Product feature extraction method and device, electronic equipment, storage medium and computer program product

Abstract

The invention relates to a product feature extraction method and apparatus, an electronic device, a storage medium and a computer program product. The method comprises the steps of obtaining a picture about a target product; extracting a plurality of keywords from the picture; obtaining a vector corresponding to each keyword in the plurality of keywords; clustering the plurality of vectors corresponding to the plurality of keywords to obtain a clustering center; determining a target vector closest to the clustering center in the plurality of vectors; and taking the target keyword corresponding to the target vector as the key feature of the target product. Therefore, the key features of the product can be extracted from the picture by adopting an automatic means, and compared with a mode of manually extracting the key features, the extraction efficiency can be greatly improved.

Description

Technical Field

[0001] The present disclosure relates to the field of computer technology, and more specifically, to a product feature extraction method, device, electronic device, storage medium and computer program product. Background Art

[0002] With the development of the Internet, various e-commerce platforms have emerged one after another. Users can purchase goods or services through e-commerce platforms, which provides great convenience for users' work and life. In addition, if the marketing methods of e-commerce platform merchants are appropriate, it can also bring greater economic benefits to merchants. Therefore, how to quickly and accurately identify the core selling points of products has become a major challenge facing enterprises and marketers.

[0003] In the related technologies, the core features of products, i.e. marketing hot spots, are mainly extracted manually. However, the marketing pictures, marketing videos, marketing text descriptions and other information on the Internet are too complex. If the core features of products are extracted manually, the extraction workload will be too large, which will lead to low efficiency and poor accuracy in extracting the core features of products. Summary of the invention

[0004] The present disclosure provides a product feature extraction method, device, electronic device, storage medium and computer program product to at least solve the problem of low extraction efficiency of core features of products extracted manually in the above-mentioned related technologies.

[0005] According to a first aspect of an embodiment of the present disclosure, a product feature extraction method is provided, comprising: obtaining a picture of a target product; extracting multiple keywords from the picture; obtaining a vector corresponding to each of the multiple keywords; clustering the multiple vectors corresponding to the multiple keywords to obtain a cluster center; determining a target vector among the multiple vectors that is closest to the cluster center; and using a target keyword corresponding to the target vector as a key feature of the target product.

[0006] Optionally, there are multiple cluster centers, and before determining the target vector among the multiple vectors that is closest to the cluster center, it also includes: for the vector corresponding to each keyword, determining the cluster center among the multiple cluster centers that is closest to the vector corresponding to each keyword; assigning the vector corresponding to each keyword to the cluster center with the closest distance, and increasing the cluster value corresponding to the cluster center with the closest distance by a preset value; sorting the multiple cluster values ​​corresponding to the multiple cluster centers in order from large to small to obtain a sorting result; selecting a preset number of cluster centers with a front ranking position from the sorting result; determining the target vector among the multiple vectors that is closest to the cluster center includes: for each cluster center among the preset number of cluster centers, determining the target vector among the multiple vectors assigned to each cluster center whose distance from each cluster center is less than the preset threshold.

[0007] Optionally, obtaining the vector corresponding to each of the multiple keywords includes: for each keyword, mapping each keyword using each text vector embedding method among a plurality of text vector embedding methods, to obtain a mapping vector corresponding to each text vector embedding method; based on the weight corresponding to each text vector embedding method, performing weighted summation on multiple mapping vectors mapped respectively by the plurality of text vector embedding methods, to obtain a vector corresponding to each keyword.

[0008] Optionally, clustering the multiple vectors corresponding to the multiple keywords to obtain cluster centers includes: setting the number of cluster centers; and clustering the multiple vectors corresponding to the multiple keywords using a K-means clustering algorithm based on the set number of cluster centers to obtain the cluster centers.

[0009] Optionally, extracting multiple keywords from the picture includes: performing text recognition on the picture to obtain multiple recognized text contents; calculating the proportion occupied by each of the multiple text contents in the picture; and determining the text content among the multiple text contents whose corresponding proportion is greater than or equal to a preset proportion threshold as the keyword.

[0010] Optionally, before extracting multiple keywords from the picture, it also includes: performing super-resolution reconstruction on the picture to obtain a reconstructed picture; extracting multiple keywords from the picture includes: extracting the multiple keywords from the reconstructed picture.

[0011] According to a second aspect of an embodiment of the present disclosure, a product feature extraction device is provided, comprising: a picture acquisition module, configured to acquire a picture of a target product; a keyword extraction module, configured to extract multiple keywords from the picture; a vector acquisition module, configured to acquire a vector corresponding to each of the multiple keywords; a clustering module, configured to cluster the multiple vectors corresponding to the multiple keywords to obtain a cluster center; a vector determination module, configured to determine a target vector among the multiple vectors that is closest to the cluster center; and a key feature determination module, configured to use a target keyword corresponding to the target vector as a key feature of the target product.

