KR-102965046-B1 - Advertising Effectiveness Prediction Method and System Using Artificial Intelligence Based on Consumer Evaluation Data

KR102965046B1KR 102965046 B1KR102965046 B1KR 102965046B1KR-102965046-B1

Abstract

The present invention relates to a method and system for predicting advertising effectiveness using artificial intelligence based on consumer evaluation data, and, It includes means for collecting consumer evaluation data and generating AI personas, means for generating multimodal storylines for advertising videos, and means for evaluating advertisements and providing results based on AI personas. According to the present invention, the time and cost required for recruiting and managing survey panels can be significantly reduced, and iterative pre-tests and A/B tests can be performed without constraints on sample size. Furthermore, by providing qualitative comments, item-specific scores, comprehensive scores, and reasoning based on thought processes, ad creators and advertisers can intuitively understand directions for improvement and make decisions.

Inventors

김용필
이보석
최진식
김영재
김지원

Assignees

(주)애드크림

Dates

Publication Date: 20260513
Application Date: 20260115

Claims (11)

A consumer evaluation database including ad evaluation comments, rating scores, and user demographic information; A persona generation unit that extracts sample user data from the above consumer evaluation database and generates an artificial intelligence persona profile by learning the sentiment, tone, and writing style characteristics of the user through natural language processing; An advertising effect prediction system using artificial intelligence based on consumer evaluation data, comprising: a storyline generation unit that acquires multimodal data including time-series scene images, audio text, and OCR text from an advertising video subject to evaluation, and generates storyline text explaining the narrative structure of the advertisement by analyzing the data; and an evaluation execution unit that analyzes an evaluation context including the storyline text and advertising metadata using an artificial intelligence language model injected with the persona profile, and generates qualitative evidence text including the chain of thought leading to the evaluation, along with quantitative scores for idea originality, production completeness, idea empathy, and impressions of the work, and stores the result in a database.
In claim 1, An advertising effect prediction system characterized by the evaluation execution unit, which calculates a comprehensive user score (user_score) by combining the quantitative scores, and performs post-processing logic that applies a **reliability coefficient (r)** calculated based on the variance of the past evaluation history of the AI persona as a weight, or identifies and corrects scores that deviate from the past evaluation pattern distribution of the persona as outliers.
In claim 1, An advertising effect prediction system characterized by the evaluation performing unit, when configuring the evaluation context, extracting past evaluation history data of the corresponding persona and past evaluation comment data of the original user in a preset standard number and including them in the evaluation prompt, thereby inducing the artificial intelligence language model to mimic the original user's evaluation tendencies and speech patterns.
In claim 3, An advertising effect prediction system characterized by the above-mentioned evaluation execution unit calculating the average value of past scores and score distribution data by score interval when the extracted past evaluation history data is greater than or equal to a minimum standard having statistical significance, and adding this to the above-mentioned evaluation prompt as evaluation pattern feedback information.
In claim 1, An advertising effect prediction system characterized by natural language processing of the persona generation unit described above including: extraction of major keywords using a morphological analyzer; determination of positive and negative sentiment using a sentiment analysis model; determination of formal and informal tone through ending analysis; and style classification based on the average length of evaluation comments.
In claim 1, An advertising effect prediction system characterized by the above-mentioned storyline generation unit storing the generated storyline text in a cache storage using a **unique key combining an ad identification code and a data type**, and, upon a subsequent request for the same ad, first querying and returning the storyline from the cache storage.
In claim 1, An advertising effect prediction system characterized by the above-mentioned evaluation unit providing an indicator of the degree of polarization of consumer reactions to an advertisement by calculating not only a simple average value but also a frequency distribution histogram by score interval for evaluation results derived from multiple artificial intelligence personas.
In claim 1, An advertising effect prediction system characterized by the above-mentioned evaluation unit selecting the most suitable model among a plurality of artificial intelligence language models according to the characteristics of the category or storyline of the advertisement to be evaluated, or consistently assigning a model using a hash function based on a persona identifier.
(Independent term - method) A method for predicting advertising effects using artificial intelligence based on consumer evaluation data, comprising: (a) collecting consumer-specific evaluation comments, evaluation scores, and demographic information from a consumer evaluation database; (b) generating an artificial intelligence persona profile by analyzing the sentiment, keywords, tone, and style of the evaluation comments through natural language processing; (c) generating a storyline text by inputting multimodal data including scene images, voice-extracted text, and screen recognition text of an advertisement to be evaluated into an artificial intelligence model; (d) having an artificial intelligence language model that receives the persona profile via a system prompt evaluate idea originality, production completeness, idea empathy, and work impressions based on the storyline text, and generating text of the logical thinking process (Thinking) for deriving the score along with the quantitative score; and (e) calculating and storing an overall user score and statistical indicators based on the evaluation results.
In claim 9, The above step (e) is characterized by including a step of calculating a weighted average by reflecting a reliability coefficient derived from the past evaluation history of the AI persona, and determining an overall user score by applying a correction logic to outlier scores that fall outside a preset statistical threshold range.
In claim 9, The above step (c) further comprises the step of caching the generated storyline text in an in-memory database, and using a key that combines an advertising identifier and a data type to prevent duplicate operations, characterized in that it is a method for predicting advertising effectiveness.

