Image-driven cross-platform e-commerce commodity price comparison processing method and system

Abstract

The invention discloses an image-driven cross-platform e-commerce commodity price comparison processing method and system, and relates to the technical field of price comparison processing. Comprising the following steps: performing structure slice partitioning processing on each commodity image, constructing each region as a coding node, and generating a structure matrix of the commodity image; performing corresponding matching on the structure matrixes of the plurality of platform commodity images, and identifying image commodity pairs with comparison potential based on a combination relationship between a regional structure and semantic features; extracting a cross feature group from the historical price sequence and attribute of each platform commodity, carrying out price-attribute-structure combined evaluation, and screening effective price comparison nodes in a layered manner through a cross combination judgment rule; and generating a multi-path price comparison set based on the effective price comparison nodes, and outputting recommendation results in stages according to the target commodity state. The method not only improves the accuracy and adaptability of price comparison processing, but also guarantees the stability and intelligent response capability of price comparison logic under the condition of facing complex e-commerce image data.

Description

Technical Field

[0001] The present invention relates to the technical field of price comparison processing, and in particular to an image-driven cross-platform e-commerce product price comparison processing method and system. Background Art

[0002] With the rapid development of e-commerce platforms, the multi-source heterogeneity of product information is becoming increasingly prominent, and users are increasingly demanding to compare product prices, attributes, and image information across different platforms. Traditional keyword-driven price comparison methods often rely on product titles or descriptions for retrieval and matching. However, in practice, inconsistent product naming conventions, redundant descriptions, or semantic biases can lead to inaccurate price comparison results or serious omissions. To address this, some studies have attempted to introduce image recognition and visual feature extraction to compensate for the shortcomings of pure text information. However, image-driven price comparison technology is still in its early stages of development, and existing methods still have significant limitations in structural modeling, cross-platform image matching, image semantic fusion, and price comparison accuracy. Furthermore, effective mechanisms for modeling structural hierarchical information and for joint reasoning of multidimensional information across images, prices, and attributes have yet to be established. More targeted and robust image price comparison solutions are urgently needed.

[0003] CN109118325A discloses a method for collecting product information and performing real-time price comparison on multiple e-commerce platforms. The main technical features are: allowing users to input product keywords or images, extracting product names based on image recognition, and performing searches on various platforms by name, calculating the cost in combination with user account information, and generating price comparison charts. This method does achieve automated information acquisition and price comparison output, but it has two shortcomings: first, it only uses images as auxiliary information, and the structure, semantics, and regional information of the image itself are not systematically modeled, resulting in insufficient image-driven characteristics for price comparison; second, it fails to achieve corresponding matching of image structures between cross-platform products, and lacks the ability to identify potential matching product pairs from a visual level. Therefore, this technology has low recognition accuracy when processing products with diverse image semantics, homogeneous brands, or ambiguous names.

[0004] CN103412938B provides a product price comparison method based on interactive multi-target extraction from images. By allowing users to select an image region on the client, image features are extracted and matched against a visual feature database constructed in an index file, which then returns similar products and price comparison ranking results. This technology attempts to construct a product image feature index and enable user interactive recognition. Its main shortcomings are: image features are limited to global or shallow visual attributes, lacking partitioning modeling and node encoding of image structures; and its price comparison logic is limited to simple price comparison ranking driven by image similarity. It does not introduce the cross-correlation between product price sequences, attribute information, and image structure, making it difficult to support multi-dimensional price comparison judgments in complex product scenarios. Summary of the Invention

[0005] In view of the problems existing in existing information retrieval and price comparison methods based on product images, the present invention is proposed.

[0006] Therefore, the problem to be solved by the present invention is how to improve the accuracy and adaptability of price comparison processing.

[0007] Furthermore, it ensures the stability and intelligent response capabilities of price comparison logic when faced with complex e-commerce image data.

[0008] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0009] In a first aspect, the present invention provides an image-driven cross-platform e-commerce product price comparison processing method.