[0012] Optionally, there are multiple cluster centers, and the product feature extraction device also includes: a cluster center determination module, configured to determine, for each keyword, a cluster center among multiple cluster centers that is closest to the vector corresponding to each keyword; a division module, configured to assign the vector corresponding to each keyword to the cluster center with the closest distance, and increase the cluster value corresponding to the cluster center with the closest distance by a preset value; a sorting module, configured to sort the multiple cluster values ​​corresponding to the multiple cluster centers in order from large to small to obtain a sorting result; a selection module, configured to select a preset number of cluster centers with a front ranking position from the sorting result; the vector determination module is configured to: for each cluster center among the preset number of cluster centers, determine the target vector among the multiple vectors assigned to each cluster center whose distance from each cluster center is less than the preset threshold.

[0013] Optionally, the vector acquisition module is configured to: for each keyword, map each keyword using each of the multiple text vector embedding methods to obtain a mapping vector corresponding to each text vector embedding method; and based on the weight corresponding to each text vector embedding method, perform weighted summation on multiple mapping vectors mapped out by the multiple text vector embedding methods to obtain a vector corresponding to each keyword.

[0014] Optionally, the clustering module is configured to: set the number of cluster centers; and based on the set number of cluster centers, cluster the multiple vectors corresponding to the multiple keywords using a K-means clustering algorithm to obtain the cluster centers.

[0015] Optionally, the keyword extraction module is configured to: perform text recognition on the image to obtain multiple identified text contents; calculate the proportion occupied by each of the multiple text contents in the image; and determine the text content among the multiple text contents whose corresponding proportion is greater than or equal to a preset proportion threshold as the keyword.

[0016] Optionally, the product feature extraction device also includes a super-resolution reconstruction module, which is configured to: perform super-resolution reconstruction on the image to obtain a reconstructed image; and the keyword extraction module is configured to: extract the multiple keywords from the reconstructed image.

[0017] According to a third aspect of an embodiment of the present disclosure, there is provided an electronic device, comprising: a processor; and a memory for storing instructions executable by the processor; wherein the processor is configured to execute the instructions to implement the product feature extraction method according to the present disclosure.

[0018] According to a fourth aspect of an embodiment of the present disclosure, a computer-readable storage medium is provided. When instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the product feature extraction method according to the present disclosure.

[0019] According to a fifth aspect of an embodiment of the present disclosure, there is provided a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the product feature extraction method according to the present disclosure is implemented.

[0020] The technical solution provided by the embodiments of the present disclosure brings at least the following beneficial effects:

[0021] In the present disclosure, an automated method can be used to extract the key features of a product from a picture, which can greatly improve the extraction efficiency compared to the manual method of extracting key features. In addition, since the present disclosure can extract key features from a large number of pictures, compared to the manual method of extracting key features, the amount of data considered by the present disclosure is larger, the data range is wider, and the data content is more comprehensive, so it can ensure that the key features that best reflect the characteristics of the product itself are extracted more accurately. In addition, since the present disclosure extracts key features through automated means, it can avoid the subjective bias caused by human intervention, thereby ensuring the objectivity and accuracy of the extracted key features.

[0022] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] The drawings herein are incorporated into and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the description are used to explain the principles of the present disclosure, and do not constitute improper limitations on the present disclosure.

[0024] Figure 1 is a flowchart illustrating a product feature extraction method according to an exemplary embodiment of the present disclosure;

[0025] Figure 2 is a block diagram showing a product feature extraction device according to an exemplary embodiment of the present disclosure;

[0026] Figure 3 is a block diagram illustrating an electronic device according to an exemplary embodiment of the present disclosure. DETAILED DESCRIPTION

[0027] In order to enable ordinary persons in the art to better understand the technical solutions of the present disclosure, the technical solutions in the embodiments of the present disclosure will be clearly and completely described below in conjunction with the accompanying drawings.

[0028] It should be noted that the terms "first", "second", etc. in the specification and claims of the present disclosure and the above-mentioned drawings are used to distinguish similar objects, and are not necessarily used to describe a specific order or sequence. It should be understood that the data used in this way can be interchanged where appropriate, so that the embodiments of the present disclosure described herein can be implemented in an order other than those illustrated or described herein. The implementation methods described in the following examples do not represent all implementation methods consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the attached claims.

[0029] It should be noted that the phrase "at least one of the items" in the present disclosure includes three types of parallel situations: "any one of the items", "a combination of any number of the items", and "all of the items". For example, "including at least one of A and B" includes the following three parallel situations: (1) including A; (2) including B; (3) including A and B. Another example is "executing at least one of step 1 and step 2" which means the following three parallel situations: (1) executing step 1; (2) executing step 2; (3) executing step 1 and step 2.

[0030] In the related technologies, the core features of products, i.e., marketing hot spots, are mainly extracted manually. However, the information such as marketing pictures, marketing videos, and marketing text descriptions on the Internet is too complex. If the core features of products are extracted manually, the extraction workload will be too large, which will lead to low extraction efficiency of the core features of products. In addition, extracting the core features of products manually will introduce more subjective factors, making it difficult to objectively and accurately extract the core features of products.

[0031] In order to solve the above-mentioned problems existing in the related art, the product feature extraction method, device, electronic device, storage medium and computer program product provided by the present disclosure can use automated means to extract the key features of the product from the picture, which can greatly improve the extraction efficiency compared to the manual extraction of key features. In addition, since the present disclosure can extract key features from a large number of pictures, compared to the manual extraction of key features, the amount of data considered by the present disclosure is larger, the data range is wider, and the data content is more comprehensive, so it can ensure that the key features that best reflect the characteristics of the product itself are extracted more accurately. In addition, since the present disclosure extracts key features through automated means, it can avoid the subjective deviation caused by human intervention, and thus can ensure the objectivity and accuracy of the extracted key features.

