag corresponding to the original tag set based on the constructed prompt words.

[0055] As another implementation, the aggregated large model can be a large language model that has been trained and fine-tuned. This fine-tuned large language model is obtained by fine-tuning the base large language model using aggregated training data. For example, the aggregated training data includes multiple aggregated training samples, each containing a set of labels and a target label sample. The set of labels is input into the base large language model to obtain the predicted target label, minimizing the difference between the predicted target label and the target label sample. This process is then used to train the base large language model, resulting in the aggregated large model. This aggregated large model is suitable for tasks involving identifying semantic repetition and selecting the best representative label. Using this approach, the aggregated large model can be more accurately adapted to the specific domains and needs of POI label aggregation.

[0056] For example, the original tags include "roast duck is delicious", "roast duck is highly rated", "barbecue is delicious", "the shop is clean", "the experience is great", "there is a toilet", "Wi-Fi is available", etc. Based on semantic similarity, the multiple original tags are divided into multiple sets of original tags.

[0057] If one of the original tag sets is ["Roast duck is delicious", "Roast duck is highly rated", "Barbecue is delicious"], the aggregation model is used to identify multiple candidate tags within this original tag set that describe the same attribute of the POI or the same service provided by the POI. In other words, the aggregation model is used to further divide the original tags within this original tag set, grouping the original tags describing the same attribute of the POI or the same service provided by the POI into one aggregation class. For the original tag set ["Roast duck is delicious", "Roast duck is highly rated", "Barbecue is delicious"], the aggregation model can obtain two aggregation classes: one aggregation class is ["Roast duck is delicious", "Roast duck is highly rated"] (the original tags included in this aggregation class are the multiple candidate tags corresponding to this aggregation class), and the other aggregation class is ["Barbecue is delicious"] (the original tags included in this aggregation class are the multiple candidate tags corresponding to this aggregation class).

[0058] Furthermore, the aggregated large model is used to output the target label that meets the preset information quality conditions among multiple candidate labels within each aggregated class. That is, for the aggregated class ["Roast duck is delicious", "Roast duck is good"], the output target label can be "Roast duck is delicious"; for the aggregated class ["Barbecue is delicious"], the output target label can be "Barbecue is delicious".

[0059] If one of the original tag sets is ["Roast duck is delicious", "Roast duck is highly recommended"], the aggregation model is used to find that all the original tags in the original tag set are used to describe the same attribute of the POI or the same service provided by the POI. That is, all the original tags in the original tag set are used as multiple candidate tags, and the aggregation model is used to output the target tag that meets the preset information quality conditions among the multiple candidate tags. That is, the output target tag is "Roast duck is delicious".

[0060] In the above example, the output target label can be retained as one or multiple. For example, when multiple candidate labels are related but have slightly different focuses (such as "cheap price" and "high cost performance"), the aggregated large model can consider that both meet the information quality conditions and thus retain them.

[0061] It's important to note that target tags are used to describe Points of Interest (POIs) on the electronic map or to match POI retrieval information to recall POIs. For example, after processing, target tags such as "delicious roast duck," "private rooms available," and "reservations available" can be associated with corresponding restaurant-related POIs. When a user searches for "delicious roast duck restaurant" on the electronic map displayed on their device, these target tags can serve as matching criteria, effectively recalling the restaurant-related POI. Simultaneously, these target tags can also be displayed to the user on the restaurant-related POI's details page, providing information for reference.

[0062] To further optimize the quality of the target labels in the above embodiments, this application can further filter the target labels obtained by using the aggregated large model to remove abnormal labels.

[0063] The abnormal tags include at least one of the following: target tags containing preset abnormal words; target tags belonging to the blacklist corresponding to the POI; target tags whose relevance to the information of the POI does not meet the relevance condition; tags whose credibility does not meet the credibility condition among multiple target tags with semantic conflicts; tags of sub-POIs contained in the target tags, wherein the geographic object corresponding to the sub-POI is located within the spatial range of the geographic object corresponding to the POI; and target tags that, after tag classification processing, fail to be classified into a preset tag category.

