Label system construction method and device, label providing method and device, equipment and storage medium

Abstract

The embodiment of the invention provides a label system construction method and device, a label providing method and device, equipment and a storage medium. The label system construction method comprises the steps of collecting comment data of various items; performing label mining on each piece of comment data of each type of articles by utilizing a label mining model to obtain a comment label of each piece of comment data of each type of articles; merging the comment labels of the comment data of each category of articles to obtain an original comment label of each category; and filtering out semantically repeated labels from the original comment labels of each category to obtain the selected comment labels of each category. According to the embodiment of the invention, the comment labels are automatically mined from the comment data by utilizing the model, so that the time and manpower input for manually creating and distributing the labels are reduced, the label creating efficiency is improved, and the number and types of the labels are expanded; the label comes from the actual expression of the user and is more in line with the language habit of the user, the display effect is more natural, and the stiff feeling is reduced.

Description

Technical Field

[0001] Embodiments of the present invention relate to the field of computer technology, and in particular to a label system construction, a label providing method, an apparatus, a device and a storage medium. Background Art

[0002] After online shopping, users will leave comments on the page, which can be used as a reference for other users to shop. Comment tags are a way to simplify user comment data. By adding tags to comments, users can quickly understand the main content of the comments.

[0003] Currently, labeling of item comments is mainly done manually by creating a set of tags manually based on the characteristics of the item, setting a mapping relationship between comments and tags, querying the mapping relationship based on the actual comments of the item to determine the hit tags, and displaying the hit tags on the page so that users can quickly understand the core content of the comments through tags when browsing items.

[0004] In the process of realizing the present invention, the inventors found that the current labeling system has at least the following problems: labels rely on manual creation, with low creation efficiency and a limited number of labels, making it difficult to cover all possible comment content and unable to adapt to the diversity of comment content, resulting in stiff labels and unnatural display effects. Summary of the invention

[0005] The embodiments of the present invention provide a label system construction, label providing method, device, equipment and storage medium, which can automatically construct labels, improve label construction efficiency and improve the quality of constructed labels.

[0006] In a first aspect, a label system construction method provided by an embodiment of the present invention includes:

[0007] Collect review data for various items;

[0008] Use the label mining model to perform label mining on each review data of each category of items to obtain the review labels of each review data of each category of items;

[0009] Combine the review labels of each review data of each category of items to obtain the original review labels of each category;

[0010] Semantically repeated tags are filtered out from the original review tags of each category to obtain the selected review tags of each category.

[0011] In a second aspect, a label providing method provided by an embodiment of the present invention includes:

[0012] Get the review data of the target item;

[0013] Identify the category to which the target item belongs and obtain the target category;

[0014] From the selected review tags of the target category, a review tag semantically matching the review data of the target item is selected to obtain a target review tag; wherein the selected review tags of the target category are constructed according to the tag system construction method described in any embodiment of the present invention;

[0015] Provide target review labels for target items.

[0016] In a third aspect, an embodiment of the present invention provides a label system construction device, including:

[0017] The collection module is used to collect review data of various items;

[0018] A mining module is used to perform label mining on each review data of each category of items using a label mining model to obtain a review label for each review data of each category of items;

[0019] The merging module is used to merge the comment labels of each comment data of each category of items to obtain the original comment labels of each category;

[0020] The deduplication module is used to filter out semantically repeated tags from the original review tags of each category to obtain the selected review tags for each category.

[0021] In a fourth aspect, an embodiment of the present invention provides a label providing device, comprising:

[0022] The acquisition module is used to obtain the review data of the target item;

[0023] The recognition module is used to identify the category to which the target item belongs and obtain the target category;

[0024] A selection module is used to select a comment tag that semantically matches the comment data of the target item from the selected comment tags of the target category to obtain a target comment tag; wherein the selected comment tags of the target category are constructed according to the tag system construction method as described in any embodiment of the present invention;

[0025] Provides a module for providing target review labels for target items.

[0026] In a fifth aspect, an electronic device provided by an embodiment of the present invention includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements a label system construction method as in any embodiment of the present invention, or implements a label providing method as in any embodiment of the present invention.

[0027] In a sixth aspect, a computer-readable storage medium provided by an embodiment of the present invention stores a computer program thereon, which, when executed by a processor, implements a label system construction method as in any embodiment of the present invention, or implements a label providing method as in any embodiment of the present invention.

[0028] In the embodiment of the present invention, the label mining model is used to automatically mine labels, which reduces the time and manpower input of manually creating and assigning labels, and improves the efficiency of label creation; the label mining model can deeply analyze the review content, automatically mine and generate a variety of labels, expand the number and types of labels, make the labels more comprehensive, and capture a wider range of user opinions and experiences; the label mining model can process various types of review data, adapt to different vocabulary, expressions and contexts, and make the generated labels more flexible and close to the actual review content. Labels are mined from the review data of the item, and the labels directly reflect the real feelings and opinions of the user after actually using the item, and the labels will be more natural and close to the user; the review data is dynamically changing, and labels are automatically mined from the review data, so that the label system can quickly respond to and adapt to new user feedback and update in a timely manner; the automatically generated labels are derived from the actual expression of the user, which is more in line with the user's language habits, and the display effect is more natural, reducing the sense of stiffness, and the label quality is higher. Since the label mining is performed according to categories, there are duplicate labels mined from the review data of items in the same category. The original review labels mined can be semantically deduplicated. Deduplication reduces the number of labels, avoids information redundancy, helps highlight the core ideas of the reviews, and makes the label display more concise and clear. By processing the review data and constructing review labels according to categories, labels can be reused for similar items, avoiding the work of repeatedly constructing labels and greatly improving the efficiency of label construction; using the same label system for items in the same category can ensure the unification and consistency of review labels, so that users can get a consistent experience when browsing different items; when introducing new items, the label system of existing similar items can be directly applied to quickly launch the review labels of new items, which is easy to expand. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solution of the present invention, the drawings required for use in the embodiments are briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0030] Figure 1 is a flow chart of a method for constructing a label system according to an embodiment of the present invention;