[0010] The method includes collecting product images of target products on multiple platforms, performing structural slicing and partitioning processing on each product image, constructing each region as a coding node, and generating a structural matrix of the product image; performing correspondence matching on the structural matrices of product images on multiple platforms, and identifying image-product pairs with comparison potential based on the combined relationship between regional structure and semantic features; based on the image-product pairs, extracting cross-feature groups from the historical price series and attributes of products on each platform, performing price-attribute-structure joint evaluation, and hierarchically screening effective price comparison nodes through cross-combination judgment rules; generating a multi-path price comparison set based on the effective price comparison nodes, and outputting recommendation results in stages according to the status of the target product.

[0011] As a preferred solution of the image-driven cross-platform e-commerce product price comparison processing method described in the present invention, the correspondence matching of the structural matrices of product images on multiple platforms includes: adding image source tags to the structural matrices corresponding to the product images collected from all platforms, classifying the structural matrices according to the source tags, and dividing the structural matrices generated for the same product on different platforms into different subsets to ensure that no pairing of homologous images occurs during combination; the structural matrices in each platform are paired with the structural matrices belonging to the same product on other platforms to generate cross-platform structural matrix pairs, each pair containing structural matrices from two different source platforms; for each pair of structural matrices, two-way structural matching groups are generated by combining them in two directions.

[0012] As a preferred solution of the image-driven cross-platform e-commerce product price comparison processing method described in the present invention, the method of identifying image product pairs with comparison potential includes: extracting a set of structure coding nodes for each group of structure matrices, and constructing an intersection node set and a union node set based on the node position index and label semantics in the structure coding nodes; calculating the ratio of the number of nodes between the intersection node set and the union node set as the original intersection-and-union ratio indicator; calculating the coverage ratio of each node set in the two structure matrices in the entire image structure as a compensation coefficient, and multiplying it with the initial intersection-and-union ratio to form an adjusted intersection-and-union ratio; if the adjusted intersection-and-union ratio is higher than the judgment threshold T s , then the structure matching group is labeled as an image-product pair.

[0013] As a preferred solution of the image-driven cross-platform e-commerce product price comparison processing method described in the present invention, the price-attribute-structure joint evaluation includes: extracting the time evolution structure matrix and corresponding price data of each platform product in the confirmed image product pairs, and constructing a structure-price coupling vector of structural changes and price fluctuations; based on the structure matching group, constructing a structure alignment map and binding the structure-price coupling vector to form a structure-price fusion feature group.

[0014] As a preferred solution of the image-driven cross-platform e-commerce product price comparison processing method described in the present invention, the cross-combination judgment rule includes: constructing a cross-stability factor map, each factor node corresponds to a joint stability index of three dimensions; wherein, the three dimensions include time slice overlap, structural coding semantic consistency and structural alignment density; according to the joint judgment of the three dimensions, each node in the map is assigned the following hierarchical label: if the structural coding similarity corresponding to the factor node is higher than a preset first threshold, the corresponding structural coding node falls in a continuous area in the structural alignment map, and the dynamic time regularization distance between the corresponding commodity price curves within the preset time window is less than a preset second threshold, it is marked as a high-credibility node; if there is fluctuation in a certain dimension indicator but other dimensions can compensate, it is marked as a conditional reference node; otherwise, it is marked as a secondary support node; taking the high-credibility node as the starting point, combined with the connected conditional reference nodes, multiple collaborative judgment paths are constructed.

[0015] As a preferred embodiment of the image-driven cross-platform e-commerce product price comparison processing method described in the present invention, the hierarchical screening of effective price comparison nodes includes: forming a path classification according to the following rules during the construction of multiple collaborative judgment paths: if the structural coding sequence corresponding to any two factor nodes in the path changes before the change in the product price sequence, and the time window overlap between the two exceeds an overlap threshold, the path is marked as a synchronous response channel; if more than 50% of the factor nodes in the path have a label semantic similarity above a first threshold and the price change slope and the structural change slope are consistent within a preset tolerance range, the path is marked as a structure-driven channel; if the structural matching degree between the factor nodes in the path is below a second threshold, but the spatial positions are clustered in adjacent areas in the structure matrix, the path is marked as a regional focus channel; and based on the collaborative path type and node level, three types of effective price comparison nodes are hierarchically output from the cross-stable factor map: high-confidence nodes in the synchronous response channel are used as accurate price comparison references; stable reference nodes in the structure-driven channel are used for trend-based price comparison judgments; and conditional reference nodes in the regional focus channel are used as regional control support nodes.

