A product marking method and apparatus

Abstract

The application discloses a product marking method and device, and relates to the technical field of computers. A specific embodiment of the method comprises the following steps: constructing a vector representation of a candidate product based on product information of the candidate product, constructing a vector representation of a label by using product information of a seed product set corresponding to the label, inputting the vector representation of the candidate product and the vector representation of the label into a prediction model, and marking the candidate product as the label in the case where a prediction result of the prediction model indicates that the candidate product meets the label. The embodiment can make full use of information of similar products in a large amount of products, information of products in a label, and global information of the label itself, solve the problems that label marking is affected by information quality of products themselves and label information is not comprehensive enough, and has low labor cost, low maintenance cost, and good marking effect.

Description

Technical Field

[0001] This invention relates to the field of computer technology, and in particular to a product marking method and apparatus. Background Technology

[0002] To further enhance the consumer experience, e-commerce platforms integrate their product systems through tag association. Currently, product tagging solutions primarily rely on similarity algorithms such as collaborative filtering and machine learning to assign tags to products.

[0003] In the process of realizing this invention, the inventors discovered at least the following problems in the prior art:

[0004] The acquisition of similar products under the label is affected by the quality of the product information itself. It does not make full use of the information of other products under the label, ignores the global information of the label, affects the diversity and comprehensiveness of the recall results, has a low labeling accuracy, high labor and maintenance costs, and poor scalability. Summary of the Invention

[0005] In view of this, embodiments of the present invention provide a product marking method and apparatus that can make full use of information of similar products in a large number of products, product information in the label, and global information of the label itself. This solves the problems of label marking being affected by the quality of information of the product itself and insufficient utilization of label information, improves label diversity and marking accuracy, has low labor and maintenance costs, good marking effect, and good scalability.

[0006] To achieve the above objectives, according to one aspect of the present invention, a product marking method is provided.

[0007] A product tagging method includes: constructing a vector representation of the candidate product based on the product information of the candidate product; constructing a vector representation of the tag using the product information of a seed product set corresponding to the tag; inputting the vector representation of the candidate product and the vector representation of the tag into a prediction model; and tagging the candidate product as the tag if the prediction result of the prediction model indicates that the candidate product conforms to the tag.

[0008] Optionally, the product information of the candidate products includes the key attribute information of the candidate products; the method further includes: for any target product, obtaining the key attribute information of the target product in the following manner: clustering products in the product library according to a preset clustering dimension using a clustering algorithm to determine the product entity of the target product, wherein the product entity of the target product is a set consisting of the target product and similar products of the target product; scoring each attribute information of each product in the product entity of the target product to obtain the importance score of each attribute information under the product entity of the target product; sorting the importance scores to filter out a preset number of attribute information with the highest importance scores as the key attribute information of the target product.

[0009] Optionally, scoring each attribute information of each product in the product entity of the target product to obtain the importance score of each attribute information under the product entity of the target product includes: for each attribute information to be scored under the product entity of the target product, calculating a first conditional probability of the occurrence of the attribute information to be scored in the product entity of the target product based on the frequency of co-occurrence of the product and the attribute information to be scored in the product entity of the target product, and the total number of products in the product entity of the target product; calculating a second conditional probability of the occurrence of the product entity of the target product under the condition that the attribute information to be scored is known, based on the frequency of co-occurrence of the product and the attribute information to be scored in the product entity of the target product, and the total frequency of the attribute information to be scored in the products in the product library; and calculating the importance score of the attribute information to be scored based on the first conditional probability and the second conditional probability.

[0010] Optionally, the product information of the candidate product further includes the title information of the candidate product; the step of constructing the vector representation of the candidate product based on the product information of the candidate product includes: segmenting the title information of the candidate product into words to obtain title words of the candidate product, wherein the title words of the candidate product include product words; generating corresponding vectors based on the key attribute information of the candidate product, product words, and title words respectively to obtain the key attribute vector, product word vector, and title word vector of the candidate product; and constructing the vector representation of the candidate product based on the key attribute vector, product word vector, and title word vector of the candidate product.

[0011] Optionally, generating corresponding vectors based on the key attribute information, product terms, and title segmentation of the candidate product to obtain the key attribute vector, product term vector, and title segmentation vector of the candidate product includes: generating corresponding vectors based on the key attribute information, product terms, and title segmentation of the candidate product; performing a weighted average operation on the vectors corresponding to each key attribute information of the candidate product, using the importance score of each key attribute information of the candidate product as the weight, to obtain the key attribute vector of the candidate product; averaging the vectors corresponding to each product term of the candidate product to obtain the product term vector of the candidate product; and averaging the vectors corresponding to each title segmentation of the candidate product to obtain the title segmentation vector of the candidate product.

[0012] Optionally, constructing a vector representation of the candidate product based on the candidate product's key attribute vector, product word vector, and title word segmentation vector includes: sequentially concatenating the candidate product's key attribute vector, product word vector, and title word segmentation vector to obtain the candidate product's vector representation.

[0013] Optionally, the product information of the seed product set includes the title information, attribute information, and key attribute information of the products in the seed product set; the step of constructing the vector representation of the tag using the product information of the seed product set corresponding to the tag includes: extracting keywords of the seed product set based on the title information and attribute information of the products in the seed product set, generating word vectors for each keyword, performing a weighted average operation on the word vectors of each keyword to obtain the keyword vector of the seed product set; segmenting the title information of the products in the seed product set to obtain title word segments of the products in the seed product set, wherein the title word segments of the products in the seed product set include product words; generating corresponding vectors based on the product words and key attribute information of the products in the seed product set to obtain product word vectors and key attribute vectors of the products in the seed product set; and constructing the vector representation of the tag based on the keyword vectors, product word vectors, and key attribute vectors of the products in the seed product set.

