Product search method, device, and storage medium by tag mining

Abstract

The application discloses a product retrieval method and device based on label mining, and a storage medium. The method comprises the following steps: performing label extraction on obtained product data to obtain original labels of products, wherein the products comprise at least one of commodities and services; analyzing the original labels one by one, determining the original labels as standard labels, or associating the original labels with corresponding standard labels to form at least one label cluster; wherein each standard label corresponds to one label cluster; adding the standard labels included in the label cluster to each product corresponding to the label cluster; taking a product corresponding to a target label as a target product, wherein the target label refers to a standard label corresponding to retrieval content. The technical scheme of the application realizes the technical effects of label standardization clustering and improving the precision and speed of product retrieval based on standard labels.

Description

Technical Field

[0001] This application belongs to the field of data retrieval, specifically involving product retrieval methods, equipment, and storage media based on tag mining. Background Technology

[0002] In this era of information overload, efficient retrieval of massive amounts of product and service data has become a core requirement for various industries. Tags, as a crucial carrier for characterizing product features and connecting search needs with target products, directly determine search efficiency and accuracy. Currently, the construction of tag systems largely relies on manual definition or simple keyword extraction, and a systematic tag standardization and clustering mechanism has not yet been formed.

[0003] The core shortcomings of existing technologies are concentrated in two points: First, the original tags lack unified standards and associated management. Product tags from different sources often have problems such as arbitrary wording and semantic repetition. There are no clear standard tags as core anchors, nor are semantically related non-standard tags clustered and integrated, resulting in a chaotic tag system. Second, product retrieval based on such non-standard tags makes it difficult to accurately match user needs and target products through unified standard tags. Often, related products are missed due to differences in tag descriptions, or irrelevant results are introduced due to chaotic tags, which seriously affects the accuracy and efficiency of product retrieval.

[0004] Therefore, how to achieve standardized clustering of tags and improve the accuracy and speed of product retrieval based on standard tags is a technical problem that urgently needs to be solved. Summary of the Invention

[0005] The purpose of this application is to achieve standardized clustering of tags and improve the accuracy and speed of product retrieval based on standardized tags.

[0006] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0007] According to one aspect of the embodiments of this application, a product retrieval method based on tag mining is provided, the method comprising:

[0008] Tag extraction is performed on the acquired product data to obtain the original tags for each product, wherein the products include at least one of goods and services;

[0009] Each of the original tags is analyzed individually, and the original tags are determined as standard tags, or non-standard tags are associated with corresponding standard tags to form at least one tag cluster; wherein each of the standard tags corresponds to a tag cluster;

[0010] Add the standard labels included in the label cluster to each product corresponding to the label cluster;

[0011] The product corresponding to the target tag is taken as the target product. The target tag refers to the standard tag corresponding to the search content.

[0012] According to one aspect of the embodiments of this application, tag extraction is performed on the acquired product data to obtain the original tags for each product, including:

[0013] Tag extraction is performed on the product data to obtain multiple initial tags;

[0014] Remove initial labels that are irrelevant to the target domain;

[0015] Initial tags that are not directly derived from the product data and have undergone semantic transformation will be filtered out.

[0016] Filter out initial tags that do not contain entity content;

[0017] The retained initial labels are generalized by modifiers to obtain the original labels.

[0018] According to one aspect of the embodiments of this application, the original labels are analyzed one by one to determine whether the original labels are standard labels or non-standard labels and associated with corresponding standard labels, including:

[0019] Select any original label, and determine the standard label with the highest similarity as the similar standard label based on the similarity between the original label and each of the standard labels;

[0020] Based on the similarity between the original tag and similar standard tags, the original tag is either used as a new standard tag or added as a non-standard tag to the tag cluster containing the similar standard tags;

[0021] Select the next original label for analysis, and continue until the analysis of all original labels is completed.

[0022] According to one aspect of the embodiments of this application, the method further includes:

[0023] If there is no standard label when analyzing the original label, the original label is directly used as the standard label.

[0024] According to one aspect of the embodiments of this application, based on the similarity between the original tag and similar standard tags, the original tag is either used as a new standard tag or added as a non-standard tag to the tag cluster containing the similar standard tags, including:

[0025] If the similarity between the original label and the similar standard label is less than the first similarity value, then the original label is used as the standard label;

[0026] If the similarity between the original tag and the similar standard tag is greater than the second similarity value, then the original tag is added as a non-standard tag to the tag cluster where the similar standard tag is located, and the first similarity value is less than the second similarity value;

[0027] If the similarity between the original label and the similar standard label is greater than or equal to the first similarity value and less than or equal to the second similarity value, then the original information of the product corresponding to the original label is obtained to correct the original label, and the original label is re-analyzed until the original label is used as a standard label or added to a label cluster.

[0028] According to one aspect of the embodiments of this application, the method further includes:

[0029] When an update event occurs in any tag cluster, the centroid of the tag cluster is calculated;

[0030] If a non-standard label is closest to the centroid of the label cluster, then the standard label of the label cluster is changed to the non-standard label; if a standard label is closest to the centroid of the label cluster, then the standard label of the label cluster is not changed.

[0031] According to one aspect of the embodiments of this application, the method further includes:

[0032] If the standard label of the label cluster changes, the non-standard label within the label cluster will be changed back to the original label for re-analysis.

[0033] According to one aspect of the embodiments of this application, the method further includes:

[0034] The non-standard label with the lowest similarity to the standard label in the changed label cluster is changed to the label to be reviewed and re-analyzed. The changed label cluster refers to the label cluster in which the standard label has been changed.

[0035] If the tag to be reviewed is added to a tag cluster outside the changed tag cluster or is used as a new standard tag, the tag to be reviewed is re-determined and analyzed in the changed tag cluster until the tag to be reviewed is added to the changed tag cluster.

[0036] According to one aspect of the embodiments of this application, a product retrieval device by tag mining is provided, including a memory, a processor, and a readable program stored in the memory, wherein the processor executes the readable program to implement the method described in any of the above.

[0037] According to one aspect of the embodiments of this application, a readable storage medium is provided, on which a readable program / instruction is stored, which, when executed by a processor, implements the method described in any one of the above-described embodiments.

[0038] In this application, tags are extracted from the acquired product data to obtain the original tags for each product, which includes at least one of goods and services; the original tags are analyzed one by one, and the original tags are determined as standard tags, or non-standard tags are associated with standard tags to form at least one tag cluster; each standard tag corresponds to a tag cluster; the standard tags included in the tag cluster are added to each product corresponding to the tag cluster; the product corresponding to the target tag is taken as the target product, and the target tag refers to the standard tag corresponding to the search content.

[0039] On the one hand, this solution establishes a unified semantic benchmark for the chaotic original tags by analyzing each original tag individually to determine standard tags and associated standard tags, forming tag clusters (each standard tag corresponds to a unique tag cluster). Standard tags, as core anchors within the same semantic category, incorporate non-standard tags with different expressions but consistent semantics into their corresponding tag clusters, completely breaking the isolated and unregulated state of the original tags. Simultaneously, it clarifies that tags cover at least one of the products and services, further adapting to the tag management needs of multiple types of objects, transforming the originally disordered tag library into a structured system centered on standard tags, directly solving the problems of chaotic tag systems and lack of unified standards in the background technology.