[0031] Figure 2ais another flow chart of the label system construction method according to an embodiment of the present invention;

[0032] Figure 2b is an example diagram of a label system construction method according to an embodiment of the present invention;

[0033] Figure 3 is a flow chart of a label providing method according to an embodiment of the present invention;

[0034] Figure 4 is a structural schematic diagram of a label system construction device according to an embodiment of the present invention;

[0035] Figure 5 is a structural schematic diagram of a label providing device according to an embodiment of the present invention;

[0036] Figure 6 It is a structural schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0037] In order to enable those skilled in the art to better understand the scheme of the present invention, the technical scheme in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of the present invention.

[0038] It should be noted that the terms "first", "second", etc. in the specification and claims of the present invention and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present invention described herein can be implemented in an order other than those illustrated or described herein. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions, for example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units that are clearly listed, but may include other steps or units that are not clearly listed or inherent to these processes, methods, products or devices.

[0039] Figure 1It is a flow chart of the label system construction method provided by an embodiment of the present invention. The label system construction method provided by an embodiment of the present invention can be applied to the scenario where an online platform provides comment labels for items. The label system construction method can be executed by the label system construction device provided by an embodiment of the present invention, and the device can be implemented in software and / or hardware. In a specific embodiment, the device can be integrated into an electronic device, and the electronic device can be a computer, a server, etc. The following embodiment is described by taking the example of integrating the label system construction device into an electronic device, see Figure 1 The label system construction method of this embodiment may include the following steps:

[0040] Step 101, collect review data of various items.

[0041] Items can be physical products or virtual products or services. For example, items can be physical products such as household appliances, electronic products, furniture, clothing, etc., virtual products such as software, online courses, digital media content, financial digital products, etc., and services such as consulting services and content subscription services. For ease of understanding, the following will be explained as an example of items being physical products. Users can obtain items through online platforms. After obtaining items, they can post comments under the corresponding items on the online platform based on the usage of the items, personal experience, etc., to form comment data. These comment data may include users' feelings and opinions on the performance, quality, appearance, usage experience, logistics and transportation of items.

[0042] When displaying items on online platforms, they are divided into one or more categories according to the characteristics, properties, and uses of the items. Items in the same category may have the same characteristics, properties, or uses. The following is an example of a multi-level category:

[0043] First-level category: electronic products;

[0044] Secondary category: mobile phone;

[0045] Level 3 categories: Brand A mobile phones, Brand B mobile phones;

[0046] Secondary category: Computer;

[0047] Third-level categories: laptops, desktop computers.

[0048] The category level for collecting comment data can be determined based on the complexity of the item, the resource allocation of the equipment, the actual operational needs, etc. Complex items may be divided into more categories, and lower-level categories can be collected, such as third-level categories and fourth-level categories; simple items are divided into fewer categories, and higher-level categories can be collected, such as first-level categories or second-level categories. If the equipment resources are limited and the amount of data that can be processed each time is small, the lower-level categories can be collected; if the amount of data that can be processed is large, the higher-level categories can be collected. If you need to analyze the data in depth and need more refined classified comment data, you can collect lower-level categories; if you just want to simply analyze the data, you can collect higher-level categories.

[0049] After determining the category level of the collected data, the review data of various items can be collected from the online platform according to the determined category level. The secondary category can usually classify the items into a more specific level. Therefore, it is generally more appropriate to collect the secondary category. Taking the collection of the secondary category as an example, the review data of mobile phones, computers, refrigerators, washing machines, etc. can be collected from the e-commerce platform. All the review data of various items can be collected, or part of the review data of various items can be collected. For example, the review data generated by the mobile phone category in a specified period of time (such as the past year or the past three months) can be collected, or part of the review data can be randomly or according to preset rules from all the review data of the mobile phone category. Specifically, network crawling technology, interface call technology, etc. can be used to collect the review data of various items from the online platform. One category of items includes multiple items. For example, items such as mobile phones may include different models of mobile phones produced by different manufacturers. Each item can have a different stock keeping unit (SKU), and each item can have multiple review data.

[0050] After collecting the review data of various items, you can pre-process the review data of various items to improve the quantity and quality. Pre-processing can include: removing duplicate data, removing invalid data (for example, the review data is a piece of content that is completely unrelated to the item), correcting erroneous data, filling missing values, normalizing data in different formats, etc.

[0051] Step 102: Use a label mining model to perform label mining on each review data of each category of items to obtain a review label for each review data of each category of items.

[0052] The label mining model can be a large language model (LLM), or a traditional machine learning model, decision tree model, or clustering model. A large language model refers to a deep learning model trained with a large amount of text data, and is mainly used for natural language processing (NLP) tasks. The label mining model can be obtained by pre-training with training data, and the sample data in the training data can be comment data, and the sample label can be the comment label added to the corresponding comment data.

[0053] Tag mining is the process of extracting or summarizing representative and useful information from comment data. Comment tags are keywords or phrases extracted or summarized from comment data, which can summarize and represent the main content and theme of the comment.

[0054] Specifically, the label mining model can be used to perform label mining on each comment data, and one or more comment labels can be mined from one comment data. For example, in a comment data of a mobile phone, comment labels such as short battery life, high screen resolution, good sound quality, and clear photos may be mined, and in a comment data of an air conditioner, comment labels such as low noise, power saving, and fast delivery may be mined.

[0055] Step 103: combine the review labels of the review data of each category of items to obtain the original review labels of each category.

