CN-122022920-A - Advertisement optimization-oriented user feedback key factor recognition analysis method and device

CN122022920ACN 122022920 ACN122022920 ACN 122022920ACN-122022920-A

Abstract

The application provides a user feedback key factor recognition analysis method and device for advertisement optimization, and relates to the field of large language models. The method comprises the steps of obtaining multi-source user comment data corresponding to a target advertisement to be optimized, conducting hierarchical semantic modeling processing on the multi-source user comment data through a pre-trained large language model, outputting a multi-dimensional semantic characterization result of the user comment, calculating user feedback comprehensive distribution based on the multi-dimensional semantic characterization result of the user comment, outputting a user feedback key influence factor set through the multi-dimensional semantic characterization result of the user comment and the user feedback comprehensive distribution, and outputting advertisement optimization decision suggestions corresponding to the target advertisement to be optimized based on the user feedback key influence factor set. The application solves the problems that the advertisement content effect after optimization in the prior art is still poor, and the specific sources which are not satisfied or approved by the users are difficult to accurately position.

Inventors

BAO ZHIHUI
CHEN CHENG
XIA KUN

Assignees

武汉卓尔数科信息科技有限公司

Dates

Publication Date: 20260512
Application Date: 20251226

Claims (10)

1. The advertisement optimization-oriented user feedback key factor identification and analysis method is characterized by comprising the following steps of: Acquiring multi-source user comment data corresponding to a target advertisement to be optimized; carrying out hierarchical semantic modeling processing on the multi-source user comment data through a pre-trained large language model, and outputting a user comment multi-dimensional semantic characterization result; calculating user feedback comprehensive distribution based on the user comment multidimensional semantic characterization result; Outputting a user feedback key influence factor set through comprehensive distribution of the user comment multidimensional semantic characterization result and the user feedback; And outputting advertisement optimization decision suggestions corresponding to the target advertisements to be optimized based on the user feedback key influence factor set.
2. The method of claim 1, wherein the obtaining multi-source user comment data corresponding to the advertisement to be optimized specifically includes: The method comprises the steps of obtaining user comment data related to the target advertisement to be optimized in different types of platforms, wherein the different types of platforms comprise an advertisement putting platform, a content distributing platform, a social interaction platform and an application service platform; And uniformly gathering the user comment data in the different types of platforms according to comment association context information, and constructing the multi-source user comment data, wherein the comment association context information comprises advertisement identification information, comment time information and platform source information.
3. The method according to claim 1, wherein the hierarchical semantic modeling process is performed on the multi-source user comment data through a pre-trained large language model, and a multi-dimensional semantic characterization result of the user comment is output, specifically including: Performing semantic coding processing on the user comment text based on the multi-source user comment data to generate a comment base semantic representation for characterizing comment context semantic relationships; On the basis of the comment basic semantic representation, carrying out semantic decoupling processing on attitude pointing information, emotion expression information and evaluation object information contained in the multi-source user comment data, so that different semantic dimensions form distinguishable but interrelated sub-semantic representations in the same semantic space; Constructing a hierarchical semantic representation based on the sub-semantic representations; And performing vectorization mapping processing on the hierarchical semantic representation to construct the user comment multidimensional semantic representation result.
4. A method according to claim 3, wherein said constructing a hierarchical semantic representation based on said sub-semantic representation comprises in particular: Carrying out semantic hierarchy division on the sub-semantic representation corresponding to the attitude pointing information, the sub-semantic representation corresponding to the emotion expression information and the sub-semantic representation corresponding to the evaluation object information; Establishing a semantic hierarchy structure based on semantic hierarchy dividing results, wherein the semantic hierarchy structure is used for carrying out joint modeling on semantic dependency relationships and co-occurrence relationships among different sub-semantic representations; the sub-semantic representations of different levels in the semantic hierarchy are combined according to a preset level mapping rule, and the hierarchical semantic representation is constructed, wherein the hierarchical semantic representation comprises high-level semantics, middle-level semantics and bottom-level semantics, the high-level semantics are used for representing feedback directions of user comments, the middle-level semantics are used for representing emotion expression features, and the bottom-level semantics are used for representing evaluation objects and attention elements.
5. The method according to claim 1, wherein calculating a user feedback integrated distribution based on the user comment multi-dimensional semantic characterization result specifically comprises: Calculating emotion polarity parameters through emotion polarity mapping processing, subjective intensity parameters through subjective intensity mapping, and attitude stability parameters through attitude consistency statistical processing based on the user comment multidimensional semantic characterization result; and carrying out joint statistical processing on the emotion tendencies of the user comments based on the emotion polarity parameter, the subjective intensity parameter and the attitude stability parameter so as to calculate the user feedback comprehensive distribution.
6. The method according to claim 5, wherein the outputting the set of key influencing factors of the user feedback through the comprehensive distribution of the multi-dimensional semantic characterization result of the user comment and the user feedback specifically comprises: Based on the user comment multidimensional semantic characterization result, carrying out semantic aggregation processing on the evaluation object information to form candidate influence factors for characterizing different evaluation objects; Analyzing the distribution difference of the emotion polarity parameter, the subjective intensity parameter and the attitude stability parameter on different candidate influence factors by combining the user feedback comprehensive distribution, and determining the influence degree of each candidate influence factor on the user feedback comprehensive distribution; And screening the candidate influence factors based on the influence degree, and taking screening results meeting preset selection blocking conditions as the user feedback key influence factor set.
7. The method of claim 6, wherein outputting advertisement optimization decision suggestions corresponding to the targeted advertisements to be optimized based on the user feedback key influence factor set specifically comprises: identifying advertisement elements corresponding to each user feedback key influence factor in the target advertisement to be optimized based on the user feedback key influence factor set; determining an optimization priority corresponding to the advertisement element based on the influence degree; Outputting the advertisement optimization decision advice according to the optimization priority, wherein the advertisement optimization decision advice comprises advertisement content adjustment advice and advertisement delivery strategy adjustment advice.
8. The advertisement optimization-oriented user feedback key factor recognition and analysis device is characterized by comprising an acquisition module and a processing module, wherein, The acquisition module is used for acquiring multi-source user comment data corresponding to the advertisement to be optimized, carrying out hierarchical semantic modeling processing on the multi-source user comment data through a pre-trained large language model, and outputting a user comment multi-dimensional semantic characterization result; The processing module is used for calculating user feedback comprehensive distribution based on the user comment multi-dimensional semantic characterization result, outputting a user feedback key influence factor set through the user comment multi-dimensional semantic characterization result and the user feedback comprehensive distribution, and outputting advertisement optimization decision suggestions corresponding to the target advertisement to be optimized based on the user feedback key influence factor set.
9. An electronic device comprising a processor, a communication bus, a user interface, a network interface, and a memory, the memory for storing instructions, the user interface and the network interface for communicating to other devices, the processor for executing instructions stored in the memory to cause the electronic device to perform the method of any of claims 1-7.
10. A non-transitory computer readable storage medium storing instructions which, when executed, perform the method of any one of claims 1 to 7.

