Labeling classification method for e-commerce commodities

Abstract

The invention relates to the field of tag classification, in particular to a tagging classification method for e-commerce commodities, which comprises the following steps: carrying out commodity matching on target commodities according to keyword similarity and keyword integrating degree to obtain a pre-matched commodity set; according to the scene keyword difference degree, determining a commodity processing mode to analyze commodity features or background features in commodity images of pre-matched commodities in the pre-matched commodity set; determining an image state according to the total number of the commodity images and the effective image proportion; according to the commodity feature similarity or the background feature similarity, video frames are selected for image supplementation; according to the background feature similarity or the commodity feature similarity, commodity tags are determined to be generated according to the keywords or the user carries out autonomous early warning processing; and determining a standard label generation mode according to the maximum same commodity label number proportion. According to the invention, commodities can be classified according to commodity usage scenes, and the identification accuracy of commodity labels is improved.

Description

Technical Field

[0001] The present invention relates to the field of label classification, and in particular to a label classification method for e-commerce products. Background Art

[0002] In today's e-commerce environment, due to the non-standardization of product information, the same product may be used in different application scenarios. When users search by product name, it is difficult to accurately match relevant products that meet user needs, resulting in a poor user experience. Therefore, how to classify products according to product usage scenarios to improve the recognition accuracy of product labels is a technical problem that technical personnel in this field urgently need to solve.

[0003] Chinese patent publication number CN108491490A discloses a product labeling and identification system for an e-commerce platform, including: a user behavior information acquisition module, used to acquire the browsing and consumption information of users on the e-commerce platform; a user behavior information classification and integration module, used to classify and integrate the acquired user behavior information; a user behavior information analysis module, used to analyze the integrated user behavior information; and a comparison module, used to compare user behavior information and product labels. It can be seen that the above technical solution has the following problems: the products are not classified according to the product usage scenarios, resulting in poor recognition accuracy of product labels. Summary of the invention

[0004] To this end, the present invention provides a labeling classification method for e-commerce products, so as to overcome the problem in the prior art that products are not classified according to product usage scenarios, resulting in poor recognition accuracy of product labels.

[0005] To achieve the above object, the present invention provides a labeling classification method for e-commerce products, comprising: matching target products according to keyword similarity and keyword fit of product texts corresponding to the target products to obtain a pre-matched product set;

[0006] Detect the difference of the scene keywords corresponding to the pre-matched product set, and determine the product processing method according to the difference of the scene keywords;

[0007] The product processing method is to analyze product features or background features in the product image of each pre-matching product in the pre-matching product set;

[0008] Determine the image status based on the total number of product images and the percentage of valid images;

[0009] When the total number of product images is less than the preset total number of product images or the effective image ratio is less than the preset effective image ratio, video frames are selected for image supplementation based on product feature similarity or background feature similarity;

[0010] Generate product tags based on keywords or perform autonomous warning processing for users based on the similarity of background features or product features;

[0011] The method of generating standard labels is determined according to the ratio of the maximum number of identical product labels, which is to convert product labels into standard labels or to use keyword combinations as standard labels.

[0012] Furthermore, product matching includes:

[0013] When matching a single target product, the product matching status is determined based on the keyword similarity and keyword fit between the target product and the reference product, and the reference product and the target product in the matching product matching status are uniformly recorded as a pre-matched product set;

[0014] When the pre-matching commodity set is established, the target commodity and the reference commodity in the set are recorded as commodities to be standardized.

[0015] Furthermore, the effective image proportion is determined according to the ratio of the number of effective images to the total number of product images.

[0016] Furthermore, when analyzing the product features in the product images of each product to be standardized in the pre-matched product set, if the image status is that the total number of product images is less than the preset total number of product images or the proportion of valid images is less than the preset proportion of valid images, video frames are selected for image supplementation according to the similarity of product features:

[0017] Record the video frames corresponding to the reference videos whose product feature similarity is greater than the preset product feature similarity as the first-category video frames, record the first-category video frames with the largest product feature similarity as the target first-category video frames, record the other first-category video frames excluding the target first-category video frames as the reference first-category video frames, perform contour difference detection on the target first-category video frames, record the target first-category video frames and the reference video frames with the largest contour difference from the target first-category video frames as a first-category combination, continue to perform difference detection on the first-category combination, add the reference first-category video frames with the largest average contour difference from each of the first-category video frames in the first-category combination to the first-category combination, until the number of first-category video frames in the first-category combination is equal to the pre-extracted number;

[0018] The reference videos include product review videos and product videos.