[0064] By filtering target tags from multiple dimensions, the system identifies and removes various abnormal situations such as preset abnormal words, blacklists, insufficient relevance, semantic conflicts, sub-POI tags, and unclassified tags. This further purifies the output based on aggregation, effectively solving the problems of narrow coverage and fixed rules in traditional quality control methods, and comprehensively improving the accuracy, security, and consistency of the filtered target tags.

[0065] Next, the exception labels for each abnormal situation will be explained.

[0066] The first method involves identifying target tags containing predefined anomalous words as anomalous tags. Specifically, anomalous words can be terms such as sensitive words or advertising terms. Target tags are quickly filtered using regular expressions and keyword matching to identify obviously violating tags, which are then identified as anomalous tags and removed from the target tags determined by the aggregated large model.

[0067] The second method involves identifying target tags whose relevance to the POI information does not meet the relevance criteria as anomalous tags. The POI information can refer to at least one of the following: name information (e.g., XX restaurant), attribute information (e.g., food, Chinese restaurant), and address information (e.g., XX district, XX street). Then, a large-scale anomalous tag identification model is invoked to determine the relevance of the target tag to at least one of the name, attribute, and address information. Target tags whose relevance does not meet the relevance criteria are identified as anomalous tags. For example, the anomalous tag identification model can determine the relevance of each target tag to at least one of the name, attribute, and address information, and identify target tags with a relevance below a relevance threshold as anomalous tags.

[0068] For example, such as Figure 4 As shown, if the target label of a POI in the restaurant category is "near the sea" or "near the subway", and the address information of the POI in the restaurant category is "XX city (non-coastal city) XX district XX street (next to the subway)", then it can be considered that the relevance of the target label "near the sea" does not meet the relevance condition, that is, "near the sea" is an abnormal label, while the relevance of the target label "near the subway" meets the relevance condition.

[0069] The anomaly label identification model can be a fine-tuned relevance identification model. This fine-tuned relevance identification model is obtained by fine-tuning the base language model using relevance training data. For example, the relevance training data includes multiple relevance training samples. Each relevance training sample includes POI information samples, multiple label samples, and irrelevant label samples. The POI information samples and multiple label samples are input into the base language model to obtain predicted irrelevant labels. This minimizes the difference between the predicted irrelevant labels and the irrelevant label samples. The base language model is then trained to obtain the relevance identification model. This relevance identification model is suitable for identifying labels in the target label that do not meet the relevance condition.

[0070] By calling a specialized large-scale anomaly label recognition model to deeply evaluate the semantic relevance between target labels and POI information (such as name information, attribute information, and address information), the powerful contextual understanding capability of the large-scale anomaly label recognition model can be used to accurately identify labels that are unrelated to or weakly related to the POI topic. This solves the problem of high misjudgment rate based on simple keyword matching or rule-based methods, and significantly improves the accuracy and semantic coverage depth of anomaly label filtering.

[0071] The third method involves identifying the labels of sub-POIs contained in the target label as anomalous labels. Specifically, the spatial extent (such as boundary coordinates) of the geographic object corresponding to the POI is obtained. Combined with Geographic Information System (GIS) data, the sub-POIs located within the spatial extent of the geographic object corresponding to the POI (e.g., canteens and libraries on a university campus, restaurants in a large shopping mall, etc.) are identified, and the labels belonging to the sub-POIs are removed from the target labels of the parent POI.

[0072] For example, if the target tag "Delicious Potato Chicken Nuggets" appears for the POI "XX University", we can combine the spatial range of the geographic object corresponding to "XX University" to determine the sub-POIs (such as the First Canteen) within the spatial range of the geographic object corresponding to the POI. Then we can determine that the target tag is a tag of a sub-POI contained in XX University and filter it out.