[0014] Optionally, constructing the vector representation of the tag based on the keyword vector of the seed product set, the product word vector of the products in the seed product set, and the key attribute vector includes: for each product in the seed product set, sequentially concatenating the key attribute vector and the product word vector of the product to obtain the vector corresponding to the product in the seed product set; averaging the vectors corresponding to each product in the seed product set to obtain the vector representation of the seed product set; and sequentially concatenating the keyword vector of the seed product set and the vector representation of the seed product set to construct the vector representation of the tag.

[0015] Optionally, the prediction model is one of the following: a vector similarity calculation model, a dual-tower model, or a binary classification model.

[0016] According to another aspect of the present invention, a product marking device is provided.

[0017] A product labeling device includes: a product vector representation construction module, used to construct a vector representation of the candidate product based on the product information of the candidate product; a label vector representation construction module, used to construct a vector representation of the label using the product information of a seed product set corresponding to the label; and a labeling module, used to input the vector representation of the candidate product and the vector representation of the label into a prediction model, and when the prediction result of the prediction model indicates that the candidate product conforms to the label, label the candidate product as the label.

[0018] Optionally, the product information of the candidate products includes the key attribute information of the candidate products; the device further includes a key attribute information generation module, used to: cluster products in the product library according to a preset clustering dimension using a clustering algorithm to determine the product entity of the target product, wherein the product entity of the target product is a set consisting of the target product and similar products of the target product; score each attribute information of each product in the product entity of the target product to obtain an importance score for each attribute information under the product entity of the target product; sort the importance scores to filter out a preset number of attribute information with the highest importance scores as the key attribute information of the target product.

[0019] Optionally, the key attribute information generation module is further configured to: for each attribute information to be scored under the product entity of the target product, calculate a first conditional probability of the attribute information to be scored appearing in the product entity of the target product based on the frequency of co-occurrence between the product and the attribute information to be scored in the product entity of the target product, and the total number of products in the product entity of the target product; calculate a second conditional probability of the product entity of the target product appearing under the condition that the attribute information to be scored is known based on the frequency of co-occurrence between the product and the attribute information to be scored in the product entity of the target product, and the total frequency of the attribute information to be scored appearing in the products in the product library; and calculate the importance score of the attribute information to be scored based on the first conditional probability and the second conditional probability.

[0020] Optionally, the product information of the candidate product also includes the title information of the candidate product; the product vector representation construction module is further configured to: segment the title information of the candidate product to obtain the title segment of the candidate product, wherein the title segment of the candidate product includes product words; generate corresponding vectors based on the key attribute information, product words, and title segment of the candidate product to obtain the key attribute vector, product word vector, and title segment vector of the candidate product; and construct the vector representation of the candidate product based on the key attribute vector, product word vector, and title segment vector of the candidate product.

[0021] Optionally, the product vector representation construction module is further configured to: generate corresponding vectors based on the key attribute information, product words, and title word segments of the candidate products; perform a weighted average operation on the vectors corresponding to each key attribute information of the candidate products, using the importance score of each key attribute information of the candidate products as the weight, to obtain the key attribute vector of the candidate products; average the vectors corresponding to each product word of the candidate products to obtain the product word vector of the candidate products; and average the vectors corresponding to each title word segment of the candidate products to obtain the title word segment vector of the candidate products.

[0022] Optionally, the product vector representation construction module is further configured to: sequentially connect the key attribute vectors, product word vectors, and title word segmentation vectors of the candidate product to obtain the vector representation of the candidate product.

[0023] Optionally, the product information of the seed product set includes the title information, attribute information, and key attribute information of the products in the seed product set; the tag vector representation construction module is further configured to: extract keywords of the seed product set based on the title information and attribute information of the products in the seed product set, generate word vectors for each keyword, perform a weighted average operation on the word vectors of each keyword to obtain the keyword vector of the seed product set; segment the title information of the products in the seed product set to obtain title word segments of the products in the seed product set, wherein the title word segments of the products in the seed product set include product words; generate corresponding vectors based on the product words and key attribute information of the products in the seed product set to obtain product word vectors and key attribute vectors of the products in the seed product set; and construct the vector representation of the tag based on the keyword vector, product word vectors, and key attribute vectors of the products in the seed product set.

[0024] Optionally, the tag vector representation construction module is further configured to: for each product in the seed product set, sequentially concatenate the key attribute vector and product word vector of the product to obtain the vector corresponding to the product in the seed product set; average the vectors corresponding to each product in the seed product set to obtain the vector representation of the seed product set; and sequentially concatenate the keyword vector of the seed product set and the vector representation of the seed product set to construct the vector representation of the tag.

[0025] Optionally, the prediction model is one of the following: a vector similarity calculation model, a dual-tower model, or a binary classification model.

[0026] According to another aspect of the present invention, an electronic device is provided.

[0027] An electronic device includes: one or more processors; and a memory for storing one or more programs, which, when executed by the one or more processors, cause the one or more processors to implement the product marking method provided in the embodiments of the present invention.

[0028] According to another aspect of the present invention, a computer-readable medium is provided.

[0029] A computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the product marking method provided in the embodiments of the present invention.

[0030] One embodiment of the above invention has the following advantages or beneficial effects: It constructs vector representations of candidate products based on their product information; it constructs vector representations of tags using product information from the seed product set corresponding to the tags; it inputs the vector representations of candidate products and tags into a prediction model, and when the prediction model indicates that a candidate product matches the tag, it tags the candidate product as the tag. This fully utilizes information from similar products in a massive product pool, product information within tags, and the global information of the tags themselves, solving the problems of tag marking being affected by the quality of product information and insufficient utilization of tag information. It improves tag diversity and accuracy, has low labor and maintenance costs, good tagging results, and good scalability.