[0056] Tag merging can be done by putting the tags mined from all the review data of the same type of items into a set, and the number of sets will be the same as the number of categories. For example, the tags mined from all the review data of all mobile phones can be put into a set to get the original review tags of mobile phones; the tags mined from all the reviews of all washing machines can be put into a set to get the original review tags of washing machines. There are multiple original review tags for each category.

[0057] Step 104 , filtering out semantically duplicated tags from the original review tags of each category to obtain selected review tags for each category.

[0058] Semantic duplication refers to different review tags that are different in literal meaning but semantically represent the same or very similar content. For example, "long battery life" and "excellent battery life" are semantically duplicated, so only one can be retained. Since similar items have the same characteristics, attributes, or uses, the user feedback content of similar items may also be the same or similar. Therefore, the review tags mined from the review data of similar items may also have semantic duplication and need to be removed.

[0059] Specifically, for the original review labels of any category, the semantic similarity between these original review labels can be calculated. For the review labels whose semantic similarity exceeds the similarity threshold, only one can be retained, thereby filtering out semantically repeated labels. The calculation of semantic similarity can adopt a similarity calculation method based on word embedding, a string similarity algorithm, etc., which is not specifically limited here. For example, for two original review labels, the two original review labels can be vectorized, and then the cosine similarity between the two vectors is calculated. If the cosine similarity exceeds the preset similarity threshold (for example, 90%, 85%), one of them is retained.

[0060] Specifically, for the original review labels of any category, these original review labels can be vectorized and clustered to find the cluster center of each cluster. The original review labels corresponding to the cluster center vector are retained, and the original review labels corresponding to other points in each cluster except the cluster center are deleted, thereby filtering out semantically repeated labels.

[0061] In this embodiment, the label mining model is used to automatically mine labels, which reduces the time and manpower input of manually creating and assigning labels, and improves the efficiency of label creation; the label mining model can deeply analyze the review content, automatically mine and generate a variety of labels, expand the number and types of labels, make the labels more comprehensive, and capture a wider range of user opinions and experiences; the label mining model can process various types of review data, adapt to different vocabulary, expressions and contexts, and make the generated labels more flexible and close to the actual review content. Labels are mined from the review data of items. The labels directly reflect the real feelings and opinions of users after actually using the items. The labels will be more natural and close to the users; the review data is dynamically changing, and the labels are automatically mined from the review data, so that the label system can quickly respond to and adapt to new user feedback and update in a timely manner; the automatically generated labels are derived from the actual expressions of users, which are more in line with the user's language habits, and the display effect is more natural, reducing the sense of stiffness, and the label quality is higher. Since the label mining is performed according to categories, there are duplicate labels mined from the review data of items in the same category. The original review labels mined can be semantically deduplicated. Deduplication reduces the number of labels, avoids information redundancy, helps highlight the core ideas of the reviews, and makes the label display more concise and clear. By processing the review data and constructing review labels according to categories, labels can be reused for similar items, avoiding the work of repeatedly constructing labels and greatly improving the efficiency of label construction; using the same label system for items in the same category can ensure the unification and consistency of review labels, so that users can get a consistent experience when browsing different items; when introducing new items, the label system of existing similar items can be directly applied to quickly launch the review labels of new items, which is easy to expand.

[0062] The following further illustrates the label system construction method provided by the embodiment of the present invention. Figure 2a and Figure 2b , Figure 2a is another flow chart of the label system construction method provided by an embodiment of the present invention, Figure 2b It is an example diagram of the label system construction method provided by an embodiment of the present invention.

[0063] like Figure 2a As shown, the method of this embodiment includes:

[0064] Step 201, collecting review data of various items.

[0065] You can collect all the review data of various items, or you can collect part of the review data of various items. For example, you can collect all the review data of various items on the online platform, or you can collect the review data of various items on the online platform in the past three months or six months. A large amount of review data will be collected for each type of item.

[0066] Step 202 , semantic segmentation is performed on each review data of each category of items to obtain semantic units of each review data of each category of items.

[0067] Semantic segmentation is to divide a piece of text into several meaningful units, such as phrases, sentences, keywords, etc., according to semantics. Through semantic segmentation, it is easy to extract the core content of the review data, which is convenient for subsequent analysis and processing. The semantic unit is the smallest meaningful text unit obtained after semantic segmentation. It can be a word, phrase or sentence, representing the specific semantic content in the review data. Performing semantic segmentation on a piece of review data can obtain at least one semantic unit. For example, a review data for a mobile phone can be segmented into specific semantic units such as "battery with good battery life", "good screen display effect", and "poor service attitude".

[0068] Step 203 , using a tag mining model to perform tag mining on the semantic units of each review data of each category of items, to obtain the comment tags of the semantic units in each review data of each category of items.

[0069] Each semantic unit can be input into the tag mining model, and the tag mining model can mine the comment tags of each semantic unit. One semantic unit can mine one or more comment tags.

[0070] The overall review may contain multiple viewpoints. If the label mining is performed directly on the entire review, it is easy to introduce noise and ambiguity. In this embodiment, the review data is divided into semantic units and then the labels are mined, so that each semantic unit corresponds to a specific viewpoint or information, which helps the label mining model understand the review data, reduces the ambiguity in the label mining process, and ensures that the mined labels are more accurate. For example, a review data is "The battery life is very good, but the screen is not big enough". This review data includes positive and negative feedback. Through semantic segmentation, it can be divided into two semantic units: "The battery life is very good" and "The screen is not big enough". In this way, two labels can be generated separately, so that both positive and negative content can be identified, ensuring that the labels are more accurate. Semantic units make the labels not only limited to the overall overview of the review, but can be specific to the details and specific features in the review. Fine-grained labels can more comprehensively reflect the user's feelings and feedback. Different users may have different views on different features. Through semantic segmentation and fine-grained labels, diverse perspectives can be captured, and the review data can be analyzed from multiple dimensions to improve the quality of labels.