[0019] Furthermore, the contour difference is determined according to an absolute value of a difference between a contour length corresponding to the target type one video frame and a contour length corresponding to the reference type one video frame.

[0020] Furthermore, when analyzing the background features in the product images of each product to be standardized in the pre-matched product set, if the image status is that the total number of product images is less than the preset total number of product images or the proportion of valid images is less than the preset proportion of valid images, a video frame is selected for image supplementation according to the similarity of the background features:

[0021] Record the video frames corresponding to the reference video whose background feature similarity is greater than the preset background feature similarity as the second-category video frames, sort the second-category video frames in descending order of background area to obtain a second-category video frame sequence, and extract a number of second-category video frames from each frame extraction point of the second-category video frame sequence;

[0022] The frame extraction points are evenly distributed in the two types of video frame sequences and the number of frame extraction points is determined according to the pre-extraction number.

[0023] Further, the product feature similarity is determined based on the ViT product feature value corresponding to the product image and the ViT product feature value corresponding to the video frame;

[0024] The background feature similarity is determined according to the ViT background feature value corresponding to the product image and the ViT background feature value corresponding to the video frame.

[0025] Furthermore, if the scene keyword difference is less than the preset scene keyword difference, the background features in the product images of each product to be standardized in the pre-matched product set are analyzed:

[0026] Perform background feature similarity detection on the products to be standardized, and determine whether to generate product labels based on keywords or perform autonomous warning processing for users based on the background feature similarity;

[0027] If the scene keyword difference is greater than or equal to the preset scene keyword difference, the product features in the product images of each product to be standardized in the pre-matched product set are analyzed:

[0028] The similarity of product features is detected for the products to be standardized, and product labels are generated based on keywords or users perform autonomous warning processing based on the similarity of product features.

[0029] Furthermore, if the maximum number of identical product labels accounts for a greater than or equal to a preset maximum number of identical product labels accounts for a, the product labels corresponding to the maximum number of identical product labels are converted into standard labels;

[0030] If the maximum ratio of the number of identical product tags is less than the preset maximum ratio of the number of identical product tags, identify the same keywords in each product tag, and convert the keyword combination consisting of the same keywords into a standard tag.

[0031] Furthermore, the proportion of the maximum number of identical product labels is determined based on the ratio of the maximum number of identical product labels to the total number of products to be standardized.

[0032] Compared with the prior art, the beneficial effect of the present invention lies in that, in the technical solution of the present invention, product matching is performed for target products according to keyword similarity and keyword fit, and the keyword similarity and keyword fit effectively reflect the similarity and fit of the target product keywords, and then the target products are classified to obtain a pre-matched product set, so that the correlation degree of each target product in the pre-matched product set is relatively high, which is conducive to the subsequent determination of the product processing method according to the difference in scene keywords.

[0033] Furthermore, the present invention determines the commodity processing method according to the difference in scene keywords, and effectively reflects the difference in usage scenarios of each target commodity in the pre-matched commodity set through the difference in scene keywords, and then adaptively selects different commodity processing methods, so that the selection of commodity processing methods is more in line with the actual application scenario, avoiding the problem that the target commodity cannot be accurately classified due to different usage scenarios, thereby improving the recognition accuracy of commodity labels.

[0034] Furthermore, in the present invention, the image status is determined according to the total number of product images and the proportion of valid images. The quality of the product images is reflected by the total number of product images and the proportion of valid images, and then according to the actual application scenario, it is determined whether it is necessary to select video frames for image supplementation according to the similarity of product features or background features. This avoids the problem of poor accuracy in analysis based on product features or background features in product images due to lack of featureness of product images, thereby accurately classifying products and improving the recognition accuracy of product labels.

[0035] Furthermore, the present invention effectively reflects the similarity of the commodities to be standardized through the similarity of background features and commodity features, and then generates commodity labels based on keywords or performs autonomous early warning processing according to the actual application scenario, thereby avoiding the problem of poor commodity similarity after establishing standard labels for each target commodity, and thus improving the recognition accuracy of commodity labels.