[0073] Alternatively, an anomaly label recognition model can be invoked. The POI information and target label are input into the anomaly label recognition model, which outputs the labels of the sub-POIs contained within the target label. This anomaly label recognition model can be a finely tuned sub-POI label recognition model. This finely tuned sub-POI label recognition model is obtained by fine-tuning the base language model using sub-POI label training data. For example, the sub-POI label training data includes multiple sub-POI label training samples. Each sub-POI label training sample includes POI information samples, multiple label samples, and sub-POI label samples. The POI information samples and multiple label samples are input into the base language model to obtain predicted sub-POI labels. The difference between the predicted sub-POI labels and the sub-POI label samples is minimized. The base language model is then trained to obtain the sub-POI label recognition model. This sub-POI label recognition model is suitable for the task of recognizing the labels of sub-POIs contained within the target label.

[0074] The fourth method involves identifying anomalous labels among multiple semantically conflicting target labels whose credibility does not meet the credibility criteria. Specifically, a large-scale anomalous label identification model can be invoked to identify multiple target labels that have semantic conflicts (such as "pet-friendly" and "no pets allowed"). These multiple semantically conflicting target labels are called conflicting labels. Then, the target labels among the conflicting labels whose credibility does not meet the credibility criteria are identified as anomalous labels. If there are multiple conflicting labels, the conflicting labels with credibility less than the credibility threshold can be identified as anomalous labels, or all conflicting labels except the one with the highest credibility can be identified as anomalous labels. For example, if there are only two conflicting labels, the conflicting label with the lower credibility can be identified as an anomalous label.

[0075] When determining the credibility of conflicting tags, one feasible approach is to base the credibility of each conflicting tag on the number of original tags that satisfy the similarity criteria with the conflicting tag. For example, the credibility of each conflicting tag can be determined based on the number of original tags whose similarity to the conflicting tag exceeds a second similarity threshold. Here, the similarity can be determined based on the cosine similarity, distance, etc., between the conflicting tag and the original tags. As a specific implementation, the number of original tags that satisfy the similarity criteria with "pet-friendly" and "no pets allowed" can be determined, and conflicting tags with fewer original tags corresponding to "pet-friendly" and "no pets allowed" can be identified as anomalous tags.

[0076] By calling the abnormal label recognition model to identify semantically conflicting label groups, and retaining or filtering them based on objective data such as the number of original labels that meet similarity conditions, the problem that traditional methods cannot effectively handle complex semantic conflicts is solved, ensuring the consistency of the final target label content and improving the reliability and decision reference value of the target label.

[0077] As another feasible approach, step 202 utilizes a large aggregation model to further aggregate the original tags within the original tag set, resulting in various aggregation classes. Each aggregation class includes multiple candidate tags. The credibility of each conflicting tag can be determined based on the number of candidate tags included in the aggregation class to which each conflicting tag belongs. As a specific implementation, the number of candidate tags included in the aggregation class to which "pet-friendly" belongs, and the number of candidate tags included in the aggregation class to which "pets are prohibited" belongs, can be determined. Conflicting tags with fewer candidate tags corresponding to "pet-friendly" and "pets are prohibited" are identified as anomalous tags.

[0078] Of course, it is also possible to identify the tags whose publication time exceeds the current time in the source information of the conflicting tags as abnormal tags, or to have the abnormal tag identification model directly determine which tag in the conflicting tags is unreasonable based on the POI information.

[0079] The anomaly label recognition model can be a fine-tuned conflict recognition model. This fine-tuned conflict recognition model is obtained by fine-tuning the base language model using conflict label training data. For example, the conflict label training data includes multiple conflict label training samples, each of which includes multiple label samples and conflict label samples. These multiple label samples are input into the base language model to obtain predicted conflict labels. The difference between the predicted conflict labels and the conflict label samples is minimized, and the base language model is then trained to obtain the conflict recognition model. This conflict recognition model is suitable for tasks involving identifying labels with semantic conflicts within the target label.

[0080] The fifth method involves identifying target tags that, after tag classification, fail to be categorized into a preset tag category as abnormal tags. Specifically, the target tags are classified to determine their respective tag categories, and each tag category corresponds to at least one application scenario.