[0071] Step 204 , combining the comment labels of the semantic units in the comment data of each category of items to obtain the comment labels of the comment data of each category of items.

[0072] All comment tags mined from the semantic unit of a comment data constitute the comment tag of this comment data. Each comment data can have multiple comment tags.

[0073] Step 205 , combining the review labels of the review data of each category of items to obtain the original review labels of each category.

[0074] All comment labels of all comment data of a category of items constitute the original comment labels of the category. There are multiple original comment labels for a category.

[0075] Step 206 , filtering out semantically duplicated tags from the original review tags of each category to obtain candidate review tags for each category.

[0076] Due to the large and diverse amount of review data, the diversity of natural language, and the same characteristics or attributes of similar items, there are a large number of semantically repeated labels in the original review labels of each category. For example, the amount of review data for each category of items is usually very large, and different users may use different words or expressions to evaluate the same item, but the information content conveyed by these reviews may be similar in substance. For example, "long battery life" and "very durable" are semantically repeated. For example, natural language is highly diverse and flexible, and the same meaning can be expressed in different language forms. Different users may use different words and sentences to describe the same feature due to personal expression habits, language style and other factors. For another example, the uses and characteristics of similar items may be the same or similar, and users may have the same or similar experience or feelings about similar items. These factors indicate that it is necessary to deduplicate labels.

[0077] Specifically, the semantic similarity between the original review tags of each category can be calculated, and for the review tags whose semantic similarity exceeds the preset similarity threshold, only one can be retained. The candidate review tags of each category are the tags remaining after filtering out semantically duplicated tags from the original review tags of the corresponding category.

[0078] Step 207, sending the candidate review tags of each category to the operation terminal, and obtaining the review results of the candidate review tags of each category from the operation terminal.

[0079] The operation terminal can be a management system or platform for operating and managing comment labels, usually used by operators. Operators can receive, view, adjust and confirm comment labels through the operation terminal. The review results are the confirmed and adjusted label results after the operator or system reviews these candidate labels. These results represent labels that have been manually reviewed and confirmed to be suitable for actual use. Specific adjustments may include deleting labels that are uncivilized, have unclear semantics, are completely irrelevant to the current category, modifying the label description, adding some new labels, etc. Figure 2b As shown, for example, when a category is unpopular and there are not enough comments, there may be insufficient labels. This can be supplemented by introducing procurement and sales expansion and third-party crawling, and adding some new labels.

[0080] Specifically, the candidate review labels for each category can be packaged and sent to the operation terminal, where the operation staff will view the candidate labels, conduct manual review and correction, and ensure the accuracy and applicability of the labels.

[0081] Adding manual review can ensure that the final generated review labels are more accurate and in line with actual user needs, reduce errors and deviations, and improve the accuracy and credibility of labels; the combination of automatic generation and manual review not only utilizes the ability of machines to efficiently process massive amounts of data, but also combines human wisdom and judgment to achieve the best results; by adding manual review, operators can also adjust and update review labels in a timely manner based on market changes and user feedback, maintaining the flexibility and adaptability of the system.

[0082] Step 208, determining the selected review tags for each category based on the review results.

[0083] Selected review tags are tags that are manually reviewed and ultimately determined to represent the characteristics of items in a specific category and user reviews. Selected tags enable users to obtain valuable information in the shortest possible time, improving browsing and shopping experience.

[0084] Suppose there is a batch of mobile phone review data. The original review tags mined by the tag mining model are, for example, "good battery life", "good screen display", "fast operation", "reasonable price", "clear photos", "fashionable body design" and "long battery life". In the original review tags, "good battery life" and "long battery life" have semantic duplication. After semantic deduplication, the candidate review tags obtained can be "good screen display", "fast operation", "reasonable price", "clear photos", "fashionable body design" and "long battery life". The candidate review tags are sent to the operation terminal for review. The operation staff finds that the tag "reasonable price" is subjective and can be deleted. Then the selected review tags can be "good screen display", "fast operation", "clear photos", "fashionable body design" and "long battery life".

[0085] After obtaining the selected review tags for each category, you can save the selected review tags for each category in the database. You can also add some sorting attributes to the selected review tags for each category for subsequent use. Sorting attributes are attributes used to sort tags. Sorting attributes can include the number of matching comments, sentiment categories, time freshness, interaction rate, posterior click data, etc.

[0086] The selected comment tags of each category can be semantically matched with the comment data of the corresponding category of items, and matching comment information can be set for the selected comment tags of each category based on the matching results. The matching comment information may include the number of matching comments and / or the specific comment data matched. Specifically, for the selected comment tags of each category, the comment data that semantically matches each selected comment tag can be found from the comment data of the corresponding category of items, and the number of matching comment data can be counted, and the number of matching comment data and / or the index or identifier of the matched specific comment data can be set as attributes to the corresponding selected comment tag. When the selected comment tags are subsequently displayed, the number of matching comments of the selected comment tags can be displayed, and the matching specific comment data can also be found based on the index or identifier, and the specific comment data matched by the selected comment tags can be displayed under the corresponding selected comment tags.

[0087] The number of matching comments reflects the frequency or proportion of the selected comment tag in user comments, which indirectly reflects the attention paid to this feature. By sorting the tags by the number of comments, you can determine the product features that users care about most. For example, "battery life is very good (150 comments)" is more popular than "screen display effect is good (100 comments)".

[0088] In addition, considering the amount of data that can be processed each time and the fact that comments will continue to be generated over time, it is also possible to obtain incremental comments and / or existing comments for each type of item. Incremental comments are comment data generated after label mining, and existing comments are comment data generated before label mining but not used for label mining; semantically match the incremental comments and / or existing comments of each type of item with the selected comment tags of the corresponding category; and update the matching comment information set to the selected comment tags of the corresponding category based on the matching results.