[0040] On the other hand, this solution adds standard tags to the corresponding products within the tag cluster, ensuring that all objects belonging to the same semantic category (regardless of how the original tags are expressed) carry a unified standard tag, eliminating retrieval omissions caused by differences in the original tags for the same type of object. Simultaneously, the retrieval process uses the standard tags corresponding to the search content as the target tags, directly matching the tag clusters and objects associated with the standard tags. This avoids the problems of relying on non-standard tags and being interfered with by expression differences in traditional retrieval—it eliminates the need to match diverse non-standard tags one by one, and avoids introducing irrelevant results due to tag expression deviations. From the source matching logic of the retrieval, it solves the shortcomings of low retrieval accuracy and poor efficiency in the background technology, ensuring that users can quickly locate the target product or service with a completely matching semantics.

[0041] Other features and advantages of this application will become apparent from the following detailed description, or may be learned in part from practice of this application.

[0042] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and do not limit this application. Attached Figure Description

[0043] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are merely some embodiments of this application, and those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0044] Figure 1 A schematic diagram of a product retrieval method based on tag mining according to an embodiment of this application is shown.

[0045] Figure 2 A flowchart illustrating the process of extracting labels from acquired product data to obtain the original labels for each product, according to one embodiment of this application, is shown.

[0046] Figure 3 A flowchart illustrating how, according to an embodiment of this application, the standard label with the highest similarity to each standard label is determined as the similar standard label based on the similarity between the original label and each standard label.

[0047] Figure 4 The flowchart illustrates a process according to an embodiment of this application, in which an original tag is added as a new standard tag or as a non-standard tag to a tag cluster containing similar standard tags based on the similarity between the original tag and similar standard tags.

[0048] Figure 5 A flowchart illustrating the updating of a standard label according to one embodiment of this application is shown.

[0049] Figure 6 A flowchart illustrating the updating of non-standard labels in a label cluster where standard labels have been modified, according to an embodiment of this application, is shown.

[0050] Figure 7 A computer device structural block diagram of a product retrieval method for tag mining according to an embodiment of this application is shown. Detailed Implementation

[0051] Exemplary embodiments will now be described more fully with reference to the accompanying drawings. However, these exemplary embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided to make this application more comprehensive and complete, and to fully convey the concept of the exemplary embodiments to those skilled in the art.

[0052] Furthermore, the described features, structures, or characteristics can be combined in any suitable manner in one or more embodiments. Numerous specific details are provided in the following description to give a thorough understanding of embodiments of this application. However, those skilled in the art will recognize that the technical solutions of this application can be practiced without one or more of the specific details, or other methods, components, apparatuses, steps, etc., can be employed. In other instances, well-known methods, apparatuses, implementations, or operations are not shown or described in detail to avoid obscuring various aspects of this application.

[0053] The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities can be implemented in software, in one or more hardware modules or integrated circuits, or in different network and / or processor devices and / or microcontroller devices.

[0054] The flowcharts shown in the accompanying drawings are merely illustrative and do not necessarily include all content and operations / steps, nor do they necessarily have to be performed in the described order. For example, some operations / steps can be broken down, while others can be combined or partially combined; therefore, the actual execution order may change depending on the specific circumstances.

[0055] In the embodiments of this application, the terms "module" or "unit" refer to a computer program or part of a computer program that has a predetermined function and works with other related parts to achieve a predetermined goal, and can be implemented wholly or partially using software, hardware (such as processing circuitry or memory), or a combination thereof. Similarly, a processor (or multiple processors or memory) can be used to implement one or more modules or units. Furthermore, each module or unit can be part of an overall module or unit that includes the functionality of that module or unit.

[0056] Please see Figure 1 , Figure 1 A schematic diagram of a product retrieval method based on tag mining according to an embodiment of this application is shown. This application embodiment provides the execution steps of a product retrieval method based on tag mining, including:

[0057] Step S110: Extract labels from the acquired product data to obtain the original labels for each product. Products include at least one of goods and services.

[0058] Step S120: Analyze each original label one by one, determine the original label as a standard label, or determine the non-standard label and associate it with the corresponding standard label to form at least one label cluster; wherein, each standard label corresponds to a label cluster;

[0059] Step S130: Add the standard labels included in the label cluster to each product corresponding to the label cluster;

[0060] Step S140: The product corresponding to the target tag is taken as the target product. The target tag refers to the standard tag corresponding to the search content.

[0061] The four steps described above are described in detail below.

[0062] In step S110, the product includes at least one of goods and services. Product data is a multi-source information collection covering physical goods and services, including structured official parameters and specifications, as well as unstructured product descriptions, user reviews, service descriptions, etc., providing basic data support for tag extraction, standardized clustering, and subsequent accurate retrieval.

[0063] It needs to be clarified that the methods for obtaining product data include: calling merchant onboarding and operational data through the platform's internal API, and obtaining structured parameters and unstructured descriptions as product data after cleaning and preprocessing.

[0064] We sign agreements with brands and industry organizations to transmit official parameters and qualification data via API or offline files, supplementing product data with professional-dimensional information.

[0065] The platform's review and Q&A APIs are called up, and OCR is used to recognize the text of customer reviews. After anonymization, real experience data is extracted and supplemented with user perspective data as product data.

[0066] Comply with the robots.txt protocol to crawl publicly available competitor or industry data, control the crawling frequency and deduplicate data, and supplement with cross-platform information as product data. Alternatively, use offline product QR code scanning or service point terminal data to synchronize with online data.

[0067] Original labels refer to the initial feature identifiers extracted from structured parameters, unstructured descriptions, and other data in product data, which are then processed through semantic parsing or rule extraction to obtain a standardized representation. In other words, original labels already provide a complete and standardized representation of the product data.

[0068] It should be clarified that, in some embodiments, each product corresponds to at least one original label.

[0069] Tag extraction is performed on each product's data to obtain the original tags for each product. For example, original tag extraction requires first parsing product data in multiple formats such as PDF and HTML using a library, and then preprocessing the image-like text using OCR. The core is based on a large language model (such as GPT-4-mini), taking unstructured JSON data as input, and extracting features from the semantic level by constraining domain vocabulary and generalizing expressions through Prompt, without relying on explicit keywords. This is supplemented by traditional methods: structured parameters are extracted using regular expressions / rules, and text-like data is combined with models such as BERT for entity recognition. Finally, the original tags are output after de-identification and deduplication.

[0070] In other embodiments, a trained label extraction model can be used to extract raw labels from each product data. In still other embodiments, raw labels can be obtained by performing core semantic recognition on the product data.

