Search

CN-121998705-A - Multi-source perception-based electronic commerce advertisement putting strategy optimization method and system

CN121998705ACN 121998705 ACN121998705 ACN 121998705ACN-121998705-A

Abstract

The application provides a method and a system for optimizing an e-commerce advertisement putting strategy based on multi-source perception, which comprises the steps of firstly extracting user interest vectors representing user interest preferences and scene vectors representing current interaction scene semantics and a user instant state; generating a preference association evaluation quantity based on a user interest vector and a candidate advertisement feature vector, generating a scene adaptation evaluation quantity representing the adaptation degree of a candidate advertisement to a current interaction scene based on the scene vector and the candidate advertisement feature vector under the guidance of an advertisement release strategy template, carrying out fusion evaluation on a current advertisement release decision through the preference association evaluation quantity and the scene adaptation evaluation quantity to obtain a strategy decision score of advertisement release, and adjusting and executing an advertisement release strategy aiming at a target user in the current interaction scene according to the strategy decision score. By adopting the scheme of the application, the dynamic decision of collaborative analysis of the user interest preference and the real-time interaction scene can be based, thereby improving the robustness of the advertisement delivery system.

Inventors

  • HU XUEJIN

Assignees

  • 宁波财经学院

Dates

Publication Date
20260508
Application Date
20251224

Claims (10)

  1. 1. The method for optimizing the e-commerce advertisement putting strategy based on multi-source perception is characterized by comprising the following steps: Acquiring long-term behavior data of a target user, and generating a user interest vector representing interest preference of the target user; Acquiring multidimensional context information of a current interaction scene, and generating a scene vector representing the semantics of the current interaction scene and the instant state of a target user; According to the scene vector, matching an optimal association historical scene from a pre-constructed historical scene strategy library, and acquiring an advertisement putting strategy template corresponding to the optimal association historical scene; For candidate advertisements, generating preference association evaluation quantities for representing the association degree of the candidate advertisements with the long-term interests of the target users based on the user interest vectors and the candidate advertisement feature vectors; Under the guidance of the advertisement putting strategy template, generating scene adaptation evaluation quantity used for representing the adaptation degree of the candidate advertisement to the current interaction scene based on the scene vector and the candidate advertisement feature vector; And carrying out fusion evaluation on the advertisement putting decision under the current interaction scene through the preference association evaluation quantity and the scene adaptation evaluation quantity to obtain a strategy decision score of advertisement putting, and further adjusting and executing the advertisement putting strategy aiming at the target user under the current interaction scene according to the strategy decision score.
  2. 2. The method of claim 1, wherein obtaining long-term behavioral data of the target user, generating a user interest vector characterizing interest preferences thereof, comprises: Collecting advertisement interaction behavior logs of a target user in a preset history period, wherein the advertisement interaction behavior logs at least comprise advertisement clicks, advertisement browsing time length and commodity collection or purchase records related to advertisement content; Cleaning and normalizing the advertisement interaction behavior log, and extracting interaction frequency and depth characteristics corresponding to different advertisement categories or topics; Inputting the interaction frequency and the depth characteristic into a pre-trained user interest model, wherein the user interest model learns implicit preference of a user on advertisement categories or topics based on a behavior log; And outputting a multidimensional vector serving as the user interest vector by the user interest model.
  3. 3. The method of claim 1, wherein collecting multi-dimensional context information for a current interaction scenario, generating a scenario vector characterizing semantics of the current interaction scenario and an instant state of a target user, comprises: Acquiring environment state information and current interactive content of a user in real time; Respectively carrying out feature extraction and vectorization on the environment state information and the current interaction content to obtain an environment state vector and a user instant interaction vector; And generating a scene vector representing the current interaction scene semantic and the instant state of the target user by fusing the environment state vector and the instant interaction vector of the user.
  4. 4. The method of claim 1, wherein matching the optimal correlation history scene from the pre-constructed history scene policy library according to the scene vector, and obtaining the advertisement delivery policy template corresponding to the optimal correlation history scene specifically comprises: calculating the similarity between the current scene vector and each historical scene vector in the historical scene strategy library; Selecting a history scene with highest similarity as the optimal association history scene; and reading a strategy parameter set corresponding to the historical scene to serve as the advertisement putting strategy template.
  5. 5. The method of claim 1, wherein generating a preference relevance score measure for characterizing a degree of relevance of a candidate advertisement to a target user long-term interest based on the user interest vector and a candidate advertisement feature vector, comprises: Determining the similarity between the user interest vector and the candidate advertisement feature vector; Correcting the similarity according to the time attenuation attribute of the user interest; And taking the corrected value as a preference association evaluation quantity for representing the association degree of the candidate advertisement and the long-term interest of the target user.
  6. 6. The method of claim 5, wherein correcting the similarity based on the temporal decay attribute of the user's interest comprises assigning decay weights to interest features in different time windows in the user interest vector, and weighting a similarity calculation based on the decay weights.
  7. 7. The method of claim 1, wherein performing a fusion evaluation on the advertisement placement decision in the current interaction scene by the preference association evaluation amount and the scene adaptation evaluation amount, the obtaining a policy decision score of the advertisement placement specifically comprises: Determining initial membership degrees of the preference association evaluation quantity and the scene adaptation evaluation quantity respectively corresponding to advertisement recommendation grades; constructing a two-dimensional evaluation relation matrix for describing the mutual influence relation between the preference association evaluation quantity and the scene adaptation evaluation quantity based on the cooperative rules defined in the advertisement putting strategy template; Mapping the numerical values in the two-dimensional evaluation relation matrix to corresponding fuzzy sets through fuzzy membership functions; constructing a fusion decision fuzzy evaluation matrix according to the initial membership and the fuzzy set; And carrying out fuzzy comprehensive judgment on the advertisement putting decision under the current interaction scene based on the fusion decision fuzzy evaluation matrix, and outputting the strategy decision score of the advertisement putting after defuzzification.
  8. 8. An electronic commerce advertisement delivery strategy optimization system based on multi-source perception, which is characterized by comprising: the acquisition module is used for acquiring long-term behavior data of a target user and generating a user interest vector representing interest preference of the target user; the acquisition module is also used for acquiring multidimensional context information of the current interaction scene and generating a scene vector representing the semantics of the current interaction scene and the instant state of the target user; The processing module is used for matching the optimal association historical scene from a pre-constructed historical scene strategy library according to the scene vector and acquiring an advertisement putting strategy template corresponding to the optimal association historical scene; The processing module is further used for generating a preference association evaluation quantity for representing the association degree of the candidate advertisement and the long-term interest of the target user for the candidate advertisement based on the user interest vector and the candidate advertisement feature vector; the processing module is further used for generating scene adaptation evaluation quantity used for representing the adaptation degree of the candidate advertisement to the current interaction scene based on the scene vector and the candidate advertisement feature vector under the guidance of the advertisement putting strategy template; And the execution module is used for carrying out fusion evaluation on the advertisement putting decision under the current interaction scene through the preference association evaluation quantity and the scene adaptation evaluation quantity to obtain a strategy decision score of advertisement putting, and further adjusting and executing the advertisement putting strategy aiming at the target user under the current interaction scene according to the strategy decision score.
  9. 9. A computer device comprising a memory storing code and a processor, wherein the processor is configured to obtain the code and perform the multi-source awareness based e-commerce advertising policy optimization method of any one of claims 1 to 7.
  10. 10. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the multi-source awareness based e-commerce advertisement placement policy optimization method of any one of claims 1 to 7.