[0036] Furthermore, in the present invention, the maximum number of identical product labels is effectively reflected by the proportion of the maximum number of identical product labels, and then the standard label generation method is adaptively selected, so that the selection of the label generation method is more in line with the actual application scenario, avoiding the problem of poor representativeness of the standard label due to too few identical labels, thereby improving the recognition accuracy of the product labels. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] Figure 1 A schematic diagram of a labeling classification method for e-commerce products of the present invention;

[0038] Figure 2 This is a flow chart of the present invention for determining a commodity processing method according to the difference between scene keywords;

[0039] Figure 3 This is a flow chart of the present invention for determining, according to an image state, whether to select a video frame for image supplementation or not according to product feature similarity or background feature similarity;

[0040] Figure 4 The present invention is a flow chart of determining a standard label generation method according to the maximum proportion of the number of identical product labels, which is to convert product labels into standard labels or to use keyword combinations as standard labels. DETAILED DESCRIPTION

[0041] In order to make the objects and advantages of the present invention more clearly understood, the present invention is further described below in conjunction with embodiments; it should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0042] The preferred embodiments of the present invention are described below with reference to the accompanying drawings. It should be understood by those skilled in the art that these embodiments are only used to explain the technical principles of the present invention and are not intended to limit the protection scope of the present invention.

[0043] It should be noted that, in the description of the present invention, terms such as "up", "down", "left", "right", "inside" and "outside" indicating directions or positional relationships are based on the directions or positional relationships shown in the drawings. This is merely for the convenience of description and does not indicate or imply that the device or element must have a specific orientation, be constructed and operated in a specific orientation. Therefore, it cannot be understood as a limitation on the present invention.

[0044] In addition, it should be noted that in the description of the present invention, unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be indirectly connected through an intermediate medium, or it can be the internal communication of two components. For those skilled in the art, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0045] See also Figures 1 to 4 As shown, the present invention provides a labeling classification method for e-commerce products, comprising:

[0046] Match the target product according to the keyword similarity and keyword fit of the product text corresponding to the target product to obtain a pre-matched product set;

[0047] Detect the difference of the scene keywords corresponding to the pre-matched product set, and determine the product processing method according to the difference of the scene keywords;

[0048] The product processing method is to analyze product features or background features in the product image of each pre-matching product in the pre-matching product set;

[0049] Determine the image status based on the total number of product images and the percentage of valid images;

[0050] When the total number of product images is less than the preset total number of product images or the effective image ratio is less than the preset effective image ratio, video frames are selected for image supplementation based on product feature similarity or background feature similarity;

[0051] Generate product tags based on keywords or perform autonomous warning processing for users based on the similarity of background features or product features;

[0052] The method of generating standard labels is determined according to the ratio of the maximum number of identical product labels, which is to convert product labels into standard labels or to use keyword combinations as standard labels.

[0053] The application scenario of the present invention is the labeling processing of product information on an e-commerce platform. The target products include product pictures, product videos, product texts and product review videos. Product pictures are pictures uploaded by merchants to the e-commerce platform containing the appearance and usage scenarios of the target products. Product videos are videos uploaded by merchants to the e-commerce platform containing the product usage process, product function demonstration or appearance display, etc. The product text is a character paragraph containing information such as the product name, product specifications and product use. The product review video is a video uploaded by consumers to the e-commerce platform after purchasing the product. This is content that is easy for technical personnel in this field to understand and will not be elaborated here.

[0054] If the scene keyword difference is greater than or equal to the preset scene keyword difference, the product processing method is to analyze the product features in the product image of each product to be standardized in the pre-matched product set;

[0055] If the scene keyword difference is less than the preset scene keyword difference, the product processing method is to analyze the background features in the product image of each product to be standardized in the pre-matched product set;

[0056] The image status includes: the total number of product images is less than the preset total number of product images or the effective image ratio is less than the preset effective image ratio, and the total number of product images is greater than or equal to the preset total number of product images and the effective image ratio is greater than or equal to the preset effective image ratio;

[0057] When the image status is that the total number of product images is less than the preset total number of product images or the effective image ratio is less than the preset effective image ratio, video frames are selected for image supplementation according to product feature similarity or background feature similarity, and product features or background features in the supplemented video frames and product images of each pre-matching product in the pre-matching product set are analyzed;

[0058] When the image status is that the total number of product images is greater than or equal to the preset total number of product images and the effective image ratio is greater than or equal to the preset effective image ratio, there is no need to supplement the images, and the product features or background features in the product images of each pre-matching product in the pre-matching product set are analyzed;

[0059] Scene keywords include but are not limited to living room, bedroom, hiking, camping, rock climbing, mountain biking, Christmas, Valentine's Day, workplace and conference room. The recognition of scene keywords can be achieved through machine vision and deep learning networks. This is easy for technical personnel in this field to understand and will not be elaborated here.