[0071] Please see Figure 2 , Figure 2 This document illustrates a flowchart illustrating the process of extracting labels from acquired product data to obtain the original labels for each product, according to an embodiment of this application. The embodiment of this application provides step S110 for extracting labels from acquired product data to obtain the original labels for each product, including:

[0072] Step S111: Extract tags from product data to obtain multiple initial tags;

[0073] Step S112: Filter out initial labels that are irrelevant to the target domain;

[0074] Step S113: Filter out the initial tags that are not directly derived from product data and have undergone semantic transformation.

[0075] Step S114: Filter out initial tags that do not have entity content;

[0076] Step S115: Perform modifier generalization on the retained initial labels to obtain the original labels.

[0077] The above five steps are described in detail below.

[0078] In step S111, labels are extracted from the acquired product data to obtain initial labels for multiple products. The original labels are initial feature identifiers extracted from structured parameters and unstructured descriptions of products (including entities and services), obtained through semantic parsing or rule extraction. They are not standardized and have characteristics such as arbitrary word choice and semantic repetition, and are the basis for determining the original labels in the subsequent steps.

[0079] Initial labels can be obtained through semantic extraction, using the extracted semantics as initial labels. Alternatively, a trained initial label extraction model can be used to extract labels from the input product data. It's important to note that each product data point should have at least one corresponding initial label.

[0080] In step S112, initial tags irrelevant to the target domain are filtered out, meaning only initial tags related to the target domain are retained. This completes the domain constraint on the initial tags, ensuring that all remaining initial tags are related to the target domain, thus achieving the technical effect of extracting tags only for products within the target domain. The target domain can be automatically determined based on user actions.

[0081] In step S113, initial tags that are not directly derived from product data content or have undergone semantic transformation are filtered out. That is, the retained initial tags are all directly derived from product data content and have not undergone semantic transformation. In other words, each word in the initial tag originates from the product data content, effectively preventing the initial tags from becoming distorted due to deviations from the product data content during the initial tag generation process.

[0082] In step S114, initial tags that do not contain entity content are filtered out, that is, initial tags representing abstract content are removed, and only initial tags containing entity content are retained. For example, if the target domain is the tourism domain, initial tags such as place names, tourism-unrelated service terms (such as "electronic vouchers"), abstract descriptions, or process arrangements are filtered out.

[0083] In step S115, the initial tags that have been filtered above are generalized to obtain the original tags. For descriptions of place names or specific locations, generalization is performed; for example, "visit to Sofia Church" is generalized to "visit to the church". It should be noted that the above generalization is only performed on initial tags with modifiers. If the initial tag has no modifier, no generalization is needed, and it is directly used as the original tag.

[0084] It should be clarified that the initial label filtering in steps S112 to S113 can be performed in any order, without any restrictions.

[0085] In this embodiment, through multi-round precise screening and targeted generalization of tag optimization logic, irrelevant tags are not only filtered out by domain constraints (automatically matching target domains based on user behavior), focusing on core domains to reduce invalid processing costs, but also the authenticity and substance of tags are guaranteed by retaining tags directly derived from product data without semantic transformation and filtering out abstract tags without entity content. This ensures that tags accurately depict the core attributes and service essence of the product and meet the core search needs of users. At the same time, generalization processing is performed on modifiers such as place names to effectively alleviate the problem of arbitrary wording in the original tags, laying a unified foundation for the subsequent determination of standard tags and the formation of tag clusters. Finally, high-quality original tags empower the standardized clustering and product retrieval in the claims, significantly improving the accuracy and efficiency of retrieval.

[0086] In step S120, each original label is analyzed to determine whether it can become a standard label. If the condition is met, the original label is used as a standard label. If the original label does not meet the condition, it is treated as a non-standard label and associated with the corresponding standard label. This process continues until all original labels have been analyzed, resulting in at least one label cluster. Each label cluster contains only one standard label. Each standard label and its associated non-standard label form a label cluster. It is important to note that a label cluster must contain one standard label, but it may not contain any non-standard labels; that is, the standard label may not have any associated non-standard labels.

[0087] It should be clarified that in this application embodiment, only the original tags that meet the conditions are used as standard tags, while other original tags that do not meet the conditions are used as non-standard tags and associated with the corresponding standard tags. No content changes are made to the original tags. Whether it is a standard tag or a non-standard tag, its essential content is still the original tag.

[0088] In this embodiment, once the standard label is determined, a label cluster is created with the standard label as the core, and the label cluster is continuously expanded when analyzing the original labels in the future.

[0089] The analysis of the original tags can include the following two methods:

[0090] First, for each original label, perform the following steps: based on the similarity between the original label and each standard label, determine the most similar standard label among the standard labels as the similar standard label; based on the similarity between the original label and the similar standard label, use the original label as a new standard label, or add it to the label cluster formed by the similar standard labels, until all original labels have been analyzed.

[0091] It's important to clarify that for the first analysis of the original labels, there were no previously analyzed original labels, and therefore no standard labels or label clusters. Therefore, the original labels from the first analysis are directly used as standard labels to form clusters.

[0092] Secondly, all existing original labels can be converted into vectors using a word vector model to construct a label similarity weighted graph (with edge weights representing semantic relevance). A community detection algorithm is then used to divide the graph into tightly connected subgraphs, with each subgraph forming a label cluster. Within each cluster, the original label vector closest to the centroid of the subgraph is determined, and the original label corresponding to this vector is designated as the standard label. Other original labels within the subgraph are considered non-standard labels.

[0093] In this embodiment, existing original tags for implementing this technical solution are uniformly processed, and newly generated original tags are analyzed one by one. For example, if the cosine similarity between an original tag vector (the vector of the original tag) and a standard tag vector (the vector of the standard tag) is greater than a set value, the original tag is assigned to the tag cluster containing the standard tag. If the cosine similarity between the original tag vector and a standard tag vector is less than or equal to a set value, the original tag is used as the standard tag.

[0094] Please see Figure 3 , Figure 3 A flowchart illustrating how, according to an embodiment of this application, the standard label with the highest similarity to each standard label is determined as the similar standard label based on the similarity between the original label and each standard label, is shown. This embodiment provides step S120, which involves determining the standard label with the highest similarity to each standard label based on the similarity between the original label and each standard label, including:

[0095] Step S121: Select any original label, and determine the standard label with the highest similarity as the similar standard label based on the similarity between the original label and each standard label.

[0096] Step S122: Based on the similarity between the original tag and the similar standard tags, the original tag is either used as a new standard tag or added as a non-standard tag to the tag cluster where the similar standard tags are located.

[0097] Step S123: Select the next original label for analysis, until the analysis of all original labels is completed.

[0098] The two steps described above are described in detail below.

[0099] In step S121, all original labels that need to be analyzed are analyzed one by one. For example, an original label is randomly selected, and the similarity between the original label and the standard label is calculated. The standard label with the highest similarity to the original label is taken as the similar standard label of the original label.