[0032] Figure 1 is a flowchart illustrating a product feature extraction method according to an exemplary embodiment of the present disclosure.

[0033] Reference Figure 1 In step 101, pictures of the target product may be obtained. Specifically, the online marketing network data source of the product may be determined first. Then, the picture data of the target product on the online marketing network data source may be crawled through a network collection method.

[0034] It should be noted that the data content covered by the aforementioned "online marketing network data source" is closely related to the product, and may include but is not limited to: online e-commerce, social media platforms, product forums, etc. The aforementioned "network collection method" may include but is not limited to: web crawler technology, API calls, data exchange protocols, etc. Exemplarily, when using web crawler technology, promotional pictures of target products can be automatically collected according to preset search rules and screening conditions. Exemplarily, the "product" in this disclosure may be but is not limited to: smart electronic products, etc.

[0035] In step 102, multiple keywords may be extracted from the image. Specifically, a natural scene text recognition method and a post-processing method may be used to extract a keyword group in the image that is closely related to the product marketing selling point.

[0036] The above-mentioned "natural scene text recognition method" can be to recognize the text in the picture through the text recognition deep learning model, and then extract the initial text content. The above-mentioned "post-processing method" can retain the text with a larger proportion as the marketing selling point keyword group of the product based on the proportion of the text characters in the picture.

[0037] According to an exemplary embodiment of the present disclosure, text recognition can be performed on a picture to obtain multiple recognized text contents. Then, the proportion of each of the multiple text contents in the picture can be calculated. Next, the text contents whose corresponding proportions among the multiple recognized text contents are greater than or equal to a preset proportion threshold can be determined as keywords of the product.

[0038] It should be noted that when introducing and promoting a product, in order to facilitate users to quickly capture the core points of the product, the core characteristics, advantages and other information of the product are generally highlighted. For example, the text description reflecting the core characteristics of the product can be highlighted and enlarged, etc. Therefore, the present disclosure can filter out the text content that occupies a large proportion from the product picture, and then use the text content that occupies a large proportion as the keyword of the product, which can ensure that the determined keywords are keywords that can comprehensively and accurately summarize the core characteristics of the product, and then ensure the accuracy and objectivity of the subsequent acquisition of the key characteristics of the product.

[0039] In addition, when highlighting the core characteristics of the product in the promotional image, in addition to the aforementioned method of enlarging the text content, other methods can also be used. For example, the text content related to the core characteristics of the product can be bolded, tilted, added with special colors, etc., to make the display of the corresponding content more prominent and eye-catching. Correspondingly, when extracting the keywords of the product using the product feature extraction method disclosed in the present invention, the text content in the promotional image that has been bolded, tilted, or added with special colors can be extracted as the keywords of the product.

[0040] According to an exemplary embodiment of the present disclosure, before extracting multiple keywords from a picture, super-resolution reconstruction can be performed on the picture, that is, the image super-resolution method can be used to enhance the marketing picture of the product, and then a reconstructed picture, that is, an enhanced picture, can be obtained. Exemplarily, a super-resolution deep learning model can be used to perform super-resolution reconstruction on the marketing picture of the product. Next, multiple keywords can be extracted from the reconstructed picture, that is, from the picture enhanced by super-resolution.

[0041] In this way, by performing super-resolution reconstruction on the product's marketing images, the resolution and clarity of the images can be improved, thereby ensuring that the product's keywords can be quickly and accurately extracted from the product's promotional images, avoiding the phenomenon of being unable to extract keywords or the extracted keywords being erroneous due to unclear images.

[0042] In step 103, a vector corresponding to each keyword in the plurality of keywords may be obtained. Specifically, a text vector embedding method of multi-representation fusion may be established to convert the extracted product keywords into a vector space.

[0043] According to an exemplary embodiment of the present disclosure, for each keyword, each text vector embedding method in a plurality of text vector embedding methods can be used to map each keyword, and a mapping vector corresponding to each text vector embedding method can be obtained. Exemplarily, the text vector embedding methods that can be used may include but are not limited to: Word2Vec, Bert, etc.

[0044] Then, based on the weight corresponding to each text vector embedding method, the weighted sum of multiple mapping vectors mapped by multiple text vector embedding methods can be performed to obtain the vector corresponding to each keyword. That is, the multiple text representation vectors obtained by multiple text vector embedding methods can be weighted fused to map the product's marketing selling point keyword group to a high-dimensional vector space.

[0045] It should be noted that the user can set the fusion weight corresponding to each text vector embedding method based on the actual situation. For example, suppose that a total of three text vector embedding methods are used, and the dimension of the vector mapped by each text vector embedding method is 1000, that is, the number of elements contained in the vector mapped by each text vector embedding method is 1000. At this time, the mapping vector obtained by the first text vector embedding method can be: (a1, a2, ..., a 1000 ); The mapping vector obtained by the second text vector embedding method can be: (b1, b2, ..., b 1000 ); The mapping vector obtained by the third text vector embedding method can be: (c1, c2, ..., c 1000 ).