[0081] As one embodiment, tag categories can include basic categories (tags that primarily describe the core categories and attributes of a POI), objective categories (tags that primarily describe objective facts or facilities), and subjective categories (tags derived from user reviews and carrying emotional or experiential connotations). For example, the target tag "provide free WiFi" can be categorized into the objective category, while "authentic taste" can be categorized into the subjective category. For target tags that are not categorized into basic, objective, or subjective categories, or that are categorized into a meaningless category (such as...), the following categories may be considered. Figure 5 As shown in the figure, for example, some garbled characters without clear semantics, words and phrases with overly vague semantics, and words and phrases with overly broad descriptions are identified as abnormal tags.

[0082] Furthermore, each tag category can have subcategories. Specifically, the basic category can include subcategories such as classification tags (used to describe the industry or category to which the POI belongs, such as hot pot restaurant, Sichuan hot pot, chain restaurant), authoritative tags (used to describe certifications, honors, or special marks from official, authoritative institutions or platforms, such as must-eat list, time-honored brand, green restaurant), etc. The objective category can include subcategories such as service facilities (used to describe the specific hardware facilities or services provided by the merchant, such as 24-hour operation, reservations available, private rooms available, parking available, free Wi-Fi, high chairs available, credit card payment accepted), experience scope (used to describe objective facts such as its business model and product line, such as takeout available, dine-in available), etc. The supervisory tag can include subcategories such as service evaluation (evaluation of service personnel and service processes, such as excellent service, very enthusiastic staff, orderly queue management), store environment (feelings about the environment such as decoration, atmosphere, and hygiene, such as clean environment, a bit noisy, modern decoration), etc.

[0083] The aforementioned label classification process can be performed by a large-scale label classification model. Specifically, each target label is input into the large-scale label classification model to obtain the label category or subcategory to which the target label belongs, as output by the model. This large-scale label classification model is obtained by fine-tuning a base large-scale language model using classification label training data. For example, the classification label training data includes multiple classification label training samples, each containing a classification label sample and a true label category. These samples are input into the base large-scale language model to obtain predicted label categories, minimizing the difference between the predicted and true label categories. This process is then used to train the base large-scale language model, resulting in the large-scale label classification model. This model is suitable for tasks involving identifying the label category to which a target label belongs.

[0084] It should be noted that the above content mentions that the tag categories correspond to at least one application scenario. Here, the application scenario can refer to the specific business environment or user demand context in which the POI tag is used or invoked. Specifically, in response to a POI retrieval request under any application scenario, based on the target tags corresponding to each POI under the tag category corresponding to that application scenario, the POIs matching the POI retrieval request are determined. For example, in an electronic map search service, in response to a POI retrieval request in the electronic map search service, the user's selected query term for service facilities (such as parking lot) is matched with the target tags under the objective categories corresponding to each POI. For example, POIs with the target tag "has parking lot" are matched.

[0085] In addition, personalized recommendations can display target tags under the subjective categories of each POI to provide users with more subjective information and help them make decisions. In information display, the target tags of each POI can be displayed separately based on each tag category.

[0086] By standardizing and classifying target tags and associating them with specific application scenarios, POI retrieval can be matched based on scenario-based tag categories. This deeply integrates the tag category with the search intent, solving the problem of insufficient accuracy in general tag retrieval results and improving the hit rate and user experience of POI retrieval in different vertical scenarios.

[0087] It should also be noted that if the tag classification process has been completed before the conflict detection, conflict detection can be performed only on target tags whose tag category is an objective category. This will identify target tags with semantic conflicts from the target tags whose tag category is an objective category, thus ensuring the content consistency of target tags under the objective category.

[0088] The sixth method: Identify target tags that belong to the blacklist corresponding to the POI as anomalous tags. Each POI corresponds to a blacklist, which can include anomalous tags identified under at least one of the above-mentioned anomalous situations.

