CN-121979526-A - Intelligent generation method and system for electronic commerce landing page

CN121979526ACN 121979526 ACN121979526 ACN 121979526ACN-121979526-A

Abstract

The invention relates to the field of data processing, in particular to an intelligent generation method and system of an electronic commerce landing page, wherein the method comprises the steps of acquiring a uniform resource locator of a target commodity and acquiring multi-mode data of the target commodity based on the uniform resource locator; the method comprises the steps of carrying out feature extraction on multi-mode data to obtain structural style feature data containing visual feature vectors and semantic feature vectors, constructing a design space generated by a landing page, modeling the design space into a directed graph structure, wherein nodes represent page component modules, edges represent connection relations between adjacent component modules, carrying out path search in the directed graph structure by utilizing a cluster search algorithm to obtain an optimal design scheme, and generating a target E-commerce landing page based on the optimal design scheme. The method can effectively improve the efficiency and quality of the generation of the landing page.

Inventors

FENG ZHONGCHENG
WANG YUJUE
YANG LIHAO
PENG FENG

Assignees

钛动科技股份有限公司

Dates

Publication Date: 20260505
Application Date: 20260209

Claims (10)

1. The intelligent generation method of the e-commerce landing page is characterized by comprising the steps of obtaining a uniform resource locator of a target commodity, and collecting multi-mode data of the target commodity based on the uniform resource locator, wherein the multi-mode data comprises a commodity picture list, text description and attribute parameters; Extracting features of the multi-mode data to obtain structured style feature data containing visual feature vectors and semantic feature vectors; Constructing a design space generated by a landing page, and modeling the design space into a directed graph structure, wherein nodes represent page component modules, and edges represent connection relations between adjacent component modules; And performing path search in the directed graph structure by using a cluster search algorithm to obtain an optimal design scheme, and generating a target e-commerce landing page based on the optimal design scheme, wherein the path search comprises the following steps: Initializing a candidate path set, expanding a next possible page component module for a current candidate path in each step of iteration, calculating the matching degree of visual and semantic features of the expanded path and the structured style feature data to obtain the score of the expanded path, wherein the path score is positively correlated with the matching degree, and keeping K paths with the highest score as the updated candidate path set until reaching a preset depth; And selecting a path with the highest score in the updated candidate path set, and resolving an optimal design scheme based on the path with the highest score, wherein the optimal design scheme comprises a page component module sequence which is arranged in sequence.
2. The intelligent generation method of the e-commerce landing page of claim 1, wherein the feature extraction is performed on the multi-modal data to obtain the structured style feature data comprising the visual feature vector and the semantic feature vector, specifically comprising the steps of extracting a main graph from the commodity picture list, performing cluster analysis on the main graph in a perceptually uniform color space by using an image processing algorithm, and extracting main color and auxiliary color distribution to form color features; inputting the commodity picture list into a pre-trained visual analysis model to identify texture material characteristics and scene type labels of pictures to form advanced visual characteristics; Inputting the text description and attribute parameters into a natural language processing model, extracting functional words and attribute words to obtain a candidate word set, and carrying out priority ordering on words in the candidate word set by combining historical click conversion data to construct the semantic feature vector; And combining the visual feature vector and the semantic feature vector to form the structural style feature data in a JSON format.
3. The intelligent e-commerce floor page generation method of claim 2, wherein the image processing algorithm adopts a K-means clustering algorithm.
4. The e-commerce floor page intelligent generation method of claim 1, wherein the score calculation expression of the extended path is: ; In the formula, For the estimated conversion output by the conversion prediction model trained based on historical data, Visual and semantic features representing a current path Structured style feature data with the target commodity The degree of cosine similarity between the two, A degree of difference indicator representing the current path and the historical success case base, Represents a readability evaluation index based on the visual saliency map, 、、 And Respectively corresponding weight coefficients.
5. The method for intelligently generating an e-commerce floor page of claim 1, wherein the edge weights in the directed graph structure are statistically derived based on historical transformation data, wherein the path searching is performed in the directed graph structure by using a cluster searching algorithm, and further comprising the steps of calculating the longest common subsequence similarity of a current candidate path and an ideal narrative sequence as a narrative continuity score; Defining an information density value of each page component module, calculating a density change rate between adjacent page component modules, and increasing a path score when the density change rate is in a preset rewarding interval; The narrative continuity score and a density change rate based score are factored into the path score.