[0046] In addition, users can also set the fusion weight corresponding to each of the above three text vector embedding methods according to actual needs. For example, the fusion weight corresponding to the first text vector embedding method can be set to 0.8, the fusion weight corresponding to the second text vector embedding method can be set to 0.1, the fusion weight corresponding to the third text vector embedding method can be set to 0.1, and so on. At this time, the vector corresponding to the keyword can be calculated in the following way:

[0047] The vector corresponding to the keyword = (a1, a2, ..., a 1000 )×0.8+(b1,b2,…,b 1000 )×0.1+(c1,c2,…,c 1000 )×0.1.

[0048] Thus, in the present disclosure, after determining the text vector embedding method and the vector fusion weight, the mapping vectors mapped by different text vector embedding methods can be weighted and summed to map the product's marketing selling point keyword group to a high-dimensional vector space. Since the present disclosure uses a variety of text vector embedding methods in combination, it can absorb the advantages of different text vector embedding methods. Moreover, since the fusion weight corresponding to each text vector embedding method can also be set based on actual needs, that is, since the size of the role played by each text vector embedding method in the process of obtaining the keyword vector can also be determined based on actual needs, it can be ensured that the advantages of each text vector embedding method are fully utilized and the disadvantages of each text vector embedding method are avoided as much as possible, thereby ensuring that the vector corresponding to the keyword finally obtained can accurately, objectively and comprehensively characterize the core characteristics of the product.

[0049] In step 104, the vectors corresponding to the above-mentioned multiple keywords may be clustered to obtain a cluster center. That is, the vectors corresponding to the multiple selling point phrases of the product may be clustered to obtain a cluster center.

[0050] According to an exemplary embodiment of the present disclosure, the number of cluster centers N can be set. Exemplarily, the number of cluster centers N input by the user can be received, and N can be, but is not limited to, 3. Then, based on the number of cluster centers set, a K-means clustering algorithm can be used to cluster multiple vectors corresponding to multiple keywords to obtain cluster centers. Exemplarily, 3 cluster centers can be obtained at this time, and each cluster center can be a vector with a dimension of 1000.

[0051] In step 105, a target vector that is closest to the cluster center among the multiple vectors may be determined. Exemplarily, the "distance" here may be, but is not limited to, Euclidean distance.

[0052] According to an exemplary embodiment of the present disclosure, the number of the above-mentioned cluster centers can be multiple, and for each keyword corresponding vector, the cluster center with the closest distance to the vector corresponding to each keyword among the multiple cluster centers can be determined. Exemplarily, as mentioned above, the number of cluster centers N can be 3, assuming that the three cluster centers are cluster center A, cluster center B, and cluster center C. And, assuming that there are 30 vectors corresponding to keywords in total. At this time, for each keyword corresponding vector, the cluster center with the closest distance to the vector corresponding to the keyword among the three cluster centers can be determined.

[0053] Then, the vector corresponding to each keyword can be classified under the cluster center that is closest to it, and the cluster value corresponding to the cluster center that is closest to it can be increased by a preset value. That is, the vectors corresponding to the product selling point phrases can be quantified based on multiple cluster centers, and the number of cluster centers mapped to each cluster center can be counted. In other words, for the vector corresponding to each keyword, the Euclidean distance between it and each of the above three cluster centers can be calculated respectively, and then the cluster center with the smallest Euclidean distance can be taken as the quantization result, and the count value of the cluster center can be increased by 1. Assume that 20 of the 30 vectors corresponding to the above 30 keywords are classified under cluster center A, 8 vectors are classified under cluster center B, and 2 vectors are classified under cluster center C.

[0054] Next, the multiple cluster values ​​corresponding to the multiple cluster centers can be sorted in order from large to small to obtain a sorting result. For example, since 20 vectors are divided under cluster center A, the cluster value corresponding to cluster center A can be 20; since 8 vectors are divided under cluster center B, the cluster value corresponding to cluster center B can be 8; since 2 vectors are divided under cluster center C, the cluster value corresponding to cluster center C can be 2. At this time, since 20>8>2, the sorting result obtained by sorting the 3 cluster values ​​corresponding to the above 3 cluster centers in order from large to small is: A—B—C.

[0055] Then, a preset number of cluster centers with a high ranking position can be selected from the above sorting results. Exemplarily, the "preset number" can be 2, in which case two cluster centers with a high ranking position can be selected from the above sorting results: cluster center A and cluster center B.

[0056] Next, for each of the preset number of cluster centers, a target vector that is closest to each cluster center among the multiple vectors classified under each cluster center can be determined. Exemplarily, for cluster center A among the two cluster centers selected above, a target vector that is closest to cluster center A in terms of Euclidean distance among the 20 vectors classified under cluster center A can be determined; for cluster center B among the two cluster centers selected above, a target vector that is closest to cluster center B in terms of Euclidean distance among the 8 vectors classified under cluster center B can be determined.