[0089] Specifically, in the process of executing the POI tag processing method, when target tags containing preset abnormal words are identified as abnormal tags, target tags whose relevance to the POI information does not meet the relevance condition are identified as abnormal tags, tags among multiple target tags with semantic conflicts whose credibility does not meet the credibility condition are identified as abnormal tags, tags containing sub-POIs within a target tag are identified as abnormal tags, and target tags that fail to be classified into a preset tag category are identified as abnormal tags, the abnormal tags identified in these abnormal situations are added to the blacklist corresponding to that POI to obtain an updated blacklist. Further, the target tags determined using the aggregated large model can be filtered based on the updated blacklist. For the next execution of the POI tag processing method, the blacklist can be updated again in the above manner, and then the target tags determined using the aggregated large model can be filtered based on the updated blacklist.

[0090] As a more complete embodiment, such as Figure 6 As shown, multiple original tags for the same POI are obtained, and the multiple original tags are divided into multiple original tag sets based on semantic similarity. The aggregation model is used to determine multiple candidate tags in each original tag set that describe the same attribute of the POI or the same service provided by the POI. The target tag that meets the preset information quality conditions is determined from the multiple candidate tags.

[0091] Next, the target tags are classified to determine their respective categories. Target tags classified into meaningless categories are identified as anomalous tags and filtered out. Based on this, a relevance identification model is invoked to determine the relevance between the target tag and at least one of the POI's name, attribute, and address information. Target tags whose relevance does not meet the relevance criteria are identified as anomalous tags and filtered out. Then, a sub-POI tag identification model is invoked to identify sub-POIs within the spatial range of the corresponding geographic object based on the POI's geographic location. Tags of sub-POIs contained in the target tags are identified as anomalous tags and filtered out. Finally, a conflict identification model is invoked to identify multiple target tags with semantic conflicts. Among these multiple semantically conflicting target tags, those whose credibility does not meet the credibility criteria among multiple original tags are identified as anomalous tags and filtered out.

[0092] In the application scenario example, suppose there is a restaurant-class POI named "XX Restaurant", and the following original tags are obtained: "Delicious", "Spicy and Flavorful", "Delicious Sichuan Cuisine", "Parking Available", "Parking Lot Available", "#￥%\n", and "The Library is Quiet". Based on semantic similarity, multiple original tags are divided into original tag sets: the first original tag set is ["Delicious", "Spicy and Flavorful", "Delicious Sichuan Cuisine"], the second original tag set is ["Parking Available", "Parking Lot Available"], the third original tag set is ["#￥%\n"], and the fourth original tag set is ["The Library is Quiet"]. Next, using an aggregated large model, the target tag for the first original tag set is "Spicy and Flavorful", the target tag for the second original tag set is "Parking Available", the target tag for the third original tag set is "#￥%\n", and the target tag for the fourth original tag set is "The Library is Quiet".

[0093] Then, the four target tags are categorized. "Spicy and savory" and "The library is quiet" can be classified into the subjective category, "Parking available" can be classified into the objective category, while "#￥%\n" does not belong to the preset tag category and can be filtered out as an abnormal tag. Next, using the relevance identification model, "The library is quiet" is determined to be a target tag whose relevance to the POI information does not meet the relevance condition and can be filtered out as an abnormal tag. The sub-POI tag identification model failed to identify tags containing sub-POIs in the target tags, and the conflict identification model also failed to identify target tags with semantic conflicts.

[0094] Therefore, the final target labels after filtering out abnormal labels are "spicy and flavorful" and "parking available".

[0095] It should be noted that, in this application, the aggregated large model or the anomaly label recognition large model can refer to any machine learning model capable of processing and understanding natural language and generating, analyzing, or reasoning about text based on context. Its core capability lies in grasping deep semantics and complex logical relationships. For example, it may include, but is not limited to: pre-trained models based on the Transformer architecture, models that have been fine-tuned by instructions to adapt to specific tasks, or specialized models that have been further fine-tuned for specific domains (such as POI label processing).

[0096] The collection, storage, use, processing, transmission, provision, and disclosure 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.