6. The method for intelligently generating the e-commerce floor page of claim 1, wherein the generating the target e-commerce floor page based on the optimal design scheme comprises generating a plurality of candidate code implementations for each page component module in the optimal design scheme by using a large language model, wherein the candidate code implementations comprise hypertext markup language codes and cascading style sheet codes; Performing static analysis on the plurality of candidate code implementations, calculating a circle complexity index and a readability index, and checking code normalization; selecting the code segment with the highest comprehensive index from the plurality of candidate code realizations to be combined; Calculating the comprehensive index of the candidate codes according to the normalized value of the circle complexity index, the normalized value of the readability index and the normalized value of the code normalization, wherein the comprehensive index is positively correlated with the normalized value of the readability index and the normalized value of the code normalization and is negatively correlated with the normalized value of the circle complexity index; selecting the code segment with the highest comprehensive index from the plurality of candidate code realizations to be combined; analyzing and rewriting the combined codes by utilizing an abstract syntax tree tool, naming unified variables, merging repeated patterns, generating the corresponding hypertext markup language codes and cascading style sheet codes, and assembling into a landing page code package. .
7. The method for intelligently generating an e-commerce floor page of claim 6, further comprising performing performance optimization processing on the floor page code package, including performing document structure analysis on the hypertext markup language code, extracting key cascading style sheet codes required by first screen rendering and connecting the key cascading style sheet codes to a document header, and marking script codes which are not first screen as delay loading; Traversing picture resources in the code packet, converting pictures in a first screen area into WebP format and adding a preloading mark, converting pictures in a non-first screen area into AVIF format and adding lazy loading attributes, and generating a responsive picture set adapting to different screen resolutions; Performing dependency analysis on script codes in the code package, removing unreferenced codes by using a rocking tree optimization algorithm, packaging the script into code blocks loaded according to the need, and compressing and confusing reserved codes; And calculating the content hash value of the static resource file in the code packet and renaming the file according to the content hash value.
8. The method for intelligently generating the e-commerce landing page of claim 1, further comprising rendering the landing page code package with a browser environment to obtain a visual page after generating the target e-commerce landing page based on the optimal design; Analyzing a document object model structure and a visual layout of the visual page by using an artificial intelligent model, and planning a test action sequence for simulating user behaviors; Executing the test action sequence on the visual page, wherein the test action sequence comprises clicking, scrolling and inputting instructions, and collecting interaction response data and operation error logs in the executing process; Intercepting a rendering screenshot of the visual page, and extracting structural visual specification data defined in the optimal design scheme, wherein the structural visual specification data at least comprises color codes, module layout coordinates and visual level definitions; And inputting the rendering screenshot and the structured visual specification data into a pre-trained visual analysis model for consistency comparison, and if the consistency score obtained by the comparison is lower than a preset threshold value, generating a test report containing the problem area labels.
9. The method for intelligently generating the e-commerce landing page according to any one of claims 1 to 8, further comprising a step of cross-device adaptive splicing, wherein the step of identifying the device type of the target release terminal comprises a personal computer end, a mobile end and a tablet end; If the equipment type is a personal computer end, adopting a framework constraint of transverse comparison and multi-column layout in the cluster searching algorithm; if the equipment type is a mobile terminal, adopting a longitudinal story flow and a single-column layout framework constraint in the cluster searching algorithm, and splitting or merging partial page component modules; Ensuring that the results generated for different device types remain consistent in color, font and brand element characteristics.
10. An intelligent electronic commerce landing page generation system comprises a processor and a memory, wherein the memory stores computer program instructions, and the intelligent electronic commerce landing page generation system is characterized in that the intelligent electronic commerce landing page generation method of any one of claims 1-9 is realized when the computer program instructions are executed by the processor.