[0057] It should be noted that the larger the cluster value corresponding to the cluster center, the more vectors are classified under the corresponding cluster center, and further, it can be explained that the corresponding cluster center can more comprehensively represent most of the vectors, that is, the corresponding cluster center is likely to be the most appropriate and comprehensive cluster center for summarizing the product characteristics; and if the cluster value corresponding to the cluster center is small, it means that the vectors classified under the corresponding cluster center are small, and further, it can be explained that the corresponding cluster center can only represent a very small part of the vectors, that is, the corresponding cluster center may have a large deviation in summarizing the product characteristics. Therefore, in the present disclosure, only the cluster centers with larger corresponding cluster values ​​that are ranked in the front positions can be retained, and further, only the keywords corresponding to the vectors closest to the cluster centers ranked in the front positions can be used as the core features of the product, which can ensure that the key features finally obtained are the most comprehensive and accurate features that summarize the product characteristics, and further, it can ensure that the user's attention is attracted to the greatest extent to improve the product promotion effect.

[0058] In step 106, the target keyword corresponding to the target vector may be used as a key feature of the target product, that is, the target keyword corresponding to the target vector may be used as a core feature of the target product.

[0059] It should be noted that the product feature extraction method provided by the present disclosure can utilize the product's marketing image data on the Internet to explore the marketing hot spots of the product, which can be applied to advertising in different scenarios as well as to the development and design of similar products.

[0060] Figure 2 is a block diagram illustrating a product feature extraction apparatus according to an exemplary embodiment of the present disclosure.

[0061] Reference Figure 2 The product feature extraction device 200 may include a picture acquisition module 201, a keyword extraction module 202, a vector acquisition module 203, a clustering module 204, a vector determination module 205 and a key feature determination module 206.

[0062] The picture acquisition module 201 can acquire pictures of the target product. Specifically, the online marketing network data source of the product can be determined first. Then, the picture data of the target product on the online marketing network data source can be crawled through a network acquisition method.

[0063] It should be noted that the data content covered by the aforementioned "online marketing network data source" is closely related to the product, and may include but is not limited to: online e-commerce, social media platforms, product forums, etc. The aforementioned "network collection methods" may include but are not limited to: web crawler technology, API calls, data exchange protocols, etc. For example, when using web crawler technology, promotional pictures of target products can be automatically collected according to preset search rules and screening conditions.

[0064] The keyword extraction module 202 can extract multiple keywords from the image. Specifically, a natural scene text recognition method and a post-processing method can be used to extract a keyword group in the image that is closely related to the product marketing selling point.

[0065] The above-mentioned "natural scene text recognition method" can be to recognize the text in the picture through the text recognition deep learning model, and then extract the initial text content. The above-mentioned "post-processing method" can retain the text with a larger proportion as the marketing selling point keyword group of the product based on the proportion of the text characters in the picture.

[0066] According to an exemplary embodiment of the present disclosure, the keyword extraction module 202 can perform text recognition on the image to obtain multiple recognized text contents. Then, the proportion of each of the multiple text contents in the image can be calculated. Next, the text content whose corresponding proportion among the multiple recognized text contents is greater than or equal to a preset proportion threshold can be determined as a keyword for the product.

[0067] It should be noted that when introducing and promoting a product, in order to facilitate users to quickly capture the core points of the product, the core characteristics, advantages and other information of the product are generally highlighted. For example, the text description reflecting the core characteristics of the product can be highlighted and enlarged, etc. Therefore, the present disclosure can filter out the text content that occupies a large proportion from the product picture, and then use the text content that occupies a large proportion as the keyword of the product, which can ensure that the determined keywords are keywords that can comprehensively and accurately summarize the core characteristics of the product, and then ensure the accuracy and objectivity of the subsequent acquisition of the key characteristics of the product.

[0068] In addition, when highlighting the core characteristics of the product in the promotional image, in addition to the aforementioned method of enlarging the text content, other methods can also be used. For example, the text content related to the core characteristics of the product can be bolded, tilted, added with special colors, etc., to make the display of the corresponding content more prominent and eye-catching. Correspondingly, when extracting the keywords of the product using the product feature extraction method disclosed in the present invention, the text content in the promotional image that has been bolded, tilted, or added with special colors can be extracted as the keywords of the product.

[0069] According to an exemplary embodiment of the present disclosure, the product feature extraction device 200 may further include a super-resolution reconstruction module.

[0070] Before extracting multiple keywords from the image, the super-resolution reconstruction module can also perform super-resolution reconstruction on the image, that is, use the image super-resolution method to enhance the marketing image of the product, and then obtain a reconstructed image, that is, an enhanced image. Exemplarily, a super-resolution deep learning model can be used to perform super-resolution reconstruction on the marketing image of the product. Next, the keyword extraction module 202 can extract multiple keywords from the reconstructed image, that is, from the super-resolution enhanced image.

[0071] In this way, by performing super-resolution reconstruction on the product's marketing images, the resolution and clarity of the images can be improved, thereby ensuring that the product's keywords can be quickly and accurately extracted from the product's promotional images, avoiding the phenomenon of being unable to extract keywords or the extracted keywords being erroneous due to unclear images.

[0072] The vector acquisition module 203 can acquire the vector corresponding to each keyword in the multiple keywords. Specifically, a text vector embedding method of multi-representation fusion can be established to convert the extracted product keywords into a vector space.