[0100] One way to calculate the similarity between the original label and each standard label is to use the semantic similarity between the original label and the standard labels as the similarity between the original label and the standard labels. Alternatively, the original label and the standard labels can be vectorized, and the similarity between the original label vector and the standard label vector can be used as the similarity between the original label and the standard labels.

[0101] It's important to clarify that for the first analysis of the original labels, there were no previously analyzed original labels, and therefore no standard labels or label clusters. Therefore, the original labels from the first analysis are directly used as the standard labels to form the cluster.

[0102] In some embodiments, each original label is vectorized to obtain an original label vector. For example, for any original label to be processed (O-Tag), a pre-trained deep language model (such as the LaBSE model) is used to perform semantic encoding on the original label (O-Tag), generating a multi-dimensional, dense semantic vector (O-Emb) that precisely corresponds to the semantic connotation of the original label. Each original label vector includes the original label number (O-Tag) and the semantic vector (O-Emb) of that original label.

[0103] When no standard label is available, any original label is taken as the first one analyzed and directly used as the original label. When an original label is determined as the standard label, the original label number (O-Tag) and semantic vector (O-Emb) of the original label vector corresponding to that original label are directly adopted as the standard label (S-Tag) and standard semantic vector (S-Emb), forming a (O-Tag, S-Emb, S-Tag) triple. That is, the standard label vector consists of three parts: the original label number, the standard semantic vector, and the standard label number. The original label number indicates which original label the standard label originally belonged to. The standard semantic vector refers to the semantic vector of the standard label. The standard label number refers to the sequence number of the standard label among all standard labels.

[0104] When calculating the similarity between the original tag and the standard tag, the similarity can be determined by the angle between the semantic vectors of the original tag and the standard tag. For example, the similarity is determined by the cosine of the angle between the semantic vectors of the original tag and the standard tag. The larger the cosine value, the greater the similarity.

[0105] In other embodiments, each standard tag is associated with multiple non-standard tags (original tags that have already been analyzed and are consistent with the original tag in form and content). The most similar standard tag is determined as follows: for the original tag to be analyzed, the similarity between the original tag and each non-standard tag and standard tag is obtained. The non-standard tag or standard tag most similar to the original tag is selected as the initial selected tag. If the initial selected tag is a non-standard tag, the standard tag of the tag cluster containing the initial target tag is used as the similar standard tag of the original tag. If the initial selected tag is a standard tag, it is directly used as the similar standard tag of the original tag.

[0106] In step S122, based on the similarity between the original label and similar standard labels, the original label is either used as a new standard label or added as a non-standard label to the label cluster containing the similar standard labels.

[0107] For example, if the similarity between the original tag and the similar standard tag is greater than the similarity threshold, the original tag is added to the tag cluster containing the similar standard tag. If the similarity between the original tag and the similar standard tag is less than or equal to the similarity threshold, the original tag is used as the standard tag to generate a tag cluster.

[0108] In step S123, the next original label is selected for analysis until the analysis of all original labels is completed. That is, until all original labels are analyzed and classified into standard labels or non-standard labels, the product label mining is completed.

[0109] In this embodiment, dynamic clustering and similarity-driven label standardization logic bring multi-dimensional benefits. First, the first original label directly serves as the standard label to form the initial cluster, solving the process startup problem when there is no benchmark label, ensuring that label clustering proceeds in an orderly manner from scratch, and avoiding logical gaps in the initial stage. Second, by measuring the association between the original label and the standard label through semantic similarity or vector similarity calculation, the deep semantic associations between labels (such as synonyms and related expressions) can be accurately captured, effectively solving the semantic dispersion problem caused by arbitrary word choice in the original labels, and providing a scientific basis for the aggregation of similar labels. Furthermore, dynamic decision-making based on similarity thresholds (entering a cluster if the threshold is exceeded, otherwise creating a new cluster) ensures the semantic uniformity within the label cluster (avoiding irrelevant labels from being mixed in) and flexibly accepts new feature labels (forming new clusters), realizing the dynamic expansion and precise convergence of the label system, taking into account both stability and adaptability. Ultimately, this process aggregates the scattered original tag system into a tag cluster with standard tags at its core, providing a structured and standardized tag foundation for subsequent steps such as product retrieval based on standard tags and tag knowledge graph construction in the claims, significantly improving tag matching efficiency and retrieval accuracy.

[0110] Please see Figure 4 , Figure 4 This document illustrates a flowchart illustrating how, according to an embodiment of this application, an original tag is added as a new standard tag or as a non-standard tag to a tag cluster containing similar standard tags based on the similarity between the original tag and similar standard tags. The embodiment of this application provides step S122, which involves adding an original tag as a new standard tag or as a non-standard tag to a tag cluster containing similar standard tags based on the similarity between the original tag and similar standard tags.

[0111] Step S1221: If the similarity between the original label and the similar standard label is less than the first similarity value, then the original label is used as the standard label.

[0112] Step S1222: If the similarity between the original label and the similar standard label is greater than the second similarity value, then the original label is added as a non-standard label to the label cluster where the similar standard label is located, and the first similarity value is less than the second similarity value.

[0113] Step S1223: If the similarity between the original label and the similar standard label is greater than or equal to the first similarity value and less than or equal to the second similarity value, then obtain the original information of the product corresponding to the original label to correct the original label and re-analyze the original label until the original label is used as a standard label or added to a label cluster.

[0114] The above three steps are described in detail below.

[0115] In step S1221, if the similarity between the original label and the similar standard label is less than the first similarity value, then the original label is used as the standard label.

[0116] In step S1222, if the similarity between the original label and the similar standard label is greater than the second similarity value, then the original label is added as a non-standard label to the label cluster where the similar standard label is located, and the first similarity value is less than the second similarity value.

[0117] In step S1223, if the similarity between the original label and a similar standard label is greater than or equal to a first similarity value and less than or equal to a second similarity value, then the original information of the product corresponding to the original label is obtained. The original label is then corrected using this product's original information. The original label is then re-analyzed until it becomes a standard label or is added to a label cluster. That is, if the similarity between the original label and the similar standard label is always greater than or equal to the first similarity value, then the original information of the product corresponding to the original label is always obtained. The original label is then corrected using this product's original information. The original label is then re-analyzed. This process continues until the set number of analyses or the set time limit is reached, and a report indicating that analysis is not possible is issued to remind the user to resolve the issue.

[0118] In this embodiment, the technical solution achieves multiple beneficial effects through hierarchical threshold judgment and dynamic correction feedback in label processing logic: First, hierarchical processing based on the first and second similarity values ​​enables accurate sorting of label classification. Low similarity is directly set as the new standard label, ensuring that new feature labels are not missed and enriching the coverage of the label system. High similarity is assigned to the corresponding cluster, ensuring high semantic uniformity within the cluster and avoiding redundant standard labels. Second, for the intermediate similarity range, a product original information correction mechanism is introduced. By tracing the essential information of the product (such as parameters and descriptions), the label semantics are calibrated, effectively solving ambiguous scenarios where the semantics are similar but the substance is different (such as easily confused functional labels), reducing the situation of mis-clustering or mis-setting standard labels, and improving the accuracy of label standardization. Finally, setting an upper limit for the number of analysis times or duration and issuing a report of failure to analyze not only avoids infinite loops caused by ambiguous labels and ensures the system's operating efficiency, but also compensates for the limitations of the algorithm through a manual intervention mechanism, forming a closed loop of automatic processing and manual fallback, ensuring the integrity and reliability of label processing.