Description

Intelligent generation method and system for electronic commerce landing page Technical Field The present invention relates to the field of data processing. More particularly, the invention relates to an intelligent generation method and system for an e-commerce landing page. Background With the vigorous development of mobile internet and electronic commerce, the electronic commerce landing page has become a core carrier for connecting advertisement traffic and commodity transformation. In a large-scale e-commerce marketing campaign, how to quickly generate high-quality landing pages is a problem to be solved in the face of massive commodities. The existing e-commerce landing page generation method mainly comprises two types of manual design development and automatic generation based on templates. The manual mode is high in flexibility, but low in production efficiency, and cannot meet the high-concurrency marketing requirement. Therefore, an automated template-based generation system is becoming mainstream. The system usually predefines a plurality of fixed page templates, and fills commodity information into the templates through a slot filling technology. However, this prior art still has the following technical limitations in practical applications: First, the data preparation phase relies on a large amount of manual intervention, and the process is cumbersome and extremely prone to error. Although template generation itself is automated, operators are often required to manually prepare the material prior to generation. The operator needs to download pictures from the background of the electronic commerce or the detail pages of the commodity one by one, copy text descriptions and input attribute parameters, and then upload the attribute parameters to the corresponding slots of the generation system. The preparation process of the data relying on manual input and carrying is extremely tedious and time-consuming, and severely restricts the mass production efficiency of the landing page. Meanwhile, the manual operation has unavoidable error, and in high-intensity repeated labor, the problems of picture pasting errors, price parameter input deviation or key attribute omission and the like are very easy to occur, so that the display information of the landing page is inconsistent with the actual commodity data, and user complaints or transaction disputes are caused. Secondly, the page topological structure is stiff, and the multi-mode characteristic difference of the commodity cannot be adapted. In the prior art, the component arrangement order of templates (i.e., the topology of pages) is usually preset fixed. However, different classes of goods have very different multimodal data characteristics (e.g., picture keytone, texture complexity, and emotional tendency of text). The template of the fixed structure cannot dynamically adjust the connection relation of the components according to the input commodity characteristics. For example, hard-fitting a digital product with extremely high information density into a cosmetic template mainly shown in a large image can lead to incomplete information display or disordered layout, and cause mismatch of visual styles and semantic logics. Again, there is a lack of global optimization mechanisms based on feature matching. Existing generation methods typically employ random selection or simple rule-based local stitching that breaks the overall relevance between page components. The inability of the system to quantitatively evaluate which combination scheme best matches the characteristics of the current commodity in a vast design combination space often results in only "available" but not "optimal" schemes, resulting in the final page performing mediocre in terms of appeal and conversion guidance. In summary, the method for generating the e-commerce landing page in the prior art has the technical problems that the generation efficiency of the landing page is low, the mismatch of the visual style and the semantic logic is caused, and the global optimum of the generated page cannot be realized in the visual style and the semantic logic. Disclosure of Invention In order to solve the technical problems that in the method for generating the e-commerce landing page in the prior art, the generation efficiency of the landing page is low, the mismatch of the visual style and the semantic logic exists, and the global optimum of the generated page cannot be realized in the visual style and the semantic logic, the invention provides a scheme in the following aspects. In a first aspect, the present invention provides an intelligent generation method for an e-commerce landing page, including: acquiring a uniform resource locator of a target commodity, and acquiring multi-mode data of the target commodity based on the uniform resource locator, wherein the multi-mode data comprises a commodity picture list, text description and attribute parameters; Extracting features of the multi-mo