[0073] According to an exemplary embodiment of the present disclosure, for each keyword, the vector acquisition module 203 can respectively use each text vector embedding method in a plurality of text vector embedding methods to map each keyword, and obtain a mapping vector corresponding to each text vector embedding method. Exemplarily, the text vector embedding methods that can be used may include but are not limited to: Word2Vec, Bert, etc.

[0074] Then, the vector acquisition module 203 can perform weighted summation on multiple mapping vectors mapped by multiple text vector embedding methods based on the weight corresponding to each text vector embedding method to obtain the vector corresponding to each keyword. That is, weighted fusion can be performed on multiple text representation vectors obtained by multiple text vector embedding methods to map the marketing selling point keyword group of the product to a high-dimensional vector space.

[0075] Thus, in the present disclosure, after determining the text vector embedding method and the vector fusion weight, the mapping vectors mapped by different text vector embedding methods can be weighted and summed to map the product's marketing selling point keyword group to a high-dimensional vector space. Since the present disclosure uses a variety of text vector embedding methods in combination, it can absorb the advantages of different text vector embedding methods. Moreover, since the fusion weight corresponding to each text vector embedding method can also be set based on actual needs, that is, since the size of the role played by each text vector embedding method in the process of obtaining the keyword vector can also be determined based on actual needs, it can be ensured that the advantages of each text vector embedding method are fully utilized and the disadvantages of each text vector embedding method are avoided as much as possible, thereby ensuring that the vector corresponding to the keyword finally obtained can accurately, objectively and comprehensively characterize the core characteristics of the product.

[0076] The clustering module 204 may cluster the multiple vectors corresponding to the multiple keywords to obtain a cluster center, that is, may cluster the vectors corresponding to the multiple selling point phrases of the product to obtain a cluster center.

[0077] According to an exemplary embodiment of the present disclosure, the clustering module 204 may set the number of cluster centers N. Exemplarily, the number of cluster centers N input by the user may be received, and N may be, but is not limited to, 3. Then, the clustering module 204 may cluster multiple vectors corresponding to multiple keywords using a K-means clustering algorithm based on the number of cluster centers set to obtain cluster centers. Exemplarily, three cluster centers may be obtained at this time, and each cluster center may be a vector with a dimension of 1000.

[0078] The vector determination module 205 may determine a target vector that is closest to the cluster center among the multiple vectors. For example, the “distance” here may be, but is not limited to, Euclidean distance.

[0079] According to an exemplary embodiment of the present disclosure, the above-mentioned product feature extraction device 200 may further include a cluster center determination module, a division module, a sorting module and a selection module.

[0080] There may be multiple cluster centers. For each vector corresponding to a keyword, the cluster center determination module may determine a cluster center that is closest to the vector corresponding to each keyword among the multiple cluster centers.

[0081] Then, the partitioning module can classify the vector corresponding to each keyword into the cluster center closest to it, and can increase the cluster value corresponding to the cluster center closest to it by a preset value. That is, the vector corresponding to the product selling point phrase can be quantified based on multiple cluster centers, and the number of each cluster center mapped can be counted.

[0082] Next, the sorting module may sort the multiple clustering values ​​corresponding to the multiple clustering centers in order from large to small to obtain a sorting result.

[0083] Then, the selection module may select a preset number of cluster centers with top ranking positions from the above sorting results.

[0084] Next, for each cluster center among the preset number of cluster centers, the vector determination module 205 may determine a target vector that is closest to each cluster center among the multiple vectors classified under each cluster center.

[0085] It should be noted that the larger the cluster value corresponding to the cluster center, the more vectors are classified under the corresponding cluster center, and further, it can be explained that the corresponding cluster center can more comprehensively represent most of the vectors, that is, the corresponding cluster center is likely to be the most appropriate and comprehensive cluster center for summarizing the product characteristics; and if the cluster value corresponding to the cluster center is small, it means that the vectors classified under the corresponding cluster center are small, and further, it can be explained that the corresponding cluster center can only represent a very small part of the vectors, that is, the corresponding cluster center may have a large deviation in summarizing the product characteristics. Therefore, in the present disclosure, only the cluster centers with larger corresponding cluster values ​​that are ranked in the front positions can be retained, and further, only the keywords corresponding to the vectors closest to the cluster centers ranked in the front positions can be used as the core features of the product, which can ensure that the key features finally obtained are the most comprehensive and accurate features that summarize the product characteristics, and further, it can ensure that the user's attention is attracted to the greatest extent to improve the product promotion effect.

[0086] The key feature determination module 206 may use the target keyword corresponding to the target vector as the key feature of the target product, that is, may use the target keyword corresponding to the target vector as the core feature of the target product.

[0087] It should be noted that the product feature extraction method provided by the present disclosure can utilize the product's marketing image data on the Internet to explore the marketing hot spots of the product, which can be applied to advertising in different scenarios as well as to the development and design of similar products.

[0088] Figure 3 is a block diagram illustrating an electronic device 300 according to an exemplary embodiment of the present disclosure.