Description

Multi-source perception-based electronic commerce advertisement putting strategy optimization method and system Technical Field The application relates to the technical field of internet advertisements, in particular to an electronic commerce advertisement putting strategy optimization method and system based on multi-source perception. Background Along with the evolution of internet advertisement services from rough display to accurate delivery, how to effectively reach target users has become a core challenge, and the current mainstream method mainly relies on user history behaviors to construct interest portraits or combines real-time contexts to carry out simple matching, so that the modes improve advertisement correlation to a certain extent, but when dealing with dynamic evolution of user interests and complex and changeable interaction scenes, the limitations of stiff adaptation and large decision fluctuation often exist, and the delivery effect is difficult to stably maintain in a changing environment. The conventional advertisement putting strategy faces a remarkable bottleneck in terms of improving robustness, and the problem is that the system fails to realize deep collaborative analysis and dynamic decision between long-term interest preference of a user and a real-time interaction scene, specifically, the conventional scheme is mainly dependent on a static user history interest model, is insufficient in instant scene intention response to the user, causes sudden drop of the putting effect when temporary deviation or scene mutation occurs in the user interest, or excessively focuses on real-time context matching, ignores a stable interest baseline of the user, causes the putting decision to be greatly influenced by accidental interaction noise, and is unstable in performance, and the two split analysis modes lead to insufficient generalization and adaptation capability and poor overall decision robustness when the system faces actual conditions such as data sparsity, interest drift, scene switching and the like. Therefore, how to dynamically decide based on collaborative analysis of user interest preferences and real-time interaction scenarios, thereby improving the robustness of the advertisement delivery system becomes a difficult problem for the industry. Disclosure of Invention The application provides an electronic commerce advertisement putting strategy optimization method and system based on multi-source perception, which can be based on dynamic decisions of collaborative analysis of user interest preference and real-time interaction scene, so that the robustness of an advertisement putting system is improved. In a first aspect, the present application provides a method for optimizing an e-commerce advertisement delivery strategy based on multi-source perception, comprising the steps of: Acquiring long-term behavior data of a target user, and generating a user interest vector representing interest preference of the target user; Acquiring multidimensional context information of a current interaction scene, and generating a scene vector representing the semantics of the current interaction scene and the instant state of a target user; According to the scene vector, matching an optimal association historical scene from a pre-constructed historical scene strategy library, and acquiring an advertisement putting strategy template corresponding to the optimal association historical scene; For candidate advertisements, generating preference association evaluation quantities for representing the association degree of the candidate advertisements with the long-term interests of the target users based on the user interest vectors and the candidate advertisement feature vectors; Under the guidance of the advertisement putting strategy template, generating scene adaptation evaluation quantity used for representing the adaptation degree of the candidate advertisement to the current interaction scene based on the scene vector and the candidate advertisement feature vector; And carrying out fusion evaluation on the advertisement putting decision under the current interaction scene through the preference association evaluation quantity and the scene adaptation evaluation quantity to obtain a strategy decision score of advertisement putting, and further adjusting and executing the advertisement putting strategy aiming at the target user under the current interaction scene according to the strategy decision score. Preferably, the obtaining long-term behavior data of the target user, and the generating the user interest vector characterizing interest preference of the target user specifically includes: Collecting advertisement interaction behavior logs of a target user in a preset history period, wherein the advertisement interaction behavior logs at least comprise advertisement clicks, advertisement browsing time length and commodity collection or purchase records related to advertisement content; Cleaning