[0060] The difference degree of scene keywords = (total number of products in the set - number of target products corresponding to the maximum same scene keyword) / total number of products in the set, where the total number of products in the set is the number of products to be standardized in a single pre-matched product set. The method for confirming the maximum same scene keyword is to perform the same scene keyword detection on each product to be standardized in the pre-matched product set. When performing the same scene keyword detection on a single product to be standardized, the product to be standardized is recorded as the target product to be standardized, and other products to be standardized in the pre-matched product set that do not include the target product to be standardized are recorded as reference products to be standardized. The reference products to be standardized and the target products to be standardized that have the same scene keywords corresponding to the target products to be standardized are recorded as the same scene keyword combination, and the same scene keyword detection is performed on the products to be standardized that are not recorded as the same scene keyword combination, until all products to be standardized are recorded as the same scene keyword combination, and the number of products to be standardized corresponding to the same scene keyword combination containing the largest number of products to be standardized is recorded as the target number of products corresponding to the maximum same scene keyword.

[0061] The value of the preset scene keyword difference can be determined by the user according to the actual application scenario. The higher the user's demand for the accuracy of setting the standard label, the smaller the value of the preset scene keyword difference. A value of the preset scene keyword difference is provided, and the preset scene keyword difference is 20%.

[0062] The total number of product images is the number of product images corresponding to a single target product. The values ​​of the preset total number of product images and the preset ratio of valid images can be determined by the user according to the actual application scenario. The higher the user's demand for the accuracy of setting standard labels, the larger the values ​​of the preset total number of product images and the preset ratio of valid images. Provided are values ​​of the preset total number of product images and the preset ratio of valid images. The preset total number of product images is the average value of the product images corresponding to each target product in the e-commerce platform, and the preset ratio of valid images is 80%.

[0063] Specifically, product matching includes:

[0064] When matching a single target product, the product matching status is determined based on the keyword similarity and keyword fit between the target product and the reference product, and the reference product and the target product in the matching product matching status are uniformly recorded as a pre-matched product set;

[0065] When the pre-matching commodity set is established, the target commodity and the reference commodity in the set are recorded as commodities to be standardized.

[0066] Keyword similarity = number of identical keywords / number of reference keywords. The number of reference keywords is determined as follows: if the number of target product keywords is greater than or equal to the number of reference product keywords, the number of reference keywords is the number of target product keywords; if the number of target product keywords is less than the number of reference product keywords, the number of reference keywords is the number of reference product keywords. The number of target product keywords is the number of keywords that can be identified in the product text corresponding to the target product, and the number of reference product keywords is the number of keywords that can be identified in the product text corresponding to the reference product.

[0067] Keyword fit = number of matching keywords / (number of reference product keywords + number of target product keywords). The number of matching keywords refers to the number of keywords in which the proportion of the number of times a single keyword corresponding to the target product is used in combination with the reference product is greater than the proportion of the number of times the preset combination is used. The proportion of times the combination is used = the number of times two keywords appear simultaneously in the product text corresponding to the target product on the e-commerce platform / the number of target products on the e-commerce platform.

[0068] The matching status of the matching product is that the keyword similarity is greater than the preset keyword similarity and the keyword fit is greater than the preset keyword fit. The values ​​of the preset keyword similarity and the preset keyword fit can be determined by the user according to the actual application scenario. The higher the user's demand for the accuracy of setting the standard label, the greater the values ​​of the preset keyword similarity and the preset keyword fit. A value of a preset keyword similarity and a preset keyword fit is provided, and the preset keyword similarity is 70%, and the preset keyword fit is 70%.

[0069] Specifically, the effective image ratio is determined according to the ratio of the number of effective images to the total number of product images.

[0070] Valid image ratio = number of valid images / total number of product images. Valid images are product images in which both the product area ratio and the background area ratio are greater than the preset area ratios. Product area ratio = product area / product image area. Background area ratio = background area / product image area. The product area is the area of ​​the target product in the product image. The background area and the product area are segmented by edge detection technology. The product area is calculated by contourArea in OpenCV. This is easy to understand for those skilled in the art and will not be elaborated here. Background area = product image area - product area. Product image area = length of product image × width of product image.