[0089] Reference Figure 3 The electronic device 300 includes at least one memory 301 and at least one processor 302. The at least one memory 301 stores instructions. When the instructions are executed by the at least one processor 302, the product feature extraction method according to the exemplary embodiment of the present disclosure is executed.

[0090] As an example, the electronic device 300 may be a PC, a tablet device, a personal digital assistant, a smart phone, or other device capable of executing the above instructions. Here, the electronic device 300 is not necessarily a single electronic device, but may also be any device or circuit capable of executing the above instructions (or instruction sets) individually or in combination. The electronic device 300 may also be part of an integrated control system or a system manager, or may be configured as a portable electronic device interconnected with a local or remote (e.g., via wireless transmission) interface.

[0091] In the electronic device 300, the processor 302 may include a central processing unit (CPU), a graphics processing unit (GPU), a programmable logic device, a dedicated processor system, a microcontroller or a microprocessor. As an example and not limitation, the processor may also include an analog processor, a digital processor, a microprocessor, a multi-core processor, a processor array, a network processor, etc.

[0092] The processor 302 may execute instructions or codes stored in the memory 301, wherein the memory 301 may also store data. Instructions and data may also be sent and received over a network via a network interface device, wherein the network interface device may employ any known transmission protocol.

[0093] The memory 301 may be integrated with the processor 302, for example, by placing RAM or flash memory within an integrated circuit microprocessor or the like. In addition, the memory 301 may include a separate device, such as an external disk drive, a storage array, or any other storage device that can be used by a database system. The memory 301 and the processor 302 may be operatively coupled, or may communicate with each other, such as through an I / O port, a network connection, etc., so that the processor 302 can read files stored in the memory.

[0094] In addition, the electronic device 300 may also include a video display (such as a liquid crystal display) and a user interaction interface (such as a keyboard, a mouse, a touch input device, etc.) All components of the electronic device 300 may be connected to each other via a bus and / or a network.

[0095] According to an exemplary embodiment of the present disclosure, a computer-readable storage medium may also be provided, and when the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to perform the above-mentioned product feature extraction method. Examples of computer-readable storage media here include: read-only memory (ROM), random access programmable read-only memory (PROM), electrically erasable programmable read-only memory (EEPROM), random access memory (RAM), dynamic random access memory (DRAM), static random access memory (SRAM), flash memory, non-volatile memory, CD-ROM, CD-R, CD+R, CD-RW, CD+RW, DVD-ROM, DVD-R, DVD+R, DVD-RW, DVD+RW, DVD-RAM, BD-ROM, BD-R, BD-R LTH, BD-RE, Blu-ray or optical disk storage, hard disk drive (HDD), solid state drive (SSD), card storage (such as, multimedia card, secure digital (SD) card or extreme digital (XD) card), magnetic tape, floppy disk, magneto-optical data storage device, optical data storage device, hard disk, solid state disk and any other device, any other device is configured to store computer programs and any associated data, data files and data structures in a non-transitory manner and provide the computer programs and any associated data, data files and data structures to a processor or computer so that the processor or computer can execute the computer program. The computer program in the above-mentioned computer-readable storage medium can be run in an environment deployed in a computer device such as a client, a host, an agent device, a server, etc. In addition, in one example, the computer program and any associated data, data files and data structures are distributed on a networked computer system, so that the computer program and any associated data, data files and data structures are stored, accessed and executed in a distributed manner by one or more processors or computers.

[0096] According to an exemplary embodiment of the present disclosure, a computer program product may also be provided, including a computer program, and when the computer program is executed by a processor, the product feature extraction method according to the present disclosure is implemented.

[0097] According to the product feature extraction method, device, electronic device, storage medium and computer program product disclosed in the present invention, the key features of the product can be extracted from the picture by automated means, which can greatly improve the extraction efficiency compared to the manual extraction of key features. In addition, since the present invention can extract key features from a large number of pictures, compared to the manual extraction of key features, the amount of data considered by the present invention is larger, the data range is wider, and the data content is more comprehensive, so it can ensure that the key features that best reflect the characteristics of the product itself are extracted more accurately. In addition, since the present invention extracts key features by automated means, it can avoid subjective deviations caused by human intervention, thereby ensuring the objectivity and accuracy of the extracted key features.

[0098] According to an exemplary embodiment of the present disclosure, the present disclosure can filter out text content that occupies a large proportion from the product image, and then use the text content that occupies a large proportion as the keyword of the product, which can ensure that the determined keywords are keywords that can comprehensively and accurately summarize the core characteristics of the product, and then ensure the accuracy and objectivity of the subsequent acquisition of the key features of the product.

[0099] According to an exemplary embodiment of the present disclosure, by performing super-resolution reconstruction on a marketing image of a product, the resolution and clarity of the image can be improved, thereby ensuring that the keywords of the product can be quickly and accurately extracted from the promotional image of the product, thereby avoiding the phenomenon that keywords cannot be extracted or the extracted keywords are erroneous due to unclear images.