[0016] As a preferred solution of the image-driven cross-platform e-commerce product price comparison processing method described in the present invention, the method includes: outputting recommendation results in stages based on the status of the target product, dividing the three types of output valid price comparison nodes into node sets according to the collaborative path types to which they belong, and constructing three types of multi-path price comparison sets; based on the change frequency and spatial clustering of the structural coding nodes, determining the latest evolutionary state of the target product in the structural matrix, identifying the structural change trend, and outputting the multi-path price comparison set accordingly according to the structural change trend.

[0017] In a second aspect, the present invention provides an image-driven cross-platform e-commerce product price comparison processing system, which includes: an image acquisition module that collects product images of a target product on multiple platforms, performs structural slicing and partitioning processing on each product image, constructs each region into a coding node, and generates a structural matrix of the product image;

[0018] The matching recognition module is used to perform correspondence matching on the structural matrices of product images on multiple platforms and identify image-product pairs with potential for comparison based on the combined relationship between regional structure and semantic features;

[0019] The feature cross-module is used to extract cross-feature groups from the historical price series and attributes of products on each platform based on image product pairs, conduct a joint evaluation of price, attribute, and structure, and hierarchically screen valid price comparison nodes based on cross-combination judgment rules;

[0020] The recommendation output module is used to generate a multi-path price comparison set based on the valid price comparison nodes and output the recommendation results in stages according to the status of the target product.

[0021] In a third aspect, the present invention provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: when the computer program instructions are executed by the processor, the steps of the image-driven cross-platform e-commerce product price comparison processing method as described in the first aspect of the present invention are implemented.

[0022] In a fourth aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein: when the computer program instructions are executed by a processor, the steps of the image-driven cross-platform e-commerce product price comparison processing method as described in the first aspect of the present invention are implemented.

[0023] The beneficial effects of the present invention are as follows: the present invention significantly improves the accuracy of image comparison between cross-platform commodities through the construction of a structural matrix and a joint modeling mechanism of structural-semantic features. Through node hierarchical division and cross-feature evaluation, it is possible to achieve deep correlation mining between structural changes, price evolution and attribute information, thereby ensuring that the price comparison reference has higher timeliness and consistency; at the same time, relying on multi-path collaborative judgment and node level stratification mechanism, the present invention can dynamically output personalized price comparison result recommendations based on the structural evolution trends of different target commodities. In summary, the present invention not only improves the accuracy and adaptability of price comparison processing, but also ensures the stability and intelligent response capabilities of price comparison logic under the conditions of complex e-commerce image data. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0025] Figure 1 A flowchart of an image-driven cross-platform e-commerce product price comparison processing method;

[0026] Figure 2 This is a structural diagram of an image-driven cross-platform e-commerce product price comparison processing system. DETAILED DESCRIPTION

[0027] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the specific embodiments of the present invention are described in detail below with reference to the accompanying drawings.

[0028] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without violating the connotation of the present invention. Therefore, the present invention is not limited to the specific embodiments disclosed below.

[0029] Secondly, the term "one embodiment" or "embodiment" herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in various places throughout this specification does not necessarily refer to the same embodiment, nor does it refer to a separate or selective embodiment that is mutually exclusive of other embodiments.

[0030] As mentioned in the background technology above, traditional keyword-driven price comparison methods often rely on product titles or descriptions for retrieval and matching. However, in actual applications, due to inconsistent product naming rules, redundant descriptions, or semantic deviations, the price comparison results are inaccurate or seriously missed. To this end, some studies have attempted to introduce image recognition, visual feature extraction, and other methods to make up for the shortcomings of pure text information. However, image-driven price comparison technology is still in its early stages of development, and existing methods still have significant limitations in structural modeling, cross-platform image matching, image semantic fusion, and price comparison accuracy. In addition, there is no effective mechanism for modeling structural hierarchical information and joint reasoning of image-price-attribute multidimensional information. There is an urgent need for more targeted and robust image price comparison processing solutions.