[0089] For example, when performing tag mining, only the review data of a certain category in the last six months is taken. However, if the online platform has the review data of the category in the past year, the review data of the remaining six months (for example, Figure 2b The existing reviews shown in the figure are semantically matched with the selected review tags of the category. After matching, the number of matching reviews corresponding to the selected review tags and the specific matched review data are updated according to the matching results. For example, after performing tag mining, a category generates new review data (for example Figure 2b If the new comment data is a daily increase in comments, the new comment data can be semantically matched with the selected comment label of the category, and the number of matching comments and the specific comment data of the corresponding selected comment label can be updated according to the matching results. That is, after the label system is built, it can also be dynamically updated according to real-time data or stock data to ensure the effectiveness of the label system.

[0090] The sentiment classification model can be used to classify the sentiment of the selected comment labels of each category, obtain the sentiment category of the selected comment labels of each category, and set the corresponding sentiment category for the selected comment labels of each category. Sentiment categories can include positive, neutral, negative, and other categories. Tags can be classified and sorted according to the sentiment category. For example, labels with positive sentiment are displayed first, followed by neutral sentiment labels, and finally negative sentiment labels.

[0091] Temporal freshness refers to the temporal characteristics of comments associated with a tag. It can be expressed by the time of the latest comment or the temporal distribution of the comments. Tags are sorted according to the freshness of the comments. Tags with more recent comments can be sorted higher to provide users with the latest comment references. For example, "Photo clear (most in the last month)" may be more worthy of attention.

[0092] The interaction rate refers to the number of times the comments related to the tag receive interactive behaviors such as likes, replies, and shares. This attribute can reflect the user interaction and influence of the tag. Tags are sorted according to the interaction rate. For example, "Fast running speed (100 likes, 50 replies)" has a higher interaction rate and is ranked higher.

[0093] Post-hoc click data refers to various behavioral data generated after users interact with tags. This data can provide in-depth insights into user behavior and interests, help optimize tag sorting and display, and thus improve the overall user experience and conversion rate. Post-hoc click data can include click-through rate, page view time after click, page interaction number, conversion rate, etc. Click-through rate can be the ratio of the number of clicks after users see a tag to the number of impressions, which can measure the attractiveness and relevance of the tag. Page view time after click can be the time a user stays on the page after clicking the tag to enter the page. A longer stay time usually indicates that the user is interested in the content. Page interaction number can be the number of further interactions (such as clicking on other products, leaving messages, etc.) on the page after users click on the tag. Conversion rate can be the proportion of users who finally complete the purchase after clicking on the tag, which is a measure of the effect of tags in truly promoting conversion. Post-hoc click data can help identify which tags can actually attract users' attention and prompt users to perform valuable operations (such as add to cart, purchase). By analyzing post-hoc click data, the display logic and order of tags can be continuously adjusted to further improve user experience and conversion rate.

[0094] In this embodiment, the label mining model is used to automatically mine labels, which reduces the time and manpower input of manually creating and assigning labels, and improves the efficiency of label creation; the label mining model can deeply analyze the review content, automatically generate a variety of labels, expand the number and types of labels, make the labels more comprehensive, and capture a wider range of user opinions and experiences; the label mining model can process various types of review data, adapt to different vocabulary, expressions and contexts, and make the generated labels more flexible and close to the actual review content. Labels are mined from the review data of items. Labels directly reflect the real feelings and opinions of users after actually using the items. The labels will be more natural and close to users; the review data is dynamically changing, and labels are automatically mined from the review data, so that the label system can quickly respond to and adapt to new user feedback and update in a timely manner; the automatically generated labels are derived from the actual expressions of users, which are more in line with the user's language habits, and the display effect is more natural, reducing the sense of stiffness, and the label quality is higher. Since the label mining is performed according to categories, there are duplicate labels mined from the review data of items in the same category. The original review labels mined can be semantically deduplicated. Deduplication reduces the number of labels, avoids information redundancy, helps highlight the core ideas of the reviews, and makes the label display more concise and clear. By processing the review data and constructing review labels according to categories, labels can be reused in similar items, avoiding the work of repeatedly creating labels and greatly improving the efficiency of label construction; using the same label system for items in the same category can ensure the unification and consistency of review labels, so that users can get a consistent experience when browsing different items; when introducing new items, the label system of existing similar items can be directly applied to quickly launch the review labels of new items, which is easy to expand.

[0095] The label system constructed in the embodiment of the present invention is mainly used to provide comment labels for items on the online platform. The label providing method of the embodiment of the present invention is described below. The label providing method is executed by an electronic device, which may be a computer, a server, or a computer. Figure 3 As shown, the following steps may be included:

[0096] Step 301, obtaining review data of a target item.

[0097] The target item can be any item that is traded via an online platform and needs to provide a review tag, and can be an item with a specific SKU. For example, the target item can be a mobile phone of a certain brand and model. The target item can have one or more review data.

[0098] Step 302, identifying the category to which the target item belongs, and obtaining the target category.

[0099] You can refer to the category system divided by the online platform to identify the category to which the target item belongs. For example, if the target item is a mobile phone of a certain brand and model, the category to which it belongs can be mobile phone.

[0100] Step 303 , selecting a comment tag that semantically matches the comment data of the target item from the selected comment tags of the target category to obtain a target comment tag.

[0101] Among them, the selected review tags of the target category are constructed according to the tag system construction method provided in the embodiment of the present invention. For the specific semantic matching method, please refer to the previous embodiment or the currently existing method, which will not be repeated here. This step is equivalent to selecting a tag from the selected review tags of a category based on the review data of an item. For example, there are 50 selected review tags in the category of mobile phones. There are currently 10 review data for the mobile phone of brand A and model X. These 10 review data are matched to 15 tags from the 50 selected review tags. The 15 matched tags are the target review tags.