[0100] According to an exemplary embodiment of the present disclosure, after determining the text vector embedding method and the vector fusion weight, the mapping vectors mapped by different text vector embedding methods can be weighted and summed to map the product's marketing selling point keyword group to a high-dimensional vector space. Since the present disclosure uses a variety of text vector embedding methods in combination, it can absorb the advantages of different text vector embedding methods. In addition, since the fusion weight corresponding to each text vector embedding method can also be set based on actual needs, that is, since the size of the role played by each text vector embedding method in the process of obtaining the keyword vector can also be determined based on actual needs, it can ensure that the advantages of each text vector embedding method are fully utilized and the disadvantages of each text vector embedding method are avoided as much as possible, thereby ensuring that the vector corresponding to the keyword finally obtained can accurately, objectively and comprehensively characterize the core characteristics of the product.

[0101] According to an exemplary embodiment of the present disclosure, the larger the cluster value corresponding to the cluster center, the more vectors are classified under the corresponding cluster center, and further, it can be explained that the corresponding cluster center can more comprehensively represent most of the vectors, that is, the corresponding cluster center is likely to be the cluster center that summarizes the product characteristics most appropriately and comprehensively; and if the cluster value corresponding to the cluster center is small, it means that the vectors classified under the corresponding cluster center are small, and further, it can be explained that the corresponding cluster center can only represent a very small part of the vectors, that is, the corresponding cluster center may have a large deviation in summarizing the characteristics of the product. Therefore, in the present disclosure, only the cluster centers with larger corresponding cluster values ​​that are ranked in the front positions can be retained, and further, only the keywords corresponding to the vectors closest to the cluster centers ranked in the front positions can be used as the core features of the product, which can ensure that the key features finally obtained are the most comprehensive and accurate features that summarize the characteristics of the product, and further, it can ensure that the user's attention is attracted to the greatest extent to improve the promotion effect of the product.

[0102] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are to be considered exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

[0103] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A product feature extraction method, characterized in that: include: Get pictures of the target product; Extracting multiple keywords from the image; Obtaining a vector corresponding to each keyword in the multiple keywords; Clustering multiple vectors corresponding to the multiple keywords to obtain cluster centers; Determine a target vector which is closest to the cluster center among the multiple vectors; The target keyword corresponding to the target vector is used as the key feature of the target product.

2. The product feature extraction method according to claim 1, characterized in that: There are multiple cluster centers, and before determining the target vector closest to the cluster center among the multiple vectors, the method further includes: For each vector corresponding to the keyword, determine a cluster center among multiple cluster centers that has the shortest distance to the vector corresponding to the keyword; The vector corresponding to each keyword is classified under the cluster center with the closest distance, and the cluster value corresponding to the cluster center with the closest distance is increased by a preset value; Sorting the multiple clustering values ​​corresponding to the multiple clustering centers in order from large to small to obtain a sorting result; Selecting a preset number of cluster centers with top ranking positions from the sorting results; The determining of the target vector that is closest to the cluster center among the multiple vectors includes: For each cluster center of the preset number of cluster centers, determine the target vector that is closest to each cluster center among the multiple vectors classified under each cluster center.

3. The product feature extraction method according to claim 1, characterized in that: The obtaining a vector corresponding to each keyword in the plurality of keywords includes: For each keyword, use each text vector embedding method among multiple text vector embedding methods to map each keyword, and obtain a mapping vector corresponding to each text vector embedding method; Based on the weight corresponding to each text vector embedding method, a weighted sum is performed on a plurality of mapping vectors respectively mapped out by the plurality of text vector embedding methods to obtain a vector corresponding to each keyword.

4. The product feature extraction method according to claim 1, characterized in that: The clustering of the multiple vectors corresponding to the multiple keywords to obtain cluster centers includes: Setting the number of cluster centers; Based on the set number of cluster centers, the multiple vectors corresponding to the multiple keywords are clustered using a K-means clustering algorithm to obtain the cluster centers.

5. The product feature extraction method according to claim 1, characterized in that: The extracting a plurality of keywords from the picture includes: Performing text recognition on the image to obtain multiple recognized text contents; Calculating the proportion of each text content in the plurality of text contents in the image; The text contents whose corresponding proportions among the multiple text contents are greater than or equal to a preset proportion threshold are determined as the keywords.

6. The product feature extraction method according to claim 1, characterized in that: Before extracting a plurality of keywords from the image, the method further includes: Performing super-resolution reconstruction on the image to obtain a reconstructed image; The extracting a plurality of keywords from the picture includes: The multiple keywords are extracted from the reconstructed image.

7. A product feature extraction device, characterized in that: include: A picture acquisition module, configured to acquire pictures of the target product; A keyword extraction module is configured to extract a plurality of keywords from the image; A vector acquisition module, configured to acquire a vector corresponding to each keyword in the plurality of keywords; A clustering module, configured to cluster the multiple vectors corresponding to the multiple keywords to obtain cluster centers; A vector determination module, configured to determine a target vector that is closest to the cluster center among the multiple vectors; The key feature determination module is configured to use the target keyword corresponding to the target vector as the key feature of the target product.

8. An electronic device, characterized in that: include: processor; a memory for storing instructions executable by the processor; The processor is configured to execute the instructions to implement the product feature extraction method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that: When the instructions in the computer-readable storage medium are executed by a processor of an electronic device, the electronic device is enabled to execute the product feature extraction method as described in any one of claims 1 to 6.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the product feature extraction method according to any one of claims 1 to 6 is implemented.

Images