[0031] The further effects of the aforementioned unconventional alternative methods will be explained below in conjunction with specific implementation methods. Attached Figure Description

[0032] The accompanying drawings are provided to better understand the invention and are not intended to unduly limit the scope of the invention. Wherein:

[0033] Figure 1 This is a schematic diagram of the main steps of a product marking method according to an embodiment of the present invention;

[0034] Figure 2 This is a flowchart illustrating the process of generating key attribute information of a product according to an embodiment of the present invention;

[0035] Figure 3 This is a schematic diagram of the process for constructing a product vector representation according to an embodiment of the present invention;

[0036] Figure 4 This is a schematic diagram illustrating the process of constructing a label vector representation according to an embodiment of the present invention;

[0037] Figure 5 This is a schematic diagram of the main modules of a product marking device according to an embodiment of the present invention;

[0038] Figure 6 This is an exemplary system architecture diagram in which embodiments of the present invention can be applied;

[0039] Figure 7 This is a schematic diagram of the structure of a computer system suitable for implementing terminal devices or servers of the present invention. Detailed Implementation

[0040] The following description, in conjunction with the accompanying drawings, illustrates exemplary embodiments of the present invention, including various details to aid understanding. These details should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the invention. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0041] Figure 1 This is a schematic diagram of the main steps of a product marking method according to an embodiment of the present invention.

[0042] like Figure 1 As shown, a product marking method according to an embodiment of the present invention mainly includes the following steps S101 to S103.

[0043] Step S101: Construct a vector representation of the candidate product based on the product information of the candidate product, wherein the product information of the candidate product includes the key attribute information of the candidate product and the title information of the candidate product.

[0044] Candidate products are those to be labeled with the corresponding tag. The vector representation of a candidate product is the vector used to represent that candidate product.

[0045] For any target product, its product information can include its title, key attributes, and attribute information, including but not limited to functional attributes, material attributes, style attributes, origin attributes, color attributes, flavor attributes, and specifications. The key attribute information of the target product can be obtained as follows: Cluster the products in the product library according to a preset clustering dimension using a clustering algorithm to determine the target product entity. The target product entity is a set consisting of the target product and its similar products. Each attribute of each product within the target product entity is scored to obtain an importance score for each attribute. The importance scores are then sorted to select the attribute information with the highest preset importance scores as the target product's key attributes. The key attribute information of candidate products can be obtained using the same method.

[0046] Clustering algorithms can be implemented using various commonly used methods. Preset clustering dimensions include, for example, product title segmentation or attribute information. Similar products to the target product can then be clustered together.

[0047] Each attribute of each product in the target product entity is scored to obtain an importance score for each attribute under the target product entity. This includes: for each attribute to be scored under the target product entity, calculating a first conditional probability of the attribute to be scored appearing in the target product entity based on the frequency of co-occurrence between the product and the attribute to be scored in the target product entity, and the total number of products in the target product entity; calculating a second conditional probability of the target product entity appearing given the attribute to be scored, based on the frequency of co-occurrence between the product and the attribute to be scored in the target product entity, and the total frequency of the attribute to be scored appearing in the products in the product library; and calculating the importance score of the attribute to be scored based on the first and second conditional probabilities.

[0048] The process of constructing a vector representation of candidate products based on their product information includes: segmenting the title information of candidate products to obtain title segments, which include product words; generating corresponding vectors based on the key attribute information, product words, and title segments of candidate products to obtain key attribute vectors, product word vectors, and title segment vectors of candidate products; and constructing a vector representation of candidate products based on the key attribute vectors, product word vectors, and title segment vectors of candidate products.

[0049] Based on the key attribute information, product terms, and title segmentation of the candidate products, corresponding vectors are generated respectively to obtain the key attribute vector, product term vector, and title segmentation vector of the candidate products. This includes: generating corresponding vectors based on the key attribute information, product terms, and title segmentation of the candidate products; performing a weighted average operation on the vectors corresponding to each key attribute information of the candidate products, using the importance score of each key attribute information of the candidate products as the weight, to obtain the key attribute vector of the candidate products; averaging the vectors corresponding to each product term of the candidate products to obtain the product term vector of the candidate products; and averaging the vectors corresponding to each title segmentation of the candidate products to obtain the title segmentation vector of the candidate products.

[0050] Based on the key attribute vectors, product word vectors, and title word segmentation vectors of the candidate products, a vector representation of the candidate products is constructed, including: sequentially connecting the key attribute vectors, product word vectors, and title word segmentation vectors of the candidate products to obtain the vector representation of the candidate products.

[0051] Step S102: Construct a vector representation of the tag using the product information of the seed product set corresponding to the tag.

[0052] The product information for a seed product collection includes the product titles, attribute information, and key attribute information. Following the method described above for obtaining the key attribute information of the target product, the key attribute information of the products in the seed product collection can be obtained.

[0053] The vector representation of a label is the vector used to represent that label.

[0054] Using product information from the seed product set corresponding to the tag, a vector representation of the tag is constructed, including: extracting keywords from the seed product set based on the title and attribute information of the products in the seed product set, generating word vectors for each keyword, and performing a weighted average operation on the word vectors of each keyword to obtain the keyword vector of the seed product set; segmenting the title information of the products in the seed product set to obtain title word segments, which include product words; generating corresponding vectors based on the product words and key attribute information of the products in the seed product set to obtain product word vectors and key attribute vectors of the products in the seed product set; and constructing a vector representation of the tag based on the keyword vector, product word vector, and key attribute vector of the seed product set.

[0055] Based on the keyword vectors of the seed product set, the product word vectors of the products in the seed product set, and the key attribute vectors, a vector representation of the tag is constructed, including: for each product in the seed product set, the key attribute vector and the product word vector of the product are sequentially concatenated to obtain the vector corresponding to the product in the seed product set; the average of the vectors corresponding to each product in the seed product set is calculated to obtain the vector representation of the seed product set; the keyword vectors and the vector representations of the seed product set are sequentially concatenated to construct the vector representation of the tag.

[0056] Step S103: Input the vector representation of the candidate product and the vector representation of the label into the prediction model. If the prediction result of the prediction model indicates that the candidate product matches the label, label the candidate product with the label.