[0102] Step 304, providing a target review tag for the target item.

[0103] The target review tags may be sorted according to the sorting attributes, and the target review tags may be provided for the target items according to the sorting results.

[0104] Each target comment tag can have one or more sorting attributes. When a target comment tag has multiple sorting attributes, the corresponding target comment tag can be scored comprehensively by combining multiple sorting attributes, and the multiple comment tags can be sorted according to the comprehensive scoring results. For example, the sorting attributes include the number of matching comments and sentiment classification. The more matching comments and the more positive sentiment comment tags have, the higher the ranking can be.

[0105] When the electronic device is a server, has no screen, and is far away from the user, the electronic device can provide a target review tag for the target item through the client. For example, the target review tag can be displayed on the target item's detail page or comment area using the client. When the electronic device is a computer with a screen on the user's side, the target review tag can be displayed directly on the target item's detail page or comment area.

[0106] In this embodiment, by selecting comment tags that semantically match the comment data of the target item from the selected comment tags of the target category, it can be ensured that the tags displayed to the user are highly relevant to the item, which helps the user quickly understand other users' opinions on the item. The target comment tags provided are mined from a large number of user comments on similar items. After semantic matching, deduplication and sorting, they can more accurately reflect the real experience and opinions of other users or consumers, and increase user trust. By sorting the tags and giving priority to displaying tags with a large number of matching comments and positive sentiment classification, the key points that can most influence user decisions can be highlighted, thereby improving conversion rates.

[0107] Figure 4 is a schematic diagram of a structure of a label system construction device provided in an embodiment of the present invention, and the device is suitable for executing a label system construction method provided in an embodiment of the present invention, such as Figure 4 As shown, the device may specifically include:

[0108] A collection module 401 is used to collect review data of various items;

[0109] A mining module 402 is used to perform label mining on each review data of each category of items using a label mining model to obtain a review label for each review data of each category of items;

[0110] A merging module 403 is used to merge the review labels of each review data of each category of items to obtain the original review label of each category;

[0111] The deduplication module 404 is used to filter out semantically repeated tags from the original review tags of each category to obtain selected review tags of each category.

[0112] In one embodiment, the mining module 402 uses a label mining model to perform label mining on each review data of each category of items to obtain the review labels of each review data of each category of items, including:

[0113] Perform semantic segmentation on each review data of each category of items to obtain semantic units of each review data of each category of items;

[0114] Use the label mining model to perform label mining on the semantic units of each review data of each category of items to obtain the review labels of the semantic units in each review data of each category of items;

[0115] The comment labels of the semantic units in the comment data of each category of items are merged to obtain the comment labels of the comment data of each category of items.

[0116] In one embodiment, the deduplication module 404 filters out semantically duplicated tags from the original review tags of each category to obtain selected review tags of each category, including:

[0117] Filter out semantically duplicated labels from the original review labels of each category to obtain candidate review labels for each category;

[0118] Sending candidate review tags for each category to the operation terminal, and obtaining the review results of the candidate review tags for each category from the operation terminal;

[0119] The featured review label for each category is determined based on the review results.

[0120] In one embodiment, a setting module is further included, and the setting module is used to:

[0121] Semantically match the selected review tags of each category with the review data of the corresponding category items;

[0122] According to the matching results, matching review information is set for the featured review tag of each category.

[0123] In one embodiment, the setting module is further used to:

[0124] Obtain incremental reviews and / or stock reviews of various items. Incremental reviews are review data generated after tag mining, and stock reviews are review data generated before tag mining but not used for tag mining.

[0125] Semantically match incremental reviews and / or stock reviews of various items with selected review tags of the corresponding categories;

[0126] Update the matching review information set for the featured review tag of the corresponding category based on the matching results.

[0127] In one embodiment, the setting module is further used for:

[0128] Use the sentiment classification model to classify the sentiment of the selected review tags of each category to obtain the sentiment category of the selected review tags of each category;

[0129] Set the corresponding sentiment category for the selected review tags of each category.

[0130] Those skilled in the art can clearly understand that for the convenience and simplicity of description, only the division of the above-mentioned functional modules is used as an example for illustration. In actual applications, the above-mentioned functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. The specific working process of the functional modules described above can refer to the corresponding process in the aforementioned method embodiment, and will not be repeated here.

[0131] The device of this embodiment uses a tag mining model to automatically mine tags, which reduces the time and manpower input of manually creating and assigning tags, and improves the efficiency of tag creation; the tag mining model can deeply analyze the review content, automatically mine and generate a variety of tags, expand the number and types of tags, make the tags more comprehensive, and capture a wider range of user opinions and experiences; the tag mining model can process various types of review data, adapt to different vocabulary, expressions and contexts, and make the generated tags more flexible and close to the actual review content. Tags are mined from the review data of the item, and the tags directly reflect the real feelings and opinions of the user after actually using the item, and the tags will be more natural and close to the user; the review data is dynamically changing, and tags are automatically mined from the review data, so that the tag system can quickly respond to and adapt to new user feedback and update in a timely manner; the automatically generated tags are derived from the actual expression of the user, which is more in line with the user's language habits, and the display effect is more natural, reducing the sense of stiffness, and the label quality is higher. Since the label mining is performed according to categories, there are duplicate labels mined from the review data of items in the same category. The original review labels mined can be semantically deduplicated. Deduplication reduces the number of labels, avoids information redundancy, helps highlight the core ideas of the reviews, and makes the label display more concise and clear. By processing the review data and constructing review labels according to categories, labels can be reused for similar items, avoiding the work of repeatedly constructing labels and greatly improving the efficiency of label construction; using the same label system for items in the same category can ensure the unification and consistency of review labels, so that users can get a consistent experience when browsing different items; when introducing new items, the label system of existing similar items can be directly applied to quickly launch the review labels of new items, which is easy to expand.