[0071] Specifically, when analyzing the product features in the product images of each product to be standardized in the pre-matched product set, if the image status is that the total number of product images is less than the preset total number of product images or the proportion of valid images is less than the preset proportion of valid images, video frames are selected for image supplementation based on the similarity of product features:

[0072] Record the video frames corresponding to the reference videos whose product feature similarity is greater than the preset product feature similarity as the first-category video frames, record the first-category video frames with the largest product feature similarity as the target first-category video frames, record the other first-category video frames excluding the target first-category video frames as the reference first-category video frames, perform contour difference detection on the target first-category video frames, record the target first-category video frames and the reference video frames with the largest contour difference from the target first-category video frames as a first-category combination, continue to perform difference detection on the first-category combination, add the reference first-category video frames with the largest average contour difference from each of the first-category video frames in the first-category combination to the first-category combination, until the number of first-category video frames in the first-category combination is equal to the pre-extracted number;

[0073] The reference videos include product review videos and product videos.

[0074] The value of the preset product feature similarity can be determined by the user according to the actual application scenario. The higher the user's demand for the accuracy of setting the standard label, the greater the value of the preset product feature similarity. A value of the preset product feature similarity is provided, and the preset product feature similarity is 0.8.

[0075] Pre-extracted quantity = total number of preset product images - number of product images. The calculation formula for the average value μ of contour difference is: Where N is the number of video frames of a class in a class combination, i is the i-th video frame of a class in a class combination, i = 1, 2, 3, ..., N, x i is the contour difference of the i-th video frame of the first category in the first category combination.

[0076] Specifically, the contour difference is determined according to the absolute value of the difference between the contour length corresponding to the target type video frame and the contour length corresponding to the reference type video frame.

[0077] The contour length is the length of the dividing line between the product area and the background area. The contour length is calculated by arcLength in OpenCV, which is easy to understand for those skilled in the art and will not be elaborated here.

[0078] Specifically, when analyzing the background features in the product images of each to-be-standardized product in the pre-matched product set, if the image status is that the total number of product images is less than the preset total number of product images or the proportion of valid images is less than the preset proportion of valid images, video frames are selected for image supplementation based on the similarity of background features:

[0079] Record the video frames corresponding to the reference video whose background feature similarity is greater than the preset background feature similarity as the second-category video frames, sort the second-category video frames in descending order of background area to obtain a second-category video frame sequence, and extract a number of second-category video frames from each frame extraction point of the second-category video frame sequence;

[0080] The frame extraction points are evenly distributed in the two types of video frame sequences and the number of frame extraction points is determined according to the pre-extraction number.

[0081] The frame extraction point is the position where the video frame is extracted from the second-category video frame sequence. If the number of frame extraction points is m, the second-category video frame sequence is divided into (m-1) equal parts, and the frame extraction points are located at the start and end of the second-category video frame sequence and the positions of the equal division points. The number of frame extraction points m is equal to the pre-extraction number.

[0082] The value of the preset background feature similarity can be determined by the user according to the actual application scenario. The higher the user's demand for the accuracy of setting the standard label, the larger the value of the preset background feature similarity. A value of the preset background feature similarity is provided, and the preset background feature similarity is 0.8.

[0083] Specifically, the commodity feature similarity is determined based on the ViT commodity feature value corresponding to the commodity image and the ViT commodity feature value corresponding to the video frame;

[0084] The background feature similarity is determined according to the ViT background feature value corresponding to the product image and the ViT background feature value corresponding to the video frame.

[0085] The method for confirming the ViT product feature value is to divide the product area into several image blocks of the same area through the ViT model, and input the image blocks into the linear projection layer to obtain a vector group that is recorded as the ViT product feature value; the method for confirming the ViT background feature value is to divide the background area into several image blocks of the same area through the ViT model, and input the image blocks into the linear projection layer to obtain a vector group that is recorded as the ViT background feature value.

[0086] The product feature similarity and background feature similarity are both calculated by using cosine similarity. The cosine value of the angle between the ViT product feature value corresponding to the product image and the ViT product feature value corresponding to the video frame is recorded as the product feature similarity, and the cosine value of the angle between the ViT background feature value corresponding to the product image and the ViT background feature value corresponding to the video frame is recorded as the background feature similarity. The value range of product feature similarity and background feature similarity is [0,1].