[0031] Figure 1 Flowchart of the image-driven cross-platform e-commerce product price comparison processing method according to an embodiment of the present invention. Figure 1 As shown, the image-driven cross-platform e-commerce product price comparison processing method includes:

[0032] S1: Collect product images of the target product on multiple platforms, perform structural slicing and partitioning processing on each product image, construct each area into a coding node, and generate a structural matrix of the product image.

[0033] Specifically, we first collect product display images of the same target product on multiple mainstream e-commerce platforms through cross-platform data interfaces or web crawling methods, giving priority to obtaining main images, front views, and clear image samples with minimal background interference to ensure that the images have a good structural expression foundation.

[0034] Each product image is then subjected to a structural slicing and partitioning process. This process divides the entire image into a number of fixed-size structural unit regions based on the image pixel size and the distribution of visually salient regions. This partitioning can be done using a regular grid method or by combining edge detection with a region segmentation algorithm. This is not a specific limitation in the present embodiment, but allows for a slicing effect that better reflects the physical configuration.

[0035] After completing the slice partitioning, the image feature descriptor is further extracted for each region. The region is then constructed into a structural coding node by combining its spatial position, texture properties, and edge features. Each structural coding node includes a node position index, label semantics, and feature value summary.

[0036] Ultimately, all structural encoding nodes are combined according to their spatial arrangement in the image to form a structural matrix corresponding to the target product image, which serves as the basic input structure for subsequent comparison and evaluation. This structural matrix not only preserves the spatial structure of the image but also provides encoding support for semantic comparison.

[0037] S2: Perform correspondence matching on the structural matrices of product images from multiple platforms, and identify image-product pairs with potential for comparison based on the combined relationship between regional structure and semantic features.

[0038] S2.1: Based on the structural matrices corresponding to product images on multiple platforms, perform bidirectional pairing and combination of the structural matrices according to the image sources to construct structural matching groups, each of which includes the structural matrices under two platforms.

[0039] Specifically, we add image source tags to the structure matrices corresponding to product images collected from all platforms, classify the structure matrices according to the source tags, and divide the structure matrices generated for the same product on different platforms into different subsets to ensure that the same source images are not paired when combined;

[0040] The structure matrix in each platform is paired with the structure matrices of the same product in other platforms to generate cross-platform structure matrix pairs. Each pair contains two structure matrices from different source platforms.

[0041] For each pair of structure matrices, two combinations are generated in two directions. For example, structure matrix A and structure matrix B will form two combinations: A→B and B→A, which serve as independent structure matching perspectives.

[0042] Conventional methods for cross-platform image comparison typically use global image features, such as SIFT and SURF, for coarse matching. This approach struggles to ensure structural matching accuracy, especially after product images have been re-edited by the platform (for example, by adding borders or occlusions). Traditional methods are prone to misjudgment. Our invention abandons traditional global image feature matching and instead employs a regional combination strategy at the structural matrix level, emphasizing regional correspondence and semantic coupling features, thereby better reflecting the consistency of the product itself.

[0043] S2.2: For each set of structure matrices, extract the set of structure encoding nodes, and construct the intersection node set and union node set based on the position index and label semantics of the structure encoding nodes.

[0044] In the specific operation, for any pair of structure matrices in the structure matching group, a node set is first extracted. Then, based on the spatial position index and label semantics of the nodes, an intersection node set and a union node set are constructed between the two sets. The intersection node set refers to the set of nodes with the same label semantics and appearing in similar locations in a pair of structure matrices, emphasizing the structural consistency of the local area; the union node set is the union of all node sets of the two structure matrices, representing the overall structural range.

[0045] In traditional structure matching methods, relying solely on spatial position (such as IoU or pixel overlap) often ignores semantic consistency. Our invention combines spatial position with label semantics as the judgment criterion, which significantly improves matching accuracy, effectively suppresses false matches caused by visual confusion or image occlusion, and enhances the discriminability of matching node sets.

[0046] S2.3: Calculate the intersection-and-union ratio index and adjust it according to the node coverage compensation coefficient to determine whether the structure matching group meets the preset judgment threshold T s , the structural matching groups that meet the conditions are marked as image-product pairs.

[0047] Specifically, based on the intersection node set and the union node set, the ratio of the number of nodes between the two is calculated as the original intersection-union ratio index. This index can measure the degree of matching between the two structure matrices in the semantic space and the structural area. The higher the value, the more similar the structures are.