[0097] The foregoing has described specific embodiments of this disclosure. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0098] According to another embodiment, a processing apparatus for POI tags is provided. Figure 7 A schematic block diagram of a processing apparatus for POI tags according to one embodiment is shown, the processing apparatus for POI tags being disposed in... Figure 1 The server side in the illustrated architecture. For example... Figure 7 As shown, the POI tag processing device 700 includes an acquisition unit 701, a segmentation unit 702, and a target tag determination unit 703, and further includes a filtering unit 704, an abnormal tag determination unit 705, and a classification unit 706. The main functions of each component are as follows: Acquisition unit 701 is configured to acquire multiple raw tags for the same POI.

[0099] The segmentation unit 702 is configured to segment multiple original labels into multiple sets of original labels based on semantic similarity.

[0100] The target label determination unit 703 is configured to use an aggregated large model to determine multiple candidate labels within each original label set that describe the same attribute of the POI or the same service provided by the POI, and to determine the target label that meets the preset information quality conditions from the multiple candidate labels. The target label is used to be displayed in the electronic map to describe the POI, or to be used to match with the POI retrieval information to recall the POI.

[0101] As one possible implementation method, the partitioning unit 702, when partitioning multiple original labels into multiple sets of original labels based on semantic similarity, can be specifically configured as follows: semantically encoding multiple original labels to obtain multiple semantic vectors corresponding to multiple original labels; determining the semantic similarity between multiple semantic vectors; and partitioning multiple original labels into multiple sets of original labels based on semantic similarity.

[0102] Furthermore, the filtering unit 704 can be specifically configured to filter out abnormal tags in the target tags.

[0103] As one possible approach, the abnormal label includes at least one of the following: a target label containing a preset abnormal word; a target label belonging to the blacklist corresponding to the POI; a target label whose relevance to the information of the POI does not meet the relevance condition; a label whose credibility does not meet the credibility condition among multiple target labels with semantic conflicts; a label containing a sub-POI, wherein the geographic object corresponding to the sub-POI is located within the spatial range of the geographic object corresponding to the POI; or a target label that, after label classification processing, fails to be classified into a preset label category.

[0104] As one possible approach, POI information includes at least one of name information, attribute information, and address information.

[0105] Furthermore, when the abnormal label determination unit 705 determines a target label whose relevance to the POI information does not meet the relevance condition, it can be specifically configured to: call the abnormal label identification model, determine the relevance of the target label to at least one of the name information, attribute information and address information, and determine the target label whose relevance does not meet the relevance condition as an abnormal label.

[0106] As one possible implementation method, the abnormal label determination unit 705 can be specifically configured to: call the abnormal label identification model, take the multiple target labels with semantic conflicts as conflict labels, and determine the credibility of each conflict label based on the number of original labels that meet the similarity conditions with the conflict labels, thereby determining the labels among the conflict labels whose credibility does not meet the credibility conditions.

[0107] As one possible implementation method, the classification unit 706 can be specifically configured to: perform tag classification processing on the target tag, determine the tag category to which the target tag belongs, and establish a corresponding relationship between the tag category and at least one application scenario; in response to a POI retrieval request under any application scenario, determine the POI that matches the POI retrieval request based on the target tags corresponding to each POI under the tag category corresponding to the application scenario.

[0108] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0109] Figure 8 A schematic block diagram of an example electronic device 800 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.

[0110] like Figure 8 As shown, device 800 includes a computing unit 801, which can perform various appropriate actions and processes based on a computer program stored in read-only memory (ROM) 802 or a computer program loaded from storage unit 808 into random access memory (RAM) 803. RAM 803 may also store various programs and data required for the operation of device 800. The computing unit 801, ROM 802, and RAM 803 are interconnected via bus 804. Input / output (I / O) interface 805 is also connected to bus 804.

[0111] Multiple components in device 800 are connected to I / O interface 805, including: input unit 806, such as keyboard, mouse, etc.; output unit 807, such as various types of monitors, speakers, etc.; storage unit 808, such as disk, optical disk, etc.; and communication unit 809, such as network card, modem, wireless transceiver, etc. Communication unit 809 allows device 800 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0112] The computing unit 801 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 801 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various 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 801 performs the various methods and processes described above, such as the processing method for POI tags. For example, in some embodiments, the processing method for POI tags may be implemented as a computer software program tangibly contained in a machine-readable medium, such as storage unit 808. In some embodiments, part or all of the computer program may be loaded and / or installed on device 800 via ROM 802 and / or communication unit 809. When the computer program is loaded into RAM 803 and executed by the computing unit 801, one or more steps of the processing method for POI tags described above may be performed. Alternatively, in other embodiments, the computing unit 801 may be configured to perform processing methods for POI tags by any other suitable means (e.g., by means of firmware).