[0057] The prediction model can be one of a vector similarity calculation model, a dual-tower model, or a binary classification model. The prediction result indicates whether a candidate product matches the label. For example, a vector similarity calculation model can calculate the similarity between the vector representation of a candidate product and the vector representation of the label. If the similarity is greater than a certain set threshold, it indicates that the two vectors are similar, meaning the candidate product matches the label; otherwise, it indicates that the candidate product does not match the label. The prediction model in this embodiment can be implemented using any general vector similarity calculation model, dual-tower model, or binary classification model.

[0058] Figure 2 This is a schematic diagram illustrating the process of generating key attribute information of a product according to an embodiment of the present invention.

[0059] like Figure 2 As shown, in one embodiment, a vector representation of a product is constructed based on massive product information in a product database. This fully utilizes information about similar products within the vast database to build a high-level semantic vector representation for each product. Product information from the product database is obtained. This information includes both unstructured and structured information. Unstructured information includes the product title, while structured information includes product attributes, etc. This product information can be manually entered by the merchant. Examples of product information are shown in Table 1.

[0060] Table 1 Example of product information

[0061]

[0062]

[0063] Taking a product from the product database as a candidate product, the title information of the candidate product is segmented into words to obtain the title words of the candidate product. The title words of the candidate product include product words. Specifically, the unstructured information of the product title is subjected to part-of-speech (POS) recognition of the title words, thereby converting them into structured POS-tagged data. The POS-tagged data indicates the category to which the title words belong. The category to which the title words belong can be defined as the following 16 categories: brand words, model words, product words, applicable events, applicable times, applicable locations, applicable people, other applicable objects, pattern descriptions, functional attributes, material attributes, style attributes, fashion attributes, place of origin attributes, color attributes, taste attributes, and specification attributes. The category to which each word belongs is determined by a sequence labeling model, where the sequence labeling model can be a Conditional Random Field (CRF) model, a BiLSTM-CRF model, etc. Table 2 shows some examples of the POS-tagged results of the product title "A Brand Casual Pants Men's Sweatpants Fleece-lined Thick Autumn Winter Sports Pants Men's Straight Leg Loose Casual Men's Pants Trendy Brand Jogger Nine-point Korean Style Trendy Embroidered Versatile K39 Floral XL".

[0064] Table 2 shows some examples of part-of-speech tagging results for products.

[0065]

[0066]

[0067] The product entity of a target product is determined by clustering products in a product library according to a preset clustering dimension using a clustering algorithm. The target product entity is a set consisting of the target product and its similar products. The target product can be any product, such as a candidate product. Specifically, a product entity can be a set of similar products sold by different vendors. Similar products are those with identical key attributes; for example, mobile phone products with identical model, brand, and memory information can be considered similar products. This invention can aggregate similar products from a massive database using machine learning clustering algorithms, including but not limited to k-means clustering, Gaussian mixture clustering, and large-scale nearest neighbor search techniques such as min-hash. The clustering dimension used in the clustering algorithm can be determined based on product title segmentation and product attributes. The distance metrics used in the clustering algorithm include, but are not limited to, Euclidean distance, cosine distance, and Jaccard distance. As shown in Table 3, taking the minhash nearest neighbor search technology as an example, the clustering dimension is selected from the product title word segmentation, and the distance metric is selected as Jaccard distance. Assume the title word segmentation set of product A is A = {'Brand X', 'Mobile Phone Y', '11', '(', 'A2223', ')', '64GB', 'Black', 'Mobile', 'Unicom', 'Telecom', '4G', 'Mobile Phone', 'Dual SIM', 'Dual Standby', '

', 'Repair', 'Worry-free', 'Year', 'Paid Version', '

[0068]

[0069] Here, "||" represents the length of the set, and the calculated Jaccard distance between product A and product B is 0.56. Following this method, the entire product set is traversed to obtain all Jaccard distances. Products with a Jaccard distance greater than a preset empirical threshold are considered similar products of product A and added to the product entity of product A.

[0070] Table 3. Calculation of Jaccard distance between Product A and Product B

[0071]

[0072] For each attribute information to be scored under the product entity of the target product, the first conditional probability of the attribute information to be scored appearing in the product entity of the target product is calculated based on the frequency of co-occurrence between the product and the attribute information to be scored in the product entity of the target product, and the total number of products in the product entity of the target product. The second conditional probability of the product entity of the target product appearing is calculated based on the frequency of co-occurrence between the product and the attribute information to be scored in the product entity of the target product, and the total frequency of the attribute information to be scored appearing in the products in the product library, given the attribute information to be scored. The importance score of the attribute information to be scored is calculated based on the first and second conditional probabilities. Specifically, product attributes include attribute information in the product title obtained based on part-of-speech tagging and attribute information filled in by the merchant. Let the set of entities in the product library be A = {E1, E2, ..., E...}. i ,...,E n}, where E i Let E represent the i-th product entity, and let E = {S1, S2, ..., S...} be the set of products under this product entity. i ,...,S n}, where S i Let S represent the i-th product in set E, where each product has k attribute information, denoted as S = {a1, a2, ..., a...}. i ,...,a n}, where a i This represents the i-th attribute information of the product. The importance score (score(a)) for each attribute information of the product under each product entity is calculated using the following formula: i ):

[0073] score(a i )=P(a i |E j )*P(E j |a i )

[0074] Wherein, P(a i |E j ) indicates that in product entity E j Attribute information a i The conditional probability of occurrence (i.e., the first conditional probability) can actually be calculated based on frequency, specifically as follows:

[0075]

[0076] Where, n(a) i E j ) indicates that in product entity E j Product and attribute information a i Co-occurrence frequency, n(E) j) represents product entity E j The total number of products in the middle. Similarly, P(E) j |a i ) indicates that the attribute information a is known. i Under the condition of product entity E j The conditional probability of occurrence (i.e., the second conditional probability) is calculated as follows:

[0077]

[0078] Where, n(a) i ) represents attribute information a i Total frequency of appearance of products across the entire site (product database).