[0119] Overall, the solution balances the scalability of the tag system, the accuracy of clustering, and the robustness of the system, providing strong support for the accurate generation of standard tags in the claims and the reasonable construction of tag clusters, thereby improving the effectiveness of tag-based product retrieval, recommendation and other services.

[0120] Please see Figure 5 , Figure 5 A flowchart illustrating the updating of a standard label according to an embodiment of this application is shown. Embodiments of this application provide steps for updating a standard label according to an embodiment of this application, including:

[0121] Step S201: When an update event occurs in any tag cluster, calculate the centroid of the tag cluster;

[0122] In step S202, if a non-standard label is closest to the centroid of the label cluster, the standard label of the label cluster is changed to the non-standard label; if the standard label is closest to the centroid of the label cluster, the standard label of the label cluster is not changed.

[0123] The two steps described above are described in detail below.

[0124] In step S201, when an update event occurs in any tag cluster, the centroid of the tag cluster that triggered the update event is calculated. The update event can be a preset event. For example, an update event can be defined as the number of standard and non-standard tags in the tag cluster reaching a first preset number. An update event can also be defined as the duration the tag cluster has existed for a preset duration. Furthermore, an update event can be defined as the number of newly added non-standard tags reaching a second preset number after an update event occurs. Both the first and second preset numbers are positive integers and can be defined by the developers.

[0125] The centroid of the label cluster can be calculated in the following way.

[0126] Firstly, standard and non-standard labels are vectorized, such as by vectorizing the features of standard and non-standard labels, or by using a vectorization model to obtain standard label vectors and non-standard label vectors.

[0127] The sum of the standard label vectors and the non-standard label vectors in the label cluster is used as the centroid of the label cluster.

[0128] In some embodiments, standard and non-standard labels are vectorized in the manner described in step S121 to obtain standard label vectors and non-standard label vectors.

[0129] The centroid of the tag cluster is the sum of the standard semantic vectors of the standard tags and the semantic vectors of the non-standard tags.

[0130] Secondly, standard and non-standard labels are mapped into three-dimensional space, so that each standard and non-standard label has a corresponding three-dimensional position. The average of the three-dimensional positions of each standard and non-standard label is used as the centroid of the label cluster.

[0131] In step S202, if a non-standard label is closest to the centroid of the label cluster, the standard label of the label cluster is changed to the non-standard label; that is, the label closest to the centroid of the label cluster is changed from a non-standard label to a standard label. The original standard label is changed to a non-standard label. If the standard label is closest to the centroid of the label cluster, the standard label of the label cluster is not changed.

[0132] When vectorizing standard and non-standard labels, the smaller the angle between the standard and non-standard label vectors and the centroid, the closer the standard and non-standard labels are to the centroid. Conversely, the larger the angle between the standard and non-standard label vectors and the centroid, the farther the standard and non-standard labels are from the centroid.

[0133] For example, if there exists a non-standard tag whose semantic vector has the smallest angle with the centroid, then the non-standard tag corresponding to that non-standard tag vector is taken as the new standard tag, and the original standard tag is changed to a non-standard tag. If there is no non-standard tag whose semantic vector has the smallest angle with the centroid, that is, the standard tag whose semantic vector has the smallest angle with the centroid, then the standard tag of the tag cluster remains unchanged.

[0134] In other embodiments, the centroid of the tag cluster is calculated each time a non-standard tag is added to the tag cluster; the difference (distance, angle) between the centroid of the tag cluster and the standard tag is calculated. If the difference is greater than a set value, step S202 is executed to perform a self-check on the rationality of the standard tags within the tag cluster.

[0135] In this embodiment, the technical solution achieves multiple beneficial effects through dynamic centroid calculation and adaptive updating of standard tags: First, centroid calculation is triggered based on update events (quantity threshold, duration threshold, number of new tags, etc.), ensuring that the centroid reflects the latest semantic distribution of the tag cluster in real time. This avoids centroid lag caused by dynamic changes in the tag cluster (such as the addition of new tags or content iteration), providing an objective benchmark for judging the rationality of standard tags. Second, centroid calculation integrates the features (vector summation or three-dimensional spatial mean) of all standard and non-standard tags within the cluster, breaking through the limitations of a single standard tag and more comprehensively representing the overall semantic core of the cluster, improving the consistency and representativeness of semantics within the tag cluster. Third, standard tags are dynamically updated based on the distance or vector angle between the centroid and the tag (when a non-standard tag is closer to the centroid, it replaces the original standard tag), ensuring that the standard tag is always the most semantically core identifier within the cluster. This avoids semantic confusion within the cluster caused by outdated or biased initial standard tags, guaranteeing the authority and accuracy of the standard tags. In addition, the self-checking mechanism for differences when adding new tags can promptly capture deviations between standard tags and cluster cores. Combined with flexible triggering conditions for update events, it can achieve dynamic optimization of tag clusters while controlling computational costs, thus balancing system efficiency and semantic accuracy.

[0136] Overall, this solution enhances the semantic stability and adaptability of tag clusters through dynamic centroid calibration and adaptive iteration of standard tags, providing reliable tag system support for efficient retrieval based on standard tags and significantly improving the processing efficiency of subsequent business operations.

[0137] In some embodiments, if the standard label of a tag cluster changes, the non-standard labels within the tag cluster are re-analyzed (see the analysis method in step S120). That is, all non-standard labels in the tag cluster are re-analyzed as unanalyzed original labels. This prevents non-standard labels that originally belonged to the tag cluster from becoming incompatible with the standard label due to the change in the standard label, allowing them to find a more suitable tag cluster or generate a new tag cluster as the standard label.

[0138] Please see Figure 6 , Figure 6 A flowchart illustrating the updating of non-standard tags in a tag cluster where standard tags have changed, according to an embodiment of this application, is shown. This application provides steps for updating non-standard tags in a tag cluster where standard tags have changed, including:

[0139] Step S301: Change the non-standard label with the lowest similarity to the standard label in the changed label cluster to the label to be reviewed and re-analyze. The changed label cluster refers to the label cluster in which the standard label has been changed.

[0140] Step S302: If the label to be reviewed is added to a label cluster outside the changed label cluster or is used as a new standard label, the label to be reviewed is re-determined in the changed label cluster for analysis until the label to be reviewed is added to the changed label cluster.

[0141] The two steps described above are described in detail below.

[0142] In step S301, the non-standard labels with the lowest similarity to the standard labels in the changed label clusters are changed to the labels to be reviewed and re-analyzed. The changed label clusters refer to the label clusters in which the standard labels have been changed.