[0087] Specifically, if the scene keyword difference is less than the preset scene keyword difference, the background features in the product images of each product to be standardized in the pre-matched product set are analyzed:

[0088] Perform background feature similarity detection on the products to be standardized, and determine whether to generate product labels based on keywords or perform autonomous warning processing for users based on the background feature similarity;

[0089] If the scene keyword difference is greater than or equal to the preset scene keyword difference, the product features in the product images of each product to be standardized in the pre-matched product set are analyzed:

[0090] The similarity of product features is detected for the products to be standardized, and product labels are generated based on keywords or users perform autonomous warning processing based on the similarity of product features.

[0091] Perform background feature similarity detection on the products to be standardized. If the background feature similarity is greater than or equal to the preset background feature similarity, generate product labels based on keywords; if the background feature similarity is less than the preset background feature similarity, the user performs autonomous warning processing;

[0092] For the products to be standardized, the product feature similarity detection is performed. If the product feature similarity is greater than or equal to the preset product feature similarity, a product label is generated according to the keyword; if the product feature similarity is less than the preset product feature similarity, the user performs an independent warning process;

[0093] Generating product labels based on keywords includes: converting each keyword in the product text corresponding to the product to be standardized into a product label;

[0094] User autonomous warning processing includes: manual review to convert product labels into standard labels.

[0095] Specifically, if the maximum number of identical product labels accounts for a greater than or equal to a preset maximum number of identical product labels, the product labels corresponding to the maximum number of identical product labels are converted into standard labels;

[0096] If the maximum ratio of the number of identical product tags is less than the preset maximum ratio of the number of identical product tags, identify the same keywords in each product tag, and convert the keyword combination consisting of the same keywords into a standard tag.

[0097] The value of the preset maximum ratio of the number of identical product labels can be determined by the user according to the actual application scenario. The higher the user's demand for the accuracy of setting standard labels, the larger the value of the preset maximum ratio of the number of identical product labels. A value of the preset maximum ratio of the number of identical product labels is provided, and the preset maximum ratio of the number of identical product labels is 80%;

[0098] Converting the product tags corresponding to the maximum number of identical product tags into standard tags includes: It can be understood that the order of keywords in the product tags corresponding to the maximum number of identical product tags may be different, and randomly selecting a product tag from the product tags corresponding to the maximum number of identical product tags as the standard tag for the pre-matched product set.

[0099] Specifically, the proportion of the maximum number of identical product labels is determined based on the ratio of the maximum number of identical product labels to the total number of products to be standardized.

[0100] The maximum proportion of the number of identical product labels = the maximum number of identical product labels / the total number of products to be standardized in the pre-matched product set. The total number of products to be standardized in the pre-matched product set is the number of products to be standardized contained in the pre-matched product set. The maximum number of identical product labels is confirmed by performing identical label detection on the product labels corresponding to each product to be standardized in the pre-matched product set. When performing identical label detection on the product label corresponding to a single product to be standardized, the product label corresponding to the product to be standardized is recorded as the target label, the product label corresponding to the product to be standardized that does not contain the target label in the pre-matched product set is recorded as the reference label, the reference label and the target label that are the same as the keywords corresponding to the target label are recorded as the same product label combination, and the reference label that is not recorded as the same product label combination is detected. Until all product labels are recorded as the same product label combination, the number of product labels corresponding to the same product label combination containing the largest number of product labels is recorded as the maximum number of identical product labels.

[0101] So far, the technical solutions of the present invention have been described in conjunction with the preferred embodiments shown in the accompanying drawings. However, it is easy for those skilled in the art to understand that the protection scope of the present invention is obviously not limited to these specific embodiments. Without departing from the principle of the present invention, those skilled in the art can make equivalent changes or substitutions to the relevant technical features, and the technical solutions after these changes or substitutions will fall within the protection scope of the present invention.

[0102] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A labeling classification method for e-commerce products, characterized in that: include: Match the target product according to the keyword similarity and keyword fit of the product text corresponding to the target product to obtain a pre-matched product set; Detect the difference of the scene keywords corresponding to the pre-matched product set, and determine the product processing method according to the difference of the scene keywords; The product processing method is to analyze product features or background features in the product image of each pre-matching product in the pre-matching product set; Determine the image status based on the total number of product images and the percentage of valid images; When the image status is that the total number of product images is less than the preset total number of product images or the effective image ratio is less than the preset effective image ratio, video frames are selected for image supplementation based on product feature similarity or background feature similarity; Generate product tags based on keywords or perform autonomous warning processing for users based on the similarity of background features or product features; The method of generating standard labels is determined according to the ratio of the maximum number of identical product labels, which is to convert product labels into standard labels or to use keyword combinations as standard labels.