[0048] The coverage ratio of each node set in the two structure matrices in the entire image structure is calculated as a compensation coefficient to reflect the spatial coverage balance of the structure matching group. The compensation coefficient is calculated by merging the corresponding areas of all nodes in the image coordinate system and then calculating the ratio of the calculated area to the total area of ​​the image.

[0049] The node coverage compensation coefficient is multiplied by the initial intersection-union ratio to form an adjusted intersection-union ratio, which is used to weaken the misjudgment of local areas with high overlap but overall inconsistency. It should be noted that if only the intersection-union ratio is used, it is easy to misjudge products with highly matched structures in several small areas in the image as the same product, while ignoring the overall framework differences. The present invention can effectively improve the global rationality of matching judgment by adding compensation coefficient constraints.

[0050] According to the preset judgment threshold T s , evaluate the adjusted intersection-over-union ratio of each structure matching group, if it is higher than the judgment threshold T s , then the matching group is considered to be a valid product image pair, that is, the product images displayed on the two platforms are very likely to be different platform versions of the same product, and they are marked as an image product pair.

[0051] S3: Based on image-product pairs, cross-feature groups are extracted from the historical price series and attributes of products on each platform to conduct a joint evaluation of price, attribute, and structure. Valid price comparison nodes are then hierarchically screened using cross-combination judgment rules.

[0052] S3.1: For the confirmed image-product pairs, extract the time evolution structure matrix and corresponding price data of the products on each platform, and construct the structure-price coupling vector of structural changes and price fluctuations.

[0053] First, for each image product, within the determined platform data sample, its historical image version sequence is obtained according to the time dimension, with priority given to collecting product main images with clear structure and stable main perspective to ensure the continuity and accuracy of structural expression. Each historical image is structurally sliced ​​and encoded according to the process in S1 to construct the structural matrix at that point in time;

[0054] After obtaining the structure matrix under the time series, the corresponding structure coding nodes in each matrix (based on one-to-one alignment of position indexes) are processed with node-level feature difference, that is, the changes in the coding values ​​of the structure coding nodes at the same position at different time points are calculated;

[0055] Aggregate these node evolution vectors in the time dimension to form the overall structural evolution trajectory matrix. At the same time, extract the commodity prices corresponding to each time point and construct a one-to-one mapping relationship between the structural coding change sequence and the price change sequence;

[0056] Based on the structural stability trajectory, we identify time slices where prices jump dramatically but the structure does not change significantly, eliminate them, and retain the time slices where the structure and price change synchronously, and construct a set of structure-price coupling vectors.

[0057] S3.2: Based on the structure matching group, construct a structure alignment graph and bind the structure-price coupling vector to form a structure-price fusion feature group.

[0058] In the image product pairs, based on the structure matching group selected in S2.3, the structure encoding node positions of the structure matrices in the two platforms are accurately aligned to construct a structure alignment graph. In the structure alignment graph, the time slices in the structure-price coupling vector in S3.1 are bound to the node pairs in the alignment graph, thereby forming a structure-price fusion feature group, which provides a basis for subsequent price comparison node screening.

[0059] S3.3: Based on the structural alignment graph and time series data, a cross-stability factor map is constructed. In the cross-stability factor map, each factor node corresponds to the joint stability index of the following three dimensions:

[0060] Time slice overlap: indicates the degree of temporal overlap of the evolutionary trajectories of the image product structures on the two platforms;

[0061] Structural encoding semantic consistency: Calculates the semantic consistency of the labels of the encoded nodes in the structure bitmap;

[0062] Structural alignment density: indicates the proportion and distribution concentration of alignment nodes in the entire structural matrix.