Description

Advertisement optimization-oriented user feedback key factor recognition analysis method and device Technical Field The application relates to the field of large language models, in particular to a user feedback key factor recognition analysis method and device for advertisement optimization. Background Nowadays, the scale of internet advertisement delivery is continuously enlarged, the advertisement reaching frequency is continuously improved, and the sensitivity and tolerance of users to advertisement content are synchronously reduced, so how to improve the advertisement quality while guaranteeing the delivery effect becomes a main research direction in the field of advertisement optimization. In the prior art, in order to improve the advertising effect, user feedback information is generally obtained by collecting user comments, scoring or simply feeding back labels, and user feedback is summarized based on means such as emotion classification, keyword statistics or manual rule analysis, so that references are provided for advertisement content adjustment or delivery strategy optimization. However, when the prior art faces to user comments with scattered sources, implicit expressions and various evaluation directions, only coarse-granularity positive and negative emotion judgment or high-frequency vocabulary statistics results can be generally given, and it is difficult to distinguish the difference among user emotion intensity, attitude persistence and specific evaluation objects, especially when the user comments contain various emotion expressions or multiple evaluation attention points, feedback at different levels is easy to be mixed, so that key factors which really influence advertising effects are covered, and further the optimized advertising content effects are still poor, and it is difficult to accurately locate specific sources which are not satisfied or approved by users. Therefore, a method and a device for identifying and analyzing user feedback key factors for advertisement optimization are needed. Disclosure of Invention The application provides a user feedback key factor recognition analysis method and device for advertisement optimization, which solve the problems that the advertisement content effect is still poor after the optimization in the prior art, and the specific sources which are not satisfied or approved by users are difficult to accurately position. The application provides an advertisement optimization-oriented user feedback key factor recognition analysis method, which comprises the steps of obtaining multi-source user comment data corresponding to an advertisement to be optimized, conducting hierarchical semantic modeling processing on the multi-source user comment data through a pre-trained large language model, outputting a user comment multi-dimensional semantic characterization result, calculating user feedback comprehensive distribution based on the user comment multi-dimensional semantic characterization result, outputting a user feedback key influence factor set through the user comment multi-dimensional semantic characterization result and the user feedback comprehensive distribution, and outputting advertisement optimization decision suggestions corresponding to the advertisement to be optimized based on the user feedback key influence factor set. The method comprises the steps of obtaining user comment data related to a target advertisement to be optimized in different types of platforms, wherein the different types of platforms comprise an advertisement putting platform, a content distribution platform, a social interaction platform and an application service platform, the user comment data in the different types of platforms are collected uniformly according to comment association context information, the multi-source user comment data are built, and the comment association context information comprises advertisement identification information, comment time information and platform source information. The method comprises the steps of carrying out hierarchical semantic modeling processing on multi-source user comment data through a pre-trained large language model, outputting a multi-dimensional semantic representation result of the user comment, carrying out semantic coding processing on the user comment text based on the multi-source user comment data to generate comment basic semantic representation for representing comment context semantic relations, carrying out semantic decoupling processing on attitude pointing information, emotion expression information and evaluation object information contained in the multi-source user comment data on the basis of the comment basic semantic representation, enabling different semantic dimensions to form distinguishable but interrelated sub-semantic representations in the same semantic space, constructing the hierarchical semantic representation based on the sub-semantic representation, and carrying out vectorization