[0143] In step S302, if the label to be reviewed is added to a label cluster outside the changed label cluster or is used as a new standard label, the label to be reviewed is re-determined in the changed label cluster for analysis until the label to be reviewed is added to the changed label cluster.

[0144] In other words, if the standard label of a tag cluster changes, the non-standard label with the lowest similarity to the standard label in the changed tag cluster is used as the label to be reviewed and re-analyzed. If the label to be reviewed does not belong to the changed tag cluster, the label to be reviewed is added to a tag cluster outside the changed tag cluster as a new standard label or as a non-standard label. In this way, non-standard labels that do not belong to the original tag cluster are filtered out. The label to be reviewed is re-determined for analysis (see the analysis method in step S120), that is, the label to be reviewed is re-determined and analyzed among the non-standard labels in the changed tag cluster until the tag cluster associated with the label to be reviewed does not change, that is, after analysis, the label to be reviewed is still associated with the changed tag cluster. At this time, the non-standard labels in the changed tag cluster are compatible with the changed tag cluster.

[0145] It is important to clarify that each standard tag is associated with multiple non-standard tags. The following method is used to determine the most similar standard tag: for the tag to be analyzed, the similarity between the tag and each non-standard and standard tag is obtained. The non-standard and standard tags most similar to the tag to be analyzed are used as the initial selected tags. If the initial selected tag is a non-standard tag, then the standard tag of the tag cluster to which the initial selected tag belongs is used as the standard tag associated with the tag to be analyzed. If the initial selected tag is a standard tag, then that initial selected tag is directly used as the standard tag associated with the tag to be analyzed.

[0146] In this embodiment, the technical solution brings multiple beneficial effects through a dynamic verification and correction mechanism after standard label changes: First, when the standard label of a label cluster changes, the original label with the lowest similarity to the new standard label is prioritized for re-analysis as the label to be reviewed. This can accurately locate labels in the original cluster that may fail to adapt due to standard label iteration (such as labels with original clustering bias). Secondary verification corrects mis-clustering problems, ensuring high semantic consistency within the label cluster and avoiding label mismatch within the cluster due to standard label updates. Second, the cyclic processing mechanism in step S302 (when the label to be reviewed does not belong to the original label cluster, the label to be reviewed is re-determined and analyzed from the non-standard labels in the original label cluster) ensures that each label to be reviewed can be reassigned to the most suitable label cluster or become a new standard label. This avoids labels being missed, discovers potential new feature labels, enriches the coverage of the label system, and achieves efficient utilization of label resources.

[0147] Overall, this solution enhances the self-optimization capability of tag clusters through proactive self-checking and dynamic adjustment after standard tag changes, ensuring that the tag system maintains accuracy and integrity throughout the iteration process. It provides a dynamically adaptable tag foundation for efficient retrieval based on standard tags in claims and reliable construction of tag knowledge graphs, further improving the processing accuracy of subsequent business operations.

[0148] In other embodiments, if the standard label of the standard label cluster changes, the top N standard labels most similar to the standard label, along with the standard label itself, are used as reference labels, where N is a positive integer. All non-standard labels in the label cluster are used as labels to be determined, and the similarity between the label to be determined and the reference labels is calculated sequentially, referring to the analysis method for the original labels. If the similarity between the label to be determined and the reference label is greater than the second similarity value, the label to be determined is added to the label cluster containing the reference label with the highest similarity.

[0149] If the similarity between the label to be determined and each reference label is less than or equal to the second similarity value, then the similarity between the label to be determined and each non-reference label is calculated; non-reference labels refer to standard labels other than reference labels. If there is a non-reference label with a similarity greater than the second similarity value to the label to be determined, then the label to be determined is added as a non-standard label to the label cluster of that standard label. If the similarity between the label to be determined and both non-reference labels is less than the first similarity value, then the label to be determined is used as a standard label. If the similarity between the label to be determined and non-reference labels is greater than or equal to the first similarity value and less than or equal to the second similarity value, then the original information of the product corresponding to the label to be determined is obtained, the label to be determined is corrected, and the label to be determined is re-analyzed until the label to be determined is used as a standard label or added to a label cluster.

[0150] It is important to clarify that each reference tag and non-reference tag is associated with multiple non-standard tags (original tags that have already been analyzed and are consistent with the original tag format and content). The following method is used to determine the most similar reference tag or non-reference tag: For the tag to be determined, the similarity between the tag to be determined and each non-standard tag, reference tag, or non-reference tag is obtained. The non-standard tag, reference tag, or non-reference tag most similar to the tag to be determined is used as the initial selected tag. If the initial selected tag is a non-standard tag, then the reference tag or non-reference tag of the tag cluster containing the initial selected tag is used as the standard tag associated with the tag to be determined. If the initial selected tag is a reference tag or non-reference tag, then the initial selected tag is directly used as the standard tag associated with the tag to be determined.

[0151] In some embodiments, non-standard labels of label clusters with standard label changes are updated in the following manner.

[0152] If the standard label of a label cluster changes, the original label with the lowest similarity to the standard label in the label cluster is taken as the label to be reviewed; at the same time, the top N standard labels most similar to the standard label, as well as the standard label, are taken as reference labels, where N is a positive integer; for the label to be analyzed in the label cluster, the similarity with the reference label is calculated in turn; based on the similarity between the label to be reviewed and the reference label, the label to be reviewed is added to the label cluster of the reference label with the highest similarity, or the label to be reviewed is used as a new standard label to form a label cluster.

[0153] Specifically, if the similarity between the label to be reviewed and the reference label is greater than the second similarity value, the label to be reviewed is added to the label cluster of the reference label with the highest similarity. If the similarity between the label to be reviewed and each reference label is less than or equal to the second similarity value, the similarity between the label to be reviewed and each non-reference label is calculated; non-reference labels refer to standard labels other than reference labels. If there is a non-reference label with a similarity greater than the second similarity value to the label to be reviewed, the label to be reviewed is added as a non-standard label to the label cluster of that non-reference label. If the similarity between the label to be reviewed and all non-reference labels is less than the first similarity value, the label to be reviewed is used as a standard label. If the similarity between the label to be reviewed and non-reference labels is greater than or equal to the first similarity value and less than or equal to the second similarity value, the original information of the product corresponding to the label to be reviewed is obtained, the label to be reviewed is corrected, and the label to be reviewed is re-analyzed until the label to be reviewed is used as a standard label or is added to a label cluster. Finally, the label to be reviewed is repeatedly obtained and analyzed until the label to be reviewed still belongs to the original label cluster.

[0154] It is important to clarify that each reference tag and non-reference tag is associated with multiple non-standard tags (original tags that have already been analyzed, consistent with the original tag format and content). The following method is used to determine the most similar reference tag or non-reference tag: For the tag to be analyzed, the similarity between the tag to be analyzed and each non-standard tag, reference tag, or non-reference tag is obtained. The non-standard tag, reference tag, or non-reference tag most similar to the tag to be analyzed is used as the initial selected tag. If the initial selected tag is a non-standard tag, then the reference tag or non-reference tag in the tag cluster to which the initial selected tag belongs is used as the standard tag associated with the tag to be analyzed. If the initial selected tag is a reference tag or non-reference tag, then the initial selected tag is directly used as the standard tag associated with the tag to be analyzed.