[0063] Based on the joint judgment of the three dimensions, each factor node in the graph is assigned the following hierarchical labels:

[0064] If the structural coding similarity corresponding to the node is higher than the preset first threshold, the corresponding structural node falls in a continuous area in the structural alignment graph, and the dynamic time warping distance between the corresponding commodity price curves within the preset time window is less than the preset second threshold, then it is marked as a high-confidence node; if there is fluctuation in a certain dimension indicator but other dimensions can be compensated, it is marked as a conditional reference node; otherwise, it is marked as a secondary support node. Among them, structural coding similarity: refers to the cosine similarity of the coding vectors of the corresponding structural nodes in the commodities of the two platforms; the preset first threshold is the structural semantic similarity judgment threshold set by experience or training samples, such as 0.85; the continuous area refers to the spatial position distance between the two nodes in the structural alignment graph within a set range (such as a 3×3 neighborhood); the dynamic time warping distance of the price curve is used to measure the similarity between two incompletely aligned time series in the time dimension; the preset second threshold is the set DTW distance upper limit, such as 1.5, which is used to determine the similarity of price fluctuation patterns.

[0065] S3.4: Starting from a high-trust node and combining it with connected conditional reference nodes, build multiple collaborative judgment paths.

[0066] During the path construction process, the path classification is formed according to the following rules:

[0067] If the structural coding sequence corresponding to any two factor nodes in the path changes before the change of the commodity price sequence, and the time window overlap between the two exceeds the overlap threshold, then the path is marked as a synchronous response channel;

[0068] If more than 50% of the factor nodes in the path have label semantic similarity higher than the first threshold and the price change slope and the structural change slope are consistent within the preset tolerance range, then the path is marked as a structure-driven channel;

[0069] If the degree of structural matching between structural coding nodes in the path (such as the cosine similarity of the coding vector) is lower than the second threshold, but the spatial positions of the nodes are concentrated in adjacent areas in the structure matrix (such as the significant edge area or the center area), the path is marked as a regional focus channel.

[0070] Finally, based on the collaborative path type and node level, three types of effective price comparison nodes are hierarchically output from the cross-stability factor map: high-confidence nodes in the synchronous response channel are used for accurate price comparison reference; stable reference nodes in the structure-driven channel are used for trend-type price comparison judgment; and conditional reference nodes in the regional focus channel are used as regional control support nodes.

[0071] S4: Generate a multi-path price comparison set based on the valid price comparison nodes, and output the recommendation results in stages according to the status of the target product.

[0072] S4.1: Classify and organize the three types of effective price comparison nodes of the hierarchical output, namely the synchronous response channel, the structure-driven channel, and the regional focus channel, to form the corresponding path-type price comparison sets.

[0073] Each type of collection maintains the original correspondence between the node's structural position, platform, time slice and price information, forming a data foundation that supports multi-angle price comparison analysis.

[0074] S4.2: Based on the change frequency and spatial clustering of the structural coding nodes, determine the latest evolution state of the target product in the structural matrix, identify the structural change trend, and output a multi-path price comparison set according to the structural change trend.

[0075] Specifically, the current structure matrix of the target product is compared with the structure matrix of the previous period. Structural coding nodes that have been added, deleted, or have had their labels changed are extracted, and the position index changes are recorded. The total number and distribution of structural coding nodes that have changed in the structure matrix are counted.

[0076] Among the nodes encoding the changed structure, we check whether they appear concentrated in a certain subregion or multiple adjacent regions of the structure matrix (the structure matrix can be divided into a 3x3 grid). If the structural changes show obvious clustering (for example, more than 70% of the changed nodes fall within a certain block), it is judged as a local evolution trend; if the distribution is uniform and spans multiple subregions, it is considered a global evolution trend.

[0077] Based on the results of the above two steps, the current structural change trends of the target products are divided into three categories, for example:

[0078] Overall rapid evolution type: the number of node changes is large and widely distributed, and it is recommended to use a synchronous response channel node set; trend maintenance type: the number of node changes is small and evenly distributed, and it is recommended to use a structure-driven channel node set; local update type: the number of node changes is concentrated in a local area, and it is recommended to use a regional focus channel node set.

[0079] Further, such as Figure 2 As shown, this embodiment also provides an image-driven cross-platform e-commerce product price comparison processing system, including:

[0080] The image acquisition module collects product images of the target product from multiple platforms, performs structural slicing and partitioning processing on each product image, constructs each area into a coding node, and generates a structural matrix of the product image;

[0081] The matching recognition module is used to perform correspondence matching on the structural matrices of product images on multiple platforms and identify image-product pairs with potential for comparison based on the combined relationship between regional structure and semantic features;

[0082] The feature cross-module is used to extract cross-feature groups from the historical price series and attributes of products on each platform based on image product pairs, conduct a joint evaluation of price, attribute, and structure, and hierarchically screen valid price comparison nodes based on cross-combination judgment rules;

[0083] The recommendation output module is used to generate a multi-path price comparison set based on the valid price comparison nodes and output the recommendation results in stages according to the status of the target product.