Description

Advertising Effectiveness Prediction Method and System Using Artificial Intelligence Based on Consumer Evaluation Data The present invention relates to the field of artificial intelligence and computer technology, and to a method and system for predicting advertising effects in advance through artificial intelligence generated using actual consumer advertising evaluation data. Traditionally, traditional consumer research methods such as panel surveys and focus group interviews (FGI) were primarily used to verify the effectiveness of TV and online video advertisements, which resulted in time and cost burdens as well as limitations on sample size. Accordingly, technologies have been proposed to predict advertising effectiveness by analyzing the visual and auditory characteristics of advertising videos and past performance data, or to automate advertising evaluation using artificial intelligence by collecting survey response data from global panels. In addition, technologies have been developed to classify persona types by analyzing review data for specific services or to predict the potential for a product's success using review data. However, the aforementioned technologies are primarily structured to (i) collect and analyze survey panel and global panel data, or (ii) remain at the level of statistically analyzing review data or classifying persona types; they have not sufficiently implemented a “virtual consumer panel simulation” structure that generates an AI persona reflecting consumer tone, sentiment, and writing style based on actual ad evaluation comments and score data, and allows this persona to evaluate the storyline of a new advertisement. FIG. 1 is an overall configuration diagram of an advertising effect prediction system using artificial intelligence based on consumer evaluation data according to an embodiment of the present invention. FIG. 2 is a natural language processing analysis flowchart of a persona generation unit according to an embodiment of the present invention. FIG. 3 is a multimodal analysis flowchart of a storyline generation unit according to an embodiment of the present invention. FIG. 4 is an advertising evaluation processing flowchart of an evaluation execution unit according to an embodiment of the present invention. 1. The Persona Module collects user-specific ad evaluation comments, scores, and demographic information from the consumer evaluation database. In particular, it identifies each user's evaluation tendencies (e.g., praise-oriented or criticism-oriented) by querying their historical average score (avg_score) and evaluation history through the PersonaRepository. Through sentiment analysis and morphological analysis, it extracts the user's tone (ending processing), frequently used keywords, and writing style (concise/detailed), and structures this data into an LLM System Prompt to generate individual AI persona profiles. The generated profiles are stored in the Persona Repository and retrieved during evaluations. 2. The Storyline Module integrates and analyzes visual and auditory information from advertising videos. (1) Video Analysis: Scene images are extracted at frame intervals set per second. Video frames and still images are processed separately based on filename prefixes (e.g., 'A', 'B'), and thumbnails and original paths are managed in a dual manner for analysis efficiency. (2) Text and Audio Analysis: Subtitle text within the video and audio scripts obtained through speech recognition (STT) are collected separately. Subtitles are used to analyze emphasized messages, while audio is used to analyze emotional tone and manner. (3) Structuring and Caching: Multimodal AI integrates the collected information to generate storyline text tagged according to the advertising planning stages (intro, problem statement, solution presentation, benefits, call to action). The generated data is stored in an in-memory cache storage such as Redis using the advertising identifier (Mcode) as the key, thereby reducing the cost of duplicate analysis for the same advertisement. 2-1. Structuring Ad Storylines Specialized The storyline module described above can tag each scene as a step-by-step component from an advertising planning perspective when generating the storyline text. For example, each scene can be classified into one or more categories such as intro (brand/situation introduction), problem statement (consumer inconvenience or need), solution presentation (product or service proposal), benefit description (key features), and call to action (inducing purchase, sign-up, or visit), and metadata can be assigned by extracting information such as whether the brand logo is exposed, whether a product is close-up, whether key message phrases are included, and the emotional state of the characters (e.g., joy, surprise, trust). These tagging results are reflected within the storyline text along with scene numbers, stage divisions, and sentiment tags, and are passed to the evaluation mod