[0155] In this embodiment, the technical solution achieves multiple beneficial effects through a mechanism of reference label anchoring + hierarchical comparison verification and cyclic optimization: First, by using the top N standard labels most similar to the new standard label and the original standard label as reference labels, a comparison benchmark with the highest semantic relevance is provided for the label to be reviewed. This not only narrows the comparison range to improve efficiency but also ensures that the label to be reviewed can be assigned to the label cluster with the closest semantics through nearest neighbor reference, reducing errors in cross-semantic clustering. Second, the hierarchical comparison logic (first reference label, then non-reference label) combined with the first and second similarity threshold judgments achieves accurate sorting of the label to be reviewed: labels with high similarity to reference labels are assigned to the corresponding cluster, labels matching non-reference labels are flexibly included in other clusters, and labels that do not match either become the new standard label. The intermediate ambiguity is corrected and calibrated through the original product information, ensuring the accuracy of label classification from multiple dimensions and avoiding label mismatch caused by changes in standard labels. Finally, the mechanism of cyclical processing until the tag to be reviewed still belongs to the original cluster forms a self-purification closed loop of the tag cluster. By repeatedly verifying and redistributing the tags to be reviewed that are most likely to be mismatched, the semantic consistency between the remaining tags in the original cluster and the new standard tags after the standard tags are changed is ensured. At the same time, potential new feature tags are explored to enrich the system, taking into account both the stability and scalability of the tag cluster.

[0156] Overall, this solution enhances the adaptive adjustment capability of the tag system during standard tag iteration, providing dynamic optimization technical support for the semantic unification of tag clusters in claims, the accuracy of standard tags, and efficient tag-based retrieval and recommendation, significantly improving the robustness and business adaptability of the tag system.

[0157] In step S130, each tag cluster has a standard tag. The standard tag serves as the core semantic anchor point within the tag cluster, associating with non-standard tags. It is uniform in expression and semantically unique, and is a key basis for matching target products during retrieval. Each standard tag corresponds to product data, and the non-standard tags associated with the standard tags also have corresponding product data. Each product data corresponds to a product (goods or services). The products corresponding to the standard tags and non-standard tags within a tag cluster are considered as the products corresponding to that tag cluster. Standard tags are added to all products corresponding to this tag cluster, completing the product tag mining.

[0158] In step S140, when the user performs a search, the system responds to the search command based on the user's input and extracts tags from the search content. The tag extraction method for the search content can employ the techniques described in step S110.

[0159] The tags extracted from the search content are used as search tags. The similarity between the search tags and each non-standard and standard tag is calculated. The non-standard tag or target tag most similar to the standard tag is used as the initial target tag. If the initial target tag is a non-standard tag, the standard tag within the tag cluster where the initial target tag is located is used as the target tag. If the initial target tag is a standard tag, it is directly used as the target tag.

[0160] In this embodiment, after determining the target tag, the similarity between the target tag and the search tag is obtained. If the similarity between the target tag and the search tag is less than or equal to the search threshold, the user is prompted to supplement the search content. If no supplementary content is received, the target product is directly determined as the search result based on the target tag.

[0161] In some embodiments, when determining the target tag, the similarity between the search tag and each standard tag is first obtained. If the similarity between the search tag and a standard tag is greater than a search threshold, then the standard tag is used as the target tag. If the similarity between the search tag and all standard tags is less than or equal to the search threshold, the similarity between the search tag and non-standard tags is obtained. If the similarity between the search tag and a non-standard tag is greater than the search threshold, then the standard tag of the tag cluster to which the non-standard tag belongs is used as the target tag. If the similarity between the search tag and all non-standard tags is less than or equal to the search threshold, the user is prompted to supplement the search content.

[0162] If no supplementary information is provided, the highest similarity between the search tag and the standard tag is taken as the first similarity, and the highest similarity between the search tag and the non-standard tag is taken as the second similarity. If the first similarity is greater than the second similarity, the standard tag with the highest similarity to the search tag is directly taken as the target tag. If the second similarity is greater than the first similarity, the standard tag of the tag cluster containing the non-standard tag with the highest similarity to the search tag is taken as the target tag. If the first and second similarities are equally high, the first similarity is considered to be greater than the second similarity. This is because when a standard tag can represent a standard tag, the non-standard tags in its cluster are more similar to the search tag.

[0163] After determining the target tags, the target products are identified as search results based on the target tags.

[0164] In some embodiments, the similarity calculation between the search tag and the standard tag, non-standard tag, and target tag can be performed using any of the methods described in this technical solution for calculating the similarity between the original tag and the standard tag.

[0165] On the one hand, this solution establishes a unified semantic benchmark for the chaotic original tags by analyzing each original tag individually to determine standard tags and associated standard tags, forming tag clusters (each standard tag corresponds to a unique tag cluster). Standard tags, as core anchors within the same semantic category, incorporate non-standard tags with different expressions but consistent semantics into their corresponding tag clusters, completely breaking the isolated and unregulated state of the original tags. Simultaneously, it clarifies that tags cover at least one of the products and services, further adapting to the tag management needs of multiple types of objects, transforming the originally disordered tag library into a structured system centered on standard tags, directly solving the problems of chaotic tag systems and lack of unified standards in the background technology.

[0166] On the other hand, this solution adds standard tags to the corresponding products within the tag cluster, ensuring that all objects belonging to the same semantic category (regardless of how the original tags are expressed) carry a unified standard tag, eliminating retrieval omissions caused by differences in the original tags for the same type of object. Simultaneously, the retrieval process uses the standard tags corresponding to the search content as the target tags, directly matching the tag clusters and objects associated with the standard tags. This avoids the problems of relying on non-standard tags and being interfered with by expression differences in traditional retrieval—it eliminates the need to match diverse non-standard tags one by one, and avoids introducing irrelevant results due to tag expression deviations. From the source matching logic of the retrieval, it solves the shortcomings of low retrieval accuracy and poor efficiency in the background technology, ensuring that users can quickly locate the target product or service with a completely matching semantics.

[0167] Figure 7 A computer device structural block diagram of a product retrieval method for tag mining according to an embodiment of this application is shown.

[0168] It should be noted that, Figure 7 The computer device 800 shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0169] like Figure 7 As shown, the computer device 800 includes a central processing unit (CPU) 801, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 802 or programs loaded from storage section 808 into random access memory (RAM). The RAM 803 also stores various programs and data required for device operation. The CPU 801, ROM 802, and RAM 803 are interconnected via a bus 804. An input / output interface 805 (I / O interface) is also connected to the bus 804.