[0132] Figure 5 is a schematic diagram of a structure of a label providing device according to an embodiment of the present invention, and the device is suitable for executing a label providing method according to an embodiment of the present invention, such as Figure 5 As shown, the device may specifically include:

[0133] An acquisition module 501 is used to acquire review data of a target item;

[0134] An identification module 502 is used to identify the category to which the target item belongs and obtain the target category;

[0135] A selection module 503 is used to select a comment tag that semantically matches the comment data of the target item from the selected comment tags of the target category to obtain a target comment tag; wherein the selected comment tags of the target category are constructed according to the tag system construction method of any embodiment of the present invention;

[0136] A module 504 is provided, for providing a target review tag for the target item.

[0137] In one embodiment, the providing module 504 is specifically used for:

[0138] Sort the target review tags according to their sorting attributes;

[0139] Provide target review labels for target items based on the ranking results;

[0140] The sorting attribute includes at least one of matching comment information and sentiment category.

[0141] The device of this embodiment can ensure that the labels displayed to users are highly relevant to the item by selecting review labels that semantically match the review data of the target item from the selected review labels of the target category, which helps users quickly understand other users' opinions on the item. The target review labels provided are mined from a large number of user reviews of similar items. After the processes of semantic matching, deduplication and sorting, they can more accurately reflect the real experience and opinions of other users or consumers, and increase user trust. By sorting the labels and giving priority to displaying labels with a large number of matching reviews and positive sentiment classification, the key points that can most influence user decisions can be highlighted, thereby improving conversion rates.

[0142] An embodiment of the present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the label system construction method provided in any of the above embodiments, or implements the label providing method provided in any of the above embodiments.

[0143] An embodiment of the present invention further provides a computer-readable medium having a computer program stored thereon, and when the program is executed by a processor, the method for constructing a label system provided in any of the above embodiments is implemented, or the method for providing a label provided in any of the above embodiments is implemented.

[0144] Reference below Figure 6 , which shows a schematic diagram of the structure of a computer system 600 of an electronic device suitable for implementing an embodiment of the present invention. Figure 6 The electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

[0145] like Figure 6As shown, the computer system 600 includes a central processing unit (CPU) 601, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 602 or a program loaded from a storage part 608 into a random access memory (RAM) 603. In the RAM 603, various programs and data required for the operation of the computer system 600 are also stored. The CPU 601, the ROM 602, and the RAM 603 are connected to each other via a bus 604. An input / output (I / O) interface 605 is also connected to the bus 604.

[0146] The following components are connected to the I / O interface 605: an input section 606 including a keyboard, a mouse, etc.; an output section 607 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 608 including a hard disk, etc.; and a communication section 609 including a network interface card such as a LAN card, a modem, etc. The communication section 609 performs communication processing via a network such as the Internet. A drive 610 is also connected to the I / O interface 605 as needed. A removable medium 611, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 610 as needed, so that a computer program read therefrom is installed into the storage section 608 as needed.

[0147] In particular, according to the embodiments disclosed in the present invention, the process described above with reference to the flowchart can be implemented as a computer software program. For example, the embodiments disclosed in the present invention include a computer program product, which includes a computer program carried on a computer-readable medium, and the computer program includes a program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 609, and / or installed from the removable medium 611. When the computer program is executed by the central processing unit (CPU) 601, the above-mentioned functions defined in the system of the present invention are executed.

[0148] It should be noted that the computer-readable medium shown in the present invention may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present invention, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present invention, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, which carries a computer-readable program code. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0149] The flow chart and block diagram in the accompanying drawings illustrate the possible architecture, function and operation of the system, method and computer program product according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0150] The modules and / or units involved in the embodiments of the present invention may be implemented in software or hardware. The modules and / or units described may also be arranged in a processor, for example, may be described as: a processor includes a collection module, a mining module, a fusion module and a deduplication module. Alternatively, it may be described as: a processor includes an acquisition module, an identification module, a selection module and a provision module. The names of these modules do not, in some cases, constitute limitations on the modules themselves.

[0151] As another aspect, the present invention further provides a computer-readable medium, which may be included in the device described in the above embodiment; or may exist independently without being assembled into the device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by a device, the device includes:

[0152] Collect review data of various types of items; use the label mining model to perform label mining on the review data of each type of item to obtain the review labels of each review data of each type of item; merge the review labels of each review data of each type of item to obtain the original review labels of each category; filter out semantically repeated labels from the original review labels of each category to obtain the selected review labels of each category.

[0153] Alternatively, when the above one or more programs are executed by a device, the device includes:

[0154] Acquire comment data of a target item; identify the category to which the target item belongs and obtain the target category; select comment tags that semantically match the comment data of the target item from selected comment tags of the target category to obtain a target comment tag; wherein the selected comment tags of the target category are constructed according to the tag system construction method of any embodiment of the present invention; provide a target comment tag for the target item.