[0084] This embodiment also provides a computer device suitable for the image-driven cross-platform e-commerce product price comparison processing method, including a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to implement the image-driven cross-platform e-commerce product price comparison processing method proposed in the above embodiment.

[0085] The computer device may be a terminal, comprising a processor, a memory, a communication interface, a display screen and an input device connected via a system bus. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device comprises a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system and a computer program. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The communication interface of the computer device is used to communicate with an external terminal in a wired or wireless manner, and the wireless manner may be achieved through WIFI, an operator network, NFC (near field communication) or other technologies. The display screen of the computer device may be a liquid crystal display or an electronic ink display screen, and the input device of the computer device may be a touch layer covering the display screen, or a button, trackball or touchpad provided on the housing of the computer device, or an external keyboard, touchpad or mouse.

[0086] This embodiment further provides a storage medium having a computer program stored thereon, which, when executed by a processor, implements the image-driven cross-platform e-commerce product price comparison processing method proposed in the above embodiment.

[0087] In summary, the present invention significantly improves the accuracy of image comparison between cross-platform products through the construction of a structural matrix and a joint modeling mechanism of structural-semantic features. After node hierarchical division and cross-feature evaluation, it can realize deep-level correlation mining between structural changes, price evolution and attribute information, thereby ensuring that the price comparison reference has higher timeliness and consistency; at the same time, relying on multi-path collaborative judgment and node level stratification mechanism, the present invention can dynamically output personalized price comparison result recommendations based on the structural evolution trend of different target products. In summary, the present invention not only improves the accuracy and adaptability of price comparison processing, but also ensures the stability and intelligent response capability of price comparison logic under the conditions of complex e-commerce image data.

[0088] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the preferred embodiments, those skilled in the art should understand that the technical solutions of the present invention may be modified or replaced by equivalents without departing from the spirit and scope of the technical solutions of the present invention, which should all be included in the scope of the claims of the present invention.

Claims

1. An image-driven cross-platform e-commerce product price comparison processing method, characterized by: include: Collect product images of the target product on multiple platforms, perform structural slicing and partitioning on each product image, construct each region into a coding node, and generate a structural matrix of the product image; Perform correspondence matching on the structural matrices of product images from multiple platforms and identify image-product pairs with potential for comparison based on the combined relationship between regional structure and semantic features. Based on image-based product pairs, we extract cross-feature groups from the historical price series and attributes of products on each platform, conduct a joint evaluation of price, attribute, and structure, and hierarchically screen effective price comparison nodes using cross-combination judgment rules. A multi-path price comparison set is generated based on the valid price comparison nodes, and the recommendation results are output in stages according to the status of the target product.

2. The image-driven cross-platform e-commerce product price comparison processing method according to claim 1, characterized in that: The corresponding matching of the structure matrices of the product images on multiple platforms includes: Image source labels are added to the structure matrices corresponding to product images collected from all platforms. The structure matrices are classified according to the source labels. The structure matrices generated for the same product on different platforms are divided into different subsets to ensure that the same source images are not paired when combined. The structure matrix in each platform is paired with the structure matrices of the same product in other platforms to generate cross-platform structure matrix pairs. Each pair contains two structure matrices from different source platforms. For each pair of structure matrices, two-way structure matching groups are generated by combining them in two directions.

3. The image-driven cross-platform e-commerce product price comparison processing method according to claim 2, characterized in that: The identifying of image-product pairs with matching potential includes: For each set of structure matrices, the structure encoding node set is extracted, and based on the node position index and label semantics in the structure encoding nodes, the intersection node set and the union node set are constructed respectively; Based on the intersection node set and the union node set, the ratio of the number of nodes between the two is calculated as the original intersection-union ratio indicator; Count the coverage ratio of each node set in the two structure matrices in the entire image structure as a compensation coefficient, and multiply it with the initial intersection-union ratio to form an adjusted intersection-union ratio; If the adjusted intersection-over-union ratio is higher than the judgment threshold T s , then the structure matching group is labeled as an image-product pair.