[0079] The importance scores are sorted to select the attribute information with the highest preset number of importance scores as the key attribute information of the target product. Specifically, after calculating the importance score of each attribute information under each product, the attributes are sorted according to the scores and the N attributes with the highest importance scores are selected to obtain the N key attribute information of the product entity. The value of N can be determined by experience.

[0080] Figure 3 This is a schematic diagram of the process for constructing a product vector representation according to an embodiment of the present invention.

[0081] like Figure 3 As shown, in one embodiment, a vector representation of the product is generated based on the product's part-of-speech tagging results and the extracted key product attribute information of the product entity. The product's vector representation includes three parts: key attribute vector, product word vector, and title word segmentation vector.

[0082] A vector representation of candidate products is constructed. Specifically, unstructured information from the titles of candidate products is used for title segmentation and part-of-speech tagging, thus converting it into structured part-of-speech labeled data. Based on the title segmentation of candidate products, corresponding vectors are generated. The average of the vectors corresponding to each title segmentation of candidate products is then calculated to obtain the title segmentation vector of the candidate product. Specifically, the product title segmentation is used as input to a word vector tool to obtain the vector corresponding to the title segmentation. Word vector tools include, but are not limited to, FastText and BERT. Let the set of title segmentation words for product P be P = {S1, S2, ..., S...}. i ,...,S n}, where S i V represents the set of vectors corresponding to the i-th word segment in the title, generated using a word vector tool. p ={v1,v2,...,v i ,...,v n}, where v iLet represent the word segmentation vector corresponding to the i-th title word segmentation. Then, for each title word segmentation of product P, the corresponding vectors v1, v2, ..., v... i ,...,v n Calculate the average to obtain the title segmentation vector Vec for product P. p for:

[0083]

[0084] Based on the key attribute information of candidate products, corresponding vectors are generated. Using the importance score of each key attribute information of the candidate product as a weight, a weighted average is performed on the vectors corresponding to each key attribute information of the candidate product to obtain the key attribute vector of the candidate product. Specifically, the vectors are generated based on the key attribute information of the product. Assume that the set of key attributes of entity E corresponding to product P is E = {a1, a2, ..., a...}. i ,...,a k The corresponding importance score is S = {s1, s2, ..., s}. i ,...,s k The vector set corresponding to each key attribute information is V. E ={v1,v2,...,v i ,...,v k}, then the key attribute vector Vec of product P E for:

[0085]

[0086] in, This represents the sum of attribute scores for a product entity.

[0087] Based on the product terms of candidate products, corresponding vectors are generated. The average of the vectors corresponding to each product term of the candidate product is then calculated to obtain the product term vector for the candidate product. Specifically, vectors are generated based on the title word segmentation labeled as "product term". Taking Table 2 as an example, the product terms are casual pants, sweatpants, and sports pants. Assume the set of product terms for product P is P = {P1, P2, ..., P...} i ,...,P m}, where the vector set corresponding to the product terms is V. p ={v1,v2,...,v i ,...,v m}, then the product word vector Vec of product P pt for:

[0088]

[0089] Vec, the key attribute vector of candidate products E Product word vectors (Vec)pt Title segmentation vector Vec p By sequentially connecting the components, we obtain the vector representation V of the candidate product. p The specific generation method is as follows:

[0090]

[0091] Figure 4 This is a schematic diagram of the process for constructing a label vector representation according to an embodiment of the present invention.

[0092] like Figure 4 As shown in one embodiment, to ensure the quality of the seed products for the tag, the set of seed products for the tag can be obtained through methods such as manual sorting and e-commerce search user behavior. An example of the seed product set tagged "camouflage style" is shown in Table 4.

[0093] Table 4 shows examples of seed product sets labeled "camouflage".

[0094]

[0095]

[0096] Based on the title and attribute information of products in the seed product set, a predetermined number of keywords are extracted from the seed product set, and word vectors for each keyword are generated. A weighted average of these word vectors is then performed to obtain the keyword vector for the seed product set. Specifically, the keyword extraction scope includes the attribute information of the seed products and the word segmentation information of the product titles. Keyword extraction methods include, but are not limited to, word frequency statistics, TFIDF keyword extraction, and LDA topic extraction algorithms. Taking word frequency statistics as an example, based on the word frequency statistics of title word segmentation in Table 4, the five keywords with the highest word frequency are ["camouflage", "five-point pants", "men's", "summer", "shorts"]. These are used as the keywords for this tag, serving as the overall description of the tag. Word vectors for each keyword are generated using word vector tools. Assume the keyword set is W = {w1, w2, ..., w...} i ,...,w n The corresponding set of keyword vectors is VW = {v1, v2, ..., v}. i ,...,v n}, then the keyword vector Vec of the seed product set w for:

[0097]

[0098] The title information of products in the seed product set is segmented into words, resulting in title word segments, which include product words. Based on the product words and key attribute information of the products in the seed product set, corresponding vectors are generated, resulting in product word vectors and key attribute vectors for the products in the seed product set. For each product in the seed product set, the key attribute vector and product word vector are concatenated sequentially to obtain the corresponding vector for that product. The average of the vectors for all products in the seed product set is then calculated to obtain the vector representation of the seed product set. Specifically, based on the product information of each product in the seed product set, the product word vector and key attribute vector for each product in the seed product set are obtained. Assume the product word vector of any product in the seed product set is Vec. pt The key attribute vector is Vec E Then the vector corresponding to the product in the seed product set is V. P =Vec pt ⊕Vec E This method can be used to obtain the vector corresponding to each product in the seed product set. Assume the seed product set is P = {p1, p2, ..., p...} i ,...,p m}, where the vector set corresponding to the products is V={V p1 V p2 ,...,V pi ,...,V pm Then, the vector representation of the seed product set is:

[0099]

[0100] The keyword vectors and vector representations of the seed product set are sequentially concatenated to construct the vector representation of the tag. The specific generation method is as follows:

[0101]

[0102] Among them, V tag The vector representation of the label.