[0155] The technical solution of the embodiment of the present invention uses a tag mining model to automatically mine tags when constructing a tag system, which reduces the time and manpower input of manually creating and assigning tags, and improves the efficiency of tag creation; the tag mining model can deeply analyze the comment content, automatically mine and generate a variety of tags, expand the number and types of tags, make the tags more comprehensive, and capture a wider range of user opinions and experiences; the tag mining model can process various types of comment data, adapt to different vocabulary, expressions and contexts, and make the generated tags more flexible and close to the actual comment content. Tags are mined from the comment data of the item, and the tags directly reflect the real feelings and opinions of the user after actually using the item, and the tags will be more natural and close to the user; the comment data is dynamically changing, and tags are automatically mined from the comment data, so that the tag system can quickly respond to and adapt to new user feedback and update in a timely manner; the automatically generated tags are derived from the actual expression of the user, which is more in line with the user's language habits, and the display effect is more natural, reducing the sense of stiffness, and the label quality is higher. Since the label mining is performed according to categories, there are duplicate labels mined from the review data of items in the same category. The original review labels mined can be semantically deduplicated. Deduplication reduces the number of labels, avoids information redundancy, helps highlight the core ideas of the reviews, and makes the label display more concise and clear. By processing the review data and constructing review labels according to categories, labels can be reused for similar items, avoiding the work of repeatedly constructing labels and greatly improving the efficiency of label construction; using the same label system for items in the same category can ensure the unification and consistency of review labels, so that users can get a consistent experience when browsing different items; when introducing new items, the label system of existing similar items can be directly applied to quickly launch the review labels of new items, which is easy to expand.

[0156] When using the tag system, by selecting the comment tags that semantically match the comment data of the target item from the selected comment tags of the target category, it can be ensured that the tags displayed to the user are highly relevant to the item, which helps the user quickly understand other users' opinions on the item. The target comment tags provided are mined from a large number of user comments on similar items. After semantic matching, deduplication and sorting, they can more accurately reflect the real experience and opinions of other users or consumers, and increase user trust. By sorting the tags and giving priority to displaying tags with a large number of matching comments and positive sentiment classification, the key points that can most influence user decisions can be highlighted, thereby improving conversion rates.

[0157] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0158] It should be noted that the collection, collection, updating, analysis, processing, use, transmission, storage and other aspects of user personal information involved in the technical solution of this disclosure are in compliance with the provisions of relevant laws and regulations, are used for legitimate purposes, and do not violate public order and good morals. Necessary measures are taken for user personal information to prevent illegal access to user personal information data and maintain the security of user personal information, network security and national security.

[0159] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions may occur depending on design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A method for constructing a label system, characterized in that: include: Collect review data for various items; Use the label mining model to perform label mining on each review data of each category of items to obtain the review labels of each review data of each category of items; Combine the review labels of each review data of each category of items to obtain the original review labels of each category; Semantically repeated tags are filtered out from the original review tags of each category to obtain the selected review tags of each category.

2. The method according to claim 1, characterized in that The label mining model is used to perform label mining on the comment data of each category of items to obtain the comment labels of each comment data of each category of items, including: Perform semantic segmentation on each review data of each category of items to obtain semantic units of each review data of each category of items; Use the label mining model to perform label mining on the semantic units of each review data of each category of items to obtain the review labels of the semantic units in each review data of each category of items; The comment labels of the semantic units in the comment data of each category of items are merged to obtain the comment labels of the comment data of each category of items.

3. The method according to claim 1, characterized in that: We filter out semantically duplicated tags from the original review tags of each category to obtain selected review tags for each category, including: Filter out semantically duplicated labels from the original review labels of each category to obtain candidate review labels for each category; Sending candidate review tags for each category to the operation terminal, and obtaining the review results of the candidate review tags for each category from the operation terminal; The featured review label for each category is determined based on the review results.

4. The method according to any one of claims 1 to 3, characterized in that: Also includes: Semantically match the selected review tags of each category with the review data of the corresponding category items; According to the matching results, matching review information is set for the featured review tag of each category.

5. The method according to claim 4, characterized in that Also includes: Obtain incremental reviews and / or stock reviews of various items. Incremental reviews are review data generated after tag mining, and stock reviews are review data generated before tag mining but not used for tag mining. Semantically match incremental reviews and / or stock reviews of various items with selected review tags of the corresponding categories; Update the matching review information set for the featured review tag of the corresponding category based on the matching results.

6. The method according to any one of claims 1 to 3, characterized in that: Also includes: Use the sentiment classification model to classify the sentiment of the selected review tags of each category to obtain the sentiment category of the selected review tags of each category; Set the corresponding sentiment category for the selected review tags of each category.

7. A label providing method, characterized in that: include: Get the review data of the target item; Identify the category to which the target item belongs and obtain the target category; Selecting a comment tag that semantically matches the comment data of the target item from the selected comment tags of the target category to obtain a target comment tag; wherein the selected comment tags of the target category are constructed according to the tag system construction method according to any one of claims 1 to 6; Provide target review labels for target items.

8. The method according to claim 7, characterized in that Provide target review tags for target items, including: Sort the target review tags according to their sorting attributes; Provide target review labels for target items based on the ranking results; The sorting attribute includes at least one of matching comment information and sentiment category.

9. A label system construction device, characterized in that: include: The collection module is used to collect review data of various items; A mining module is used to perform label mining on each review data of each category of items using a label mining model to obtain a review label for each review data of each category of items; The merging module is used to merge the comment labels of each comment data of each category of items to obtain the original comment labels of each category; The deduplication module is used to filter out semantically repeated tags from the original review tags of each category to obtain the selected review tags for each category.

10. A label providing device, characterized in that: include: The acquisition module is used to obtain the review data of the target item; The recognition module is used to identify the category to which the target item belongs and obtain the target category; A selection module is used to select a comment tag that semantically matches the comment data of the target item from the selected comment tags of the target category to obtain a target comment tag; wherein the selected comment tags of the target category are constructed according to the tag system construction method according to any one of claims 1 to 6; Provides a module for providing target review labels for target items.

11. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the program, it implements the label system construction method as claimed in any one of claims 1 to 6, or implements the label providing method as claimed in claim 7 or 8.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, it implements the label system construction method according to any one of claims 1 to 6, or implements the label providing method according to claim 7 or 8.

Images