[0170] The following components are connected to the input / output interface 805: an input section 806 including a keyboard, mouse, etc.; an output section 807 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 808 including a hard disk, etc.; and a communication section 809 including a network interface card such as a local area network card, modem, etc. The communication section 809 performs communication processing via a network such as the Internet. A drive 810 is also connected to the input / output interface 805 as needed. A removable medium 811, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 810 as needed so that computer programs read from it can be installed into the storage section 808 as needed.

[0171] Specifically, according to embodiments of this application, the processes described in the various method flowcharts can be implemented as computer software programs. For example, embodiments of this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 809, and / or installed from removable medium 811. When the computer program is executed by central processing unit 801, it performs the various functions defined in the device of this application.

[0172] It should be noted that the computer-readable medium shown in the embodiments of this application can be a computer-readable signal medium, a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor device, apparatus, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM), flash memory, optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution device, apparatus, or apparatus. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such transmitted data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. The computer-readable signal medium can also be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution device, apparatus, or apparatus. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to wireless, wired, etc., or any suitable combination thereof.

[0173] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of devices, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based device that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0174] It should be noted that although several modules or units for the device used to perform actions have been mentioned in the detailed description above, this division is not mandatory. In fact, according to the embodiments of this application, the features and functions of two or more modules or units described above can be embodied in one module or unit. Conversely, the features and functions of one module or unit described above can be further divided and embodied by multiple modules or units.

[0175] Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein can be implemented by software or by combining software with necessary hardware. Therefore, the technical solutions according to the embodiments of this application can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (such as a CD-ROM, USB flash drive, external hard drive, etc.) or on a network, including several instructions to cause a computing device (such as a personal computer, server, touch terminal, or network device, etc.) to execute the methods according to the embodiments of this application.

[0176] Other embodiments of this application will readily occur to those skilled in the art upon consideration of the specification and practice of the invention disclosed herein. This application is intended to cover any variations, uses, or adaptations of this application that follow the general principles of this application and include common knowledge or customary techniques in the art not disclosed herein.

[0177] It should be understood that this application is not limited to the precise structure described above and shown in the accompanying drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is limited only by the appended claims.

Claims

1. A product retrieval method based on tag mining, characterized in that, The method includes: Tag extraction is performed on the acquired product data to obtain the original tags for each product, wherein the products include at least one of goods and services; Each of the original tags is analyzed individually, and the original tags are either identified as standard tags or non-standard tags are associated with corresponding standard tags to form at least one tag cluster; wherein each of the standard tags corresponds to a tag cluster; wherein, When an update event occurs in any tag cluster, the centroid of the tag cluster is calculated; if a non-standard tag is closest to the centroid of the tag cluster, the standard tag of the tag cluster is changed to the non-standard tag; if the standard tag is closest to the centroid of the tag cluster, the standard tag of the tag cluster is not changed. If the standard label for clustering changes, the top N standard labels most similar to the standard label, along with the standard label itself, will be used as reference labels, where N is a positive integer. All non-standard tags in the tag cluster are treated as tags to be determined, and the similarity between the tags to be determined and the reference tags is calculated sequentially. If the similarity between the tags to be determined and the reference tags is greater than the second similarity value, the tags to be determined are added to the tag cluster containing the reference tag with the highest similarity. If the similarity between the tag to be determined and each reference tag is less than or equal to the second similarity value, then the similarity between the tag to be determined and each non-reference tag is calculated. The non-reference tag refers to the standard tag other than the reference tag. If there is a non-reference tag with a similarity greater than the second similarity value to the tag to be determined, then the tag to be determined is added as a non-standard tag to the tag cluster of the non-reference tag. If the similarity between the label to be determined and the non-reference label is less than the first similarity value, then the label to be determined is used as the standard label; the first similarity value is less than the second similarity value; if the similarity between the label to be determined and the non-reference label is greater than or equal to the first similarity value and less than or equal to the second similarity value, then the original information of the product corresponding to the label to be determined is obtained, the label to be determined is corrected, and the label to be determined is re-analyzed until the label to be determined is used as the standard label or added to a label cluster. Add the standard labels included in the label cluster to each product corresponding to the label cluster; The product corresponding to the target tag is taken as the target product. The target tag refers to the standard tag corresponding to the search content.

2. The method according to claim 1, characterized in that, Tag extraction is performed on the acquired product data to obtain the original tags for each product, including: Tag extraction is performed on the product data to obtain multiple initial tags; Remove initial labels that are irrelevant to the target domain; Initial tags that are not directly derived from the product data and have undergone semantic transformation will be filtered out. Filter out initial tags that do not contain entity content; The retained initial labels are generalized by modifiers to obtain the original labels.

3. The method according to claim 1, characterized in that, Each of the original tags is analyzed individually, and the original tags are identified as standard tags, or identified as non-standard tags and associated with corresponding standard tags, including: Select any original label, and determine the standard label with the highest similarity as the similar standard label based on the similarity between the original label and each of the standard labels; Based on the similarity between the original tag and similar standard tags, the original tag is either used as a new standard tag or added as a non-standard tag to the tag cluster containing the similar standard tags; Select the next original label for analysis, and continue until the analysis of all original labels is completed.

4. The method according to claim 3, characterized in that, The method further includes: If there is no standard label when analyzing the original label, the original label is directly used as the standard label.

5. The method according to claim 3, characterized in that, Based on the similarity between the original tag and similar standard tags, the original tag is either used as a new standard tag or added as a non-standard tag to the tag cluster containing the similar standard tags, including: If the similarity between the original label and the similar standard label is less than the first similarity value, then the original label is used as the standard label; If the similarity between the original tag and the similar standard tag is greater than the second similarity value, then the original tag is added as a non-standard tag to the tag cluster where the similar standard tag is located, and the first similarity value is less than the second similarity value; If the similarity between the original label and the similar standard label is greater than or equal to the first similarity value and less than or equal to the second similarity value, then the original information of the product corresponding to the original label is obtained to correct the original label, and the original label is re-analyzed until the original label is used as a standard label or added to a label cluster.

6. The method according to claim 1, characterized in that, The method further includes: If the standard label of the label cluster changes, the non-standard label within the label cluster will be changed back to the original label for re-analysis.

7. The method according to claim 1, characterized in that, The method further includes: The non-standard label with the lowest similarity to the standard label in the changed label cluster is changed to the label to be reviewed and re-analyzed. The changed label cluster refers to the label cluster in which the standard label has been changed. If the tag to be reviewed is added to a tag cluster outside the changed tag cluster or is used as a new standard tag, the tag to be reviewed is re-determined and analyzed in the changed tag cluster until the tag to be reviewed is added to the changed tag cluster.

8. A product retrieval device based on tag mining, comprising a memory, a processor, and a readable program stored in the memory, characterized in that, The processor executes the readable program to implement the method of any one of claims 1 to 7.

9. A readable storage medium, characterized in that, It stores a readable program / instruction that, when executed by a processor, implements the method of any one of claims 1 to 7.

Images