[0113] 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), complex 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.

[0114] 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.

[0115] 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.

[0116] 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).

[0117] 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.

[0118] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0119] 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 disclosed in this disclosure can be achieved, and this is not limited herein.

[0120] 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 processing POI tags, comprising: Retrieve multiple raw labels for the same POI; The multiple original labels are divided into multiple sets of original labels based on semantic similarity; The aggregation model is used to determine multiple candidate tags within each of the original tag sets that describe the same attribute of the POI or the same service provided by the POI. Target tags that meet preset information quality conditions are then determined from the multiple candidate tags. These target tags are used to display the POI on an electronic map to describe the POI, or to match it with POI retrieval information to recall the POI.

2. The method according to claim 1, wherein, The process of dividing the multiple original tags into multiple sets of original tags based on semantic similarity includes: Semantic encoding is performed on the multiple original labels to obtain multiple semantic vectors corresponding to the multiple original labels; Determine the semantic similarity among the plurality of semantic vectors; Based on the semantic similarity, the multiple original tags are divided into multiple sets of original tags.

3. The method according to claim 1, wherein, The method further includes: Filter out abnormal tags from the target tags; The anomaly label includes at least one of the following: The target tag containing preset abnormal words; The target tag that belongs to the blacklist corresponding to the POI; The target label whose relevance to the information of the POI does not meet the relevance condition; Among the multiple target tags that have semantic conflicts, the tags whose credibility does not meet the credibility condition; The target label includes the sub-POI label, wherein the geographic object corresponding to the sub-POI is located within the spatial range of the geographic object corresponding to the POI; The target label that failed to be classified into the preset label category after label classification processing.

4. The method according to claim 3, wherein, The POI information includes at least one of name information, attribute information, and address information. Target tags whose relevance to the POI information does not meet the relevance criteria are determined as follows: The abnormal label identification model is invoked to determine the correlation between the target label and at least one of the name information, attribute information and address information, and the target label whose correlation does not meet the correlation condition is identified as the abnormal label.

5. The method according to claim 3, wherein, The following method is used to identify tags whose credibility does not meet the credibility condition among multiple target tags with semantic conflicts: The abnormal label identification model is invoked, and multiple target labels with semantic conflicts are identified as conflict labels. The credibility of each conflict label is determined based on the number of original labels whose similarity with the conflict labels meets the similarity condition. In this way, the labels whose credibility does not meet the credibility condition among the conflict labels are identified.

6. The method according to any one of claims 1-5, wherein, The method further includes: The target tags are classified to determine the tag category to which the target tags belong, and the tag category is associated with at least one application scenario; In response to a POI retrieval request in any application scenario, the POI that matches the POI retrieval request is determined based on the target tags corresponding to each POI under the tag category corresponding to that application scenario.

7. A processing apparatus for POI tags, comprising: The acquisition unit is configured to acquire multiple raw tags for the same POI; The partitioning unit is configured to divide the plurality of original labels into a plurality of original label sets based on semantic similarity; The target label determination unit is configured to use an aggregated large model to determine multiple candidate labels within each of the original label sets for describing the same attribute of the POI or the same service provided by the POI, and to determine target labels that meet preset information quality conditions from the multiple candidate labels. The target labels are used to be displayed in an electronic map to describe the POI, or to be used to match with POI retrieval information to recall the POI.

8. An electronic device, comprising: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-6.

9. A non-transitory computer-readable storage medium storing computer instructions, wherein, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-6.

10. A computer program product comprising a computer program that, when executed by a processor, implements the method according to any one of claims 1-6.

Images