4. The image-driven cross-platform e-commerce product price comparison processing method according to claim 1, characterized in that: The price-attribute-structure joint evaluation includes: For the confirmed image-product pairs, we extract the temporal evolution structure matrix and corresponding price data of the products on each platform, and construct the structure-price coupling vector of structural change and price fluctuation. Based on the structure matching group, a structure alignment graph is constructed and the structure-price coupling vector is bound to form a structure-price fusion feature group.

5. The image-driven cross-platform e-commerce product price comparison processing method according to claim 4, characterized in that: The cross-combination judgment rules include: Construct a cross-stability factor map, where each factor node corresponds to a joint stability index of three dimensions; the three dimensions include time slice overlap, structural encoding semantic consistency, and structural alignment density; Based on the joint judgment of the three dimensions, each factor node in the graph is assigned the following hierarchical labels: If the structural coding similarity corresponding to the node is higher than the preset first threshold, the corresponding structural coding nodes fall in a continuous area in the structural alignment map, and the dynamic time warping distance between the corresponding commodity price curves within the preset time window is less than the preset second threshold, then it is marked as a high-confidence node; if there is fluctuation in one dimension indicator but other dimensions can compensate, it is marked as a conditional reference node; otherwise, it is marked as a secondary support node; Starting from a highly trusted node and combining it with connected conditional reference nodes, multiple collaborative judgment paths are constructed.

6. The image-driven cross-platform e-commerce product price comparison processing method according to claim 5, characterized in that: The hierarchical screening of valid price comparison nodes includes: During the construction of multiple collaborative judgment paths, path classification is formed according to the following rules: If the structural coding sequence corresponding to any two factor nodes in the path changes before the change of the commodity price sequence, and the time window overlap between the two exceeds the overlap threshold, the path is marked as a synchronous response channel; If more than 50% of the factor nodes in the path have label semantic similarity above the first threshold and the price change slope and the structural change slope are consistent within the preset tolerance range, the path is marked as a structure-driven channel; If the degree of structural matching between factor nodes in the path is lower than the second threshold, but the spatial positions are clustered in adjacent regions in the structure matrix, the path is marked as a regional focus channel; Based on the collaborative path type and node level, three types of effective price comparison nodes are hierarchically output from the cross-stability factor map: the high-confidence nodes in the synchronous response channel are used for accurate price comparison reference; the stable reference nodes in the structure-driven channel are used for trend-type price comparison judgment; and the conditional reference nodes in the regional focus channel are used as regional control support nodes.

7. The image-driven cross-platform e-commerce product price comparison processing method according to claim 1, characterized in that: Outputting recommendation results in stages according to the target product status includes: The three types of valid price comparison nodes are divided into node sets according to the collaborative path types to which they belong, and three types of multi-path price comparison sets are constructed; Based on the change frequency and spatial clustering of the structural coding nodes, the latest evolution state of the target product in the structural matrix is ​​determined, the structural change trend is identified, and a multi-path price comparison set is output according to the structural change trend.

8. An image-driven cross-platform e-commerce product price comparison processing system, based on the image-driven cross-platform e-commerce product price comparison processing method according to any one of claims 1 to 7, characterized in that: Also includes: The image acquisition module collects product images of the target product from multiple platforms, performs structural slicing and partitioning processing on each product image, constructs each area into a coding node, and generates a structural matrix of the product image; The matching recognition module is used to perform correspondence matching on the structural matrices of product images on multiple platforms and identify image-product pairs with potential for comparison based on the combined relationship between regional structure and semantic features; The feature cross-module is used to extract cross-feature groups from the historical price series and attributes of products on each platform based on image product pairs, conduct a joint evaluation of price, attribute, and structure, and hierarchically screen valid price comparison nodes based on cross-combination judgment rules; The recommendation output module is used to generate a multi-path price comparison set based on the valid price comparison nodes and output the recommendation results in stages according to the status of the target product.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the image-driven cross-platform e-commerce product price comparison processing method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the image-driven cross-platform e-commerce product price comparison processing method according to any one of claims 1 to 7 are implemented.

Images