[0103] In one embodiment, the vector representations of the candidate product and the label are input into the prediction model. If the prediction model indicates that the candidate product matches the label, the candidate product is labeled. Specifically, the vector representations of the candidate product and the label are used as input, and the prediction model determines whether the candidate product belongs to the label. The prediction model includes, but is not limited to, directly calculating vector similarity, a dual-tower deep learning model, and a machine learning binary classification model. Before using the prediction model, sample data can be obtained based on experience, and the prediction model can be trained accordingly.

[0104] Figure 5 This is a schematic diagram of the main modules of a product marking device according to an embodiment of the present invention.

[0105] like Figure 5 As shown, a product marking device 500 according to an embodiment of the present invention mainly includes: a product vector representation construction module 501, a label vector representation construction module 502, and a marking module 503.

[0106] Product vector representation construction module 501 is used to construct vector representations of candidate products based on the product information of candidate products.

[0107] The tag vector representation construction module 502 is used to construct a vector representation of the tag using product information from the seed product set corresponding to the tag.

[0108] The labeling module 503 is used to input the vector representation of the candidate product and the vector representation of the label into the prediction model. When the prediction result of the prediction model indicates that the candidate product matches the label, the candidate product is labeled as the label.

[0109] In one embodiment, the product information of the candidate products includes key attribute information of the candidate products; the apparatus further includes a key attribute information generation module, used to: cluster products in the product library according to a preset clustering dimension using a clustering algorithm to determine the product entity of the target product, wherein the product entity of the target product is a set consisting of the target product and similar products of the target product; score each attribute information of each product in the product entity of the target product to obtain an importance score for each attribute information under the product entity of the target product; sort the importance scores to select a preset number of attribute information with the highest importance scores as the key attribute information of the target product.

[0110] In one embodiment, the key attribute information generation module is further configured to: for each attribute information to be scored under the product entity of the target product, calculate a first conditional probability of the attribute information to be scored appearing in the product entity of the target product based on the frequency of co-occurrence of products and the attribute information to be scored in the product entity of the target product, and the total number of products in the product entity of the target product; calculate a second conditional probability of the product entity of the target product appearing under the condition that the attribute information to be scored is known, based on the frequency of co-occurrence of products and the attribute information to be scored in the product entity of the target product, and the total frequency of the attribute information to be scored appearing in the products in the product library; and calculate the importance score of the attribute information to be scored based on the first conditional probability and the second conditional probability.

[0111] In one embodiment, the product information of the candidate product also includes the title information of the candidate product; the product vector representation construction module is further used to: segment the title information of the candidate product to obtain the title segment of the candidate product, wherein the title segment of the candidate product includes product words; generate corresponding vectors based on the key attribute information of the candidate product, product words, and title segment, respectively, to obtain the key attribute vector, product word vector, and title segment vector of the candidate product; and construct the vector representation of the candidate product based on the key attribute vector, product word vector, and title segment vector of the candidate product.

[0112] In one embodiment, the product vector representation construction module is further configured to: generate corresponding vectors based on the key attribute information, product words, and title word segments of the candidate product; perform a weighted average operation on the vectors corresponding to each key attribute information of the candidate product, using the importance score of each key attribute information of the candidate product as the weight, to obtain the key attribute vector of the candidate product; calculate the average of the vectors corresponding to each product word of the candidate product to obtain the product word vector of the candidate product; and calculate the average of the vectors corresponding to each title word segment of the candidate product to obtain the title word segment vector of the candidate product.

[0113] In one embodiment, the product vector representation construction module is further used to: sequentially connect the key attribute vectors, product word vectors, and title word segmentation vectors of the candidate product to obtain the vector representation of the candidate product.

[0114] In one embodiment, the product information of the seed product set includes the title information, attribute information, and key attribute information of the products in the seed product set. The tag vector representation construction module is further used to: extract keywords of the seed product set based on the title information and attribute information of the products in the seed product set, generate word vectors for each keyword, perform a weighted average operation on the word vectors of each keyword to obtain the keyword vector of the seed product set; segment the title information of the products in the seed product set to obtain the title word segmentation of the products in the seed product set, which includes product words; generate corresponding vectors based on the product words and key attribute information of the products in the seed product set to obtain the product word vector and key attribute vector of the products in the seed product set; and construct the vector representation of the tag based on the keyword vector, the product word vector, and the key attribute vector of the products in the seed product set.

[0115] In one embodiment, the tag vector representation construction module is further configured to: for each product in the seed product set, sequentially concatenate the key attribute vector and product word vector of the product to obtain the vector corresponding to the product in the seed product set; average the vectors corresponding to each product in the seed product set to obtain the vector representation of the seed product set; and sequentially concatenate the keyword vector and the vector representation of the seed product set to construct the vector representation of the tag.

[0116] In one embodiment, the prediction model is one of the following: a vector similarity calculation model, a dual-tower model, or a binary classification model.

[0117] Furthermore, the specific implementation details of the product marking device in this embodiment of the invention have been described in detail in the above product marking method, so they will not be repeated here.

[0118] Figure 6 An exemplary system architecture 600 is shown that can be applied to the product marking method or product marking apparatus of the present invention.

[0119] like Figure 6 As shown, system architecture 600 may include terminal devices 601, 602, and 603, a network 604, and a server 605. Network 604 serves as the medium for providing communication links between terminal devices 601, 602, and 603 and server 605. Network 604 may include various connection types, such as wired or wireless communication links or fiber optic cables, etc.

[0120] Users can use terminal devices 601, 602, and 603 to interact with server 605 via network 604 to receive or send messages, etc. Various communication client applications can be installed on terminal devices 601, 602, and 603, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social media platform software, etc. (for example only).

[0121] Terminal devices 601, 602, and 603 can be various electronic devices with displays and web browsing capabilities, including but not limited to smartphones, tablets, laptops, and desktop computers.