2. The labeling classification method for e-commerce products according to claim 1 is characterized in that: Product matching includes: When matching a single target product, the product matching status is determined based on the keyword similarity and keyword fit between the target product and the reference product, and the reference product and the target product in the matching product matching status are uniformly recorded as a pre-matched product set; When the pre-matching commodity set is established, the target commodity and the reference commodity in the set are recorded as commodities to be standardized.

3. The labeling classification method for e-commerce products according to claim 2 is characterized in that: The effective image ratio is determined based on the ratio of the number of effective images to the total number of product images.

4. The labeling classification method for e-commerce products according to claim 3 is characterized in that: When analyzing the product features in the product images of each product to be standardized in the pre-matched product set, if the image status is that the total number of product images is less than the preset total number of product images or the proportion of valid images is less than the preset proportion of valid images, video frames are selected for image supplementation based on the similarity of product features: Record the video frames corresponding to the reference videos whose product feature similarity is greater than the preset product feature similarity as the first-category video frames, record the first-category video frames with the largest product feature similarity as the target first-category video frames, record the other first-category video frames excluding the target first-category video frames as the reference first-category video frames, perform contour difference detection on the target first-category video frames, record the target first-category video frames and the reference video frames with the largest contour difference from the target first-category video frames as a first-category combination, continue to perform difference detection on the first-category combination, add the reference first-category video frames with the largest average contour difference from each of the first-category video frames in the first-category combination to the first-category combination, until the number of first-category video frames in the first-category combination is equal to the pre-extracted number; The reference videos include product review videos and product videos.

5. The labeling classification method for e-commerce products according to claim 4 is characterized in that: The contour difference is determined according to the absolute value of the difference between the contour length corresponding to the target type video frame and the contour length corresponding to the reference type video frame.

6. The labeling classification method for e-commerce products according to claim 5 is characterized in that: When analyzing the background features in the product images of each product to be standardized in the pre-matched product set, if the image status is that the total number of product images is less than the preset total number of product images or the proportion of valid images is less than the preset proportion of valid images, select video frames for image supplementation based on the similarity of background features: Record the video frames corresponding to the reference video whose background feature similarity is greater than the preset background feature similarity as the second-category video frames, sort the second-category video frames in descending order of background area to obtain a second-category video frame sequence, and extract a number of second-category video frames from each frame extraction point of the second-category video frame sequence; The frame extraction points are evenly distributed in the two types of video frame sequences and the number of frame extraction points is determined according to the pre-extraction number.

7. The labeling classification method for e-commerce products according to claim 6 is characterized in that: The product feature similarity is determined based on the ViT product feature value corresponding to the product image and the ViT product feature value corresponding to the video frame; The background feature similarity is determined according to the ViT background feature value corresponding to the product image and the ViT background feature value corresponding to the video frame.

8. The labeling classification method for e-commerce products according to claim 7 is characterized in that: If the scene keyword difference is less than the preset scene keyword difference, the background features in the product images of each product to be standardized in the pre-matched product set are analyzed: Perform background feature similarity detection on the products to be standardized, and determine whether to generate product labels based on keywords or perform autonomous warning processing for users based on the background feature similarity; If the scene keyword difference is greater than or equal to the preset scene keyword difference, the product features in the product images of each product to be standardized in the pre-matched product set are analyzed: The similarity of product features is detected for the products to be standardized, and product labels are generated based on keywords or users perform autonomous warning processing based on the similarity of product features.

9. The labeling classification method for e-commerce products according to claim 8 is characterized in that: If the maximum number of identical product labels accounts for a greater than or equal to the preset maximum number of identical product labels, the product labels corresponding to the maximum number of identical product labels are converted into standard labels; If the maximum ratio of the number of identical product tags is less than the preset maximum ratio of the number of identical product tags, identify the same keywords in each product tag, and convert the keyword combination consisting of the same keywords into a standard tag.

10. The labeling classification method for e-commerce products according to claim 9 is characterized in that: The proportion of the maximum number of identical product labels is determined based on the ratio of the maximum number of identical product labels to the total number of products to be standardized.

Images