[0122] Server 605 can be a server that provides various services, such as a backend management server that supports shopping websites browsed by users using terminal devices 601, 602, and 603 (for example only). The backend management server can analyze and process data such as received product information query requests, and feed back the processing results (such as target push information, product information - for example only) to the terminal devices.

[0123] It should be noted that the product marking method provided in this embodiment of the invention is generally executed by server 605, and correspondingly, the product marking device is generally set in server 605.

[0124] It should be understood that Figure 6 The number of terminal devices, networks, and servers shown is merely illustrative. Depending on implementation needs, any number of terminal devices, networks, and servers can be included.

[0125] The following is for reference. Figure 7 It shows a schematic diagram of the structure of a computer system 700 suitable for implementing terminal devices or servers of the present invention. Figure 7 The terminal device or server shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present invention.

[0126] like Figure 7 As shown, the computer system 700 includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 702 or programs loaded from storage section 708 into random access memory (RAM) 703. The RAM 703 also stores various programs and data required for the operation of the system 700. The CPU 701, ROM 702, and RAM 703 are interconnected via a bus 704. An input / output (I / O) interface 705 is also connected to the bus 704.

[0127] The following components are connected to the I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card, modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to the I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 710 as needed so that computer programs read from it can be installed into the storage section 708 as needed.

[0128] In particular, according to the embodiments disclosed in this invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit (CPU) 701, it performs the functions defined above in the system of this invention.

[0129] It should be noted that the computer-readable medium shown in this invention can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this invention, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this invention, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. Computer-readable signal media can also be any computer-readable medium other than computer-readable storage media, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0130] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in a block diagram or flowchart, and combinations of blocks in a block diagram or flowchart, may be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0131] The modules described in the embodiments of the present invention can be implemented in software or hardware. The described modules can also be housed in a processor; for example, a processor may be described as including a product vector representation construction module, a tag vector representation construction module, and a tagging module. The names of these modules do not necessarily limit the module itself; for example, the product vector representation construction module may also be described as "a module for constructing vector representations of candidate products based on product information of candidate products."

[0132] In another aspect, the present invention also provides a computer-readable medium, which may be included in the device described in the above embodiments; or it may exist independently and not assembled into the device. The computer-readable medium carries one or more programs that, when executed by the device, cause the device to: construct a vector representation of the tag using product information from a set of seed products corresponding to the tag; input the vector representations of candidate products and the vector representations of the tag into a prediction model; and, if the prediction result of the prediction model indicates that a candidate product conforms to the tag, label the candidate product as a tag.

[0133] According to the technical solution of this invention, a vector representation of the tag is constructed using product information from a seed product set corresponding to the tag. The vector representations of candidate products and the tag are input into a prediction model. If the prediction model indicates that a candidate product matches the tag, the candidate product is tagged. This approach fully utilizes information from a vast number of similar products, product information within the tag, and the global information of the tag itself. It solves the problems of tagging being affected by the quality of product information and insufficient utilization of tag information, improving tag diversity and accuracy. It also features low labor and maintenance costs, good tagging results, and good scalability.

[0134] The specific embodiments described above do not constitute a limitation on the scope of protection of this invention. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can occur depending on design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this invention should be included within the scope of protection of this invention.

Claims

1. A method of product marking, characterized by include: Construct a vector representation of the candidate product based on the product information of the candidate product; Using the product information of the seed product set corresponding to the tag, construct a vector representation of the tag; The vector representations of the candidate products and the vector representations of the labels are input into the prediction model. If the prediction result of the prediction model indicates that the candidate product matches the label, the candidate product is labeled with the label. The step of constructing a vector representation of the candidate product based on the product information of the candidate product includes: segmenting the title information of the candidate product into words to obtain title words of the candidate product, wherein the title words of the candidate product include product words; generating corresponding vectors based on the key attribute information, product words, and title words of the candidate product to obtain the key attribute vector, product word vector, and title word vector of the candidate product; and constructing a vector representation of the candidate product based on the key attribute vector, product word vector, and title word vector of the candidate product. And / or, The step of constructing a vector representation of the tag using product information from a seed product set corresponding to the tag includes: extracting keywords from the seed product set based on the title and attribute information of the products in the seed product set, generating word vectors for each keyword, and performing a weighted average operation on the word vectors of each keyword to obtain a keyword vector for the seed product set; segmenting the title information of the products in the seed product set to obtain title word segments for the products in the seed product set, wherein the title word segments for the products in the seed product set include product words; generating corresponding vectors based on the product words and key attribute information of the products in the seed product set to obtain product word vectors and key attribute vectors for the products in the seed product set; and constructing a vector representation of the tag based on the keyword vectors, product word vectors, and key attribute vectors of the products in the seed product set.

2. The method of claim 1, wherein, The product information of the candidate products includes the key attribute information of the candidate products; The method further includes: For any target product, the key attribute information of the target product is obtained in the following manner: The products in the product library are clustered according to a preset clustering dimension using a clustering algorithm to determine the product entity of the target product. The product entity of the target product is a set consisting of the target product and its similar products. Each attribute of each product in the product entity of the target product is scored to obtain the importance score of each attribute under the product entity of the target product. The importance scores are sorted to filter out a preset number of attribute information with the highest importance scores as the key attribute information of the target product.

3. The method of claim 2, wherein, The scoring of each attribute information of each product in the product entity of the target product to obtain the importance score of each attribute information under the product entity of the target product includes: For each attribute information to be scored under the product entity of the target product, a first conditional probability of the attribute information to be scored appearing in the product entity of the target product is calculated based on the frequency of co-occurrence between the product and the attribute information to be scored in the product entity of the target product, and the total number of products in the product entity of the target product. A second conditional probability of the product entity of the target product appearing is calculated based on the frequency of co-occurrence between the product and the attribute information to be scored in the product entity of the target product, and the total frequency of the attribute information to be scored appearing in the products of the product library, given that the attribute information to be scored is known. The importance score of the attribute information to be scored is calculated based on the first conditional probability and the second conditional probability.

4. The method of claim 1, wherein, The process of generating corresponding vectors based on the key attribute information, product terms, and title segmentation of the candidate products, respectively, to obtain the key attribute vector, product term vector, and title segmentation vector of the candidate products, includes: Based on the key attribute information of the candidate products, product terms, and title word segmentation, corresponding vectors are generated respectively. Using the importance score of each key attribute information of the candidate product as the weight, a weighted average operation is performed on the vectors corresponding to each key attribute information of the candidate product to obtain the key attribute vector of the candidate product; the average of the vectors corresponding to each product word of the candidate product is calculated to obtain the product word vector of the candidate product; the average of the vectors corresponding to each title word segment of the candidate product is calculated to obtain the title word segment vector of the candidate product.

5. The method of claim 1, wherein, The step of constructing a vector representation of the candidate product based on the candidate product's key attribute vector, product word vector, and title word segmentation vector includes: The key attribute vectors, product word vectors, and title word segmentation vectors of the candidate products are sequentially concatenated to obtain the vector representation of the candidate products.

6. The method of claim 1, wherein, The construction of the vector representation of the tag based on the keyword vector of the seed product set, the product word vector of the product in the seed product set, and the key attribute vector includes: For each product in the seed product set, the key attribute vector and product word vector of the product are sequentially concatenated to obtain the vector corresponding to the product in the seed product set. The average of the vectors corresponding to each product in the seed product set is then calculated to obtain the vector representation of the seed product set. The keyword vector of the seed product set and the vector representation of the seed product set are sequentially concatenated to construct the vector representation of the tag.

7. The method according to claim 1, characterized in that, The prediction model is one of the following: vector similarity calculation model, dual-tower model, or binary classification model.

8. A product marking device, characterized in that, include: The product vector representation construction module is used to construct the vector representation of the candidate product based on the product information of the candidate product. The tag vector representation construction module is used to construct the vector representation of the tag using product information of the seed product set corresponding to the tag; The tagging module is used to input the vector representation of the candidate product and the vector representation of the label into the prediction model, and when the prediction result of the prediction model indicates that the candidate product matches the label, the candidate product is tagged with the label. The product vector representation construction module is further configured to: segment the title information of the candidate product to obtain the title segment of the candidate product, wherein the title segment of the candidate product includes product words; generate corresponding vectors based on the key attribute information, product words, and title segment of the candidate product to obtain the key attribute vector, product word vector, and title segment vector of the candidate product; and construct the vector representation of the candidate product based on the key attribute vector, product word vector, and title segment vector of the candidate product. And / or, The tag vector representation construction module is further configured to: extract keywords from the seed product set based on the title information and attribute information of the products in the seed product set, generate word vectors for each keyword, perform a weighted average operation on the word vectors of each keyword to obtain the keyword vector of the seed product set; segment the title information of the products in the seed product set to obtain title word segments of the products in the seed product set, wherein the title word segments of the products in the seed product set include product words; generate corresponding vectors based on the product words and key attribute information of the products in the seed product set to obtain product word vectors and key attribute vectors of the products in the seed product set; and construct the vector representation of the tag based on the keyword vector, product word vectors, and key attribute vectors of the products in the seed product set.

9. The apparatus according to claim 8, characterized in that, The product information of the candidate products includes the key attribute information of the candidate products; The device further includes a key attribute information generation module, used for: The products in the product library are clustered according to a preset clustering dimension using a clustering algorithm to determine the product entity of the target product. The product entity of the target product is a set consisting of the target product and its similar products. Each attribute of each product in the product entity of the target product is scored to obtain the importance score of each attribute under the product entity of the target product. The importance scores are sorted to filter out a preset number of attribute information with the highest importance scores as the key attribute information of the target product.

10. The apparatus according to claim 9, characterized in that, The key attribute information generation module is also used for: For each attribute information to be scored under the product entity of the target product, the first conditional probability of the attribute information to be scored appearing in the product entity of the target product is calculated based on the frequency of co-occurrence of the product and the attribute information to be scored in the product entity of the target product, and the total number of products in the product entity of the target product. Based on the frequency of co-occurrence of the product and the attribute information to be scored in the product entity of the target product, and the total frequency of the attribute information to be scored appearing in the products of the product library, a second conditional probability of the occurrence of the product entity of the target product is calculated under the condition that the attribute information to be scored is known; the importance score of the attribute information to be scored is calculated based on the first conditional probability and the second conditional probability.

11. The apparatus according to claim 8, characterized in that, The product vector representation construction module is also used for: Based on the key attribute information of the candidate products, product terms, and title word segmentation, corresponding vectors are generated respectively. Using the importance score of each key attribute information of the candidate product as the weight, a weighted average operation is performed on the vectors corresponding to each key attribute information of the candidate product to obtain the key attribute vector of the candidate product; the average of the vectors corresponding to each product word of the candidate product is calculated to obtain the product word vector of the candidate product; the average of the vectors corresponding to each title word segment of the candidate product is calculated to obtain the title word segment vector of the candidate product.

12. The apparatus according to claim 8, characterized in that, The product vector representation construction module is also used for: The key attribute vectors, product word vectors, and title word segmentation vectors of the candidate products are sequentially concatenated to obtain the vector representation of the candidate products.

13. The apparatus according to claim 8, characterized in that, The label vector representation construction module is also used for: For each product in the seed product set, the key attribute vector and product word vector of the product are sequentially concatenated to obtain the vector corresponding to the product in the seed product set. The average of the vectors corresponding to each product in the seed product set is then calculated to obtain the vector representation of the seed product set. The keyword vector of the seed product set and the vector representation of the seed product set are sequentially concatenated to construct the vector representation of the tag.

14. The apparatus according to claim 9, characterized in that, The prediction model is one of the following: vector similarity calculation model, dual-tower model, or binary classification model.

15. An electronic device, characterized in that, include: One or more processors; Storage device for storing one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in any one of claims 1-7.

16. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method as described in any one of claims 1-7.

Images